Biological Data Analysis Using R

Rodney J. Dyer, PhD

Department of Biology
Center for the Study of Biological Complexity
Virginia Commonwealth University
2
Biological Data Analysis Using R
Contents
Preface xi
I Basic Usability 1
1 Getting R 3
1.1 What Is R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Where Do I Get It? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2 Language & Grammar 5
2.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Function Quickie . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.4 Data Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.5 Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.6 Useful Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
II Biologically Motivated Topics 23
3 Data Frames 25
3.1 Data Input/Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2 Slicing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.3 Complex Selections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.4 Useful Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4 Summary Statistics 43
4.1 Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.2 Random Number Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.3 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.4 Relationships Between Pairs of Variables . . . . . . . . . . . . . . . . . . . . . 63
4.5 Useful Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5 Contingency Tables 71
5.1 One Random Sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
i
ii CONTENTS
5.2 Paired Observations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.3 Several Random Samples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.4 The Formula Notation & Box Plots . . . . . . . . . . . . . . . . . . . . . . . . 83
5.5 Useful Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
6 Linear Models 89
6.1 The t-test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
6.2 Regression With A Single Variable . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.3 Multiple Regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
6.4 Analysis of Variance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
6.5 Useful Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
7 Working With Images 109
7.1 Image Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
7.2 Loading The Image Into R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
7.3 Components of A Pixmap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
7.4 Image Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
7.5 Creating Images Programatically . . . . . . . . . . . . . . . . . . . . . . . . . . 117
7.6 Useful Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
7.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
8 Matrix Analysis 121
8.1 Matrices In R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
8.2 Stage-Classied Matrix Models . . . . . . . . . . . . . . . . . . . . . . . . . . 132
8.3 Useful Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
8.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
9 Working With Strings 147
9.1 Parsing Text Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
9.2 Producing Formatted Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
9.3 Plotting Special Characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
9.4 Useful Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
9.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
III Extending R 165
10Basic Scripts 167
10.1Writing Scripts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
10.2Evaluating Scripts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
10.3Adding Comments To Your Code . . . . . . . . . . . . . . . . . . . . . . . . . . 171
10.4Useful Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
10.5Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
11Programming 175
11.1Looping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
11.2Conditional Statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
Biological Data Analysis Using R
CONTENTS iii
11.3Outlining A Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
11.4Creating A Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
11.5Synopsis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
11.6Useful Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
11.7Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
12Functions 189
12.1Function Syntax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
12.2Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
12.3Useful Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
12.4Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
A Answers to Exercises 197
B Installing Additional Libraries 199
B.1 Library Availability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
B.2 Installing Libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
Bibliography 205
Index 205
Biological Data Analysis Using R
iv CONTENTS
Biological Data Analysis Using R
List of Tables
2.1 Common constants you will run across in R . . . . . . . . . . . . . . . . . . 11
4.1 Some useful additional commands to customize the appearance of a gure.
For a complete listing of possible values that can be customized, try the ?par
command. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.2 Graphics devices for output of gures . . . . . . . . . . . . . . . . . . . . . . 51
5.1 Diversity of enrolled undergraduate students at Virginia Commonwealth
University in the College of Humanities & Sciences between the academic
years 1998-2008 as reported by the Center for Institutional Effectiveness
(http://www.vcu.edu/cie/analysis/reports/sets.html). . . . . . . . . . . . . 75
8.1 Table of life history values separated into A Fertility estimates (the f
X
items)
and B transition probabilities depicting the movement between stages and
within stages. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
9.1 Caption For Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
v
vi LIST OF TABLES
Biological Data Analysis Using R
List of Figures
1 Example scatter plot. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvi
4.1 Values for the density function for the
2
distribution with 1, 2, and 3 de-
grees of freedom. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.2 A graphical depiction of the critical value of the
2
distribution for = 0.05
and df = 3. The shaded region constitutes a proportion of the area under
the curve equal to . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.3 Some example graphs with alternate values for symbols, line types, widths,
colors, and titles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.4 Plot of two data sets using the par(new=T command but not taking into con-
sideration the axis limits of the two data sets before plotting. . . . . . . . . . 50
4.5 Plot of two variables on the same axis after correcting for the range of each
data set. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.6 Image of colored Poisson distribution that was copied from the graphics
device to a jpeg le. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.7 Examples of the densities of two normal distributions; the red one is drawn
from a random normal distribution with default values of = 0 and = 1
and another in blue that has = = 5. . . . . . . . . . . . . . . . . . . . . . . 55
4.8 Histogram with labels and main title changed. . . . . . . . . . . . . . . . . . 56
4.9 Histogram of 1000 random numbers drawn from a Poisson distribution
with the parameter set to 5. The red line indicates the density of the values. 57
4.10Example locations for rst two moments of a Normal (N(0, 1)) distribution. . 59
4.11Negative (left) and positive (right) distributions. In both of these examples
the dotted line connects the mode of the distribution (the top peak) to the
mean (on the x axis). The direction of this lean determines if the distribution
has a negative (left) or positive (right) skew. . . . . . . . . . . . . . . . . . . . 60
4.12Three distributions )exponential, normal, and logistic) showing different
levels of kurtosis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.13Matrix of four plots created from random numbers sampled from the nor-
mal, poisson, exponential, and the logistic distributions. . . . . . . . . . . . 62
4.14Distribution of random number drawn from rpois(1000,5). . . . . . . . . . . . . 64
4.15Scatter plot of some semi-random points. . . . . . . . . . . . . . . . . . . . . 65
4.16Example plot of two variables used to test correlations. . . . . . . . . . . . . 66
5.1 Undergraduate diversity at Virginia Commonwealth University during aca-
demic years 1998, 2003, & 2008. . . . . . . . . . . . . . . . . . . . . . . . . . 77
vii
viii LIST OF FIGURES
5.2 Boxplot of Pinus echinata germination data partitioned by timber extraction
treatment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
6.1 Plot of single variable regression values. . . . . . . . . . . . . . . . . . . . . . 92
6.2 Regression model added to plot of points using abline function. . . . . . . . . 94
6.3 Regression model with tted line and formula. . . . . . . . . . . . . . . . . . 96
6.4 A 2x2 matrix plot of some diagnostic tools associated with a linear model.
They include a plot of the residuals (e
ij
) as a function of the tted values ( y
i
)
to see if there are systematic biases in the model (upper left), a Q-Q plot to
examine normality of the residuals (upper right), a scale location plot (lower
left), and a leverage plot to look for outliers (lower right). . . . . . . . . . . . 97
6.5 Boxplot of germination percentages for Pinus echinata as a function of treat-
ment. A colored rug was added to the right side to show the actual values
within treatments (see rug. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.6 Condence intervals for difference in mean germination rates for Pinus echi-
nata families. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
7.1 The image represented in the r.pbm le. This image has been scaled up
to make it large enough to see it on the page using the program GIMP
(www.gimp.org). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.2 A PBM le that was programatically created in R . The image is rotated
because of the default location of the origin. . . . . . . . . . . . . . . . . . . . 112
7.3 The image represented by the dog.pgm le. This image has been scaled up
to make it large enough to see it on the page using the program GIMP
(www.gimp.org). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
7.4 The image represented in the Libbie.ppm le. This image has been scaled
up to make it large enough to see it on the page using the program GIMP
(www.gimp.org). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
7.5 The original image along with ones where only the red, green, and blue
channel turned on. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
7.6 The greyscale translation of the PPN image, a histogram of the grey values
and the image resulting from reducing all the grey values in the image by
half. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
7.7 A random image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
7.8 A random image with a square doughnut hole in the middle. . . . . . . . . . 118
8.1 Image depicting two vectors v
red
= [4, 2] and v
blue
= [2, 1] that are projecting
in the same direction but have different magnitudes. . . . . . . . . . . . . . 131
8.2 The A graphical depiction of the life history stages in the ctitious plant
Grenus growii . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
8.3 Effects of the instantaneous growth rate as a function of time for both
exponential growth (
blue
= 1.2) and exponential decay (
red
= 0.8). . . . . . . 136
8.4 Examples of two different calls to the plotting function barplot(). The param-
eters used to create these plots is given in the R code. . . . . . . . . . . . . . 138
8.5 Example of a stacked bar plot with multiple categories represented in each
Treatment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
8.6 Size of the four stage classes through time. . . . . . . . . . . . . . . . . . . . 142
Biological Data Analysis Using R
LIST OF FIGURES ix
8.7 Differences in estimated proportions of individuals in each stage from what
was expected through time. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
9.1 Histogramof distance estimates among all sequences using the K90 model
of substitutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
9.2 Neighbor joining tree based upon the trnL-trnF intergenic spacer sequences
and the K90 model of sequence evolution. . . . . . . . . . . . . . . . . . . . 156
9.3 The html printout of a xtable as interpreted in Firefox. You can also import
tables saved as html into popular word processors and use them as normal
table items in the creation of your documents. . . . . . . . . . . . . . . . . . 159
9.4 Example of using the expression function to annotate a graphic. . . . . . . . . 161
11.1Hemispherical photograph of winter roosting habitat at Monarch Biosphere
Reserve, Mexico. Photo by S.B. Weiss made available by the Creative Com-
mons Atribution 2.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
11.2The blue channel of the canopy picture displayed as a greyscale image. . . . 181
11.3A histogram of values in the blue channel (Figure 11.2). . . . . . . . . . . . . 181
11.4Intensity of blue channel values in the image as taken through a slice of
the image (at pixel row 230 as indicated by red dashed line). . . . . . . . . . 182
B.1 Example of CRAN mirror window as viewed on Linux . . . . . . . . . . . . . 201
B.2 All packages that can be installed from the selected mirror server on my
machine. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
Biological Data Analysis Using R
x LIST OF FIGURES
Biological Data Analysis Using R
Preface
This manuscript was written to scratch a particular itch that I felt was not being sati-
ated.Increasingly students in biological research programs, both at the undergraduate
and the graduate level, are dealing with data sets that are both enormous in size and
varied in representation. Image data, sequence data, counts of species in communities,
nutrient ux, reaction networks, and a whole host of other kinds of data are encountered
on a daily basis in biological sciences. In order to drink from this rehose of data, it is
important that we have the correct kinds of tools; the spreadsheet metaphor is no longer
valid.
After spending a few years encouraging students to learn a tool, any tool, that would
help them deal with the complexity of data we encounter, I decided to put together a
course focusing on how R can be used to deal with many different kinds of data. This
course was designed for incoming graduate students in Biology at Virginia Common-
wealth University with the goal of getting them familiar with R from the beginning of
their graduate work. Many of the graduate faculty in Biology use R in their courses
and nd that a non-trivial amount of time needs to be spent on introducing students
to R in each course which is taking away from the focus of the course. However, if a
student had taken a short course in R when they began their graduate work, then it
would be possible to spend more time in our individual courses focusing on the topic at
hand.
This manuscript is not designed to be one of the Biological discipline X in R kind of
offerings; there are already a lot of those kinds of books available. My goal here is to
introduce the reader to a wide variety of data types that we deal with in Biology and
give a brief introduction to how R can be used to interact with, and perhaps perform
analyses, on these data. The treatment of any one kind of data is relatively shallow,
as I am assuming that students are going to take a specic course on that topic in the
future. And when they do, they will have already seen how R will make their life easier.
In my own research, I use tools such as R in many different circumstances and feel
that students can only benet from a broad understanding of how R can assist in their
research. With this focus, it is no coincidence that the kinds of data introduced in this
text are pulled directly from the graduate courses that our students will take, such as
Community Ecology, Population Genetics, Population Ecology, Evolution & Speciation,
Biological Complexity, Molecular Genetics, Landscape Genetics, Bioinformatic Technolo-
gies, Ecological Genetics, and Quantitative Ecology.
Give the range of topics covered herein, I think this manuscript has a broad audience
as I assume that the reader of this text will not have much previous experience using R.
xi
xii Preface
Obviously, incoming graduate students are my primary audience. However, I also feel
that this would also be a good beginning text for one who is already working in the eld
and would like to gain a broader introduction to how R can be used in their particular
discipline.
Contents
This manuscript has been partitioned into four separate sections. The rst section intro-
duces R as a language and a tool and covers some basic topics that are required to get
one going. The next section contains eleven chapters that target some particular aspect
of biological inquiry from the perspective of the kind of data that will be analyzed. The
third section focuses on how you can extend the R environment developing scripts and
dening your own functions and libraries. The nal section of this text is an appendix
that includes the answers to odd-numbered questions from the exercises in each chap-
ter as well as some additional information on installing additional libraries or groups of
libraries.
There are some common elements to each chapter that make it easy for the reader to
get the larger picture of the topics being introduced. At the beginning of each chapter, a
specic list of topics and skills that are to be covered provided. As topics are introduced,
the R code is provided and keywords from the R programming language are highlighted
to help the reader follow along.
At the end of each chapter all the R functions that were used in the chapter as well as a
brief denition of the arguments passed to each function is provided as a quick reference
source. Each chapter also contains a set of exercises that can test the readers under-
standing of chapter topics. Answers to odd numbered exercise problems are provided in
Appendix A. Throughout the text, all of the R functions used are also indexed so that
the reader can easily nd instances where they were used.
Part 1: Basic Usability
The rst part of this manuscript contains the basic information that is required to install
and begin using R for data analysis. This section has the following chapters:
Chapter 1: Getting R This chapter provides information on how to download the latest
binary release for R as well as compiling it from source code. Particular attention is paid
to the differences associated with installing R on different platforms.
Chapter 2: Language & Grammar This chapter begins introducing the R programming
language by focusing on the different kinds of data types that are used (e.g., integers,
decimal values, factors, . Topic covered include a basic overview of what a function is,
an introduction to the most commonly used data types in R , and general operations on
these data types.
Biological Data Analysis Using R
Preface xiii
Part 2: Biologically Motivated Topics
The second section of this manuscript contains the main content.
Chapter 3: Data Frames The data frame is a fundamental object in R . This chapter
builds upon the basic understanding of data frames (introduced in Chapter 2) by in-
troducing several methods for putting your data into new and existing data frames,
persistent storage of data frames. This chapter also introduces the concept of using the
data frame data type as a light-weight database object. This includes an introduction to
making slices of a data set, the methods required to make complex selections of subsets
of data, and joining data from multiple data frames.
Chapter 4: Summary Statistics This chapter introduces the reader to general summary
statistics for continuous data, statistical distributions, and random number generation.
This chapter also provides the reader a rst introduction to creating publication-quality
graphics in R . General graphics include scatter and line plots, histograms, density
plots, plotting several graphical objects on the same set of axes, creating matrices of
plots, and saving graphics to le.
Chapter 5: Categorical Data This chapter focuses on the analysis of categorical data
and contingency tables. Give the ubiquity of the
2
test in Biology, a general treatment
of contingency tables is provided with examples demonstrating how to examine genetic
linkage disequilibrium, Hardy-Weinberg equilibrium, and demographic analysis testing
for equality of population diversity. Both parametric and non-parametric approaches are
introduced with examples.
Chapter 6: Linear Models This chapter introduces the concept of linear models from sim-
ple correlations through single and multiple regression and ANOVA (which is introduced
as regression with categorical predictors). Data for this chapter is derived from my own
thesis working with the consequences of landscape modication on reproductive success
in canopy trees. Examples of model diagnostics, model selection and post-hoc tests are
also covered.
Chapter 7: Working With String Data This chapter uses genetic sequence data as an
example of string-related data that can be manipulated in R . Basic skills in string
searching and replacements are augmented with a short discussion of genetic sequence
alignments, the use of online genetic databases such as NCBI, and the creation of a
phylogenetic trees using different algorithms is demonstrated.
Chapter 8: Image Data This chapter focuses on image creation, importation, analysis,
and manipulation. After a basic overview of image formats and manipulations, hemi-
spheric canopy photos are used as an analysis topic on which several analyses are
preformed.
Chapter 9: Matrix Analysis Matrix analysis is a general tool used in a variety of biological
disciplines. In this chapter, the topic of life history analysis and population projection is
used as an example for matrix operations in R .
Chapter 10: Multivariate Data Ordination techniques are a broad class of methodologies
that seek to understand the structure of multivariate data. In this chapter, vegetation
data is used as an example of how one conducts and interprets basic ordination.
Biological Data Analysis Using R
xiv Preface
Chapter 11: Classication This chapter focuses on how morphological shape analysis
can be used for classication purposes. Morphological data from the bark beetle species
complex, Araptus attenuatus is used as an example for comparison with genetic classi-
cation schemes.
Chapter 12: Spatial Data In this chapter, the analysis of spatial data is introduced.
Topics covered include, conversion of GPS way points and GIS data les into R data
formats, plotting georeferenced raster and vector maps, and basic spatial analysis.
Chapter 13: Genetic Data This chapter focuses on how one can represent genetic data
in R and perform basic analyses on genetic structure. Examples include the analysis
of inbreeding, population structure, association mapping, and population assignment
tests.
Part 3: Extending R
The chapters in this section only require a basic understanding of R and can be used at
any time as they are stand-alone. In fact, it is suggested that after you get familiar with
R , you should look into these chapters because they contain valuable information that
will make your life easier.
Chapter 14: Creating Basic Scripts This chapter addresses how to you create basic R
scripts so that you can reuse your code and analyses as well as have persistence across
your R sessions.
Chapter 15: Programming R This chapter covers basic programming, ow control, and
decision control statements.
Chapter 16: Functions This chapter demonstrates how the user can create individual
functions from their scripts so that calling complex analyses and operations can be
simplied.
Appendices
The last part of this manuscript includes supplementary material in support of the con-
tents.
Appendix A: Answers to Exercises This appendix provides answers to the odd numbered
problems located at the end of each chapter.
Appendix B: Installing Additional Libraries There are a broad range of libraries that the
R community provides and this appendix shows you how to nd and install additional
libraries to your local copy.
Typographic Conventions
The developers of R have worked very hard to make sure that you can interface with R
on any platform without worrying about which operating system you are using. However,
Biological Data Analysis Using R
Preface xv
there are some times when things are slightly different on alternate platforms. When
there are platform specic issues to be dealt with, I will make a notation in the margins
with the name of the operating system next to the text to indicated specic issues. OS
The book is not going to show you how to interact with R using a GUI, because in my
opinion GUIs are for babies. If you want to learn how to use R , you will have to learn
how to interact with it from the command line and write scripts for R to analyze your
data. If you want to a point-and-click interface for a statistical analysis program then
perhaps you should check out SPSS (Statistical Package for Social Sciences) or similar
offerings. It is my belief that you will learn more about programming and data analysis
if you learn the R language. There are only so many options that GUI-based analyses
can provide but with R on the command-line, you will be have the most exibility in the
analysis of your data. Moreover, when you create scripts to perform your analysis, you
will have a persistent record of how you analyzed the data instead of just some data and
results. Increasingly, peer-reviewed journals are suggesting that your analysis scripts
be included in your supplementary materials for general consumption.
Throughout this book, I will provide examples of code in a box format. You will be able to
tell what is code that can be entered in R because it will be separated from the main text
and in an alternate font, slightly shaded, and with R keywords colored appropriately.
For example, the commands:
> x < seq(0,100,by=2)
> y < rnorm( 51)
> pl ot ( x, y , xlab="X Axis" , ylab="Y Axis")
create a scatter plot for the variables x, a sequence of even numbers from 0 100 and
y which are random numbers sampled from a normal distribution. The result is given
in a new graphics window with a plot similar to what is shown in Figure 1. How plots
are made and saved to a le for subsequent use is covered in depth though out the
book. I have decided to sprinkle instructions of how to create graphics into the text at
locations that are appropriate for the content being discussed rather than creating one
or more chapters on Graphics with made up data presented out of context with how that
particular graphical representation is appropriate.
In all code provided in this text will have text highlighting showing R keywords in dark
blue and strings in red (see Chapter 2 for more information on these commands). If you
are using a good editor to write your scripts, you will see this kind of text highlighting
in your own work. In these code listings the > character is the prex given by R and is
not typed. I provide it here because I want to differentiate between code you type and
answers that are given by R , which will not have the > character in it such as:
> 2 6
[ 1] 12
> rnorm( 10)
[ 1] 1.08495736 1.25010428 0.76237538 0.08486045 1.62145675 0.54872689
[ 7] 0.64345848 0.43850325 0.26551658 0.41362136
> pi /2
[ 1] 1.570796
where the answers are given in the line immediately following what was entered. Along
with the answer is also an index for the answer or answers. For example, the second
Biological Data Analysis Using R
xvi Preface
Figure 1: Example scatter plot.
example gets 10 random numbers from a normal distribution but can only give 6 on a
line before it wraps around. The [7] tells you that the rst number on the second line
is the seventh in the sequence. When you operate on vectors or matrices, these indices
are relatively important and allow you to easily nd specic indices rapidly.
Acknowledgments
There are several people I would like to acknowledge for their assistance in this work.
This has been possible primarily due to the exibility of my Department in allowing me
to experiment on our graduate students. Next, I wish to thank Dr. James Vonesh who
has goaded me into putting this together and been my colleague in crime as we continue
to push R as a general tool in our curricula. Members of my laboratory Stephen Baker,
Daniel Carr, Candace Dillion, Crystal Meadows, and Cathy Viverette sat through the rst
iteration of the course and have provided insightful feedback on the both the focus and
the content. I would also like to thank the developers of R, L
A
T
E
X, Grass GIS, Emacs, and
Vim who have provided a set of tools that facilitate good research.
Rodney J. Dyer
Richmond
June 2009
Biological Data Analysis Using R
Part I
Basic Usability
1
Chapter 1
Getting R
I am not going to spend much time on how you go about getting and installing R on
your computer. If you are going to use a machine on campus, it should have it already
installed on it. If not, VCU does not allow students to install programs on their ma-
chines so this Chapter is somewhat irrelevant anyways. However, if you are using your
own computer (which is always the best idea), the internet has a much more in-depth
and complete iteration of how to get and install the R environment for your particular
machine. Reproducing that here would be a waste of paper and both of our times as it
would probably be out of date before long.
1.1 What Is R
R is both a language and an interface for statistical analysis, programming, and graph-
ics. R is modeled after the S language that was originally created by AT&T and in many
cases scripts written for R can be run in S with little to no modication. R has be-
come a standard interface for statistical analysis in biological sciences due in part to its
openness, ability to be extended by users and it vibrant user base.
The R environment is a command-line interface that allows easy manipulation of data,
calculation of parameters related to that data, an easy to understand grammar that fa-
cilitates rapid program creation, and the ability to produce publication quality graphics.
Moreover, you can create R scripts that describe how you analyzed your data so that in
the future you can pick up where you left off. Increasingly, entities such as NSF and
prominent research journals are making R scripts a normal component of the Supple-
mentary materials that you upload along with your research results and nal reports. It
is my opinion that the sooner you start documenting your data and creating a history of
how you perform analyses on this data, the better you will be in the long run.
3
4 CHAPTER 1. GETTING R
1.2 Where Do I Get It?
The main webpage for R is located at http://www.r-project.org/ Here you can nd in-
formation on the latest version of R available for your platform. Moreover, you can nd
some nice screenshots, nd out what is new in the R community, nd links to man-
uals, newsletters, wikis, and books on R . There is a lot of information in the online
community and in general, they are a friendly lot. Since R has been around for quite a
while, most of your most basic questions can be answered by a quick google search of
the mailing list repositories. It is always a good idea to check these out prior to posting
to a discussion board or email list so you do not get the old RTFM treatment...
1.2.1 Installation From Binaries
The CRAN site maintains pre-compiled binary distributions for Linux, Mac OSX and
Windows. These binaries are the latest stable versions of the software and contain the
basic libraries that you need to run R on your operating system. Depending upon your
platform, the package will contain an installer that allows you to clickity-click your way
through the process and have a base R installation on your machine.
Connected from the main R site is also the CRAN repository where people make avail-
able extensions to R that you can download and use. There is a tremendous variety of
solutions available for you and it is always in your best interest to try to see if someone
has already tackled the problem you are working with. There is no reason to reinvent
the wheel, your time is too valuable.
1.2.2 Compiling
If you know what a compiler is and have one on your computer then you are probably
able to compile the latest version of R on your machine. If you fall into this category
then you do not need me to tell you how to proceed, there is a lot of good documentation
on this found on the R website.
Biological Data Analysis Using R
Chapter 2
Language & Grammar
R is a language that has its own grammar and in this chapter you will be exposed
to some basic concepts regarding these. In this and all subsequent Chapters, it is
important for you to remember that computers do exactly what you tell them to, and
often not what you had wanted them to do. So learning the grammar is an important
step in understanding R .
In this chapter, you will focus on the following topics:
Learn basic data types and how to create them in R
Understand various operators and how they can be used.
Understand variable naming and be able to create, manipulate, and destroy
This is a pretty short list of things but it will take you a bit of time to get through it. The
main goal here is to understand a small subset of the different kinds of data that can
be produced in R and how we interact with them. Later, we will become more procient
with them and add new data types as we move forward.
2.1 Overview
R itself consists of an underlying engine that takes commands and provides feedback
on these commands. From a technical perspective R is called a Function Language as
each command you give the R engine is either an:
Expression An expression is a statement that you give the R engine. R will evaluate
the expression, give you the answer and not keep any reference to it for future use.
Some examples include:
> 2 + 6
[ 1] 8
> sqrt ( 5)
[ 1] 2.236068
> 3 ( pi /2) 1
[ 1] 3.712389
5
6 CHAPTER 2. LANGUAGE & GRAMMAR
In each of these examples, R evaluates the expression and gives you an answer.
When you use it like this, R is acting as a gloried calculator.
Assignment An assignment causes R to evaluate the expression and stores the result
in a variable. This is important because you can use the variable in the future. An
example of an assignment is:
> x < 2+6
> myCoolVariable < sqrt ( 5)
> another one number23 < 3 ( pi /2) 1
> x
[ 1] 8
> myCoolVariable
[ 1] 2.236068
> another one number23
[ 1] 3.712389
Notice here the use of the assignment operator <-. This is made with a less
than character and a minus character. As for the expression, the variables x,
myCoolVariable, and another one number23 are all the names of variables whose
value was assigned with the expression. Also notice that to retrieve the value of a
variable, just type it into the command line and it will provide the current value.
2.2 Function Quickie
This chapter will introduce you to several conventions, the main one of which is the
function. A function in R is a collection of statements bound together to make it easier
to use. In the previous example, I used the function sqrt(x), which is the function that
gives the square-root of the argument being passed (or an error if there is one).
Some functions are easy to understand and others are relatively complicated. We will
spend a whole chapter on functions later in the book (see Chapter 12) when you be-
gin to write your own. However, in the interim, you need to know a few things about
functions.
1. A function has two parts; (1) a unique name, and (2) the stuff (e.g., variables) passed
to it within the parentheses. Not all functions need any additional variables. For
example, the function ls () shows which variables R currently has in memory and
does not require any parameters.
2. If you forget to put the parentheses on the function and only use its name, by
default R will show you the code that is inside the function (unless it is a compiled
function). This is because each function is also a variable. This is why you should
not use function names for your variable names (see 2.3 for more on naming).
3. To nd the denition of a function, the arguments passed to it, details of the imple-
mentation, and some examples, you can use the ? shortcut. To nd the denition
for the sqrt() function type ?sqrt and R will provide you the documentation for that
function. If R cannot nd the function you may have to do a more thorough search
using the help.search("functionName") approach. This searches throughout the docu-
mentation system and even uses some cool fuzzy searching techniques. For more
Biological Data Analysis Using R
2.3. VARIABLES 7
info on how to use help.search(), type ?help.search() (recursive logic recurses...).
4. Functions can be organized into libraries and only loaded when needed. At the
time of this writing there are just over 1600 different packages containing different
libraries on http://cran.r-project.org. There is no reason to have every conceivable
library loaded and in fact if they were to be loaded would probably leave little mem-
ory for you to work with your data on. As a rule of thumb, only load the libraries
that you need when you need them. More on libraries as we go forward.
5. Functions may have more than one parameter passed to it. Often if there are
a lot of parameters given then there will be some default values provided. For
example, the log() function provides logarithms. The denition of the log function
show log(x, base=exp(1)) (say from ?log). Playing around with the function shows:
> l og ( 2)
[ 1] 0.6931472
> l og ( 2 , base=2)
[ 1] 1
> l og ( 2 , base=10)
[ 1] 0.30103
where without the optional base= parameter, it is clear that the log() function returns
the natural log (in fact if you ?ln there is nothing found).
2.3 Variables
A variable is something that can hold an item for you. While this is a little bit of Dyer-
speak, and I am sure that there are more elegant denitions, it is important to under-
stand that variables are things that you will interact with. For example, you may have
a predictor and a response variable you want to nd a correlation between. It is your
responsibility to dene these variables and then you can subsequently use them in your
analyses.
There are some naming conventions that you can follow to make your life a bit eas-
ier.
1. It is a pretty good idea for you to start your variable name with a letter. You cannot
use a number or punctuation as the rst character of a variable (N.B. you can use
a period to start it but the variable will be hidden from you and you cannot see it
with ls () so unless you know what you are doing, dont to this).
2. Variable names cannot have spaces in them although it is possible to use periods
(.), underscores ( ), or you can use what is called camel case (e.g., NumberOfDogsInHouse;
notice the use of upper and lower case letters to smush words together and make
it readable).
3. Try to name your variables something that makes sense to you. Using a,b,c,d,e,
and f as variables is probably not as informative to you when you are reading the
code as Rate, number of items, foodDataForNovember.
Biological Data Analysis Using R
8 CHAPTER 2. LANGUAGE & GRAMMAR
4. In R when you make a new variable such as x <sqrt(2) then that variable is in mem-
ory. You can recall it by typing its name and hitting return, you can use it later in
functions or calculations, and you can manipulate it (e.g., x <x/2 to decrease it by
half).
5. The function ls () provides you a list of all variables that you have dened. It is a very
helpful function. You can remove a variable from memory using the rm(variableName)
function.
2.4 Data Types
R recognizes about a dozen different types of data. While it is important to know the
differences between these data types, you will probably use only a fraction of them. All
of the data types are characterized by what R calls classes. As such, every data type has
three common functions associated with it; a constructor that creates a specied type,
as introspection function that tells you if any variable is a particular type, and a casting
function that allows you to coerce the contents of a variable into a specic type (a more
complete discussion of functions can be found in ??). This may sound a bit confusing
but in reality it is pretty straight forward. For example, the constructor is the function
type(x) will create a vector of x types ( where type is the data type will create), is.type(x) to
determine if x is that particular type of variable, and as.type(x) will return x translated into
a type of variable. Confused yet? It really isnt that bad, examples for each data type
below will discuss the specics.
To determine the type of any variable you can use the built-in function class(x). This will
tell you what kind of variable x is and is relatively important in the discussions we are
going to have below about coercion. This is an important concept for understanding
data types. What follows is a brief discussion of each data type and where appropriate
an example of the use of one, how to access it, and how we can operate on it.
2.4.1 Integers
An integer is a common counting number (e.g., one without a fractional part). Techni-
cally, integers can range from however, in practice there is a limited amount of
integers that can be dened on the range 2 10
9
. The integer type is typically used in
the development of R libraries who need to pass succinct integers to C or FORTRAN code
and is not typically used by the normal R end user.
Check out the code listing below and see how one can create, coerce, and use an inte-
ger.
> i nteger ( 5)
[ 1] 0 0 0 0 0
> x < as . i nteger ( 5)
> x
[ 1] 5
> i s . i nteger ( x )
[ 1] TRUE
> cl ass ( x )
Biological Data Analysis Using R
2.4. DATA TYPES 9
[ 1] "integer"
> x + 2
[ 1] 7
> cl ass ( x+2)
[ 1] "numeric"
> y < i nteger ( 3) + 2
> i s . i nteger ( y )
[ 1] FALSE
> y < i nteger ( 3) + as . i nteger ( 2)
> i s . i nteger ( y )
[ 1] TRUE
There are some things to notice about this:
1. The command integer(5) produces a vector (see 2.4.8) of ve integers.
2. All of the items returned from the listing(5) function were assigned a value of zero
(0), which is the default value for an integer until its value is changed to something
else.
3. The variable x is assigned a particular integer, in this case 5, and is veried by the
class(x) statement.
4. You can perform operations on integers you need to make sure that you use other
integers. For example, adding 2 to the vector of integers represented by the variable
y produces a numeric type, not an integer. Whereas the integer(3) + as.integer(2) state-
ment does return an integer type. This is your rst example of coercion, where
one data type is magically turned into another type. There are rules for these
transformations and the rst one you should recognize is that the number 2 is not
considered an integer. By default, numbers are coerced into numeric values (see
2.4.2) as integers are not used that often.
5. When adding an integer as.integer(2) to a vector of integers every element is assigned
the same number. There are a few more subtle things to know about adding things
to vectors and Ill leave that until 2.4.8.
As I said above, the integer type is not used that often and is only provided here for
completeness.
2.4.2 Numeric
Numeric types represent the majority of number valued items you will deal with. When
you assign a number to a variable in R it will most likely be a numeric type (unless you
specify otherwise such as dened in 2.4.5 and 2.4.6). Numeric data types can either
be displayed with or without decimal places depending if the value(s) include a decimal
portion. For example:
> x < numeric ( 4)
> x
[ 1] 0 0 0 0
> x[ 1] = 2.4
> x
[ 1] 2.4 0.0 0.0 0.0
Biological Data Analysis Using R
10 CHAPTER 2. LANGUAGE & GRAMMAR
Notice this is an all or nothing deal here. Also notice (especially those who have some
experience in programming other languages) that dimensions in vectors (and matrices)
start at 1 rather than 0.
Operations on numeric types proceed as you would expect but since the numeric type
is the default type, you dont really have to go around using the as.numeric(x) function. For
example:
> i s . numeric ( 2. 4)
[ 1] TRUE
> as . numeric ( 2) + 0.4
[ 1] 2.4
> 2 + 0.4
[ 1] 2.4
shows that no matter how you do it, 2.4 is a numeric data type. In general, programmers
are lazy people who try to do things that minimize the amount of typing they have to do
(since they do a lot of typing to begin with) and as such the numeric type is the easiest
to use.
2.4.3 Character
The character data type is the one that handles letters and letter-like representations of
numbers. For example, observe the following:
> x < "some sequence of letters of length 37"
> cl ass ( x )
[ 1] "character"
> y < 23
> cl ass ( y )
[ 1] "numeric"
> z < as . character ( y )
> z
[ 1] "23"
> cl ass ( z )
[ 1] "character"
Notice how the variable y was initially designated as a numeric type but if we use the
as.character(y) function, we can coerce it into a non-numeric representation of the number...
There will be times when you need to translate various things into characters, such as
when making titles and axis labels and this will come in handy.
You need to think of the numeric type as a sequence of letters, numbers, symbols, or
other stuff you can produce by pushing keys on your keyboard that are enclosed in
either single or double quotations. It doesnt really make much sense to perform any
operations on a character type (e.g., what would you expect hello*3 to accomplish)
although you can paste() them together. For example,
> x < "I am"
> y < "not"
> z < a looser
> x
[ 1] "I am"
> y
[ 1] "not"
Biological Data Analysis Using R
2.4. DATA TYPES 11
> z
[ 1] "a looser"
> paste ( x, y , z )
[ 1] "I am not a looser"
> paste ( x, z )
[ 1] "I am a looser"
It is important to note that if you are a really anal person for perfection that the paste()
function by default separates the individual variables you give it with a single space.
However, this can be modied by telling the function what to use as the separator).
> paste ( x, z , sep=" not ")
[ 1] "I am not a looser"
> paste ( x, z , sep=", ")
[ 1] "I am, a looser"
2.4.4 Constants
Constants are variables that have a particular value associated with them that cannot
be changed. They are mostly here for convienence so that we do not have to go look
up values for common things. Below are listed some common constants that you will
probably encounter as you play with R .
Table 2.1: Common constants you will run across in R
Constant Description
pi The mathematical constant, representing the ratio of a circles circumference
to its diameter.
NULL The absence of a type. This is the oubliette, complete nothingness, /dev/null
Richmond on a Wednesday night... This is commonly used by functions that return
undened responses.
nan Not a number.
Inf Innity () as well as -Inf for .
NA Typically used to represent something that is not there or missing. You can use it
for missing data if you like.
For the non-numerical constants, there are commands such as is.NULL(), is.nan(), is. innite
(and its cousin is. nite () ), and is.na() to help you gure out if particular items are of that
constant type if you like. At times this can be handy such when you have missing data
and you want to set it to some meaningful value (e.g, is.na(X) <32 will set all NA values
in X to 32). Well get into this more in depth at a later time.
2.4.5 Complex Numbers
Complex numbers are those that can be written in the form a + bi where a is the real
part and the product bi being the imaginary part with i =

1. The code snippet below
shows you how to create and query the class of a complex number.
Biological Data Analysis Using R
12 CHAPTER 2. LANGUAGE & GRAMMAR
> w < complex ( 3)
> w
[ 1] 0+0i 0+0i 0+0i
> x < complex( 3 , 4 , 5)
> y < 4+5i
> x
[ 1] 4+5i 4+5i 4+5i
> y
[ 1] 4+5i
> i s . complex ( x )
[ 1] TRUE
> i s . complex ( y )
[ 1] TRUE
The main differences here in the constructor complex() from the other ones we have seen
so far is that it can take default values. For example, when called as complex(3), it returns
three complex numbers whose real and imaginary parts are set to zero. However, calling
the function as complex(3,4,5) makes a three complex numbers each assigned a four to
the real part and a ve to the imaginary part. As shown, you can also create complex
numbers by simply typing them directly on the command line as a + bi as shown and is
probably the easiest way to do it.
2.4.6 Raw
The raw data type is a hexadecimal data type bound on the inclusive range [0 255].
Raw numbers are represented as a two digit sequence of hex numbers. Valid hex digits
include 0 9 as well as a, b, c, d, e, & f. The listing below gives you some examples of how
to create some raw data types.
> raw( 3)
[ 1] 00 00 00
> as . raw(255)
[ 1] f f
> as . raw( 13)
[ 1] 0d
> as . raw(256)
[ 1] 00
Warning message:
outofrange values treated as 0 in coercion to raw
> i s . raw( 13)
[ 1] FALSE
> i s . raw( 0d)
Error : unexpected symbol in "is.raw(0d"
> x < 0d
Error : unexpected symbol in "x < 0d"
There are several important points to make here.
1. If you try to create a raw number outside the its allowable range, R will issue you
a warning and then assign the variable the default value, 00.
2. The digits 13 while valid raw digits are not considered raw given by themselves. This
is because all numbers are considered numeric data types (see 2.4.2) by R , even
in the case of 0d which is denitely a raw hex number, R doesnt coerce it into a
raw type but leaves it as the characters 0d and then chokes on it. This is probably
good behavior.
Biological Data Analysis Using R
2.4. DATA TYPES 13
3. Similar to what was shown for integers, raw numbers must be constructed from
the constructor raw() function and cannot be directly created by simply pairing up
valid digits.
2.4.7 Logical
Logical data types are boolean variables with a value of TRUE or FALSE. Obviously, these
two values are the opposites of each other (e.g., not TRUE is FALSE, etc.). You will en-
counter logical data types in two primary situations; (1) when you are writing a condi-
tional statement that requires you to know the truth about something (e.g., if x == 0 you
probably shouldnt try to divide by x because for some reason mathematicians havent
gured out how to divide by zero yet...), or (b) if you are tying to select some subset of your
data by using a particular condition (e.g., select all entries where color == blue).
The interesting thing about logical variables is that numbers can be coerced into a logical
variable. For example the number zero, as an integer, numeric, complex, or raw data
type, is considered to be FALSE whereas any non-zero value is considered TRUE.
2.4.8 Vectors
R is a vector language and as you begin to learn more and more of it you will appreciate
the fact that you can easily work with vectors of numbers as well as single ones. In fact,
I suppose it is probably better to think of a single number as a vector of length 1, which
is why the R command line interface puts the [1] after every answer...
A vector is a sequence of items that can be created using the function vector(). However,
since a vector is simply a sequence, it can be a sequence of any type of data. For example,
I may have a vector of integers or a vector of complex numbers, or whatever. To specify
the data type for a vector, must tell it what type to use. Here is an example using the
numeric data type.
> x < vector ( "numeric" , 3)
> x
[ 1] 0 0 0
> i s . vector ( x )
[ 1] TRUE
> i s . numeric ( x )
[ 1] TRUE
Notice that it assigns default values for each entry as would be expected. However, it is
also important to notice that not only is x a vector but it is also numeric! So in actuality,
in all the preceding cases where we have used the constructor to create a new data
type they are also creating vectors! Blows you mind doesnt it! This is why it is safe to
consider R as a vector language.
Because you will use vectors so much, there is an easier way to create the using the
c() function (c for combine). This is a short-hand version and R tries to determine the
type of variables that you pass to the c() function to do the right thing
c
. Here are some
examples:
Biological Data Analysis Using R
14 CHAPTER 2. LANGUAGE & GRAMMAR
> x < c( 1 , 2 , 3)
> x
[ 1] 1 2 3
> y < c (TRUE,TRUE,FALSE)
> y
[ 1] TRUE TRUE FALSE
> z < c ( "I" ,"am" ,"not" ,"a" ,"looser")
> z
[ 1] "I" "am" "not" "a" "looser"
> notGoingToWork < c(00,0b, f f )
Error : unexpected symbol in "notGoingToWork < c(00,0b"
The only caveat here is that if the data type cannot be determined unambiguously, then
R will choke and tell you so, as shown in the last example where I was trying to make a
vector of raw data types. For cases such as these, use the normal data type constructor
(e.g., raw(3)) and then assign values to each element.
To access an element in a vector, R uses square brackets ([]) as demonstrated here:
> x < vector ( "numeric" , 3)
> x
[ 1] 0 0 0
> x[ 1] < 2
> x[ 3] < 1
> x
[ 1] 2 0 1
> x[ 2]
[ 1] 0
Since working with a vector is such a common thing, there are a number of helper
function that you can use to make vectors.
> x < 1:6
> x
[ 1] 1 2 3 4 5 6
> y < seq( 1 , 6)
> y
[ 1] 1 2 3 4 5 6
> z < seq(1,20, by=2)
> z
[ 1] 1 3 5 7 9 11 13 15 17 19
> rep( 6 , 4)
[ 1] 6 6 6 6
The notion x : y provides a vector of whole numbers from x to y. In a similar fashion the
function seq(x,y,by=z) provides a sequence of numbers from x to y but can also have the
optional parameter by= to determine how the sequence is made (in this case the by 2s for
all the odd numbers from 1 to 20). The function rep(x,y) repeats x a total of y times. These
are some real time saving options and you will probably be using them often.
2.4.9 Matrices
Matrices are 2-dimensional vectors and can be created using the default constructor
matrix() function. However, since they have 2-dimensions, you must tell R the size of the
matrix that you are interested in creating by passing it a number for nrow and ncol for the
number of rows and columns.
Biological Data Analysis Using R
2.4. DATA TYPES 15
> matrix ( nrow=2, ncol =2)
[ , 1] [ , 2]
[ 1 , ] NA NA
[ 2 , ] NA NA
> matrix(23,nrow=2, ncol =2)
[ , 1] [ , 2]
[ 1 , ] 23 23
[ 2 , ] 23 23
If you do not give matrix() a default value to put in each cell, it will ll them with NA, which
is the way R indicates a missing value.
Matrices can be created from vectors as well.
> x < c( 1 , 2 , 3 , 4)
> x
[ 1] 1 2 3 4
> i s . vector ( x )
[ 1] TRUE
> i s . matrix ( x )
[ 1] FALSE
> matrix ( x )
[ , 1]
[ 1 , ] 1
[ 2 , ] 2
[ 3 , ] 3
[ 4 , ] 4
> y < matrix ( x, nrow=2)
> y
[ , 1] [ , 2]
[ 1 , ] 1 3
[ 2 , ] 2 4
> i s . matrix ( y )
[ 1] TRUE
> i s . vector ( y )
[ 1] FALSE
Be default, if you do not provide any dimension to the matrix() function, it will produce
one with a single column of data. If you provide one of the dimensions then it will try
to determine how many of the other dimension is needed by looking at the length of the
vector that you passed (e.g., here nrow=2 was given and it gured out that it should have
two columns as well).
There is a slight gotcha here if you are not careful.
> x < 1:4
> matrix ( x, nrow=4, ncol =2)
[ , 1] [ , 2]
[ 1 , ] 1 1
[ 2 , ] 2 2
[ 3 , ] 3 3
[ 4 , ] 4 4
> matrix ( x, nrow=3)
[ , 1] [ , 2]
[ 1 , ] 1 4
[ 2 , ] 2 1
[ 3 , ] 3 2
Warning message:
In matrix ( x, nrow = 3) :
data length [ 4] i s not a submultiple or multiple of the number of rows [ 3]
> matrix ( seq( 1 , 8) , nrow=4)
[ , 1] [ , 2]
Biological Data Analysis Using R
16 CHAPTER 2. LANGUAGE & GRAMMAR
[ 1 , ] 1 5
[ 2 , ] 2 6
[ 3 , ] 3 7
[ 4 , ] 4 8
Notice here that R added the values of x to the matrix until it got to the end. However,
it did not ll the matrix so it started over again. In the rst case the size of x was
a multiple of the size of the matrix whereas in the second case it wasnt but it still
assigned the values (and gave a warning). Finally, as shown in the last case, if they are
perfect multiples, then it lls up the matrix in a column-wise fashion.
To access values in a matrix you use the square brackets just as was done for the vector
types. However, for matrices, you have to use two indices rather than one.
> X < matrix ( c( 1 , 2 , 3 , 4 , 5 , 6) ,nrow=2)
> X
[ , 1] [ , 2] [ , 3]
[ 1 , ] 1 3 5
[ 2 , ] 2 4 6
> X[ 1 , 3]
[ 1] 5
> X[ 2 , 2] < 3.2
> X
[ , 1] [ , 2] [ , 3]
[ 1 , ] 1 3.0 5
[ 2 , ] 2 3.2 6
> X[ 1 , ]
[ 1] 1 3 5
> X[ , 3]
[ 1] 5 6
We will use matrices quite a bit but will delay the commentary on matrix algebra and
operations until Chapter 8. However, the last two operations provide a hint as to some
of the power associated with manipulating matrices. These are slice operations where
only one index is given (e.g., X[1,]) provide a vector as a result for the entire row or
column.
2.4.10 Factors
Factors are a particular kind of data that is used in statistics and sampling. You can
think of a factor as a categorical treatment type that you are using in your experiments
(e.g., Male vs. Female or Treatment A vs. Treatment B vs. Treatment C). Factors can
be ordered or unordered depending upon how you are setting up you experiment.
Most factors are given in as characters so that naming isnt a problem. Below is
an example of ve observations where the categorical variable sex of the organism is
recorded.
> sex < f actor ( c ( "Male" ,"Male" ,"Female" ,"Female" ,"Unknown" ) )
> l evel s ( sex )
[ 1] "Female" "Male" "Unknown"
> tabl e ( sex )
sex
Female Male Unknown
2 2 1
> sex [ 5] < "Male"
Biological Data Analysis Using R
2.4. DATA TYPES 17
> sex
[ 1] Male Male Female Female Male
Levels : Female Male Unknown
Here the table() function takes the vector of factors and makes a summary table from it.
Also notice that the levels () function tells us that there is still an "Unknown" level for the
variable even though there is no longer a sample that has been classied as "Unknown" (it
just currently has zero of them in the data set).
2.4.11 Lists
A list is a convienence data type whose function is to group other data items.
> theLi st < l i s t ( x=seq( 2 , 30) , dog=LETTERS[ 1: 5] , hasStyle=l ogi cal ( 5) )
> summary( theLi st )
Length Class Mode
x 29 none numeric
dog 5 none character
hasStyle 5 none l ogi cal
> theLi st
$x
[ 1] 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
[ 26] 27 28 29 30
$dog
[ 1] "A" "B" "C" "D" "E"
$hasStyle
[ 1] FALSE FALSE FALSE FALSE FALSE
> theLi st$x
[ 1] 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
[ 26] 27 28 29 30
> theLi st$x[ 2]
[ 1] 3
> theLi st$x[ 2]< 22
> theLi st$x
[ 1] 2 22 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
[ 26] 27 28 29 30
> theLi st$dog [ 2]
[ 1] "B"
> theLi st$MyFavoriteNumber < 2.9 + 3i
> theLi st$MyFavoriteNumber
[ 1] 2.9+3 i
As you can see, a list can contains a range of different types of data. The summary()
function gives, not to surprisingly, a summary of the items within the list. These data
are grouped together by the list but you can access them and manipulate them just as
you would if they were a stand alone variable with the exception of the list name and the
dollar sign. R uses the dollar sign $ frequently to designate something that is contained
within something else. You will nd when you conduct analyses and assign the results
to a variable that variable will be a list and to access predicted values, or error terms, or
other components of that analysis you will do so by using the $ nomenclature.
It is important to remember that lists are general groupings of variables and these vari-
ables do not necessarily have any relationship between them other than my need to
Biological Data Analysis Using R
18 CHAPTER 2. LANGUAGE & GRAMMAR
group them as it makes sense to me to do so. This is different than what is found in the
next data type, the data frame.
2.4.12 Data Frames
Data frames are kind of like lists in that they can have named items within them, how-
ever, it is easiest for me to think of a data frame as a spreadsheet. It has rows of items,
and each row has one or more columns. As in a spreadsheet, each column has a variable
name (say height or NumberOfBumps). There is an inherent relationship between the columns
of data that have the same row in that it is an observation of some sort. This is the
distinction between data frames and lists, the i
th
row of a data frame can be considered
a single observation across all columns of variables.
Typically when I load data into R from an external source, you do so by creating a data
frame. There are other ways to load data but I nd this to be the most convenient. The
topic of data frames is large enough such that I will delay discussion of it until Chapter
3 when we discuss it depth and provide some analogies to how a data frame is like a
database.
2.5 Operators
R recognizes proper orders of operation for mathematical expressions. As in normal
notation, you can override the normal order of operations by using parenthesis in ap-
propriate areas. What follows is a brief discussion of some basic kinds of operators.
2.5.1 Assignment Operators
As described above, assignments are made using the assignment operator, <- and ac-
tually can be assigned the other way with the operator ->. Examples of assignments
include:
> x < 23
> 56 > y
> x
[ 1] 23
> y
[ 1] 56
Again, it is important to note that (a) under assignment, there is nothing printed out
form the R engine, and (b) to see the value of a variable, just type its name on the
command line.
2.5.2 Numerical Operators
Numerical operators are dened as operations on variables. These include the normal
set of operators including addition (+), subtraction (-), mutliplication (), division (), and
Biological Data Analysis Using R
2.5. OPERATORS 19
exponents (

). Examples of these operators are:

> x2
[ 1] 46
> y5
[ 1] 51
> xy
[ 1] 33
> x2
[ 1] 529
> x/y
[ 1] 0.4107143
> x
[ 1] 23
> y
[ 1] 56
Notice here that these expressions did not change the values of the variables because
there was no assignment involved.
2.5.3 Logical Operators
Often times we need to run comparisons between variables. These operators determine
the true of a statement and return a boolean (e.g., TRUE or FALSE). Operators include
equality (==; notice this is two equals signs), explicit relations (< and >), range rela-
tions (>= for equal to or greater than and <= for less than or equal to), and inequality
(! =).
> x < 23
> y < 56
> x==y
[ 1] FALSE
> x<y
[ 1] TRUE
> x>y
[ 1] FALSE
> x>=y
[ 1] FALSE
> x ! =y
[ 1] TRUE
> y<=x
[ 1] FALSE
These operators are commonly found in conditions but can also be used to select a
subset of values from a data vector (see ??).
Biological Data Analysis Using R
20 CHAPTER 2. LANGUAGE & GRAMMAR
2.6 Useful Functions
The following functions were introduced in this chapter and you will be required to use
them for the exercises. To get more information on any of these functions, use the R
help system.
class(x) This function will return the kind of variable that x is. We have been using
it all along in the discussions of data types but you will probably not use it very
much.
dim(x) This function returns the dimension of x. This function returns the number
of rows and columns in x which is appropriate for matrices and data frames. The
result is returned as a vector of length=2 with the number of rows in the rst index
and the number of columns in the second. This function will return NULL for all
other data types.
length(x) This will return the length of x which means different things depending
upon the kind of variable that x is.
If x is an integer, numeric, logical, character, complex, raw, or logical, this
function will return the number of distinct items in x. Essentially, this will tell
you if you have a single data point or a vector of data points. Remember the
default constructors of these data types allow you to make a vector of item so
it treats all these data types as a vector and returns the length of the vector.
If x is a list or a data frame then it will return the number of variables in that
list or data frame. For example, assume that theList is dened as in 2.4.11,
then length(theList) would return the number 3.
For matrices the function returns the number of elements in the matrix. So a
matrix with 3 rows and 2 columns would have a length of 6.
paste(x,y) This function concatenates items into a character string. By default, this
function puts a space between the items in x and y, although you can change this
behavior by setting a value for the optional sep parameter passed to the function.
rep(x,n) This function repeats the value x a total of n times and returns it as a
vector.
seq(f,t,by=b) This function returns a sequence of numeric types from f to t by b.
summary(x) This function will return an overview of the variable x.
If x is contains numerical values then it will provide the following quantitative
measures: Minimum, 1
st
Quantile, the Median, the Mean, the 3
rd
Quantile,
and the Maximum.
If x is a list or data frame then it provides a summary of each variable in x.
Biological Data Analysis Using R
2.7. EXERCISES 21
2.7 Exercises
The following exercises are meant to help you understand the items presented in this
Chapter.
1. Create two variables, x and y of type integer using the as.integer function and assign
them the values of 4 and 5. Add the x by y and store the result in a third variable
named z. What kind of variable is z?
2. Create two variables, x and y of type integer using the as.integer function and assign
them the values of 3 and 2. Divide the x by y and store the result in a third variable
named z. What kind of variable is z? Whis is this different than the answer in the
previous question?
3. Coerce x <- 23 into other data types to see which are amenable using the as. functions
for each data type.
4. What numeric values are considered TRUE when coerced into a logical data type using
the function as.logical () ?
5. Create a sequence of numbers ranging from 1 10 by 0.1 and assign it to the variable
x.
6. Create a sequence of numbers from 100 down to 50 by 2 and assign it to the variable
y.
7. Turn the vector of character items "Control", "Control", "Control", "Ear Removal", "Ear
Removal", "Ear Removal", "Ear Removal", "Fake Ear Removal", "Fake Ear Removal", "Fake Ear Removal",
"Fake Ear Removal" into a Factor variable and create a table from it to show the number
of entries in each treatment.
8. Create a vector of character variables that contains 25 a, 15 b, and 58 c in-
stances. What is the length of this vector? Create a table from the entries.
9. Create a variable that is a list. In the list add variables for your name, email address,
and height.
10. How is a data frame different than a list?
Biological Data Analysis Using R
22 CHAPTER 2. LANGUAGE & GRAMMAR
Biological Data Analysis Using R
Part II
Biologically Motivated Topics
23
Chapter 3
Data Frames
In this chapter we will be learning about data frames and how we can use them to
our benet. Data frames are useful as they are a single object within which we can
store data (to disk or databases), perform statistical analyses, and perform complicated
selections.
In my interactions with R , the vast majority of time that I spend working with data
that is contained with a data frame. This is because I typically keep my data in either
spreadsheets or in databases, both of which force me to coerce my observations into
something like:
Population,Height,Sex
A,23.4,Female
A,32.9,Female
A,29.7,Female
A,38.2,Male
A,32.7,Male
B,28.4,Female
B,27.3,Male
B,27.7,Male
B,30.1,Female
This format is relatively rigid but is amenable to several types of observations. The rst
row is a header row with the name of each variable spelled out. The second and all
subsequent rows are observations with a value for each column of data. Columns of
data are also separated by some kind of delimiter. Here I am using a comma (and the le
is probably saved as a csv le) but tabs, spaces, and other characters can also be used.
For the rest of this chapter, we will use the data above as an example to show how to
interact with and manipulate data frames.
In this Chapter you will learn the following skills:
Enter data into a data frame.
Load a data frame from an existing le.
Save a data frame to a le.
25
26 CHAPTER 3. DATA FRAMES
Manipulate data within a data frame.
Perform complex queries and joins on data frames.
3.1 Data Input/Output
Data can be input into data frames in two different ways; you can enter it directly or
load it from an external le. The former method is good if you have just a little bit of
data whereas the later is probably better if you have persistent data.
3.1.1 Entering Data Directly
After reading Chapter 2 discussing different data types should be all you need to under-
stand how to put data in manually. To recreate the example data set you could:
> Pop < c ( "A" ,"A" ,"A" ,"A" ,"A" ,"B" ,"B" ,"B" ,"B" )
> Ht < c(23.4,32.9,29.7,38.2,32.7,28.4,27.3,27.7,30.1)
> Sx < c ( "Female" ,"Female" ,"Female" ,"Male" ,"Male" ,"Female" ,"Male" ,"Male" ,"Female")
Once you have these variables entered into R , you can put them into a single data frame
by:
> myData < data . frame ( Population=Pop, Height=Ht , Sex=Sx)
> myData
Population Height Sex
1 A 23.4 Female
2 A 32.9 Female
3 A 29.7 Female
4 A 38.2 Male
5 A 32.7 Male
6 B 28.4 Female
7 B 27.3 Male
8 B 27.7 Male
9 B 30.1 Female
> summary( myData)
Population Height Sex
A:5 Min. :23.40 Female:5
B:4 1st Qu.:27.70 Male :4
Median :29.70
Mean :30.04
3rd Qu.:32.70
Max. :38.20
Notice how the data are already numbered by observation. The names that you pass to
the data.frame() function will be the names of the variables in the data frame and the names
of the variables you previously dened for them will be thrown away (e.g., there is not a
variable named Pop in myData).
Once you have created a data frame, you can access elements within it as you would for
a list (and even as a matrix to some extent).
Biological Data Analysis Using R
3.1. DATA INPUT/OUTPUT 27
3.1.2 Loading Data From A File
It is relatively common for you to already have data on hand and it is a bit of a waste
of time for you to re-enter the data into R (this would also cause a high probability of
errors as you type these values in). Getting data into R is pretty easy.
The data format of the le is a relatively important item. There are methods available to
import normal Excel les into R but will not go into them because the le format for this
program changes with each release and it is not portable across platforms (e.g., there is
no Excel on unix). Moreover, there are a lot of other places that you can get data such
as online databases, data loggers, etc. and a more general approach will be followed
here.
I will assume that you can get your data le into a text format. What matters for the
import are the following items:
1. Does the data have a row of variable names (headers) in the rst row? If you do not
have a row of headers then R will assign them as V 1, V 2, . . ..
2. What character do you use to separate columns of data? Is it tab, space, comma,
or some other character that separates you data columns?
3. Do you have any items that are in quotes? Some programs will output text wrapped
in quotes. This is not that common but you should be aware of it.
4. You need to either have the data le in the same directory that you are working
in when you started R or know the full path to the le (e.g., /Desktop/data.txt or
C:Whatever).
It is important for you to realize that the data you enter into a data frame have to have Note!
the same number of data columns for every observation. In the example data le above,
there are three observations for each row. If you do not have the same number of
observations for each row, R will barf up some errors. Be careful here, some times when
you export from a particular spreadsheet program (that shall remain nameless) you can
get extra columns of data that will screw up your import. You may want to open the text
le in a text editor to look to make sure if you get some odd errors. If you forget to add
one of the additional options to the read.table() function, R may actually load the le but
it wont be as you expect. For example, it the example below where I did not tell R that
the data le uses a comma as a column separator, it loads every row as a single text
observation (and considers it a factor) rather than three column of data.
> data < read . tabl e ( "DataFrame1.txt" , header=T)
> data
Population . Height . Sex
1 A, 23. 4 , Female
2 A, 32. 9 , Female
3 A, 29. 7 , Female
4 A, 38. 2 , Male
5 A, 32. 7 , Male
6 B, 28. 4 , Female
7 B, 27. 3 , Male
8 B, 27. 7 , Male
9 B, 30. 1 , Female
> data [ 1 , ]
[ 1] A, 23. 4 , Female
Biological Data Analysis Using R
28 CHAPTER 3. DATA FRAMES
9 Levels : A, 23. 4 , Female A, 29. 7 , Female A, 32. 7 , Male . . . B, 30. 1 , Female
> data < read . tabl e ( "DataFrame1.txt" , header=TRUE, sep=",")
> data
Population Height Sex
1 A 23.4 Female
2 A 32.9 Female
3 A 29.7 Female
4 A 38.2 Male
5 A 32.7 Male
6 B 28.4 Female
7 B 27.3 Male
8 B 27.7 Male
9 B 30.1 Female
> summary( data )
Population Height Sex
A:5 Min. :23.40 Female:5
B:4 1st Qu.:27.70 Male :4
Median :29.70
Mean :30.04
3rd Qu.:32.70
Max. :38.20
The options passed to the read.table() are the le name (with path if necessary), a header
parameter (TRUE or FALSE) to indicate if the data le has a header row, and sep to indicate
what character is used for a separator. Other separators are tab (indicated as sep="\t")
and as space sep="". Baring any errors that I made in typing in the data in the last section
(3.1.1) the printing of the data frame should be identical.
3.1.3 Adding Data To An Existing Data Frame
Once you have a data frame in R , you an add data to it relatively easily using. To add
additional rows of data you use the function rbind() (as in row bind). What you add to the
data frame must be another list or data frame that has the same variables in it as in
your original data frame. If you do not have all the variables in the thing you are adding
R will give you an error. Here is an example.
> rbind ( data , data . frame ( Population="B" , Height =31.3,Sex="Female" ) )
Population Height Sex
1 A 23.4 Female
2 A 32.9 Female
3 A 29.7 Female
4 A 38.2 Male
5 A 32.7 Male
6 B 28.4 Female
7 B 27.3 Male
8 B 27.7 Male
9 B 30.1 Female
10 B 31.3 Female
> data
Population Height Sex
1 A 23.4 Female
2 A 32.9 Female
3 A 29.7 Female
4 A 38.2 Male
5 A 32.7 Male
6 B 28.4 Female
7 B 27.3 Male
8 B 27.7 Male
9 B 30.1 Female
Biological Data Analysis Using R
3.1. DATA INPUT/OUTPUT 29
Notice that the addition of the data B 31.3 Female items were not retained in the data
object. That is because this function does not change the data frame that is passed to
it, rather it returns a brand new data frame that is identical to the original one but has
the additional data appended on the bottom. If you want to permanently change your
existing data frame then you need to use the assignment operator as:
> data < rbind ( data , l i s t ( Population="A" , Height=32,Sex="Male" ) )
> data
Population Height Sex
1 A 23.4 Female
2 A 32.9 Female
3 A 29.7 Female
4 A 38.2 Male
5 A 32.7 Male
6 B 28.4 Female
7 B 27.3 Male
8 B 27.7 Male
9 B 30.1 Female
10 A 32.0 Male
To add additional columns of data you use the function cbind() (as in column bind). This
amounts to adding another variable to all the observations in your current data set.
Again, for this to work, you should provide as many items as there are rows of data in
the data frame.
> cbind ( data , l i s t ( SizeClass = c( 1 , 1 , 1 , 2 , 2 , 1 , 2 , 2 , 1 , 2) ) )
Population Height Sex SizeClass
1 A 23.4 Female 1
2 A 32.9 Female 1
3 A 29.7 Female 1
4 A 38.2 Male 2
5 A 32.7 Male 2
6 B 28.4 Female 1
7 B 27.3 Male 2
8 B 27.7 Male 2
9 B 30.1 Female 1
10 A 32.0 Male 2
Again, if you want to make the additions to your data frame permanent then you need
to use the assignment operator.
> data < cbind ( data , l i s t ( SizeClass = c( 1 , 1 , 1 , 2 , 2 , 1 , 2 , 2 , 1 , 2) ) )
> data
Population Height Sex SizeClass
1 A 23.4 Female 1
2 A 32.9 Female 1
3 A 29.7 Female 1
4 A 38.2 Male 2
5 A 32.7 Male 2
6 B 28.4 Female 1
7 B 27.3 Male 2
8 B 27.7 Male 2
9 B 30.1 Female 1
10 A 32.0 Male 2
The reason that these two functions do not change the data frame that you passed to
them is because you may want to make a temporary data frame with some additional
variables or copy the data frame
Biological Data Analysis Using R
30 CHAPTER 3. DATA FRAMES
3.1.4 Copying Data Frames
To copy a data frame, use the assignment operator. This make a new copy of the data
frame that is independent. For example, in the listing below, newData is made as a copy
of data. Then the Population variable for the rst row is changed from A to B. Notice how
changes to newData are independent of entries in data.
> newData[ 1 , ]
Population Height Sex SizeClass
1 A 23.4 Female 1
> newData[ 1 , 1] < "B"
> newData[ 1 , ]
Population Height Sex SizeClass
1 B 23.4 Female 1
> data [ 1 , ]
Population Height Sex SizeClass
1 A 23.4 Female 1
3.1.5 Removing Data From A Data Frame
How you remove items from a data frame depends upon if you are removing columns
or rows of data. To remove a row of data (e.g., a whole set of variables for a single
observation) you an use a negative sign in front of the index.
> data[ 10,]
Population Height Sex SizeClass
1 A 23.4 Female 1
2 A 32.9 Female 1
3 A 29.7 Female 1
4 A 38.2 Male 2
5 A 32.7 Male 2
6 B 28.4 Female 1
7 B 27.3 Male 2
8 B 27.7 Male 2
9 B 30.1 Female 1
> data
Population Height Sex SizeClass
1 A 23.4 Female 1
2 A 32.9 Female 1
3 A 29.7 Female 1
4 A 38.2 Male 2
5 A 32.7 Male 2
6 B 28.4 Female 1
7 B 27.3 Male 2
8 B 27.7 Male 2
9 B 30.1 Female 1
10 A 32.0 Male 2
Again, this returns a data frame without the given index. If you want to make this
permanent you must make an assignment as before. You can also pass an array of
indices to remove more than one at a time (see also the function subset() in 3.3.1).
> data < data[ 10,]
> data
Population Height Sex SizeClass
1 A 23.4 Female 1
2 A 32.9 Female 1
3 A 29.7 Female 1
Biological Data Analysis Using R
3.1. DATA INPUT/OUTPUT 31
4 A 38.2 Male 2
5 A 32.7 Male 2
6 B 28.4 Female 1
7 B 27.3 Male 2
8 B 27.7 Male 2
9 B 30.1 Female 1
> data[c ( 2 , 4 , 6 , 8) , ]
Population Height Sex SizeClass
1 A 23.4 Female 1
3 A 29.7 Female 1
5 A 32.7 Male 2
7 B 27.3 Male 2
9 B 30.1 Female 1
Deleting a column of data can also be accomplished by the same manner or by assigning
the variable the value of NULL.
> data < data[ , 4]
> data
Population Height Sex
1 A 23.4 Female
2 A 32.9 Female
3 A 29.7 Female
4 A 38.2 Male
5 A 32.7 Male
6 B 28.4 Female
7 B 27.3 Male
8 B 27.7 Male
9 B 30.1 Female
> data$Sex < NULL
> data$Population < NULL
> data
Height
1 23.4
2 32.9
3 29.7
4 38.2
5 32.7
6 28.4
7 27.3
8 27.7
9 30.1
3.1.6 Saving Data Frames to Files
There comes a time when you have to save some data you have been working on. In fact,
it is quite often. There are several ways to save data in R . First, you can have R save
every variable in memory. When you quit R using the q() function, it will ask if you want
to save:
> q ( )
Save workspace image? [ y/n/c ] : y
If you do, there will be a .RData le saved in the directory you are working with that
contains all the data you currently have in memory. When you restart R , it will load
these data back into memory for you. Fairly easy and direct way of getting your data to
disk and back and it is cross-platform. If you are going to use this kind of data saving,
you should create a new folder for any data set you are working with. This will keep
Biological Data Analysis Using R
32 CHAPTER 3. DATA FRAMES
the raw data le(s) in the same location as the data entered and formatted in R . The
main drawback to this is that the name of the saved data le (.RData) starts with a period
(.) and will therefore be invisible to you when you look in the folder with your normal
Finder, File Browser, or whatever. You can easily overwrite it or throw it away since it
isnt immediately visible. It is also a bit inefcient in that if you have a bunch of other
variables in memory you may not want to save them all. If I just merged a bunch of data
frames (see 3.3.2), I may only want to save the nal data.
The second way that you can save your data frame is to save the data frame directly.
This allows you to save different data frames with different names and you can save
them where ever and named what ever you like.
> save ( data , f i l e ="MyNewSavedData.Rdata")
You can also save several variables at once by passing their names as a list to the save()
function. Here is an example:
> g < 1:20
> otherData < f actor ( c ( T, T, T, T, F, F) )
> g
[ 1] 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
> otherData
[ 1] TRUE TRUE TRUE TRUE FALSE FALSE
Levels : FALSE TRUE
> save ( l i s t =c ( "data" ,"g" ,"otherData") , f i l e ="DataType2.RData")
It is common for saved data from R to have the le sufx of .Rdata so lets not buck
tradition...
Once you have saved the data frame, you can load it back into memory at any time
by:
> l s ( )
[ 1] "data
> rm(data)
> ls()
character(0)
> load("MyNewSavedData. Rdata")
> ls()
[1] "data"
Notice here I use ls () to see what is in memory, rm() to remove data from memory (and
check, then reload the data using the load() function.
3.1.7 Deleting Data Frame
Removing a data frame from memory is no different than removing any other variable.
You simply use the rm() function as:
> rm( data )
If you have a lot of different data les in memory, you can delete them individually, as a
group, or delete everything in memory at once as shown below:
Biological Data Analysis Using R
3.1. DATA INPUT/OUTPUT 33
> l s ( )
[ 1] "elvis genotypes" "kent hovinds secret data"
[ 3] "myCoolData" "x"
[ 5] "y" "yourNotLooserData"
> rm( "x")
> l s ( )
[ 1] "elvis genotypes" "kent hovinds secret data"
[ 3] "myCoolData" "y"
[ 5] "yourNotLooserData"
> rm( l i s t =c ( "y" ,"myCoolData") )
> l s ( )
[ 1] "elvis genotypes" "kent hovinds secret data"
[ 3] "yourNotLooserData"
> rm( l i s t =l s ( ) )
> l s ( )
character ( 0)
To delete individual variables, you must name them but when you delete several vari-
ables you need to tell the rm() command that you are going to pass it a list of variable
names to delete (the list =) parameter. The nal example shows you how you can tell it
to delete everything in memory (e.g., delete this list and this list is all the data that are
currently in memory.
3.1.8 Components of a Data Frame
A data frame has a few distinct components in addition to the data points. Using the
function attributes() shows the things that are make up a data frame. This function returns
a list containing the variables names, class, and row.names.
> attri butes ( data )
$names
[ 1] "Population" "Height" "Sex"
$cl ass
[ 1] "data.frame"
$row. names
[ 1] 1 2 3 4 5 6 7 8 9
> dataAttributes < attri butes ( data )
> dataAttributes$row. names
[ 1] 1 2 3 4 5 6 7 8 9
There are also other ways to access these attributes. In Chapter 2, you were introduced
to the class(x) function and we will not need to go over that again here. There are corre-
sponding functions names(x) and row.names(x) that you can easily use to get access to these
components of a data frame. You can also use these functions to assign new values to
an existing data set. For example:
> data
Population Height Sex
1 A 23.4 Female
2 A 32.9 Female
3 A 29.7 Female
4 A 38.2 Male
5 A 32.7 Male
6 B 28.4 Female
7 B 27.3 Male
Biological Data Analysis Using R
34 CHAPTER 3. DATA FRAMES
8 B 27.7 Male
9 B 30.1 Female
> names( data ) < c ( "Group" ,"DistanceFromGround" ,"Gender")
> data
Group DistanceFromGround Gender
1 A 23.4 Female
2 A 32.9 Female
3 A 29.7 Female
4 A 38.2 Male
5 A 32.7 Male
6 B 28.4 Female
7 B 27.3 Male
8 B 27.7 Male
9 B 30.1 Female
> row. names( data ) < seq( 9 , 1 , by=1)
> data
Group DistanceFromGround Gender
9 A 23.4 Female
8 A 32.9 Female
7 A 29.7 Female
6 A 38.2 Male
5 A 32.7 Male
4 B 28.4 Female
3 B 27.3 Male
2 B 27.7 Male
1 B 30.1 Female
3.2 Slicing
Grabbing portions of your data frame is pretty easy. Below are some examples of how
you can access some of your data components:
> data [ , 1]
[ 1] A A A A A B B B B
Levels : A B
> data [ , 2]
[ 1] 23.4 32.9 29.7 38.2 32.7 28.4 27.3 27.7 30.1
> data [ 1: 4 , ]
Population Height Sex
1 A 23.4 Female
2 A 32.9 Female
3 A 29.7 Female
4 A 38.2 Male
> data$Sex
[ 1] Female Female Female Male Male Female Male Male Female
Levels : Female Male
> data$Population
[ 1] A A A A A B B B B
Levels : A B
Here are some rules that you need to keep in mind:
1. To access a data frames items by index, you use the square brackets [] along with
the indices of the components separated by a comma ,.
2. R uses indices for all its data types in what is called row major format. That is to
say that the rst index is for the row and the second index is for the column. For
example data[1,2] will provide access to the 1
st
row and the 2
nd
column.
Biological Data Analysis Using R
3.3. COMPLEX SELECTIONS 35
3. To get all the items in a given row or column you can leave out the index. For
example, the command data[i,] returns all rows of data from the i
th
row whereas
data[,j] returns the data in all rows for the j
th
column.
4. You can also index the data for a particular column by calling its name. For exam-
ple, the example data set has variables named Population, Height, and Sex. You can
get all the data in one of these variables by using the notation data$VariableName as in
data$Population.
5. To get a range of values on one or the other index such as the 2
nd
through 5
th
entries in the height variable you put the range of indices separated by a colon as
in data[2:5,2]. You can also combine this with the naming of the variables, which
may be able to make it a bit easier to read, as data$Height[2:5]. This can work
in both directions, as shown above when retrieving all the data for the rst four
records (data[1:4,]).
3.3 Complex Selections
R data frames can be thought of as pseudo databases. There is a standard language that
has been adapted by both the American National Standards Institute (ANSI) and later the
International Organization for Standardization (ISO). If you ever interact directly with a
database, you use the Standard Query Language (SQL) to interact with the data. R does
allow you to interact with databases through one of its many database libraries but I will
not be covering that in this chapter. However, if you are familiar with some basic SQL
operations you will nd this section rather easy. If not, I will be spending a little extra
time trying to convince you that it is probably in your best interest to understand how to
query your data frames because it gives you a lot of power and exibility. After all, being
agile with your data is a key skill I hope you will be learning in this course.and show you
how to use a data frame as a lite-database. Even if you do not ever use a database, this
section is really important as it will allow you to think about interacting with your data
in interesting and complex way.
To understand SQL you need to understand that in a database data is contained within
tables. And tables have rows and columns of data, just like a data frame. Each table also
has a name. You can think of a database table as a worksheet in a spreadsheet program
if that helps (though real database gurus are probably cringing as they read that). The
SQL language is very easy to understand and I will partition this section into commands
that query the database and those that create new data frames by the combination of
two or more existing data frames that have a common data column.
3.3.1 Queries
Queries are essentially what we have been doing in 3.2 with indices so I wont go over
the basic stuff that we have already covered other than to show the SQL equivalents
in case you need to know them. I will however delve a bit into how the function subset()
works because it is pretty powerful.
Biological Data Analysis Using R
36 CHAPTER 3. DATA FRAMES
To select all observations in SQL, you use the statement SELECT
*
FROM tableName, which in
R is simply what we have been doing by tying the name of the data frame (hereafter I
will use data to refer to the name of the table for similarity with our previously loaded
data frame).
> data
Population Height Sex
1 A 23.4 Female
2 A 32.9 Female
3 A 29.7 Female
4 A 38.2 Male
5 A 32.7 Male
6 B 28.4 Female
7 B 27.3 Male
8 B 27.7 Male
9 B 30.1 Female
In these SQL statements I use words in all capitol letters to indicate SQL language
components and lowercase words to indicate table names or variables. Also, in SQL the
asterisk means everything (as in all variables).
The strength of SQL and databases lies in the fact that you can do complicated selections
from the tables. For example, in SQL you can select by row number and column number
using the statement SELECT
*
FROM data WHERE rownum==x AND colnum==y. Using the logical oper-
ator AND adds a lot of power to this statement. However, in R we have been doing this
using the indices directly and the square bracket notation as (with x = 1 and y = 2):
> data[ 1 , 2]
[ 1] 23.4
Several rows or columns can be selected in SQL by SELECT
*
from data WHERE rownum>=5 AND
rownum<=7 is accomplished in R as:
> data [ 5: 7 , ]
Population Height Sex
5 A 32.7 Male
6 B 28.4 Female
7 B 27.3 Male
To get only a subset of the variables in each row, you can indicate which variables you
are interested in selecting in SQL as SELECT height, sex FROM data and in R we can either
slice both indices as:
> data [ , 2: 3]
Height Sex
1 23.4 Female
2 32.9 Female
3 29.7 Female
4 38.2 Male
5 32.7 Male
6 28.4 Female
7 27.3 Male
8 27.7 Male
9 30.1 Female
Or we can use the subset() function as:
Biological Data Analysis Using R
3.3. COMPLEX SELECTIONS 37
> subset ( data , sel ect =c ( "Height" ,"Sex" ) )
Height Sex
1 23.4 Female
2 32.9 Female
3 29.7 Female
4 38.2 Male
5 32.7 Male
6 28.4 Female
7 27.3 Male
8 27.7 Male
9 30.1 Female
Often times you will have rather large data sets in R that you will be working with and
it may be easier to grab parts of your data set by using names of variables rather than
by using column indices (it is up to you).
You can also get a bit more specic and only look for components in your data set using
relational operations. For example, the SQL statements SELECT
*
FROM data WHERE height>30
and SELECT
*
FROM data WHERE height>30 AND columnnum==2 is accomplished in R by:
> data [ data$Height >30,]
Population Height Sex
2 A 32.9 Female
4 A 38.2 Male
5 A 32.7 Male
9 B 30.1 Female
> data [ data$Height >30,2]
[ 1] 32.9 38.2 32.7 30.1
Notice how in the last example here I mixed the use of selecting subsets of observations
using the relational operator > and subsets of column using the numeric index. Also
notice how using the 2 in the position after the comma gives only the second column of
data.
You can combine conditions in a SELECT-like query such as SELECT
*
FROM data WHERE height>30
AND sex="Male" by using the unary & operator as:
> data [ data$Height>30 & data$Sex=="Male" , ]
Population Height Sex
4 A 38.2 Male
5 A 32.7 Male
This complicated statement needs to be dissected to reduce confusion. The part in the
square brackets [] consists of the stuff on the left side of the comma (data$Height>30 &
data$Sex=="Male") and the stuff on the right side (which happens to be empty in this case).
There are some things to remember when doing compound statements like this:
1. The & operator in between the requires that the things on both sides of it are TRUE.
2. The equality operator == must be a double equals sign.
3. I nd it easy to take a few passes at these compound statements to make sure I am
getting them correct.
In addition to the AND operator in the SELECT statements you there is also an OR operator. It
is valid to say in SQL SELECT
*
FROM data WHERE sex=="FEMALE" OR population=="A". This can also
be done in R using the OR operator .
Biological Data Analysis Using R
38 CHAPTER 3. DATA FRAMES
> data [ data$Sex=="Female" | data$Population=="A" , ]
Population Height Sex
1 A 23.4 Female
2 A 32.9 Female
3 A 29.7 Female
4 A 38.2 Male
5 A 32.7 Male
6 B 28.4 Female
9 B 30.1 Female
If the selection of subsets of your data become more complicated than this, you can use
parenthesis to separate out conditions. This makes it easier for you to read and since
you are the one that will be writing this code and coming back later and looking at it, it
pays to be as un-convoluted as possible. Here is a whack example from the SQL SELECT
*
FROM data WHERE (population=="A" AND sex=="Female") OR (population=="B" AND height<30).
> data [ ( data$Population=="A" & data$Sex=="Female")
+ | ( data$Population=="B" & data$Height<30) , ]
Population Height Sex
1 A 23.4 Female
2 A 32.9 Female
3 A 29.7 Female
6 B 28.4 Female
7 B 27.3 Male
8 B 27.7 Male
Note: I split the command across two lines at the OR operator. In R when you do this, it
gives you the little + sign and you can continue typing as if it were on a single line. I had
to do this because the command is longer than the width of this paper...
3.3.2 Joins
OK, so now that you have everything you want to know about how to select stuff from
within a single data frame with an arbitrary level of complexity lets move into joins. A
join is an operation where you have two or more tables (or data frames) and you are
going to create a new one based upon the merging of the two, provided that they both
have a variable in them you can use as a common index. Here are two examples that we
will be using. The rst table is the data table we have been working with thus far.
> data
Population Height Sex
1 A 23.4 Female
2 A 32.9 Female
3 A 29.7 Female
4 A 38.2 Male
5 A 32.7 Male
6 B 28.4 Female
7 B 27.3 Male
8 B 27.7 Male
9 B 30.1 Female
The second table is one that has characteristics of the Populations themselves. It is in the
example data sets and is called PopulationAttributes.txt and we can load it into R as:
> popData < read . tabl e ( "PopulationAttributes.txt" , header=T, sep=",")
Biological Data Analysis Using R
3.3. COMPLEX SELECTIONS 39
> popData
Population LongName State North East Elevation
1 A Richmond Vi rgi ni a 37.53300 77.4670 45.7
2 B Seattl e Washington 47.60972 122.3331 0.0
If you look at these two tables, there is the common variable Population. So in essence,
I could add the data from popData and data to create a new data set that has all this
information. It is common in databases to have tables split like this. It saves space
(imagine having the 5 extra data columns for each row in data, it would be repetitive and
for large data sets may max out the memory of your computer. It is also common to nd
biologists who have programmed software to do some kind of analysis that requires you
to put some kinds of data in one le another kind in a second le, etc. Joins allow you
to take these different data frames and join them (catchy name, no?).
To join two tables you will use the function merge() on the data sets. In SQL this would be
SELECT
*
FROM data, popData WHERE data.Population == popData.Population. Fortunately, is is a bit
easier to do this in R , here is an example:
> merge( data , popData)
Population Height Sex LongName State North East Elevation
1 A 23.4 Female Richmond Vi rgi ni a 37.53300 77.4670 45.7
2 A 32.9 Female Richmond Vi rgi ni a 37.53300 77.4670 45.7
3 A 29.7 Female Richmond Vi rgi ni a 37.53300 77.4670 45.7
4 A 38.2 Male Richmond Vi rgi ni a 37.53300 77.4670 45.7
5 A 32.7 Male Richmond Vi rgi ni a 37.53300 77.4670 45.7
6 B 28.4 Female Seattl e Washington 47.60972 122.3331 0.0
7 B 27.3 Male Seattl e Washington 47.60972 122.3331 0.0
8 B 27.7 Male Seattl e Washington 47.60972 122.3331 0.0
9 B 30.1 Female Seattl e Washington 47.60972 122.3331 0.0
> cl ass ( merge ( data , popData ) )
[ 1] "data.frame"
As you can see, it returns a new data frame with all the data included. I think this has
gotten you enough exposure so that you can probably be dangerous. The best way to get
comfortable with these methods is to actually use them.
Biological Data Analysis Using R
40 CHAPTER 3. DATA FRAMES
3.4 Useful Functions
The following functions were introduced in this chapter and you will be required to use
them for the exercises. To get more information on any of these functions, use the R
help system.
cbind(x) This function binds a column onto the right side of x. This only works with
some kinds of data types (e.g., those where an operation of appending on a column
of data makes sense).
rbind(x) This functions binds a row of data onto the end of x. Again for those data
types that this operation makes sense.
load(x) If x is the name of a .Rdata data le then it will load the contents into memory.
merge(x) This function takes two data frames and merges themon a common variable
name. If there are more than one common variable name you can specify which
one and if there are no commonly named variables then you are out of luck (unless
you have variables that hold the same data but are just named differently).
rm(x) This function removes x from memory. Gone. Auf wiedersehen. Cant get it
back.
save(x,filename=y) This function saves the R object x to le named y.
subset(x) This function returns a slice of your data frame where you can specify
which variables to use. You can also do this with creative use of conditional opera-
tors and variable names.
Biological Data Analysis Using R
3.5. EXERCISES 41
3.5 Exercises
The following exercises are meant to help you understand the items presented in this
Chapter.
1. Create three different variables, a logical one, one that is a numeric type, and a vector
of characters. Use these to create a data frame named theData.
2. In the folder for this Chapter there is a text le named GuinneaPigData.csv. Load it into
memory and print out a summary.
3. How do you indicate a missing data point in a data le?
4. Add a numeric data column to the existing data frame, theData. Provide a summary of
the data.
5. How would you save the data frame, theData, to a le named newData.Rdata.
6. What is the difference between row major indexing and column major?
7. Using index numbers, select the 2
nd
and 3
rd
rows of the data set theData.
8. Read in the data le PersonData.csv from the class data set. What kind of data type is
the variable Names? How can you change this to a character type and then change the
name of the third entry in the data frame, theData, to Thomas?
9. Create a newdata set with a two variables, one that is Order = 1:4 and the other that is
Home=c("Olympia", "Juanita", "Centralia", "Tacoma", "Olympia", "Olympia"). Merge this data frame
with the one named theData and assign it the name combinedData.
10. How would you perform a query of the combined data set to select all records that
have Order >= 3 or Home == Olympia.
Biological Data Analysis Using R
42 CHAPTER 3. DATA FRAMES
Biological Data Analysis Using R
Chapter 4
Summary Statistics
In this chapter you will explore some of the methodologies that R has for describing
your data. R is an excellent platform for exploring data, looking at relationships among
variables, and graphically portraying results.
In this Chapter you will learn the following skills:
Learn about some common numerical distributions.
Learn about commonly used statistical distributions.
Understand parametric summary statistics.
Explore non-parametric summary statistics.
Use the table() function as an entry point into contingency table analysis.
Create single and multiple line gures.
Create histograms and density plots.
4.1 Distributions
R and its various sub-packages contain more numerical distributions than you will
probably ever need to use. Moreover, they provide them in a clear and concise inter-
face that has a consistent format. To my knowledge, all the distributions provide the
following four components:
1. A density function that is of the form dNameOfDistribution (e.g., dnorm(), df () & dchisq()).
2. A distribution function that is called as pNameOfDistribution (e.g., pnorm(), pf () & pchisq()).
3. A quantile function named qNameOfDistribution (e.g., qnorm(), qf () & qchisq()).
4. A function that produces random numbers sampled from the distribution that is
named rNameOfDistribution (e.g., rnorm(), rf () & rchisq()).
These are specically helpful in a number of situations. For example, you may be run-
ning a test and calculating a
2
statistics on some table of data and want to know if the
43
44 CHAPTER 4. SUMMARY STATISTICS
value of your observed statistic,
2
Obs
is large given the particular degrees of freedom that
you have at your disposal. Now typically, we have memorized due to the sheer number
of times that we have used it, what the critical value for a
2
statistic with a single degree
of freedom should be ( 3.841459 right?). However, what if we have 8 degrees of freedom
and
2
Obs
= 15.507? You could go nd that old stats book on the shelf and page through
the back of it to nd the correct Appendix that has the right table (How do you read
those tables again?). Or you could use the various functions in R .
In this section, three aspects of using distributions within a statistical context will be
introduced. First, you will learn how to determine critical values for the
2
distribution
as used in formal hypothesis testing using the quantile functions. Then you will see how
the distribution function can tell you the probability of a particular estimation of the
2
test statistic.
4.1.1 Finding Critical Values
In formal hypothesis testing, there is a specic test statistic that is proposed. Moreover,
the estimation of a value for that statistic is compared to a known cutoff set by the
degrees of freedom in the model and the Type I error rate that you have chosen (e.g., the
value). For some reason, as a biologist we have settled on an = 0.05 value to have
some kind of special meaning. Now, this is probably an over simplication of things that
was used initially as a teaching aid for understanding the meaning of Type I errors. There
is nothing intrinsically interesting about = 0.05 and it is probably more informative for
me to know the real probability of your calculated test statistic rather than if it exceeds
dome arbitrary cutoff. I mean, is it really that different an interpretation if P = 0.049
versus P = 0.051? That being said, lets jump into understanding how we nd critical
values for some pre-dened value for in different distributions.
The most commonly used distribution observed as an undergrad is probably the
2
distribution. The distribution itself is shown in Figure 4.1 for three different values for
the degrees of freedom. This and other statistical distributions require that you provide
the degrees of freedom before it can give you any information.
For any one particular set for the parameters and df, there is a dened cutoff. The
value of the cutoff is dened as the point along the xaxis at which there is 1 of the
are under the curve to the left of the point and of the are under the curve from that
point and beyond. While this is a very non-technical denition but I think you get the
point when you consider the shaded region in Figure 4.2 and the 1 region that is
unshaded.
To determine the critical value of the
2
distribution you use the qchisq() function. If you
were to look up the signature of this function (by typing ?qchisq into R ), you would see
that it accepts the following options:
qchisq ( p, df , ncp=0, lower . t ai l = TRUE, l og . p = FALSE)
There are two required parameters for this function, p and df. You can tell by looking
at this signature that they are required because they do not have an = sign next to
them and a default value given. If a parameter has a variable=value format in a function
Biological Data Analysis Using R
4.1. DISTRIBUTIONS 45
Figure 4.1: Values for the density function
for the
2
distribution with 1, 2, and 3 de-
grees of freedom.
Figure 4.2: A graphical depiction of the
critical value of the
2
distribution for =
0.05 and df = 3. The shaded region con-
stitutes a proportion of the area under the
curve equal to .
signature then the value will be assigned to variable if you do not give it a value when you
call it. Default values are very helpful and save a lot of typing on your part.
The parameter p is the 1 cutoff you are interested in nding. In the classic case,
this would be 1 0.05 = 0.95. At rst, it seems a little backwards to use 1 instead
of but if we look at the graphical depiction of this distribution in Figure 4.2, we see
that the point in question is where we actually have 95% of the area under the curve and
we are interested in the extreme portion. The next required parameter is df, which
corresponds to the degrees of freedom. As shown in Figure 4.1, this parameter controls
both the shape and location of the
2
values.
There are several optional parameters that you can pass to the qchisq() function and I
will briey mention them here for completeness. If you are interested in a more in depth
discussion of these parameters, look up the qchisq() function and read the documentation.
The ncp=0 option species a non-centrality parameter allowing you to get the critical val-
ues for a non-central
2
distribution. The lower.tail=TRUE indicates that you are interested
in the p proportion of the data in the lower tail of the distribution (e.g., P[x < 1]) rather
than the the 1 portion of the other side of the distribution (e.g., P[x > 1 ]). The
default value here is what we expect since we are interested in nding the proportion
on the right side of the distribution not on the left side of the distribution (which would
be all the values less than or equal to 0.03518). Finally, the log.p=FALSE option allows you
to query using the log of p rather than p directly.
There are several other statistical distributions that you can query in R for particular
critical values. Common ones that you will be playing with in the Exercises portion of
Biological Data Analysis Using R
46 CHAPTER 4. SUMMARY STATISTICS
this chapter include Students t from qt and Fishers F from qf.
Scatter & Line Plots
Creating a simple plot of a line (or points in a sequence) is accomplished using the plot ()
function. This function has a signature (e.g., the things that you can pass to the function
and the things it expects) is:
pl ot ( x, y , . . . )
This listing is not very informative! Dont worry, they get more interesting as we go along.
The plot () function is kind of a dummy function that allows you to plot lots of different
kinds of things, and if things can be plotted, they should know how to plot themselves.
Well, that is the theory at least. Lets jump into this graphing stuff by staring off with
a more basic approach to creating graphs and building up to what we see in Figure
4.1.
When you begin to create a plot, there are some default characteristics of the plot that
you may want to override. For example, the R code plot( rnorm(10) ) produces the graph
shown in the leftmost panel of Figure 4.3 consisting of a sequence of 10 random points
selected from a normal probability distribution (we will discuss these random functions
later in Section 4.2). If you try it will look different that is why they are random...
The function rnorm(x) returns x random numbers selected from a normal probability dis-
tribution with = 0 and = 1.0 (you can change these values, check the documentation
on this function using the ?rnorm command). When you look at this plot, it is rather plain
and does not convey any more information than 10 little circles. It may be of interest to
you to be able to change some of the properties of this plot. For example, you may want
to modify:
The shape of the symbols
The color of the symbols
Add a line to connect the symbols, and perhaps modify the color, width, shape of
that line.
Provide more meaningful axis labels.
Remove the box around the plot (my pet peeve)
To do this, we must understand what a graph consists of, how to access the various
components, and how to nd more information on the appropriate levels that can be set
to these components. This chapter will be very long because it takes a lot of page real
estate to show a graph, but I think youll be happy with the results when you can whip
out a nice looking graph of your data. When possible, I will use random numbers to
create these graphs so as you go through and attempt to recreate them, yours will look
slightly different than mine.
To customize any of these values, you need to pass additional information to the plot ()
function. This is what the ... part of the function signature that is shown above. Table
Biological Data Analysis Using R
4.1. DISTRIBUTIONS 47
4.1 shows a list of additional commands that can be passed to the plot () function to
customize plot appearances.
Here are some examples of how you would use some of these optional parameters with
graphs shown in Figure 4.3:
> pl ot ( x, y , xlab="X Label" , ylab="Y Label" ,pch=3, col ="green" , bty="l")
> pl ot ( x, y , xlab="X Label" , ylab="" ,pch=2, col ="blue" , type="b" , bty="n" , lwd=2)
> pl ot ( x, y , xlab="X Label" , ylab="" ,main="Title" ,sub="subtitle" ,pch=2, col ="red" ,
+ type="l" , bty="n" , lwd=5)
Figure 4.3: Some example graphs with alternate values for symbols, line types, widths, colors,
and titles.
When creating complicated graphs, I nd it easy to build them up incrementally. Start
with a plain plot () command to see what the output looks like. Then customize the labels
and titles and plot it again to see it. Then continue to add parameters and review the
plot.
Biological Data Analysis Using R
4
8
C
H
A
P
T
E
R
4
.
S
U
M
M
A
R
Y
S
T
A
T
I
S
T
I
C
S
Table 4.1: Some useful additional commands to customize the appearance of a gure. For a complete listing of possible values that can be
customized, try the ?par command.
Command Usage Description
bg bg="red" Colors the background of the gure the specied color.
bty bty="x" Sets the style of the box type around the graph. Useful values are o for
complete box (the default), l, 7, c, u, ] which will make a box with
sides around the plot area resembling the upper case version of these letters,
and n for no box (my preference)
cex cex=1.0 Magnies the default font size by the corresponding factor.
col col="blue" colors the line and symbols the given color.
fg fg="blue" Colors the foreground of the image to the set color.
lty lty=x Species the line type (0 = none, 1 = solid, 2 = dashed, 3 = dotted, etc.)
lwd lwd=x Species the width of the line ( 1 = default ).
main main="Title for Graph" Sets a title along the top of the graph.
mfrow mfrow=c(nr,nc) Creates a matrix of plots that can potentially have a number of rows
(nr) and columns (nr; see 4.3.1 for example).
pch pch=x Sets the symbol that is plotted on the gure.
sub sub="Subtitle on Graph" Adds a subtitle just under main on the top of the graph.
type type="x" Sets the plot type. Plot types can be p for points (the default), l for
lines and b for both lines and points.
xlab xlab="label for x-axis" Set the label on the x-axis.
ylab ylab="label for y-axis" Set the label on the y-axis.
B
i
o
l
o
g
i
c
a
l
D
a
t
a
A
n
a
l
y
s
i
s
U
s
i
n
g
R
4.1. DISTRIBUTIONS 49
Overlaying Plots
There are times where it is desirable to produce several plots on a common background
(e.g., the different values for df in Figure 4.1. R allows you a lot of leeway to mix up dif-
ferent types of graphs in the same plot (see Figure 11.4 for a rather complex combination
of images and plots overlayed on the same area).
To overlay two graphs, you use the par(new=T) command to tell R that the following com-
mand is going to apply to the currently active graphics device. This function allows you
to adjust a lot of different graphical parameters and the plotting of a new image onto an
existing one is only one of the things that you can adjust. For a full discussion of other
options that par() accepts type ?par in R . You use it as follows:
pl ot ( x1, y1)
par ( new=T)
pl ot ( x2, y2)
This will take the plot for the second set of variables and plot it on the same graphics
device as the previous one. When you overlay more than one plot on the same graphing
area, you must take into consideration the different scales that the graphs have. By
default R will try to maximize the are that is being plotted by changing the default
ranges of the x and yaxes. For example, if I have data such as: x
1
= [0, 1, 2, 3]; y
1
=
[10, 11, 12, 13] and plot it will automatically scale the axes to have limits of xlim=c(0,3) and
ylim=c(10,13), which means that the xaxis will start and end at 0 and 3 and the yaxis will
start and end at 10 and 13.
This is what would be expected to happen and works nicely until you try to put another
plot. If your other data has values of x
2
= [11, 12, 13, 14]; y
1
= [23, 22, 21, 20] and you try to
simply overlay the two plots by simply typing:
> x1 < c( 0 , 1 , 2 , 3)
> y1 < c(10,11,12,13)
> x2 < c(11,12,13,14)
> y2 < c(23,22,21,20)
> pl ot ( x1, y1, col ="red")
> par ( new=T)
> pl ot ( x2, y2, col ="blue")
You get the image shown in Figure 4.4.
There are several obvious issues with this image.
1. You cannot read the axis labels. The two images are put right on top of each other
and the axes are individually scaled to t the data in each plot () command.
2. It is difcult to tell the relationship among the data. If you look at the raw data, the
x1[1] = x2[1] but in the plot it appears that they are equal.
3. The labels on the axes are typed over each other.
To overcome these issues you need to rst nd the appropriate limits for the values in
both of the data sets and for both plot () statements, we need to set the xlim and ylim values
(see Table 4.1) to the appropriate values. These appropriate values will tell R what the
Biological Data Analysis Using R
50 CHAPTER 4. SUMMARY STATISTICS
Figure 4.4: Plot of two data sets using the par(new=T command but not taking into consideration
the axis limits of the two data sets before plotting.
minimum and maximum values for the x and yaxes should be. Here is some code
that does this.
> x1 < c( 0 , 1 , 2 , 3)
> y1 < c(10,11,12,13)
> x2 < c(11,12,13,14)
> y2 < c(23,22,21,20)
> yLimit < range ( c ( y1, y2) )
> yLimit
[ 1] 10 23
> xLimit < range ( c ( x1, x2) )
> xLimit
[ 1] 0 14
Here I combined the y values for both data sets and used the range() function to tell me
what the range of these values are. Then I did the same thing of the x values in both
data sets. Now, if I make the plot, I can make it for each pair of x & y variables scaling
the axes so that both data sets will be displayed on the same Figure.
> pl ot ( x1, y1, xlab="X" , ylab="Y" , bty="n" , xlim=xLimit , ylim=yLimit , col ="red")
> par ( new=T)
> pl ot ( x2, y2, xlab="X" , ylab="Y" , bty="n" , xlim=xLimit , ylim=yLimit , col ="blue")
Notice how the optional arguments xlim and ylim make sure the axes are scaled correctly
(Figure 4.1.1). I also use the bty="n" because I just hate the box that it puts around the
plot area by default and this option does not draw any box at all.
As long as you add a par(new=T) between each successive plot () command, you can add as
many plots to the same gure as you would like.
Biological Data Analysis Using R
4.1. DISTRIBUTIONS 51
Figure 4.5: Plot of two variables on the same axis after correcting for the range of each data set.
Saving Images To Disk
While it is rather cool to be able to create rather hansom graphics in R it is entirely
useless if you do not know how to save it for later use. You could take a screenshot of
the image and then crop it down a bit but that is not quite the easiest method to use
here. Almost all the images in this book were created in R and I was able to save them
into a format that made it easy to import them into this document.
R considers the little popup window that shows your graph as a graphics device. De-
pending upon which platform you are using (e.g., Linux, OSX, Windows), the kinds of
output you may be able to produce may change. At present the following types are
available:
Device What receives these graphing commands
bmp A Windows bitmap device
cairo pdf A PDF device based upon the Cairo drawing libraries
jpeg A JPEG bitmap device
pdf A PDF le
pictex A L
A
T
E
X graphics command le
png A PNG bitmap device
postscript A postscript le
quartz An OSX graphics window
tiff A TIFF bitmap device
X11 A graphics window on a system running X-Windows (unix some OSX)
Table 4.2: Graphics devices for output of gures
Biological Data Analysis Using R
52 CHAPTER 4. SUMMARY STATISTICS
When you type the command plot () a graphics window pops up showing you the image of
the gure. What is happening here is that R is looking for the default graphics device
and if you have not specied one, then the default value of show it to the user as a
window is use.
Creating The Plot And Saving To File: This is the method that I used for all the gures in
this text. I rst created the gure to look the way that I wanted and then I had R copy
the gure to a le. You should be aware that when you copy the image, it will only copy
the ACTIVE graphics device. If you have more than one graphics window open, only one of
them will say ACTIVE in the window title. Be careful of this or you could be copying the
wrong gure.
Once you have the graphic the way you like, you can use the dev.copy() command to copy
the current graphics device to a le. For this book, I have been saving all the images as
JPEG les so I pass the function the device=jpeg option and then specify the name of the
le. If you want to save yourself some heartache down the road, use meaningful names
for the graphics you create. You can quickly get a lot of different plots that you may
want to go through at some time in the future and it sure helps to have them named
nicely.
> hi st ( rpoi s (1000,2) , xlab="Counts" , ylab="Frequency" ,main="" , col =topo . col ors ( 8) )
> dev . copy ( device=jpeg , f i l e ="ColoredHistogramOfPoissonDistribution.jpeg")
jpeg
3
> dev . of f ( )
X11cairo
2
Once the dev.copy() function is nished, you must call the dev.off () function to tell R that you
are nished copying things to that particular le and you no longer want to keep it open
and ready for subsequent graphing. The output after the dev.off () command shows which
graphics device is now active and what kind of device it is (in general, you can ignore
this). The image produced from this plot is shown in Figure 4.6
I also passed the plot command the optional col=topo.colors(8). The function topo.colors(x) re-
turns x evenly spaced colors from a palette that is used for plotting topo maps. There are
other default palettes in R you can use (see ?topo.colors for a list) in coloring parts of your
gures. By default, I new that the hist () function would return 8 bins of data from the
rpois(1000,2) distribution (I plotted it rst and counted) so I added 8 evenly spaced colors to
the plot just to make it look a bit more cheesy.
Plotting Directly To A File: Plotting to a graph window and copying it to a le is not
necessarily the only way you can get your graphics saved. You could just write them
directly to a le using one of the graphics devices listed in Table 4.2 without looking at it
in a window. I nd this less appealing since I would like to see what I am plotting before
saving it, but if you are chugging through lots of data and creating hundreds of images,
perhaps you would be better served to make the plots directly and view them later. At
any rate, here is how it is done.
jpeg ( )
pl ot ( rnorm(1000) , xlab="index" , ylab="value" , bty="n")
dev . of f ( )
Biological Data Analysis Using R
4.1. DISTRIBUTIONS 53
Figure 4.6: Image of colored Poisson distribution that was copied from the graphics device to a
jpeg le.
and R will open the a jpeg() graphics device. This device is generally a le in the local di-
rectory that is named RPlotXXX.jpeg (where the XXX values are incremental numbers such
as 001, 002, . . .). Then when you call the plot () function it sends the plotting commands
to the image itself in the le.
1
You can add as many plotting commands as you like
and it will continue to send them to the le you specied. When you are done, you can
nalize the image by calling dev.off () to turn of the graphics device. To change the default
incremental numbering of the les, you can pass a le name to the jpeg() function (or any
of the other ones) as we did in the previous section using dev.copy().
4.1.2 What Probability?
The outcome of a statistical analysis is the estimation of a particular test statistic. For
example, when you calculate a
2
statistic, you need to look up a the probability that a
value as large or larger than the observed one is expected to occur. In 4.1.1 we deter-
mined how to calculate the cutoff value from a particular distribution given a specied
1
Actually it keeps them in a buffer and not in the le directly.
Biological Data Analysis Using R
54 CHAPTER 4. SUMMARY STATISTICS
Type I error rate (the value). Here we are interested in not asking if our calculated
value exceeds some particular cutoff, rather we are interested in understanding what
the probability of observing a value as large or larger than the one we see.
In keeping with the current examples from the
2
statistic, we can determine the prob-
ability associated with a particular estimation of
2
Calc
by using the distribution function
pchisq(). The arguments to pchisq() are almost identical to those for the qchisq() function
discussed in 4.1.1 with the exception that we do not pass it the 1 as the rst param-
eter, rather we pass it the estimated
2
Calc
value and it will return the answer in terms of
P[X x]. For example:
> chiCritAt0.05 < qchisq( 0. 95 , 1)
> pchisq ( chiCritAt0.05, 1 )
[ 1] 0.95
> pchisq ( 7.23, 3)
[ 1] 0.9350828
The functions qchisq() and pchisq() give us the opposite answers from each other with one
telling us what the critical value (or P[X <= x]), and the other takes a value for
2
and
tells us what the cumulative area under the curve up to and including that point.
4.2 Random Number Generation
There are often times when you need to generate some random numbers (playing poker,
picking lottery numbers, etc.). Random numbers can be drawn from any of the distri-
butions that are in R using the rdistribution function. For example, to draw a random
number from a normal distribution (N(, )) you would call the rnorm(x,\mu,\sigma) function.
The parameters and signify the mean and standard deviation of the distribution from
which you are drawing. An example of how this inuences the outcome, check out Figure
4.2.
There are a large number of random number distributions that you can run across.
Below are some commonly encountered ones:
Normal The normal distribution has a density function of P(x|, ) =
1

2
e

(x)
2
2
2
.
Exponential The exponential density has a continuous density function of P(x|) =
1 e
x
.
Poisson The Poisson distribution is a discrete distribution whose density function is
P(k|) =
e

k
k!
.
Later in the Exercises you will get to use some of these distribution.
Histograms
A histogram is a graphical display of data that has been tallied into bins (e.g., specic
buckets). How you dene the bucket locations and sizes are up to you. You can specify
that there should be a specic number of buckets and R will make them equal sized, or
Biological Data Analysis Using R
4.2. RANDOM NUMBER GENERATION 55
Figure 4.7: Examples of the densities of two normal distributions; the red one is drawn from a
random normal distribution with default values of = 0 and = 1 and another in blue that has
= = 5.
you can dene ranges yourself. The function signature for the hist () function by typing
?hist in R :
hi st ( x, breaks = "Sturges" ,
f req = NULL, probabi l i ty = ! freq ,
include . lowest = TRUE, ri ght = TRUE,
density = NULL, angle = 45, col = NULL, border = NULL,
main = paste ( "Histogram of" , xname) ,
xlim = range ( breaks ) , ylim = NULL,
xlab = xname, ylab ,
axes = TRUE, pl ot = TRUE, l abel s = FALSE,
nclass = NULL, . . . )
There are several things we should notice about this function signature. First, this is the
rst time that weve looked into a particular function and seen all the options. You can
see that several of the parameters are given what we call default values (e.g., the =VALUE
portions). That way if we do not provide a particular value for a parameter such as main,
it will ll it in for you.
Biological Data Analysis Using R
56 CHAPTER 4. SUMMARY STATISTICS
The rst thing that you typically want to change in a graphic is the default values for
the axis labels and the title of the graph. It is not commonly accepted practice to provide
titles on graphs for most publication-quality graphics, but some times it is helpful when
you are putting together a talk or just analyzing the data and making graphics for your
own interpretation. To change the default values of the axis labels and set an empty title
you would do the following (shown in Figure 4.8):
> hi st ( rnorm( 100) , xlab="My Defined Bin Categories" , ylab="Frequency" , main="")
Figure 4.8: Histogram with labels and main title changed.
Again, I am using the function rnorm() to generate the data from a random normal distri-
bution here. It is perfectly OK to give empty values to things like titles and such.
Density Plots
A density plot is one where the probability density is calculated and turned into a line
across the domain rather than a histogram. Here I will combine the histogram and
density plots to show how to overlay two graphs on the same values.
> data < rpoi s ( lambda=5,n=1000)
> den < density ( data )
> den
Cal l :
density . def aul t ( x = data )
Data: y (1000 obs . ) ; Bandwidth bw = 0.5061
x y
Min. :1.518 Min. :3.567e05
1st Qu. : 2.491 1st Qu.:8.145e03
Biological Data Analysis Using R
4.2. RANDOM NUMBER GENERATION 57
Median : 6.500 Median :3.973e02
Mean : 6.500 Mean :6.229e02
3rd Qu.:10.509 3rd Qu.:1.219e01
Max. :14.518 Max. :1.689e01
> yrange < range ( den$y )
> xrange < range ( den$x )
> hi st ( data , ylim=yrange , xlim=xrange , xlab="Value of Random Poisson" ,
+ ylab="Frequency" ,main="" , probabi l i ty=T, bty="n")
> par ( new=T)
> pl ot ( den, col ="red" , lwd=2, xlab="" , ylab="" ,main="" , bty="n")
Figure 4.9: Histogram of 1000 random numbers drawn from a Poisson distribution with the
parameter set to 5. The red line indicates the density of the values.
There are some things to point out with this plot.
1. I save the values of data as a variable because I needed to plot the same set of
random variables as a histogram and as a density plot. Had I not saved them, I
would be using a different collection of random numbers for each plot and they
wouldnt match.
2. I used the function density() to calculate the probability density function for the values
of data. The density() function has two components, an x variable and a y variable.
The the probability density is calculated as a probability rather than as a frequency
count (as the .
Biological Data Analysis Using R
58 CHAPTER 4. SUMMARY STATISTICS
4.3 Descriptive Statistics
Descriptive statistics are valuable tools in understanding particular patterns in your
data. For the purposes of this section, we will assume that your the experiments that
are producing your data yield one of two different data types. First, observations from
your data could be considered random variables; a measurement that produces a real
number. Examples of random variables may be body size, dissolved oxygen, available
light, etc. A collection of random variables will be denoted as X with elements x
i
; i =
1 . . . N (e.g., indexing across all N individual observations). The other kind of data we
will be examining here are categorical data. Your observations are grouped into distinct
categories and consist of relative counts of each category. Examples of this include
stage-dependent demographic tallies, gender of your study organisms, some types of
genetic data, disease prevalence, etc. Categorical data will be denoted as Y , consisting
of K categories and the number of counts observed in each category will be referred to
as y
i
; i = 1 . . . K.
There are two general properties of random variables that we will spend a little time
discussing because they form the basis of how we examine our data. First, the mean
of a random variable, usually denoted by the symbol is a measure of the central
tendency of your variable (a center of gravity, so to speak). We are all familiar with
the concept of mean, but in a general sense, the mean is just one of several moments of
a distribution and now we turn to this particular moment and then discuss some of the
higher moments.
4.3.1 Moments
There are several properties of random variables that we may be interested in estimating.
Notice that here I used the term estimate rather than compute, this is on purpose. We
will be making estimates of real parameters of the data and we do so because in most
cases we do not have all the data at our disposal. Rather, we have created a sample of
our data from which we make inferences. To get all the data, we would have to sample
EVERY single instance out there and in most cases this is not possible.
There are two common properties that you will probably recognize immediately (I hope)
and use all the time. These are the mean and variance of the data and are estimated in R
using the functions: mean() and var(). Figure 4.10 shows what is being measured by these
estimators. This gure was created using the density() function from rnorm(1000000).
The mean, shown by the dashed line and the symbol is located at the center of gravity
of the data. In R, you can calculate the mean of the data by using the function mean().
The image also shows the standard deviation (which is the square root of the variance
=

2
) as indicated by the dotted line. R has a function for both the variance var(), and
the standard deviation sd().
There are two more measures of distributions that we should discuss while we are here.
2
These are the skew and kurtosis of the distribution. In R these functions are not loaded
2
Actually all four of these measures are known as the rst four moments of the distribution. The rst for
moments,
k
; k = 1 . . . 4 can be calculated by
k
= E[(X )
k
].
Biological Data Analysis Using R
4.3. DESCRIPTIVE STATISTICS 59
Figure 4.10: Example locations for rst two moments of a Normal (N(0, 1)) distribution.
into memory by default and we must load the moments library to gain access to them. To
load these libraries type:
> l i brary ( moments)
If R gives you a warning, this means that the moments library is not installed by default.
In this case, see Appendix B for instructions on how to add libraries to your installation
of R.
The skew of a distribution is a measure of how pushed-over the main lump of the
distribution (again not a very statistical denition here). Distributions can either have a
positive or negative skew, compare the images in Figure 4.11
A distribution is said to have a negative skew if the direction of the longer tail is to
the left. In these cases the mean < median < mode. Conversely, a distribution has a
positive skew if the tail is on the right and the mean > median > mode. Distributions
where these measures are equal is said to not have any skew. Skew is estimated in R
using the function skewness()
The kurtosis of a distribution is a measure of the peakedness of a distribution. This
Biological Data Analysis Using R
60 CHAPTER 4. SUMMARY STATISTICS
Figure 4.11: Negative (left) and positive (right) distributions. In both of these examples the dotted
line connects the mode of the distribution (the top peak) to the mean (on the x axis). The direction
of this lean determines if the distribution has a negative (left) or positive (right) skew.
term comes from the Greek word kurtos that means bulging. A simple example of how
kurtosis looks is found in Figure 4.12 with three different distributions (the normal,
logistic, and uniform), each with a different level of kurtosis.
In general, the function for kurtosis is:
K =

4

4
3
The correction factor (the - 3 part of the equation is a normalizing constant that allows
the kurtosis of a normal distribution to be equal to zero. Below are the raw data and the
kurtosis estimates used in producing Figure 4.12.
> normData < rnorm(100000)
> l ogi sti cData < r l ogi s (100000)
> unifData < runi f (100000)
> kurtosis ( normData) 3
[ 1] 0.02320046
> kurtosis ( l ogi sti cData ) 3
[ 1] 1.219505
> kurtosis ( unifData ) 3
[ 1] 1.197009
The discrepancy here in the estimates showing the normal distribution not quite equal to
zero is because the data were created by drawing randomnumbers rather then specifying
the distribution directly. One benet of the - 3 correction factor is that it allows you to
quickly tell the different types of kurtosis by looking at the value of the estimate. In
general, the following types of kurtosis are available:
Platykurtic Curves that have negative excess kurtosis (e.g., the kurtosis()3 < 0).
Biological Data Analysis Using R
4.3. DESCRIPTIVE STATISTICS 61
Figure 4.12: Three distributions )exponential, normal, and logistic) showing different levels of
kurtosis.
Mesokurtic Curves that do not have excess kurtosis (e.g., the kurtosis()3 = 0).
Leptokurtic Curves that have positive excess kurtosis (e.g., the kurtosis()3 > 0).
The last summary statistic we will cover here is the range(), which returns a two-item
vector containing the minimum and maximum values. In fact, the range() function calls
the min() and max() directly. There is little to discuss about this particular set of func-
tions...
Creating a matrix of Plots
It is often desireable to create more than one plot on a graphic but not overlayed on
top of each other as was explained in Section 4.1.1. To do this, we need to adjust one
of the graphics properties using the function par(). The property we need to change is
mfrow=c(nr,nc). This will create a matrix of plots that has nr rows and nc columns.
An example of creating a matrix of plots is given in the code below and depicted in Figure
4.13.
Biological Data Analysis Using R
62 CHAPTER 4. SUMMARY STATISTICS
Figure 4.13: Matrix of four plots created from random numbers sampled from the normal, pois-
son, exponential, and the logistic distributions.
> par ( mfrow=c ( 2 , 2) )
> hi st ( rnorm(100000))
> hi st ( rpoi s (100000,1))
> hi st ( rexp(100000))
> hi st ( r l ogi s (100000))
Subsequent calls to plotting functions will reuse this graphic gure and replot the
graphs in the nr x nc matrix. This graphic window will have the nr x nc matrix of plots
until it is either closed or you change the mfrow property to something else.
4.3.2 Non-Parametric Parameters
Non-parametric statistics are generally concerned with the analysis of data that does
not make assumptions about the underlying statistical distributions. There are several
commonly known non-parametric statistics such as the Binomial Test, Goodness of Fit,
the Mann-Whitney Test, and the Kruskal-Wallis test. In this section, we will explore
some of the methods that R can use to describe data without assuming an underlying
Biological Data Analysis Using R
4.4. RELATIONSHIPS BETWEEN PAIRS OF VARIABLES 63
distribution.
The rst summary statistic outline here will be the quantile. While you have probably not
heard of this particular descriptive statistic, you most likely will have run across terms
such as a median, quartile, or percentile. All of these are particular kinds of quantiles
that will be obvious when we consider the formal denition of a quantile.
Quantile A p
th
quantile is the value x
p
that when considering the data (X) the probability
P(X < x
p
) p and the probability P(X > x
p
) = 1 p.
While this may be statsy, it generally says that the 50
th
quantile is the the value x
50
in
the distribution where 50% of the data is less than x
50
and 50% is greater than x
50
. Thus
far, you have probably call this the median (and R has a median() function if you like to
call it that). More generally though, we can consider the 95
th
quantile analogous to what
we were discussing in Section 4.1.1 when we were trying to gure out critical regions
of the
2
distribution. The main distinction here is in Section 4.1.1 we implicitly used
the known distributional form of the
2
function to nd the critical value whereas in
non-parametric approaches, we typically apply the approach of putting everything into
a vector, sorting it, and counting to where quantile is located in the list. As a result, the
50
th
quantile (or median) can be considered a measure of central tendency of the sorted
data.
Quantiles can also be used to look at the dispersion of data. In parametric statistics
we discussed parameters such as the variance and standard deviation that dene the
dispersion of values around the mean. The notion of Quantiles can be used in a similar
way. The values of x that give the upper and lower quartiles (e.g., the 25
th
and 75
th
quantiles) provide a range of the data X where the inner 50% of the values lie. These
are often called the inner quartiles of the data. To illustrate the use of the quantile
function, consider the data in Figure 4.14 consisting of 1000 numbers drawn from a
Poisson random distribution with a centrality parameter k = 5.
The quantile() function in R by default provides the 0
th
quantile (e.g., the minimum), the
25
th
quantile, the 50
th
quantile (the median), the 75
th
quantile, and the 100
th
quantile
(e.g., the maximum). For the data that produced the histogram in 4.14, the quantiles
are:
> x < rpoi s (1000,5)
> quantile ( x )
0% 25% 50% 75% 100%
0 3 5 6 12
showing that the center of dispersion is 5 and the inner quartile ranges from 3 6.
4.4 Relationships Between Pairs of Variables
There is often times when we are interested in knowing about the simultaneous changes
in two or more variables. Individually, we can estimate the mean, variance, skew, kur-
tosis, and various ranges but this does not tell us about how the variables interact
together. For this we need to look at measures that explain the relationship between
variables.
Biological Data Analysis Using R
64 CHAPTER 4. SUMMARY STATISTICS
Figure 4.14: Distribution of random number drawn from rpois(1000,5).
4.4.1 Covariance & Correlation
The covariance of two variable is dened as:
c
ij
= E[(X
X
)(Y
Y
)]
and measures the degree to which one variable X changes as another Y changes. Co-
variance estimates may be positive or negative as long as the two variables are not the
same, in which case it is a variance and there is no such thing as a negative variance.
Two variables that have a covariance equal to zero are said to be uncorrelated (although
if you dont know what a correlation is this moniker is kinda sucky).
In R the covariance between two vectors of values is estimated by the function cov().
Needless to say, the length of the two variables must be the same or R will rightly com-
plain.
> X < c(1,34,5,23,6,43,56,28,33,7)
> Y < runi f (10,1,100)
> Y
Biological Data Analysis Using R
4.4. RELATIONSHIPS BETWEEN PAIRS OF VARIABLES 65
[ 1] 90.112843 47.236585 17.148708 3.861546 54.871332 57.234582 8.072745
[ 8] 6.000811 84.546069 17.960688
> pl ot ( X, Y)
> cov ( X, Y)
[ 1] 2231.952
Figure 4.15: Scatter plot of some semi-random points.
So here I just pounded on my numeric keypad and made up the numbers for X (not
quite random but pretty good) and then had R make some numbers for Y by drawing
from a uniform distribution runif() selecting 10 values in the range 1 100. You can see
that the values that I used produced a smattering of points (Figure 4.15 )
4.4.2 Tests For Correlation
There are parametric and non-parametric methods for looking at the relationship among
pairs of variables. In general, all correlations between two random variables (X, Y )
should have the following characteristics:
The value of a correlation is strictly bound on the interval [1, 1].
Biological Data Analysis Using R
66 CHAPTER 4. SUMMARY STATISTICS
If larger values of X tend to be associated with larger values of Y then the cor-
relation should approach +1 as the association becomes stronger. We call this a
positive correlation.
If smaller values of X tend to be associated with larger values of Y then the cor-
relation should approach 1 as the association becomes stronger. We call this a
negative correlation.
If there is no general relation between the variables X and Y then the correlation
statistic should approach 0. We call this a relationship where the variables are
uncorrelated.
The most commonly used measure of correlation is Pearsons product moment correla-
tion, r, that is calculated as:
r =

N
i=1
(X
i
x)(Y
i
y)

N
i=1
(X
i
x)

N
i=1
(Y
i
y)
(4.1)
where the x and y values are the mean of the N sampled variables in X and Y .
Figure 4.16: Example plot of two variables used to test correlations.
Biological Data Analysis Using R
4.4. RELATIONSHIPS BETWEEN PAIRS OF VARIABLES 67
In R the test for correlation is performed with the cor.test () function. To demonstrate, we
will use the following data shown in Figure 4.16:
> X < 1:20
> Y < c(17, 7, 12, 12, 4, 11, 10, 2, 35, 31, 34, 49, 27, 33, 45, 32, 36, 38, 58, 44)
> cor . t est ( X, Y)
Pearson productmoment correl ati on
data : X and Y
t = 7.3194, df = 18, pvalue = 8.489e07
al t ernat i ve hypothesis : true correl ati on i s not equal to 0
95 percent confidence i nt erval :
0.6848344 0.9456427
sample estimates :
cor
0.8651642
The correlation between these two variables is r = 0.865, which is both large and positive
as expected by looking at the graph. By default when you use cor.test () , it will use the
Pearson product moment approach. There are two additional approaches for estimating
correlation, approaches developed by Spearman and Kendal but these two are consid-
ered non-parametric methods based upon ranks rather than that shown in Eqn. 4.1
and will be left until 5.2.1 when we can fully discuss how it works. The output also
includes a signicance test and a display of the 95% condence intervals which are very
useful.
Biological Data Analysis Using R
68 CHAPTER 4. SUMMARY STATISTICS
4.5 Useful Functions
The following functions were introduced in this chapter and you will be required to use
them for the exercises. To get more information on any of these functions, use the R
help system.
dchisq(x,df) Returns the density of the
2
distribution with df degrees of freedom.
df(x,df1,df2) Returns the density of the F distribution with df1 and df2 degrees of
freedom.
dnorm(x) Returns the density of a normal distribution at x.
mean() Calculates the mean of the values in x.
pchisq(x,df) Returns the distribution of the
2
distribution with df degrees of free-
dom.
pf(x,df1,df2) Returns the distribution of the F distribution with df1 and df2 degrees
of freedom.
plot(x) This is the main wrapper function that creates a graphical display of the
variable(s) that you pass to it. Depending upon the variables passed, it will create
different types of plots.
pnorm(x) Returns the distribution of a normal distribution at x.
qchisq(x,df) Returns the quantile of the
2
distribution with df degrees of freedom.
qf(x,df1,df2) Returns the quantile of the F distribution with df1 and df2 degrees of
freedom.
qnorm(x) Returns the quantile of a normal distribution at x.
rchisq(x,df) Returns x random numbers from the
2
distribution with df degrees of
freedom.
rf(x,df1,df2) Returns x random numbers from the F distribution with df1 and df2
degrees of freedom.
rnorm(x) Returns x random numbers from the normal distribution.
sd(x) Returns the sample standard deviation of data in x.
table(f) This function takes the list of levels in the factor f and makes a table from
it.
var(x) Estimates the sample variance, s
2
, from the variables in x.
Biological Data Analysis Using R
4.6. EXERCISES 69
4.6 Exercises
The following exercises are meant to help you understand the items presented in this
Chapter.
1. What are the critical values for a
2
distribution with df = 8 if you are assuming that
= [0.2, 0.1, 0.01, 0.001]?
2. Create a scatter plot using the variables x<rnorm(10) and y<rpois(10,1). Label the axes
Jaw Size and Number of Kids.
3. For the probabilities p = seq(0.1,0.9,by=.1) create a graph that has a red line representing
the quantile function for the Poisson distribution (qpois with = 1) and a blue one
representing the quantile function for the
2
distribution (qchisq with df = 1). Make
sure to have your axes labeled and drawn properly. Save the image and include it in
your answer.
4. In a Platykurtic distribution what is the relationship between the mean, mode, and
median?
5. Create a histogram of 1000 random numbers drawn from the F-distribution with
parameters df1 = 1 & df2 = 10. On this plot, overlay the density using the density
function. Label the axes appropriately.
6. What is the inner-quartile of the data x <rnorm(200,3)?
7. Is the data from the command x <rf(1000,1,10), lepto, meso, or platykurtic? How do you
know?
8. Explain what is happening with the command data <LETTERS[ rpois(23, 2 ) ]. Create a new
variable that is a table of the results of this command, show me the table, and show
how you would access the B element in the table.
9. What is the range of possible values you can get for a Pearsons Product-Moment
Correlation?
10. There is a data set named HWCorrelationData.csv in the folder. Load this data into R ,
plot it an appropriate graphic, and then test the hypothesis H
O
:Height is independent
of Weight.
Biological Data Analysis Using R
70 CHAPTER 4. SUMMARY STATISTICS
Biological Data Analysis Using R
Chapter 5
Contingency Tables
In this chapter we will examine non-parametric methodologies that are available for
the analysis of random variables. It is not uncommon in Biology to encounter the notion
that non-parametric approaches are only to be used with categorical (e.g., nominal) data.
However, non-parametric analyses are just as applicable to normal ordinal and interval
data that we commonly come into contact with and in this Chapter we will go over a few
examples of how you can use general non-parametric statistical approaches in your own
research.
In this Chapter you will learn the following skills:
Non-parametric analysis of data single categorical data set (x
1
, x
2
, . . . , x
N
) using a

2
test.
Non-parametric analysis of paired data ( (x
1
, y
1
), (x
2
, y
2
), . . . , (x
N
, y
N
)) using the Fisher
Exact for small data and the general
2
test for large data sets.
Non-parametric analysis of several random samples using the Kruskal-Wallis test.
For most of the exercises in this chapter you will need to load the stats library by issuing
the command: library(stats).
5.1 One Random Sample
For this section, we will assume that your data consist of N observations made on a
single variable, X = [x
1
, x
2
, . . . , x
N
].
5.1.1 Goodness of Fit
The
2
test for goodness of t is the typical
2
test that we have all had a million times
as an undergraduate and a graduate student. The data for this test consists of N obser-
vations that can be categorized into K discrete Categories. In R we will use the factor
data type (see 2.4.10 for more on the factor type).
71
72 CHAPTER 5. CONTINGENCY TABLES
The assumptions of this test are:
1. All the observations are selected randomly.
2. You can assign an observation to one of the K categories without error.
The test statistic for this analysis is the calculated
2
Calc
which is:

2
Calc
=
K

i=1
(O
i
E
i
)
2
E
i
(5.1)
(5.2)
The underlying distribution of
2
Calc
will be approximated using the
2
-distribution with
K 1 degrees of freedom. From the discussion of this distribution and its depiction in
Figure 4.2, it is large values of
2
Calc
that will lead to the rejection of the null hypothesis,
H
O
.
Example Problem: Assume that we have captured a sample of the Marbled Salamander,
Ambystoma opacum, from the Rice Center for Environmental Studies (a eld station for
Virginia Commonwealth University). On each of these individuals we have classied
their marbling pattern as either Little White (N
A
= 24), Moderate White Marbling (N
B
=
47), and Mostly White (N
C
= 29). A separate crossing experiment has suggest that the
marbling on an individual may be under the control of a limited number of genetic loci
and has predicted that the frequency of these types would be 1 : 2 : 1 in populations at
equilibrium. Do the proposed mechanisms predict a distribution of phenotypes that you
sampled from the wild? To test the hypothesis, H
O
:Phenotypes occur at a ratio of 1 : 2 : 1
in R we would:
> Phenotypes < as . f actor ( c ( rep ( "Little White" , 24) ,
+ rep ( "Marbled" , 47) , rep ( "Mostly White" , 29) ) )
> p < c ( 1, 2, 1)
> p < p / sum( p) # makes p a vector of probabi l i t i es
> tabl e ( Phenotypes )
Phenotypes
Li t t l e White Marbled Mostly White
24 47 29
> chisq . t est ( tabl e ( Phenotypes ) , p = p )
Chisquared t est f or given probabi l i t i es
data : tabl e ( Phenotypes )
Xsquared = 0.86, df = 2, pvalue = 0.6505
So here, the observed and expected values were relatively close to each other producing
a
2
Calc
(in R called X-squared) of 0.86, which with df = 2 has a P-value of 0.6505. Not
something that would be considered rare. As a result, we fail to reject H
O
that the ratio
of phenotypes is 1 : 2 : 1.
Here the thing that was passed to the chisq.test function was an object of class table. This
is only one way that you can pass data to to the chisq.test function. See ?chisq.test for more
information on other ways to pass your data to this function.
Biological Data Analysis Using R
5.1. ONE RANDOM SAMPLE 73
5.1.2 Binomial Test
The binomial test evaluates the support for the probability (p) that an observation was
categorized into one of two groups. The following assumptions are inherent in the bino-
mial test:
1. Each observation has the ability to be characterized as either Category A or Cate-
gory B and the probably of assigning to A is denoted as p (and B as 1 p).
2. Each of the N observations are mutually independent.
The binomial test tests to see if the number of items you have classied as Category A
is rare given a specied probability, p. The test itself is performed using the binom.test()
function. In the example below, I am considering the situation where a coin was ipped
20 times and was found to have shown Heads only six times. The hypothesis is: H
O
:
p = 0.5. The function itself need a few pieces of data; the number of times Category A
was observed (as x), the total number of trials (as n), and the hypothesized probability p.
Call it with these data would be done as:
> binom. t est ( x=6, n=20, p=0.5 )
Exact binomial t est
data : 6 and 20
number of successes = 6, number of t r i al s = 20, pvalue = 0.1153
al t ernat i ve hypothesis : true probabi l i ty of success i s not equal to 0.5
95 percent confidence i nt erval :
0.1189316 0.5427892
sample estimates :
probabi l i ty of success
0.3
These results suggest that even with only 6 observed Heads in 20 ips, we cannot reject
H
O
that it is a fair coin. However, the 95% condence intervals show that there is a large
range of values we cannot reject...
5.1.3 General Contingency Tables
For this next application of a contingency tables we will focus on data describing the
diversity of students in the College of Humanities & Sciences at Virginia Commonwealth
University. These data are reported by all public institutions and can be found for VCU at
the webpage http://www.vcu.edu/cie/analysis/reports/sets.html and are summarized
in Table 5.1.
In general, we are going to create a contingency table that has the general form:
Col 1 Col 2 Col 3 Col c Totals
Row 1 O
11
O
12
O
13
O
1c
R
1
Row 2 O
21
O
22
O
23
O
2c
R
2
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Row r O
r1
O
r2
O
r3
O
rc
R
r
Totals C
1
C
2
C
3
C
c
N
Biological Data Analysis Using R
74 CHAPTER 5. CONTINGENCY TABLES
with r rows of data and c columns. Each of the entries in the rxc contingency table (the
O
ij
values) are counts of the number of observations that were classied as belonging
to the category in the i
th
row and the j
th
column. Above, when we looked at the
2
test, it was a smaller version of this table and the test statistic for analyses in general
contingency tables are the same as above:

2
Calc
=
r

i=1
c

j=1
(O
ij
E
ij
)
2
E
ij
The only distinction here is that our expected values are based upon row and column
totals such that:
E
ij
=
R
i
C
j
N
where R
i
and C
j
are the respective row and column total.
There are two specic assumptions that are required to conduct a general contingency
table test such as this:
1. The sample of N samples are drawn randomly from the larger population.
2. Each observation can be classied into exactly one of the possible r and c categories
according to single and independent criteria (e.g., there is no correlation between
the row and column variables).
Biological Data Analysis Using R
5
.
1
.
O
N
E
R
A
N
D
O
M
S
A
M
P
L
E
7
5
Table 5.1: Diversity of enrolled undergraduate students at Virginia Commonwealth University in the College of Hu-
manities & Sciences between the academic years 1998-2008 as reported by the Center for Institutional Effectiveness
(http://www.vcu.edu/cie/analysis/reports/sets.html).
Group 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008
Non-resident Aliens 186 158 188 208 206 235 272 375 512 577 673
Black non-Hispanic 2985 3094 3282 3332 3387 3456 3633 3797 3983 4158 4193
American Indian or Alaskan Native 91 80 83 86 90 113 109 116 124 131 131
Asian or Pacic Islander 1103 1139 1132 1175 1231 1437 1632 1764 1970 2148 2330
Hispanic 279 305 362 400 449 521 559 623 709 761 822
White, non-Hispanic 8688 8586 9013 9373 9916 10077 10757 11088 11180 11170 11202
Race/ethnicity unknown 0 188 208 279 387 665 849 928 1019 1287 1642
Total 13332 13550 14268 14853 15666 16504 17811 18691 19497 20232 20993
B
i
o
l
o
g
i
c
a
l
D
a
t
a
A
n
a
l
y
s
i
s
U
s
i
n
g
R
76 CHAPTER 5. CONTINGENCY TABLES
To demonstrate this analysis we will analyze the 1998, 2003 and 2008 enrollment data from
Table 5.1 to see if the diversity of students at VCU has changed over the last decade.
These data are present in a text le named VCUCommonData.csv in the folder for this Chapter.
It is loaded into R with the following commands.
> data < read . tabl e ( "VCUCommonData.csv" , header=T, sep=" " )
> summary( data )
Yr1998 Yr1999 Yr2000 Yr2001
Min. : 0.0 Min. : 80 Min. : 83 Min. : 86.0
1st Qu. : 138.5 1st Qu. : 173 1st Qu. : 198 1st Qu. : 243.5
Median : 279.0 Median : 305 Median : 362 Median : 400.0
Mean :1904.6 Mean :1936 Mean :2038 Mean :2121.9
3rd Qu.:2044.0 3rd Qu.:2116 3rd Qu.:2207 3rd Qu.:2253.5
Max. :8688.0 Max. :8586 Max. :9013 Max. :9373.0
Yr2002 Yr2003 Yr2004 Yr2005
Min. : 90.0 Min. : 113 Min. : 109.0 Min. : 116
1st Qu. : 296.5 1st Qu. : 378 1st Qu. : 415.5 1st Qu. : 499
Median : 449.0 Median : 665 Median : 849.0 Median : 928
Mean :2238.0 Mean : 2358 Mean : 2544.4 Mean : 2670
3rd Qu.:2309.0 3rd Qu. : 2446 3rd Qu. : 2632.5 3rd Qu. : 2780
Max. :9916.0 Max. :10077 Max. :10757.0 Max. :11088
Yr2006 Yr2007 Yr2008
Min. : 124.0 Min. : 131 Min. : 131.0
1st Qu. : 610.5 1st Qu. : 669 1st Qu. : 747.5
Median : 1019.0 Median : 1287 Median : 1642.0
Mean : 2785.3 Mean : 2890 Mean : 2999.0
3rd Qu. : 2976.5 3rd Qu. : 3153 3rd Qu. : 3261.5
Max. :11180.0 Max. :11170 Max. :11202.0
Once the entire data set is loaded into R , we can extract only the values that we are
going to use.
> Obs < as . matrix ( cbind ( data$Yr1998, data$Yr2003, data$Yr2008 ) )
> Obs
[ , 1] [ , 2] [ , 3]
[ 1 , ] 186 235 673
[ 2 , ] 2985 3456 4193
[ 3 , ] 91 113 131
[ 4 , ] 1103 1437 2330
[ 5 , ] 279 521 822
[ 6 , ] 8688 10077 11202
[ 7 , ] 0 665 1642
> colnames ( Obs ) < c ( "1998" ,"2003" ,"2008")
> rownames( Obs ) < c ( "Non-resident Aliens" , "Black non-Hispanic" ,
+ "American Indian or Alaskan Native" , "Asian or Pacific Islander" ,
+ "Hispanic" , "White, non-Hispanic" , "Race/ethnicity unknown")
> Obs
1998 2003 2008
Nonresi dent Aliens 186 235 673
Black nonHispanic 2985 3456 4193
American Indian or Alaskan Native 91 113 131
Asian or Paci f i c Isl ander 1103 1437 2330
Hispanic 279 521 822
White , nonHispanic 8688 10077 11202
Race/ethni ci ty unknown 0 665 1642
With these data we will be specically testing the hypothesis that across years there is
no differences in the relative distributions of self-identied racial and ethnic group.
In some texts, this (7x3) contingency test is called a
2
Test for Independence and in R
is conducted using the chisq.test(). To begin with, we can plot the categories as the barplot
(see 8.2.1 for how to make these plots yourself) as represented in Figure 5.1.
Biological Data Analysis Using R
5.1. ONE RANDOM SAMPLE 77
Figure 5.1: Undergraduate diversity at Virginia Commonwealth University during academic years
1998, 2003, & 2008.
> test1 < chisq . t est ( Obs )
> test1
Pearsons Chisquared t est
data : Obs
Xsquared = 1704.417, df = 12, pvalue < 2.2e16
> summary( test1 )
Length Class Mode
st at i st i c 1 none numeric
parameter 1 none numeric
p. value 1 none numeric
method 1 none character
data .name 1 none character
observed 21 none numeric
expected 21 none numeric
resi dual s 21 none numeric
Notice here that I actually assigned the results of the statistical test to the variable
test1. I did this because there are many reasons why you may be interested in looking
a various aspects of the analysis. By printing the contents of the test itself, we see that
Biological Data Analysis Using R
78 CHAPTER 5. CONTINGENCY TABLES
the calculated statstic
2
Calc
= 1704.417, which with (r 1) (c 1) = 6 2 = 12df produces
a very small Pvalue. If you look back at Figure 4.2, our observed value is way out to
the right with a very small likelihood that that you would get a value this large if it were
not signicant.
As shown using the function summary(test1) shows, the analysis itself returns a list that
has all the components as list items. There are a lot of different reasons why you may be
interested in using various components of the analysis. For example, you may want to
create a table of the observed or expected values, you may need to run this test a large
number of times and store
Caveats
There are some caveats that need to be made with respect to general use of contingency
tables. First, they are very robust as long as you have a moderate amount of samples
in each of the cells. The test statistic we have been using,
2
Calc
with (r 1) (c 1)df is
actually an approximation that is good only with good representation. If the values in the
cells are small then the approximation that we use to nd the Type I error (the value)
is poorly estimated. OK but what is moderate? Here are some general guidelines:
1
1. If any of the E
ij
estimates are less than 1 the approximation will be poor.
2. If more than 20% of the E
ij
values are less than 5 then the approximation will be
poor.
So what do you do if you have some small expected values? First, you can try to col-
lapse some of your row or column categories and recalculate. It really depends upon
your knowledge of the biology of the system if this can be done without making it a
meaningless analysis.
Second, you can try to use Fishers Exact Test. This uses combinatorial theory to esti-
mate the probabilities of the test statistic rather than asymptotic assumptions. This is
an excellent choice but has the problem that since it use combinatorial theory, at some
point you will have to perform an operation like N! which when N > 170 the computer
cannot calculate a number that large. There is also the restriction that product of the
row marginals (the R
i
values in the table) must be strictly less than 2
31
1 but he N < 170
rule is a bit easier to remember.
5.2 Paired Observations
Analyses in this section will be concerned with data that is collected in a pair-wise
fashion (e.g., for each observation, there are two values collected).
1
These guidelines are a bit on the conservative side and you may want to see a text on non-parametric
statistics for a more complete discussion of how far you can stray from these and still not get laughed at.
Biological Data Analysis Using R
5.2. PAIRED OBSERVATIONS 79
5.2.1 Rank Correlation
In 4.4.2 we looked at how you use the cor.test function to get a parametric estimate of the
correlation between two sets of variables. This is possible as well using a non-parametric
approach by adopting a ranking methodology. Non-parametric correlation methods in-
clude Spearmans and Kendals , among others but the interface in R is identical (and
the same as we already saw for the Pearson product moment correlation) so I will only
cover the Spearman approach and leave you to look into the differences.
Spearmans correlation statistic, , is calculated as:
=

N
i=1
R[X
i
]R[Y
i
] N
_
N+1
2
_
2
_

N
i=1
R[X
i
]
2
N
_
N+1
2
_
2
_1
2
_

N
i=1
R[Y
i
]
2
N
_
N+1
2
_
2
_1
2
(5.3)
where the terms R[X
i
] is the rank of the i
th
element in X. These ranks are computed
in comparison to other values in X. For example R[X
i
] = 1 is the smallest value of X,
R[X
i
] = 2 would be the second smallest, etc. So what is begin done here is that we are
replacing the actual values of the variables by the relative ranks.
Using the same data as in 4.4.2 you specify the use of the Spearman approach using
ranks by passing it as an additional option to the cor.test function.
> X < 1:20
> Y < c(17, 7, 12, 12, 4, 11, 10, 2, 35, 31, 34, 49, 27, 33, 45, 32, 36, 38, 58, 44)
> cor . t est ( X, Y, method="spearman")
Spearmans rank correl ati on rho
data : X and Y
S = 198, pvalue < 2.2e16
al t ernat i ve hypothesis : true rho i s not equal to 0
sample estimates :
rho
0.8511278
Notice here that the correlation is signicant although the correlation statistic is a bit
smaller. There is some loss of information by putting the data into ranks rather than
using the raw values.
So why use this instead of the parametric approaches? Well the calculation of Pearsons
r statistic depends upon the bivariate distribution of X and Y . If there is no known
joint distribution for these variables then the density function of r is undened. What
does this mean to you? It means that if your data can be assumed to be normal or then
go ahead and use the Pearson approach. However, if you cannot assume that they are
normal or they you know they are not, then a rank approach may be more appropriate.
For me, I consider the non-parametric approaches as appropriate for all data, whereas
the parametric ones as only good for a subset of the data that we encounter.
Biological Data Analysis Using R
80 CHAPTER 5. CONTINGENCY TABLES
5.2.2 Wilcoxon Test
The Wilcoxon test is also known as the Mann-Whitney test and a ranks based method
analogous to the a paired t-test. This approach tests the null hypothesis that samples
drawn from two different populations are essentially the same (e.g., they are as likely as
samples drawn from one or the other population). Data here are drawn randomly from
two different treatments to see if the application of either produces a signicant shift
in the values of one set of observations.
As was discussed for Spearmans , samples will be ranked in increasing order for this
analysis. If the ranks in sample X tend to be generally larger or smaller than those
observations in Y then we can reject the null hypothesis H
O
: X = Y . In general your
data should look like:
Treatment 1 Treatment 2
X
1
Y
1
X
2
Y
2
. . . . . .
X
n
Y
m
In this analysis, we do not assume that both X and Y have the same number of obser-
vations and in general will consider X to have n observations while Y has m and denote
N = n +m. Samples are lumped together and assigned ranks based upon the combined
N observations. In the case of ties where two or more samples have the exact same
value, it is recommended to assign the average rank to all the tied observations. For-
tunately for us, the internal R code takes care of this for us (and will provide warnings
when appropriate) so we can focus on our tasks and let R focus on the specics.
Assumptions
The Wilcoxon test has the following assumptions:
1. Both sets of samples (the X and Y observations) are drawn randomly form each
population.
2. There is an expected mutual independence between the X and Y values as well.
3. The variables are at least ordinal.
The test statistic for this analysis is the sum of the ranks of the X variables:
W =
n

i=1
R[X
i
]
If the observations in X and Y are drawn from a single population, as stated in the null
hypothesis, then the sum of the ranks of X should be just as large as expected for the
sum of the ranks for Y . If the treatments are producing differences in either X or Y then
the test statistic will be unusually large given N.
Biological Data Analysis Using R
5.2. PAIRED OBSERVATIONS 81
To show how to conduct the Wilcoxon test, I will use the pine germination data that is in
the folder for this Chapter. These data are from my thesis and record the average germi-
nation rates for offspring arrays of Pinus echinata families who were sampled in continu-
ous (CTRL), selectively cut (SEL), and stands where all the trees around P. echinata were
clear-cut (CLR). Here we will use the Wilcoxon to see if there is a signicant difference
in germination rates between the control (CTRL) and clear-cut treatments (CLR). Here is
how to load the data into R and extract just the treatments of interest.
> pineData < read . tabl e ( "PineGerminationData.txt" , header=T)
> summary( pineData )
GERM TRT
Min. :0.0000 CLR :15
1st Qu.:0.1800 CTRL:23
Median :0.3700 SEL :15
Mean :0.3625
3rd Qu.:0.5700
Max. :0.9400
> X < pineData$GERM[ pineData$TRT=="CLR" ]
> Y < pineData$GERM[ pineData$TRT=="CTRL" ]
> length (X)
[ 1] 15
> length ( Y)
[ 1] 23
> X
[ 1] 0.67 0.64 0.94 0.40 0.01 0.45 0.58 0.00 0.80 0.81 0.21 0.36 0.82 0.35 0.41
> Y
[ 1] 0.63 0.29 0.37 0.56 0.19 0.02 0.06 0.07 0.11 0.18 0.03 0.64 0.21 0.00 0.00
[ 16] 0.53 0.00 0.00 0.00 0.00 0.35 0.39 0.37
> mean(X)
[ 1] 0.4966667
> mean( Y)
[ 1] 0.2173913
> range (X)
[ 1] 0.00 0.94
> range ( Y)
[ 1] 0.00 0.64
You can see that there are different numbers of samples in each treatment but that they
have overlapping ranges. To run the Wilcoxon test, use the function wilcox.test and pass it
the two variables.
> wilcox . t est ( X, Y)
Wilcoxon rank sum t est with conti nui ty correcti on
data : X and Y
W = 269.5, pvalue = 0.003835
al t ernat i ve hypothesis : true l ocati on shi f t i s not equal to 0
Warning message:
In wilcox . t est . def aul t ( X, Y) : cannot compute exact pvalue with t i es
According to our test, the data in X and Y appear to be different. The test statistic, W =
269.5 which gives it a P-value of 0.004. There are some error messages that you should
be aware of. Apparently in the data, there were ties and this causes some problems
in calculating the signicance of the parameter. These ties are for families that did not
produce any offspring. From a biological perspective, these are valid responses and you
would have to just live with the fact that ties existed because throwing out all the 0.00
values changes the interpretation of what happened.
Biological Data Analysis Using R
82 CHAPTER 5. CONTINGENCY TABLES
In general, the Wilcoxon test is rather powerful in determining the equality of samples
drawn from two different populations. It is essentially the non-parametric version of the
normal t-test.
2
Situations where you may favor a Wilcoxon approach over the t-test are
when you have non-normal data or data with several outlier points.
5.3 Several Random Samples
The nal section in this chapter is focused on data that is collected from multiple treat-
ments. In the previous discussion of the Wilcoxon test, the data had k = 2 treatments
and it was introduced as a rank based analog of the t-test. Here we will introduce the
Kruskal-Wallis test which allows for the analysis of k > 2 treatments and we could again
consider it a rank-based analog of an analysis of variance (ANOVA) approach.
5.3.1 Kruskal-Wallis Tests
The Kruskal-Wallis test examines the differences among k different treatments using a
rank-based approach similar to that discussed for the Wilcoxon test. In fact, this test is
just an extension of the Wilcoxon test using the same sum or ranks approach.
Data for this test is not assumed to be of equal sizes. Each treatment may have a
different number of observations in it with a total sample size of: N =

k
i=1
n
i
. You
should be able to make a list of your data by treatment such as:
Treatment 1 Treatment 2 Treatment k
X
11
X
21
X
k1
X
12
X
22
X
k2
.
.
.
.
.
.
.
.
.
.
.
.
X
1n
1
X
2n
2
X
kn
k
The test statistic for this test is a
2
approximation with k 1 degrees of freedom
Assumptions
There are several assumptions associated with this test:
1. All samples are randomly drawn from their perspective treatments.
2. Treatments are independent of each other.
3. The observations are at least ordinal in nature.
As an example using this analysis, we will examine the same Pinus echinata data set
that we used to demonstrate the Wilcoxon test. The default method for performing this
analysis looks like kruskal.test(x, g, ...) where the variable x is the raw data and the g one is
another variable that has the groupings. In the code below I separate out the variables
2
Actually if you do a t-test on the ranks you will get the same answer as the Wilcoxon, the approaches
are identical except for how the data are encoded; raw or as ranks.
Biological Data Analysis Using R
5.4. THE FORMULA NOTATION & BOX PLOTS 83
and then pass them to the function with Germination as the response and grouped by the
factor Treatment. I also conduct the analysis and assign it to the variable named germTest
so you can see that this analysis also returns a list of results.
> pineData < read . tabl e ( "PineGerminationData.txt" , header=T)
> GerminationRates < pineData$GERM
> Treatment < as . f actor ( pineData$TRT )
> Treatment
[ 1] CTRL CTRL CTRL CTRL CTRL CTRL CTRL CTRL CTRL CTRL CTRL CTRL CTRL CTRL CTRL
[ 16] CTRL CTRL CTRL CTRL CTRL CTRL CTRL CTRL SEL SEL SEL SEL SEL SEL SEL
[ 31] SEL SEL SEL SEL SEL SEL SEL SEL CLR CLR CLR CLR CLR CLR CLR
[ 46] CLR CLR CLR CLR CLR CLR CLR CLR
Levels : CLR CTRL SEL
> GerminationRates
[ 1] 0.630 0.290 0.370 0.560 0.190 0.020 0.060 0.070 0.110 0.180 0.030 0.640
[ 13] 0.210 0.000 0.000 0.530 0.000 0.000 0.000 0.000 0.350 0.390 0.370 0.580
[ 25] 0.490 0.450 0.380 0.510 0.570 0.240 0.290 0.620 0.520 0.200 0.240 0.615
[ 37] 0.760 0.300 0.670 0.640 0.940 0.400 0.010 0.450 0.580 0.000 0.800 0.810
[ 49] 0.210 0.360 0.820 0.350 0.410
> germTest < kruskal . t est ( GerminationRates , Treatment )
> summary( germTest )
Length Class Mode
st at i st i c 1 none numeric
parameter 1 none numeric
p. value 1 none numeric
method 1 none character
data .name 1 none character
> germTest
KruskalWal l i s rank sum t est
data : GerminationRates and Treatment
KruskalWal l i s chisquared = 12.539, df = 2, pvalue = 0.001893
When looking at the results of the test, we see that the estimated test statistic was
relatively large suggesting that it is unlikely that the three timber extraction treatments
do not differentially inuence the germination percentages.
5.4 The Formula Notation & Box Plots
If you look at the function signature for the kruskal.test (by typing ?kruskal.test into R ), you
can see several alternate ways you can pass your data to it.
kruskal . t est package : stats R Documentation
KruskalWal l i s Rank Sum Test
Description :
Performs a KruskalWal l i s rank sum t est .
Usage:
kruskal . t est ( x, . . . )
## Default S3 method:
kruskal . t est ( x, g , . . . )
## S3 method f or cl ass formula :
kruskal . t est ( formula , data , subset , na. action , . . . )
Biological Data Analysis Using R
84 CHAPTER 5. CONTINGENCY TABLES
When discussing the relationship between the raw germination data and the grouping
variable, I used the statement ...is a function of... This notation is the formula notation
that is indicated in the last option for calling the kruskal.test function. In R you can often
use the formula notation to perform analyses and plots and here we will spend a little
bit of time on how that is done. In Chapter 6 you will use this notation quite a bit when
writing out linear models.
The formula notation in R consists of the response variable (or variables that Ill call
Y ), the predictor variable (or variables which will be denoted as X), and the tilde sign
showing the relationship. For example, a simple function would be denoted as Y X
stating that Y is a function of X. Using the function notation for the kruskal.test would
look like:
> kruskal . t est ( GerminationRates Treatment )
KruskalWal l i s rank sum t est
data : GerminationRates by Treatment
KruskalWal l i s chisquared = 12.539, df = 2, pvalue = 0.001893
Figure 5.2: Boxplot of Pinus echinata germination data partitioned by timber extraction treatment.
Biological Data Analysis Using R
5.4. THE FORMULA NOTATION & BOX PLOTS 85
It is even possible (and perhaps better because we are rather lazy in our typing) to use
the function notation of the variable names within a data.frame without having to make the
other variables (GerminationRates and Treatments). However, when you do this, you will have
to pass an additional parameter to the analysis function to tell it which data to look into
for those variable names. For example, with the pineData data set you can type:
> kruskal . t est ( GERM TRT, data=pineData )
KruskalWal l i s rank sum t est
data : GERM by TRT
KruskalWal l i s chisquared = 12.539, df = 2, pvalue = 0.001893
Another common place to nd the function notation is in plotting. Thus far, we have
called scatter plots by the function plot(x,y). It is just as easy to call the plot as plot(y x)
and you will get the same results if the variable x is a continuous variable. However, if
x is a categorical variable you will not get a normal scatter plot. What you will get is a
box plot as depicted in Figure 5.2 which was created by calling the function
3
:
> pl ot (GERM TRT, data=data , xlab="Treatment" , ylab="GerminationRate")
3
To adjust additional parameters on the box plots see the function bxp which is the actual plotting
function that the plot function is handing the data off to. You can adjust many other components of the plot
including notches, box colors, etc.
Biological Data Analysis Using R
86 CHAPTER 5. CONTINGENCY TABLES
5.5 Useful Functions
The following functions were introduced in this chapter and you will be required to use
them for the exercises. To get more information on any of these functions, use the R
help system.
as.factor(x) Coerces the data in x into a factor data type.
as.matrix(x) Coerces the data in x into a matrix data type if possible.
binom.test(x,n,p) Performs a binomial test to see if observing x occurrances of one
category of data in n trials is consistent with the likelihood of it occuring with a
frequency of p.
c(x,y) The concatinate function that munges all the items together and returns
them as a vector.
cbind(x,y,...) Binds together the data in x, y, etc. by columns.
colnames(x) Access the column names in the item x. This only works for matrices
and data.frames.
cor.test(x) Tests for a signicant (e.g., = 0) correlations.
chisq.test(t) Performs a
2
test on the values in the table t.
kruskal.test(x,g) Performs the Kruskal-Wallis Rank Sum test for the data in x as
partitioned into groups dened by g.
length(x) Returns the length of x.
mean(x) Returns the mean of the items in x.
range(x) Returns a two-element vector containing the minimum and maximum val-
ues in x.
read.table() Reads in a raw data into R .
rownames(x) Access the row names in x. This only works ofr matrices or data.frames.
summary(x) Returns a general summary of the data in x.
table(f) This function takes the list of levels in the factor f and makes a table from
it.
wilcox.test(x,y) Performs the Wilcoxon Rank Sum Test on the variables in x and y.
Biological Data Analysis Using R
5.6. EXERCISES 87
5.6 Exercises
The following exercise are meant to help you understand the items presented in this
Chapter
1. Calculate the relative proportions of each group in the 1999 VCU data and use the
goodness of t approach (as in 5.1.1 to see if the 2008 student class has the same
relative proportions as are predicted by the 1999 class.
2. Compare the enrollment freshmen enrollment in the College of Humanities & Sciences
at VCU (from Table 5.1) during the 2006-2007 academic year for Degree-Seeking Un-
dergraduates to the three Universities listed below. Is the student diversity across
these institutions the same? These data sets are prepared each academic year by
each public institution and can be found by searching for Common Data Sets and
looking at Enrollment & Persistence. Below are the places you can get this informa-
tion for three Universities in our region.
Auburn University https://oira.auburn.edu/cds/2006/sectionb.aspx
University of Virginia http://www.web.virginia.edu/IAAS/data catalog/institutional/cds/current/enrollment.htm
Virginia Tech http://www.ir.vt.edu/common ds 2006.htm
3. Use the wilcoxon.test to see if the germination rates observed in the SEL and CLR treat-
ments are signicantly different. Provide some interpretation of your results.
4. Load the data into R that is found in the le CornOutput.csv (Note: this data is tab-
delimited so you will have to adjust the separator you use in the read.table function),
These data represent the output in numbers of bushels per acre of corn with three dif-
ferent fertilizer treatments. Create a density plot showing the distribution of bushels
yielded by each treatment.
5. Test the equality of the fertilizers in the data loaded from the last question using a
Kruskal-Wallis test. Interpret your results.
6. What are the inner-quartiles of the three fertilizer yields?
7. From a total of N = 15 students in this course, if 14 pass, is the probability of passing
this course equal to p = 0.65?
8. What does the optional parameter rescale.p change in the chisq.test function? Why would
you want to use this option?
9. Assume that you observed phenotypes in the following amounts: n
spots
= 12 individ-
uals with spots, n
silky
= 22 with silky fur, n
Smooth
= 15 smooth coated, and n
aguti
= 8
aguti. Do these data t the hypothesis that the probability of any one of these phe-
notypes is equal?
10. Create a data three variables named First, Second, and Third and assign each of them
the value of runif(3). Now, create a bar plot of these data assuming that the rst entry
in each data set represents Category A, the second Category B, and the third Category C.
Make it look something like Figure 5.1 with the Categories used as the partitioning
variable along the x-axis. Feel free to provide your own colors.
Biological Data Analysis Using R
88 CHAPTER 5. CONTINGENCY TABLES
Biological Data Analysis Using R
Chapter 6
Linear Models
This chapter focuses on the analysis of linear models in R . The term linear model is a
general one that will be used a bit loosely. In general, a linear models is one that can be
written down in the form:
y = x
Some variable, or set of variables, y, are predicted to have a particular relationship with
some predictor variable (or variables) denoted in x. In the simplest case when both x
and y are continuous variables, the analysis is called a regression analysis, if x has
more than one predictor variable then it is called a multiple regression, and if y is binary
it is a logistic regression. However, if the predictor variable is categorical the model
is called an analysis of variance with many variants depending upon the number and
relationship of categorical predictor variables in x. Finally, if predictor variables consist
of categorical and continuous variables then it is called an analysis of covariance. There
are many different ways of introducing these different kinds of analysis but we are going
to focus on the functional form and the kinds of variables that make up the predictor
x.
In this Chapter you will learn the following skills:
Learn to analyze data using a simple regression approach.
Be able to incrementally build a multiple regression model using Type III sums of
squares.
Perform an analysis of variance (ANOVA) analysis for both 1-way and factorial mod-
els.
6.1 The t-test
6.1.1 One-Sample Tests
The rst linear model we will deal with is the t-test. The functional form of this is:
89
90 CHAPTER 6. LINEAR MODELS
y =
where we believe that the observations sampled in y have some particular mean value
and the variation around that mean value is simply the natural variation there is is the
kind of samples we are measuring. The function that performs the one-sample ttest in
R is (not surprisingly) called t.test and has the following options available to it.
t . t est ( x, y = NULL,
al t ernat i ve = c ( "two.sided" , "less" , "greater") ,
mu = 0, paired = FALSE, var . equal = FALSE,
conf . l evel = 0.95, . . . )
For a one-sampled test, we will pass the response variable and a value for the parameter
mu to the function. By default, it will test the null hypothesis H
O
: y = (the mu in
the signature) using a two.sided alternate hypothesis. This means that we can reject
the null if y < and if y > using a

2
rejection region. If you have reason to believe
that the observations are supposed to increase or decrease over some particular value,
something along the lines of say the addition of fertilizer should increase yield, then
you should be using a one-tailed test instead that only examines an -sized region one
end.
In the data below, we are testing the hypothesis that H
O
: y = 15 with the given data.
> Y < c(19,25,14,15,24,17,19,27,29,25)
> test1 < t . t est ( Y,mu=15)
> summary( test1 )
Length Class Mode
st at i st i c 1 none numeric
parameter 1 none numeric
p. value 1 none numeric
conf . i nt 2 none numeric
estimate 1 none numeric
nul l . value 1 none numeric
al t ernat i ve 1 none character
method 1 none character
data .name 1 none character
> pri nt ( test1 )
One Sample tt est
data : Y
t = 3.8523, df = 9, pvalue = 0.003892
al t ernat i ve hypothesis : true mean i s not equal to 15
95 percent confidence i nt erval :
17.64182 25.15818
sample estimates :
mean of x
21.4
You can see that I assigned the results of the analysis to the variable named test1. Just
as in the contingency tables examples (5.1.3 & 5.2.2) the results of an analysis are a
list containing all the parameters that were used to perform the analysis as well as
intermediary materials and results. Of particular mention are the parameters p.value,
conf.int, and statistic. Overall, the analysis found that we can reject the null hypothesis
H
O
: y = 15 with a P-value of 0.004. This is fairly good support for the notion that the
mean of these observations is not equal to 15.
Biological Data Analysis Using R
6.2. REGRESSION WITH A SINGLE VARIABLE 91
6.1.2 Paired Tests
The t-test can also be used in a paired fashion. This analysis consists of two sets of
variables, X and Y that are observations that are taken in such a manner as to think
that the differences between them are negligible. For example, perhaps you think that
parasite load has inuenced the development of young warblers so you measure the
lengths of the primary feathers. Overall the null hypothesis for this is H
O
: X = Y .
Another way to write this hypothesis is: H
O
: (X Y ) = 0, in which case this becomes
identical to the one-sampled test. An example of this in R (with entirely contrived data)
would be:
> X < round( runi f (10,min=12,max=20) )
> Y < round( runi f (10,min=12,max=20) )
> X
[ 1] 12 18 18 13 14 15 15 16 17 19
> Y
[ 1] 14 17 20 13 17 12 16 17 17 15
> t . t est ( X, Y, paired=T)
Paired tt est
data : X and Y
t = 0.1416, df = 9, pvalue = 0.8905
al t ernat i ve hypothesis : true di f f erence in means i s not equal to 0
95 percent confidence i nt erval :
1.697808 1.497808
sample estimates :
mean of the di f f erences
0.1
Notice that since these are paired, they must be taken from the same experimental unit,
which is why we added the paired=T option to the parameters we passed to t.test.
6.2 Regression With A Single Variable
A linear regression seeks to see if the values in the response variable y can be predicted
to change systematically with the predictor variable x. The general form of a regression
model is:
y
ij
=
0
+
1
x
i
+e
j
where the response variable y
ij
is hypothesized to be a function of three independent
components:
1. The intercept,
0
.
2. A slope coefcient,
1
that determines at what rate y changes with changes in x.
3. The error term, e
j
, is the latent variation that every observed value has around the
predicted regression line.
The methods by which the parameters
0
and
1
are estimated are varied. The most
common approach is the least squares approach which tries to nd estimates for these
Biological Data Analysis Using R
92 CHAPTER 6. LINEAR MODELS
two parameters that minimizes the sum of squared error terms (e.g.,

N
i=1
e
i
). In R we
can use the function lm to construct the linear model. Here is an example data set with
the values plotted in Figure 6.1.
Figure 6.1: Plot of single variable regression values.
> X < 1:10
> X
[ 1] 1 2 3 4 5 6 7 8 9 10
> Y < c(19,25,14,15,24,17,19,27,29,25)
> Y
[ 1] 19 25 14 15 24 17 19 27 29 25
> pl ot ( YX, xlab="X" , ylab="Y" , bty="n" , col ="red" ,pch=19, ylim=c( 0 , 30) , xlim=c ( 0 , 10) )
To plot these, I used the functional form (see 5.4 for a discussion of how this works)
with Y X, set the labels, the plot colors, the ranges of the x and yaxes, and the
plot characters with the pch option.
1
By eye-balling the image, do you think there is a
relationship between these variables?
> f i t 1 < lm( YX)
> f i t 1
1
To see all the different characters that you can use as plot symbols type plot(1:25,pch=1:25) and it will
plot each symbol along the x = y line.
Biological Data Analysis Using R
6.2. REGRESSION WITH A SINGLE VARIABLE 93
Cal l :
lm( formula = Y X)
Coef f i ci ent s :
( I ntercept ) X
16.3333 0.9212
I start by assigning the response of the analysis to the variable fit1. Printing the contents
of the analysis shows that the intercept term (the
0
) has been estimated to be 16.333
whereas the slope term (R calls this by the variable name you use for it and above we
called it
1
) as 0.92. So for each increment of X, there is almost a corresponding increase
in Y (OK since the points do kinda point upwards). But is this signicant? You can have
a non-zero estimate for a non-signicant relationship. To see a slightly more detailed
printout of the components in fit1 use the summary function.
> summary( f i t 1 )
Cal l :
lm( formula = Y X)
Residuals :
Min 1Q Median 3Q Max
5.097 4.591 0.600 3.238 6.824
Coef f i ci ent s :
Estimate Std. Error t value Pr( >| t | )
( I ntercept ) 16.3333 3.2258 5.063 0.000973
X 0.9212 0.5199 1.772 0.114348

Si gni f . codes : 0 0.001 0.01 0.05 . 0.1 1

Residual standard error : 4.722 on 8 degrees of freedom
Multiple Rsquared: 0.2819, Adjusted Rsquared: 0.1921
Fst at i st i c : 3.14 on 1 and 8 DF, pvalue : 0.1143
Here we see several components:
1. The formula that was used to call the lm function.
2. A summary of the residuals (the e
ij
terms)
3. The coefcients themselves with standard errors and probabilities.
4. A summary of the test statistic, F, the df, and the probability.
Overall, it does not appear that the regression line is signicant. If you are interested in
printing out a more standard ANOVA table for this model, you can pass the variable fit1
to the anova function and it will print out the more normal results.
> anova( f i t 1 )
Analysis of Variance Table
Response : Y
Df Sum Sq Mean Sq F value Pr(>F)
X 1 70.012 70.012 3.1398 0.1143
Residuals 8 178.388 22.298
This printout is probably more like what you will be putting into your manuscripts.
Again, the trend does not seem to be signicant.
Biological Data Analysis Using R
94 CHAPTER 6. LINEAR MODELS
Plotting the Regression Model onto Your Points
It is possible plot the regression model onto a display of the predictor and response
variables. This can sometimes be helpful when visualizing your data. The abline function
overlays a line on your current plot. To use the abline function on an existing graph does
not require you to call par(new=T) rst as it takes care of that already.
Figure 6.2: Regression model added to plot of points using abline function.
> pl ot ( YX, xlab="X" , ylab="Y" , bty="n" , col ="red" ,pch=19, ylim=c( 0 , 30) , xlim=c ( 0 , 10) )
> abline ( f i t 1 , l t y =2)
In addition to passing a variable that is a regression model (e.g., the class(t1) = "lm"), the
function abline can also be called by passing it raw values for the slope and intercept. This
means you can add an arbitrary line to any plot you like. As shown above, the function
also takes additional parameters that allow you to customize the look of the line. You
may want to revisit Table 4.1 as a reminder.
Biological Data Analysis Using R
6.2. REGRESSION WITH A SINGLE VARIABLE 95
Adding Text To A Graph
While we are customizing this image of our non-signicant regression model, it is prob-
ably a good time to look at the text () function. This function allows you to add arbitrary
text to your plot. The basic call of this function will include the x and y coordinates of
where you want to put the text and the characters string that you will be putting on the
graph.
To illustrate how this is done, we will add the regression formula to the plot. First, we
will determine where in the fit1 variable you can nd the regression coefcients. You
could type out the regression equation yourself and for a one-off image it may be easier
for you to do it this way, but if the data are already embedded in the fit1 variable then it
is a more versatile approach for you to use.
> names( f i t 1 )
[ 1] "coefficients" "residuals" "effects" "rank"
[ 5] "fitted.values" "assign" "qr" "df.residual"
[ 9] "xlevels" "call" "terms" "model"
> f i t 1$coef f i ci ent s
( I ntercept ) X
16.3333333 0.9212121
> f i t 1$coef f i ci ent s [ 1]
( I ntercept )
16.33333
> f i t 1$coef f i ci ent s [ 2]
X
0.9212121
So we can access the values estimated for
0
and
1
using the t1coefcients[1]andt1coefficients[2].
Now we need to make a single string that has the regression equation y =
0
+
1
x. The
text parts, we can write out but the variables should come from fit1. To do this, we
use the paste function. This function takes a list of items and mushes them together
into a single character string More can be found on the paste function and general string
manipulation in Chapter 9.
> formula < paste ( "y = " , f i t 1$coef f i ci ent s [ 1] , " + " , f i t 1$coef f i ci ent s [ 2] , "x")
> formula
[ 1] "y = 16.3333333333333 + 0.921212121212122 x"
> text ( 5 , 12.5 , formula )
6.2.1 Regression Diagnostics
It is possible to attempt to t any model to a set of data. However, just because R will
happily (in most cases) provide you an answer to a model tting, it does not mean that
it is the right model for the data. For example, your data may not be linear, however it
is still possible for you to t a line to non-linear data. R includes some easy methods
that you can use to examine the appropriateness of your model and here we will focus
on some of the built-in diagnostics. These focus on the single specied model and allow
you to make decisions on the appropriateness of your proposed model. Later in ?? we
will cover methods that allow you to determine if one model is better than another for
describing your data.
Biological Data Analysis Using R
96 CHAPTER 6. LINEAR MODELS
Figure 6.3: Regression model with tted line and formula.
One of the rst things you should do when you specify a linear model is look at the
residuals. The residuals are the e
ij
components of the model in the general formula.
These represent the variation that is not explained by your tted line. The things you
are looking for in the residuals are:
1. Systematic changes in the residuals when plotted as a function of the predicted val-
ues. This would indicate that there is something else that is changing the response
variable that you are not taking into consideration.
2. Non-linearity in the residuals when plotted against the predicted values. This would
suggest that perhaps your data are not linear to start with and the tting of a linear
model to it may not be appropriate.
3. Normality of the residuals. These values are expected to be N(0,
2
). If they are not,
it may not be appropriate to be tting this model to your data.
4. Outliers. Do you have any evidence that once you t your model to the data that
there are particular entries that are obviously not part of the trend. There can
be many reasons for outliers. First, they may just be an outlier and it is a real
observation that should be kept in the model. However, it is also possible that
Biological Data Analysis Using R
6.3. MULTIPLE REGRESSION 97
Figure 6.4: A 2x2 matrix plot of some diagnostic tools associated with a linear model. They include
a plot of the residuals (eij) as a function of the tted values ( yi) to see if there are systematic biases
in the model (upper left), a Q-Q plot to examine normality of the residuals (upper right), a scale
location plot (lower left), and a leverage plot to look for outliers (lower right).
there was an equipment malfunction, you entered the data point incorrectly into
the computer, etc. It is always good to check and see if you screwed up.
R provides a series of four plots for you to look at when you plot a variable specied
by lm(). These plots are displayed in Figure ??. You can see these plots by using the
command plot( t1) (or whatever your model variable name is) and R will show you a series
of plots examining the distribution of the residuals. For a more in depth discussion of
model verication you should probably consult a text book on regression analysis.
6.3 Multiple Regression
There are several occasions where we may be interested in how well several predictor
variables can explain the variation in a response variable. This is called multiple regres-
sion and has a linear model with the form:
Biological Data Analysis Using R
98 CHAPTER 6. LINEAR MODELS
y
i
=
0
+
1
X
1
+
2
X
2
+. . . +
k
X
k
+e

Here you have up to k different predictor variables, each of which contributing to the
observed value in y. When approaching a multiple regression,
The null hypothesis for a multiple regression is H
O
:
i
= 0; i and states that all the beta
regression terms are zero. To address this hypothesis, we build a linear model and then
determine how much of the observed variation can be explained by the model in.
In R we can use the same lm function as for a single predictor regression but this time
we need change how we put the function equation into it to accommodate two variables.
For this example, we can use the data shown in Table 6.3.
i Y X
1
X
2
1 4.26 1.00 0.89
2 20.74 2.00 0.41
3 14.95 3.00 0.72
4 -5.55 4.00 0.20
5 21.29 5.00 0.40
6 33.49 6.00 0.37
7 32.15 7.00 0.61
8 45.95 8.00 0.09
9 38.94 9.00 0.74
10 48.27 10.00 0.69
These values can be put into R as:
> Y < c( 4. 26 , 30.74, 14.95, 5.55, 21.29, 33.49, 32.15, 45.95, 38.94, 48.27)
> X1 < 1:10
> X2 < c( 0. 88 , 0.41, 0.72, 0.19, 0.40, 0.37, 0.61, 0.09, 0.74, 0.68)
> cbind ( Y, X1,X2)
Y X1 X2
[ 1 , ] 4.26 1 0.88
[ 2 , ] 30.74 2 0.41
[ 3 , ] 14.95 3 0.72
[ 4 , ] 5.55 4 0.19
[ 5 , ] 21.29 5 0.40
[ 6 , ] 33.49 6 0.37
[ 7 , ] 32.15 7 0.61
[ 8 , ] 45.95 8 0.09
[ 9 , ] 38.94 9 0.74
[ 10 , ] 48.27 10 0.68
And then we can create a linear model using the notation lm( Y X1 + X2 ).
> f i t 2 < lm( Y X1 + X2 )
> summary( f i t 2 )
Cal l :
lm( formula = Y X1 + X2)
Residuals :
Min 1Q Median 3Q Max
24.8394 2.7430 0.8989 4.1369 20.0461
Biological Data Analysis Using R
6.3. MULTIPLE REGRESSION 99
Coef f i ci ent s :
Estimate Std. Error t value Pr( >| t | )
( I ntercept ) 1.170 12.801 0.091 0.9297
X1 4.460 1.422 3.137 0.0164
X2 1.473 16.763 0.088 0.9324

Si gni f . codes : 0 0.001 0.01 0.05 . 0.1 1

Residual standard error : 12.85 on 7 degrees of freedom
Multiple Rsquared: 0.5857, Adjusted Rsquared: 0.4673
Fst at i st i c : 4.948 on 2 and 7 DF, pvalue : 0.04578
> anova( f i t 2 )
Analysis of Variance Table
Response : Y
Df Sum Sq Mean Sq F value Pr(>F)
X1 1 1631.66 1631.66 9.8875 0.01628
X2 1 1.27 1.27 0.0077 0.93244
Residuals 7 1155.16 165.02

Si gni f . codes : 0 0.001 0.01 0.05 . 0.1 1

As we can see, the estimates for
0
= 1.17,
1
= 4.46, and
2
= 1.47. Overall, it appears that
the only term that has is like to not be zero is the term for
1
for variable X1. However,
even with the
0
and
2
terms in the model for the intercept and the slope coefcient for
the variable X2, the overall mode is signicant (see the anova table).
Adding Interactions
Some times it is preferable to run models that show the interaction between variables
as well as the inuence of individual variables. This is appropriate when you have some
reason to believe that the combination of predictor variables will inuence the response
in a non-additive method. The linear model for this is:
y
ij
= +
1
X
1
+
2
X
2
+
3
(X
1
X
2
) +e
ij
where the
3
coefcient determines the strength of the interaction. If
3
= 0 then there
is no interaction.
In R interaction terms are indicated by the colon operator. For example, the full model
in our example data with the interaction would be specied as
> f i t 2 < lm( Y X1 + X2 + X1:X2 )
> summary( f i t 2 )
Cal l :
lm( formula = Y X1 + X2 + X1:X2)
Residuals :
Min 1Q Median 3Q Max
22.882 2.267 1.007 4.168 22.401
Coef f i ci ent s :
Estimate Std. Error t value Pr( >| t | )
( I ntercept ) 8.500 26.951 0.315 0.763
X1 6.204 4.459 1.391 0.213
Biological Data Analysis Using R
100 CHAPTER 6. LINEAR MODELS
X2 16.270 39.803 0.409 0.697
X1:X2 2.732 6.569 0.416 0.692
Residual standard error : 13.68 on 6 degrees of freedom
Multiple Rsquared: 0.5973, Adjusted Rsquared: 0.3959
Fst at i st i c : 2.966 on 3 and 6 DF, pvalue : 0.1192
> anova( f i t 2 )
Analysis of Variance Table
Response : Y
Df Sum Sq Mean Sq F value Pr(>F)
X1 1 1631.66 1631.66 8.7194 0.02552
X2 1 1.27 1.27 0.0068 0.93692
X1:X2 1 32.37 32.37 0.1730 0.69192
Residuals 6 1122.78 187.13

Si gni f . codes : 0 0.001 0.01 0.05 . 0.1 1

There is a shorthand method that indicates that you are interested in having all interac-
tions between predictor variables and that is:
> f i t 2Al t ernat e < lm( Y X1X2 )
> summary( f i t 2Al t ernat e )
Cal l :
lm( formula = Y X1 X2)
Residuals :
Min 1Q Median 3Q Max
22.882 2.267 1.007 4.168 22.401
Coef f i ci ent s :
Estimate Std. Error t value Pr( >| t | )
( I ntercept ) 8.500 26.951 0.315 0.763
X1 6.204 4.459 1.391 0.213
X2 16.270 39.803 0.409 0.697
X1:X2 2.732 6.569 0.416 0.692
Residual standard error : 13.68 on 6 degrees of freedom
Multiple Rsquared: 0.5973, Adjusted Rsquared: 0.3959
Fst at i st i c : 2.966 on 3 and 6 DF, pvalue : 0.1192
You can see that this gives the exact same response. You should be careful with this
notation when you are working with several predictor variables because it will do all the
linear interactions including the three- and four-way (and higher) ones if you have that
many variables. This may or may not be what you are interested in testing.
Models Without Intercept Terms
Some times it is of interest to test the t of a model that does not have an interaction
term. Perhaps you have already subtracted the mean of the response variable y = y y
and as such there is not predicted to be any interaction, or as in the case of our model in
the previous section, perhaps the model does not support the addition of an interaction
term. At any rate, it is possible to indicate to the lm function that you want to run the
analysis without estimating the interaction. The linear model for this would be:
Biological Data Analysis Using R
6.3. MULTIPLE REGRESSION 101
y
i
=
1
X
The formula that you pass to lm( Y X 1). The -1 addition to the function is the part that
tells R how to run properly. Running the data again but only including the variable X1
and the response variable Y without the interaction term gives:
> f i t 3 < lm( Y X1 1 )
> summary( f i t 3 )
Cal l :
lm( formula = Y X1 1)
Residuals :
Min 1Q Median 3Q Max
24.4756 2.0177 0.1422 4.0652 21.2772
Coef f i ci ent s :
Estimate Std. Error t value Pr( >| t | )
X1 4.7314 0.5798 8.16 1.89e05

Si gni f . codes : 0 0.001 0.01 0.05 . 0.1 1

Residual standard error : 11.38 on 9 degrees of freedom
Multiple Rsquared: 0.8809, Adjusted Rsquared: 0.8677
Fst at i st i c : 66.59 on 1 and 9 DF, pvalue : 1.889e05
> anova( f i t 3 )
Analysis of Variance Table
Response : Y
Df Sum Sq Mean Sq F value Pr(>F)
X1 1 8618.7 8618.7 66.587 1.889e05
Residuals 9 1164.9 129.4

Si gni f . codes : 0 0.001 0.01 0.05 . 0.1 1

Overall, this model explains much more of the variation that the full model lm(Y X1 + X2 )
or the interaction model lm(Y X1X2), just compare the Multiple R-Squared values.
6.3.1 Comparing Models
So in the previous subsection we have developed three different models that we have
proposed to explain our data. They are, in the order of reverse complexity, give as:
The full model with the interaction terms lm( Y X1 + X2 + X1:X2).
The full model without the interaction terms lm( Y X1 + X2 ).
The partial model with only X1, lm(Y X1).
The minimal model with only X1 and without an intercept term, lm(Y X1 1).
There are several methods that you should use to determine which of these models you
would like to consider to be the most appropriate.
1. Look at the overall anova signicance. If the overall models are not signicant, then
there is no use in discussing them. In our examples, the full interaction model was
not signicant and should be disregarded.
Biological Data Analysis Using R
102 CHAPTER 6. LINEAR MODELS
2. Examine the relative signicance of each of the terms in the models as is shown
by the summary function. This can give some indication of which terms may be
important. Our various models suggested that the predictor variable X2 did not
help in explaining the variation in the response variable.
3. Look at the relative R-squared values. These indicate the proportion of variation
explained by the model and are given by the summary function.
4. Use a statistically based method to test the differences between two models such
as:
anova You can use the anova function and pass it two models that have been t to
the same data and it will perform an analysis to see if the additional term(s)
are signicant. Here is an example using the models having only the variable
X1 to see if the addition of the intercept term is signicant.
> anova( f i t 3 , f i t 4 )
Analysis of Variance Table
Model 1: Y X1 1
Model 2: Y X1
Res. Df RSS Df Sum of Sq F Pr(>F)
1 9 1164.91
2 8 1156.43 1 8.48 0.0587 0.8147
AIC There are other statistical methods that you can use to see if the additional
terms are signicant in your model. One of these is the stepwise method using
the AIC (Akaike Information Criterion). In R you can do this by passing the
largest model to the function step and it will perform the analysis for you. The
AIC statistics will decrease as the estimated predictive power of your model
increases. So you want to look for the smallest values of AIC. Here is an
example using the full model (including the interaction).
Start : AIC=55.21
Y X1 + X2 + X1:X2
Df Sum of Sq RSS AIC
X1:X2 1 32.37 1155.16 53.49
<none> 1122.78 55.21
Step: AIC=53.49
Y X1 + X2
Df Sum of Sq RSS AIC
X2 1 1.27 1156.43 51.51
<none> 1155.16 53.49
X1 1 1624.25 2779.40 60.27
Step: AIC=51.51
Y X1
Df Sum of Sq RSS AIC
<none> 1156.43 51.51
X1 1 1631.66 2788.09 58.31
Cal l :
lm( formula = Y X1)
Coef f i ci ent s :
( I ntercept ) X1
Biological Data Analysis Using R
6.4. ANALYSIS OF VARIANCE 103
1.989 4.447
As you can see, the AIC values decrease until the nal model which only has
the X1 term and is missing an intercept.
You should consider a wide range of these methods when attempting to put together a
good regression model.
6.4 Analysis of Variance
The analysis of variance is a common method for examining the equality of observations
that can be partitioned into categorical treatments. In all reality, an ANOVA is simply
a regression with categorical predictor variables (e.g., the values of x are not continu-
ous).
6.4.1 1-Way ANOVA
The simplest ANOVA model is one in which a single treatment has been applied and you
have collected a single set of observations. The linear model can be presented as:
y
ij
= +
i
+e
ij
where the
i
is the treatment effect. You can think of this as the deviation from the
overall mean that can be attributed to an observation being in a particular treatment.
The e
ij
term is again the error term.
In 5.3.1, we used the Pinus echinata germination data to illustrate how to perform a
Kruskal-Wallis test. At that time, I had suggested that the Kruskal-Wallis test was a
rank-based version of an analysis of variance (ANOVA). Here will use the same data
again to demonstrate the parametric equivalent of the Kruskal-Wallis test; the one-way
ANOVA.
As a reminder, the data consist of family germination rates for Pinus echinata (perhaps
one of the homeliest looking conifer in existence) separated by timber treatment. In the
Ozark mountains of Missouri, control, selectively cut, and clear cut treatments were ap-
plied to previously continuous forest stands. No P. echinata individuals were removed so
in essence the treatments were modications of other species around the resident pines.
A summary of germination data is presented in Figure ?? showing the average germi-
nation rate lowest in the control stands and highest in the stands where heterospecics
were selectively removed from around the target species.
The null hypothesis for this model is: H
O
: NoTreatmentEffects (which is like saying

Control
=
Selective
=
ClearCut
).
> pineData < read . tabl e ( "PineGerminationData.txt" , header=T)
> anova1 < aov ( GERM TRT, data=pineData )
> anova1
Cal l :
Biological Data Analysis Using R
104 CHAPTER 6. LINEAR MODELS
Figure 6.5: Boxplot of germination percentages for Pinus echinata as a function of treatment. A
colored rug was added to the right side to show the actual values within treatments (see rug.
aov ( formula = GERM TRT, data = pineData )
Terms:
TRT Residuals
Sum of Squares 0.8717943 2.6520868
Deg. of Freedom 2 50
Residual standard error : 0.2303079
Estimated ef f ect s may be unbalanced
> anova( anova1)
Analysis of Variance Table
Response : GERM
Df Sum Sq Mean Sq F value Pr(>F)
TRT 2 0.87179 0.43590 8.218 0.0008207
Residuals 50 2.65209 0.05304

Si gni f . codes : 0 0.001 0.01 0.05 . 0.1 1

From these results, we can see that there is a treatment effect, and it appears to be
highly signicant. But in looking at the plot in Figure 6.5 are these results supposed to
lead us to believe that all the treatments are signicantly different or just some subset
Biological Data Analysis Using R
6.4. ANALYSIS OF VARIANCE 105
of them?
One way to get to this is to look at the 95% condence intervals for the treatment means
and see if they overlap. One way to do this is to use the Tukey Honest Signicant
Differences (or TukeyHSD) function. This function takes the aov analysis as an argument
and prints out the condence intervals for the differences in the means of the treat-
ments.
Figure 6.6: Condence intervals for difference in mean germination rates for Pinus echinata fam-
ilies.
> postHoc < TukeyHSD( anova1 )
> postHoc
Tukey multiple comparisons of means
95% familywise confidence l evel
Fi t : aov ( formula = GERM TRT, data = pineData )
$TRT
di f f lwr upr p adj
CTRLCLR 0.27927536 0.46389755 0.09465318 0.0017640
SELCLR 0.04566667 0.24879523 0.15746190 0.8504882
SELCTRL 0.23360870 0.04898651 0.41823088 0.0098768
> pl ot ( postHoc )
Biological Data Analysis Using R
106 CHAPTER 6. LINEAR MODELS
The postHoc anlaysis can also be plotted by calling plot( postHoc ) showing the condence
in the differences in treatment levels (those that overlap the zero are not signicantly
different) as presented in Figure 6.6. These results suggest that the signicance in the
ANOVA model is due to the differences between the control and the other two treatments
and that both of the cutting treatments had essentially the same germination rate (just
larger than families in the control stands).
Biological Data Analysis Using R
6.5. USEFUL FUNCTIONS 107
6.5 Useful Functions
The following functions were introduced in this chapter and you will be required to use
them for the exercises. To get more information on any of these functions, use the R
help system.
abline(x) Draws a line on the currently active graphics device. You can either specify
the intercept and slope or pass this a tted linear model.
anova(x) Creates the Analysis of Variance Tables for the models passed in x.
aov(x) Performs the analysis of variance on the formula in x.
cbind(x,y,... Puts the variables x, y, ... into a single column-bound variable.
lm(func) Tests the model func using linear least-squares.
t.test(x) This function performs the t-test for either a single data set and a predicted
mean or a paired t-test using two data sets.
round(x) Rounds the value of x to the nearest integer.
pch Optional parameter for the plot function that will designate the type of symbol
plotted using the plot command.
runif(n,mn,mx) Returns n random numbers drawn uniformly from the range [mn, mx].
step(x) Evaluates the terms in the model x for inclusion in the model using the AIC
criteria.
summary(x) Provides description of x.
text(x,y,c Plots the text in c on the graph at the coordinates (x, y).
TukeyHSD(x) Performs Tukeys Honest Signicant Difference post-hoc test on the aov
model in x.
Biological Data Analysis Using R
108 CHAPTER 6. LINEAR MODELS
6.6 Exercises
The following exercises are meant to help you understand the items presented in this
Chapter.
1. Load the data set Temperature.csv from the chapter folder. These data represent the
measured brood chamber temperature for a wood-boring beetle. Test the hypothesis
H
O
: Mean temperature is 61

.
2. Load the data set ClutchSizes.csv from the le. Using a paired t-test, test the hypoth-
esis H
O
: There is no difference in reproductive output between habitat types.
3. Load the data le, SingleRegresssion.RData from the le into R
2
. Fit the regression
model, Y X. Is it signicant? Show the regression equation and the anova table.
4. Plot the regression model fromthe previous example and indicate the tted regression
line with a dotted red line in the plot.
5. Fromthe single regression model, add the regression equation to the graph indicating
the coefcients that were estimated.
6. Does a plot of the residuals as a function of the predicted values from the estimated
regression model suggest that the model is appropriate?
7. Load the data set MultipleRegression.RData from the le, it will contain a data frame
named multReg. Use the variables in this data frame, Y,X1,X2,X3 to t a multiple re-
gression model. Show the summary and the anova table in your results. What is the
predicted regression equation?
8. Fit another model to the multReg data that has all the interaction terms amongst the
X predictor variables. Use the anova procedure to see which of these models is more
appropriate.
9. Load the data le VarroaCounts.RData, it will be a data frame named BeeData. These
data represent counts of the parasite Varroa destructor a common pest of domesti-
cated honey bees. Test the hypothesis using an analysis of variance that there is no
difference in mite counts between the different lines of bees.
10. Perform the TukeyHSD test on the parasite data from the previous question.
2
Use the load function.
Biological Data Analysis Using R
Chapter 7
Working With Images
In this chapter, you will focus on the following topics:
Gain a basic understanding of open image formats
Learn how to import image data into R
Manipulate image data at the pixel level.
7.1 Image Data
There are several different methods that are available to you to import image data into R
. As I was writing this document over Winter break and updating it in the fall, the main
image processing library for R , rimage, was broken and could caused a few problems
when installed. I am sure it will be xed in the near future and recommend that you
look at that library when you next have the need to do some image manipulation because
it has a lot of funcitonality. However, at the present, it is not going to be used. The
consequences of not having rimage is that it appears that importing jpeg, tiff, and bmp
image formats is beyond our grasp. Lucky for us, there are a ton of other image formats
out there and we can easily convert the image shown in Figure 11.1 into another format
and use it just as easily. Perhaps when I update this manuscript the next time around,
Ill change this section. I think it is also important that you understand the internal
workings of images and for right now, these more simple image formats will serve our
purposes nicely and everything you learn here will be easily transferable to those other
image formats when you need to deal with them in the future.
7.1.1 PNM Image Format
Images on computers have specic formats in which the color information and other
meta data is stored in the le. Some of the methods are relatively easy to use and can be
manipulated directly in a text editor. Others are more of a pain and some are owned by
some company who has patented the way the information is stored in the le and you
have to pay royalties to them to view it. For example, the ubiquitous GIF image format
109
110 CHAPTER 7. WORKING WITH IMAGES
uses an algorithm that was patented and owned by a company and if you were to write
a viewer for it in some countries you would have to pay a royalty to use it... Lame.
The PNM image format (short for portable anymap) is an open format for the exchange
of image information. Actually, there are three different formats that fall under the PNM
specication as detailed below.
Portable Bitmap Format (PBM)
This format stores bitmaps images. A bitmap can be thought of as an image whose pixels
are either turned on or off (say black and white). The representation of a PBM le can be
given as a simple text le with the extension .pbm. An example text le for a bitmap le
that encodes for the uppercase letter R would be:
P1
# This is an example bit map file r.pbm
5 8
1 1 1 1 0
1 0 0 0 1
1 0 0 0 1
1 0 0 0 1
1 1 1 1 0
1 0 0 1 0
1 0 0 0 1
1 0 0 0 1
In this le, the rst line is a special code to tell the computer how many bits per pixel to
use. The second line is a comment line that you can put anything you like into (but has
to start with the # character). The third line tells how many columns and rows of data
that the image has. Note, this is a column-major notation here where the rst number
is the number of columns and the second number is the number of rows, which is the
opposite of which we use (row-major) in R for interacting with matrices of data. The rest
of the le consists of the actual bit matrix where 1 represents a pixel that is turned on
and 0 represents a pixel that is turned off. The image represented in this le is given in
Figure 7.1.
You can make this image programatically, by creating the matrix in R and using the
image function. Here is an example creating the image of the letter T.
> x < matrix ( 0 , nrow=8, ncol =5)
> x
[ , 1] [ , 2] [ , 3] [ , 4] [ , 5]
[ 1 , ] 0 0 0 0 0
[ 2 , ] 0 0 0 0 0
[ 3 , ] 0 0 0 0 0
[ 4 , ] 0 0 0 0 0
[ 5 , ] 0 0 0 0 0
[ 6 , ] 0 0 0 0 0
[ 7 , ] 0 0 0 0 0
[ 8 , ] 0 0 0 0 0
> x[ 1 , ] < 1
> x[ , 3] < 1
> x
Biological Data Analysis Using R
7.1. IMAGE DATA 111
Figure 7.1: The image represented in the r.pbm le. This image has been scaled up to make it
large enough to see it on the page using the program GIMP (www.gimp.org).
[ , 1] [ , 2] [ , 3] [ , 4] [ , 5]
[ 1 , ] 1 1 1 1 1
[ 2 , ] 0 0 1 0 0
[ 3 , ] 0 0 1 0 0
[ 4 , ] 0 0 1 0 0
[ 5 , ] 0 0 1 0 0
[ 6 , ] 0 0 1 0 0
[ 7 , ] 0 0 1 0 0
[ 8 , ] 0 0 1 0 0
> col ors < c ( "black" ,"grey")
> image ( x, col =colors , axes=F)
Here I created the matrix that had all 0 in it and set the top row and the middle column
equal to 1. Then the image function was used to plot it. The image function takes a number
of optional arguments and here I have supplied it the colors and the option to not show
the axes. Since I have two values in the matrix, a two element vector will be sufcient to
handle all the different colors. The image shown in Figure 7.2 shows this matrix. There
seems to be a small problem with it in that it is rotated 90

counter-clockwise. This is
because the origin of the plot that is created by the image function is in the lower left-hand
corner. Conversely, most images that are stored on the computer (like the desktop image
in the background), assume that the origin is at the upper left hand corner of the image.
Obviously these two do not mesh well together.
Portable Graymap Format (PGM)
This format is for graymap images where the term graymap refers to the lack of color
in the image. In terms of complexity, this is slightly more information contained in the
data le as each pixel is not either ON or OFF, rather there is a percentage of ONNESS... (is
that a word?).
P2
# The PGM file for dog.pgm
24 7
5
Biological Data Analysis Using R
112 CHAPTER 7. WORKING WITH IMAGES
Figure 7.2: A PBM le that was programatically created in R . The image is rotated because of the
default location of the origin.
0 1 1 1 1 0 0 0 0 0 0 5 5 5 0 0 0 0 0 4 4 4 0 0
0 1 0 0 0 1 0 0 0 0 5 0 0 0 5 0 0 0 4 0 0 0 4 0
0 1 0 0 0 0 1 0 0 5 0 0 0 0 0 5 0 0 4 0 0 0 0 0
0 1 0 0 0 0 1 0 0 5 0 0 0 0 0 5 0 0 4 0 0 0 0 0
0 1 0 0 0 0 1 0 0 5 0 0 0 0 0 5 0 0 4 0 0 4 4 0
0 1 0 0 0 1 0 0 0 0 5 0 0 0 5 0 0 0 4 0 0 0 4 0
0 1 1 1 1 0 0 0 0 0 0 5 5 5 0 0 0 0 0 4 4 4 0 0
The rst three lines of the le are the same as for the PBM format. The fourth line in the
le gives the maximum value representing the the most white in the image. In this case,
the a black pixel will be represented by the number 0 and the white would be represented
by 5 and values in between would be
1
5
increments of whiteness. The remaining portions
of the le have the actual image represented in a pixel-by-pixel matrix of values. You
can see that the majority of the image is
0
5
black and the letters are varying shades of
gray (Figure 7.3).
The number of shades of gray you use in a PGM le is up to you as long as it does not
exceed 255 (I think). These are easy les to create and you could imagine how you could
Biological Data Analysis Using R
7.1. IMAGE DATA 113
Figure 7.3: The image represented by the dog.pgm le. This image has been scaled up to make it
large enough to see it on the page using the program GIMP (www.gimp.org).
create a matrix of integers from some analysis and save it as a pgm le and view it
directly.
Portable Pixmap Format (PPM)
The last le format, PPM, is one that handles pixmaps, which means that you have colored
pixels in the image. The le format is identical to that of the PGM with the exception that
the code on the rst line is P3, which represents 24-bits per pixel; 8 of which are for red,
8 for green, and 8 for blue. An example of the PPM le shown in Figure 7.4 is:
P3
# This image contains an image of my daughter Libbie (from Libbie.ppm).
180 240
255
188
219
253
189
220
252
In this le, the pixel values are placed one per line instead of next to each other. Starting
at line number 5 with a value of 188 the following 180x240 = 43, 200 lines contain an
integer whose value is between 0 and 255 (the maximum all color as depicted on line
4) for the color red followed by another 43, 200 lines of numbers for the color green, and
then another 43, 200 lines for the blue. When we begin looking at manipulating images
you will nd that you can interact with each color channel independently.
One drawback to these image formats are that they are not very efcient. For example,
the image of my daughter in Figure 7.4 has 129, 604 lines of information in it, which on
my computer makes it 465K in size. The exact same image saved as a jpeg le is only
25K in size. The compression used to make jpeg, tiff, gif, png, and other compressed le
formats is why they are used on the internet. But for our purposes, the lack compression
and inefciency in storages sizes are relatively irrelevant.
Biological Data Analysis Using R
114 CHAPTER 7. WORKING WITH IMAGES
Figure 7.4: The image represented in the Libbie.ppm le. This image has been scaled up to make
it large enough to see it on the page using the program GIMP (www.gimp.org).
7.2 Loading The Image Into R
OK, now that the basics of how one kind of image is represented in the data les, it is
time to load one into R and see what we have to work with. To load a PNM le, you must
rst import the pixmap library then you can use the function read.pnm() to load the le into
a local variable and plot it using the plot () function.
> l i brary ( pixmap)
> photo < read .pnm( f i l e ="Libbie.ppm")
Read 129600 items
> pl ot ( photo )
The plot () function will open a new image window and show the loaded image.
7.3 Components of A Pixmap
We can learn a little bit more about what kind of data type the variable we call photo is
by using the class() function.
> cl ass ( photo )
[ 1] "pixmapRGB"
at t r ( , "package")
[ 1] "pixmap"
> names( attri butes ( photo ) )
[ 1] "size" "cellres" "bbox" "bbcent" "channels" "red" "green"
[ 8] "blue" "class"
Biological Data Analysis Using R
7.4. IMAGE OPERATIONS 115
This variable is a pixmapRGB class that comes from the pixmap package. A class is a self con-
tained data structure that has both attributes and data. The command names(attributes(photo))
tells us the names of the attributes that the variable has.
There are some issues that we should touch on when dealing with classes. They differ
from what we have been using thus far such as data frames in that we cannot access
the contents of a class using the $ notation. This is because things like lists and data
frames are not classes, they are just objects. To access attributes of classes we use the
notation. For example:
> photo@size
[ 1] 240 180
> photo@channels
[ 1] "red" "green" "blue"
> dim( photo@red )
[ 1] 240 180
> photo@red[ 1 , 1]
[ 1] 0.7372549
> range ( photo@red)
[ 1] 0 1
Here we can get to the size, channels, and red components of the class directly. We can also
see that the red channel that determines the amount of redness in each pixel has been
standardized on the range [0, 1]. This is important to know if we are going to manipulate
the image directly.
7.4 Image Operations
7.4.1 Extracting Channels
So now we know how to make some alterations of the image and see what happens. In
the next example, I rst copy the photo to make three additional photos, named redPhoto,
bluePhoto, and greenPhoto. Then for each of the new variables I remove all the data in each
of the corresponding channels by making the channel contain a matrix of zeros the same
size as the original matrix.
> redPhoto < photo
> bluePhoto < photo
> greenPhoto < photo
> redPhoto@size
[ 1] 240 180
> redPhoto@blue < redPhoto@green < matrix ( 0 , nrow=240, ncol =180)
> bluePhoto@red < bluePhoto@green < matrix ( 0 , nrow=240, ncol =180)
> greenPhoto@red < greenPhoto@blue < matrix ( 0 , nrow=240, ncol =180)
> par ( mfrow=c ( 1 , 4) )
> pl ot ( photo )
> pl ot ( redPhoto )
> pl ot ( greenPhoto )
> pl ot ( bluePhoto )
Note that I used the sequential assignment A <B <C <D as a shorthand here. This will
assign the value of D to the variable C then C to B and then B to A. This a lazy trick but one
that you will probably use as it saves a bit of time and typing.
Biological Data Analysis Using R
116 CHAPTER 7. WORKING WITH IMAGES
Then I make a 1x4 matrix of plots so that I can plot all four images in the same frame (see
?? for more on how this is done) and in each of the four slots, I plot one of the images
yielding a gure similar to what is presented in Figure 7.5.
Figure 7.5: The original image along with ones where only the red, green, and blue channel turned
on.
In some cases, it is helpful if you can extract the color information and generalize the
image as a greyscale image (as you will in Chapter 11). Here we use the information
from each channel, weighed equally, in the creation of the image.
> gphoto < pixmapGrey( photo@red+photo@blue+photo@green)
> pl ot ( gphoto )
> names( attri butes ( gphoto ) )
[ 1] "size" "cellres" "bbox" "bbcent" "channels" "grey" "class"
> gphoto@grey[ 1 , 1]
[ 1] 0.8627451
> range ( gphoto@grey )
[ 1] 0 1
The function pixmapGrey() takes a matrix of data, of which we just use the element-wise
addition of each channel in the color photo. You can also see that in the creation of the
new grey image, the values were again standardized.
For the moment, lets examine the contents of this grey image and play around with it a
bit. Lets make it a bit darker by shifting all the grey values down (to make it more black).
We can do this by performing operations on the matrix of grey values in the class. For
simplicity, I will make a copy of the image rst and then perform operations on the copy
rather than the original one. Then we will look at the distribution of grey values that
make the image.
> darkerGphoto < gphoto
> darkerGphoto@grey < darkerGphoto@grey / 2
> par ( mfrow=c ( 1 , 3) )
> pl ot ( gphoto )
> hi st ( gphoto@grey , xlim=c ( 0 , 1) , xlab="Grey" ,main="")
> pl ot ( darkerGphoto )
We can see that the vast majority of values are towards the light end of the distribution.
To darken this up, we should scale these values to be closer to zero by dividing them by
2 and then replotting the image to see the result (see results in Figure 7.6).
Biological Data Analysis Using R
7.5. CREATING IMAGES PROGRAMATICALLY 117
7.5 Creating Images Programatically
Images can be made programatically once you understand how images are represented.
There are some helper functions that can help you in creating new images. For the
purposes of this section, we will focus on greyscale images and allow the analysis of
colored images for you to play with on your own time.
Lets start by making an image where each pixel is randomly assigned a greyscale value.
For convenience, Ill make it the same size as the photo named gphoto from 7.4.1.
> randomImageMatrix < matrix ( rnorm(240180) ,nrow=240, ncol =180)
> gray < grey(1:100/100)
> image ( randomImageMatrix , col =gray )
Here I use the rnorm() function to create 240 180 = 43, 200 random numbers in a matrix
that has 240 rows and 180 columns. I then use the grey() function to create 100 different
shades of grey ranging from white to black at equal intervals. When the image is made,
the range of random numbers is used to divide the pixels into the 100 different grey
colors (e.g., the image() function scales the values in randomImageMatrix into length(gray) distinct
groups for plotting). The results is shown in Figure 7.7.
This image can be manipulated by changing the values in the matrix randomImageMatrix. In
the next example, I replace the center 40x40 block with the white (which would be the
largest value from randomImageMatrix).
> randomImageMatrix[100:140,70:110] < max( randomImageMatrix)
> image ( randomImageMatrix , col =gray )
The result is shown in Figure 7.8 resembling a square doughnut (mmmmdoughnuts...).
Figure 7.6: The greyscale translation of the PPN image, a histogram of the grey values and the
image resulting from reducing all the grey values in the image by half.
Biological Data Analysis Using R
118 CHAPTER 7. WORKING WITH IMAGES
Figure 7.7: A random image Figure 7.8: A random image with a square
doughnut hole in the middle.
7.6 Useful Functions
The following functions were introduced in this chapter and you will be required to use
them for the exercises. To get more information on any of these functions, use the R
help system.
cat() This function dumps the passed arguments out to the terminal.
grey(x) This function returns the grey color associated with the value of x. It is
assumed that that 0 x 1.
image(x) Can be used to create an image as either grey or colors for the values in the
matrix x.
max(x) Returns the maximum value contained in x.
rnorm(x) Returns x random numbers from a N(, ).
Biological Data Analysis Using R
7.7. EXERCISES 119
7.7 Exercises
The following exercises are meant to help you understand the items presented in this
Chapter.
1. Create a Portable Bitmap Format le (*.pbm) exactly like the one that is shown for the
letter R but make it represent the letter L.
2. Why is Figure 7.2 not right-side-up?
3. Make your L image correct by changing the values of the underlying matrix such that
when it is plot using the image command it is in the correct orientation.
4. What is the purpose of the PX number on the rst line of the PNM le formats?
5. Load your own copy of the image Libbie.ppm into R using the read.pnm function as
demonstrated in the Chapter. Create three copies of the image and for each copy
remove the values in one channel (e.g., make one of the color matrices a zero). Plot
these images in a three-paned graphic using the function par(mfrow=c(1,3) option.
6. Replot the randomImageMatrix using a color palette instead of the grey palette shown.
(Hint: See ?rainbow for ve of the stock palettes available to you.)
7. What is the default palette used in the image plot function?
8. What is the purpose of the optional argument bbox in the pixmapGrey function?
9. Create the greyscale version of the image shown in the leftmost box in Figure 7.6.
The grey channel is composed of greyscale values that must be between [0, 1]. Can
you invert the colors in this image? (Hint: If you cant gure out how to do this, see
the footnote at the end of this sentence but only as a last resort.
1
10. Why do you have to use the @ notation to access components of the pixmaps in this
chapter?
1
Are you sure you want a hint? Take 1 minus the grey channel to make the values ipped in the [0, 1]
interval.)
Biological Data Analysis Using R
120 CHAPTER 7. WORKING WITH IMAGES
Biological Data Analysis Using R
Chapter 8
Matrix Analysis
Matrices are used in a wide variety of biological studies. In this Chapter I will use the
example of stage-classied matrix models to introduce you to how matrix manipulation
operates in R . There are some issues that need to be addressed with respect to basic
operations on matrices that if you havent had a course on Matrix Algebra, you may not
fully appreciate.
In this chapter, you will focus on the following topics:
Understand matrix operations in R .
Create stage-classied matrix models.
8.1 Matrices In R
As shown in 2.4.9, a matrix is a fully recognized data type in R . In fact, R does a
wonderful job of working with matrices and is much faster at doing vector and matrix
operations directly than looping through matrices of values using a for()-loop (see 11.1
for a complete discussion of looping R ).
In specic terms for this Chapter, a matrix can be dened as a 2-dimensional object that
holds numeric values. Matrices can be created by hand using the matrix() function and
the elements within them can be accessed using the square bracket notation (e.g., X[i,j])
as:
> X < matrix ( 0 , nrow=4, ncol =4)
> X[ 1 , 2] < 23
> X[ 1 , 4] < 42
> X
[ , 1] [ , 2] [ , 3] [ , 4]
[ 1 , ] 0 23 0 42
[ 2 , ] 0 0 0 0
[ 3 , ] 0 0 0 0
[ 4 , ] 0 0 0 0
You can also wrap the as.matrix() function around the read.table() function and read the
data from a matrix in a le into a variable directly. For a review of these two func-
121
122 CHAPTER 8. MATRIX ANALYSIS
tions see 2.4.9 and 3.1.2. In the online data sets for this chapter, there is a le called
ExampleMatrix.csv that was exported from a spreadsheet. If
> A < as . matrix ( read . tabl e ( "ExampleMatrix.csv" , header=F, sep="\t" ) )
> A
V1 V2 V3 V4 V5 V6 V7 V8 V9
[ 1 , ] 0.00000 2.00000 2.00000 5.00000 4.00000 2.00000 7.00000 2.603310 2.000000
[ 2 , ] 2.00000 0.00000 4.00000 6.00000 3.00000 4.00000 7.00000 3.603310 4.000000
[ 3 , ] 2.00000 4.00000 0.00000 6.00000 4.00000 3.00000 7.00000 1.603310 1.000000
[ 4 , ] 5.00000 6.00000 6.00000 0.00000 3.00000 1.00000 1.00000 3.694210 6.000000
[ 5 , ] 4.00000 3.00000 4.00000 3.00000 0.00000 3.00000 4.00000 1.966940 4.000000
[ 6 , ] 2.00000 4.00000 3.00000 1.00000 3.00000 0.00000 2.00000 2.148760 3.000000
[ 7 , ] 7.00000 7.00000 7.00000 1.00000 4.00000 2.00000 0.00000 4.694210 7.000000
[ 8 , ] 2.60331 3.60331 1.60331 3.69421 1.96694 2.14876 4.69421 0.000000 0.603306
[ 9 , ] 2.00000 4.00000 1.00000 6.00000 4.00000 3.00000 7.00000 0.603306 0.000000
[ 10 , ] 4.00000 5.00000 4.00000 4.00000 4.00000 2.00000 3.00000 3.421490 4.000000
[ 11 , ] 3.00000 5.00000 3.00000 5.00000 6.00000 2.00000 4.00000 3.603310 3.000000
[ 12 , ] 3.00000 4.00000 3.00000 5.00000 3.00000 3.00000 6.00000 1.421490 2.000000
V10 V11 V12
[ 1 , ] 4.00000 3.00000 3.00000
[ 2 , ] 5.00000 5.00000 4.00000
[ 3 , ] 4.00000 3.00000 3.00000
[ 4 , ] 4.00000 5.00000 5.00000
[ 5 , ] 4.00000 6.00000 3.00000
[ 6 , ] 2.00000 2.00000 3.00000
[ 7 , ] 3.00000 4.00000 6.00000
[ 8 , ] 3.42149 3.60331 1.42149
[ 9 , ] 4.00000 3.00000 2.00000
[ 10 , ] 0.00000 1.00000 3.00000
[ 11 , ] 1.00000 0.00000 4.00000
[ 12 , ] 3.00000 4.00000 0.00000
There are a few things to notice here:
1. R wraps values for matrices so that only a portion of each row can be viewed at a
time.
2. The columns of data that were read in the le did not have a header row so R
assigned them the values V1 - V12. This is the default behavior.
3. If there is one value in the matrix that has a decimal portion to it, all the values will
be displayed with the same number of decimal places (e.g., compare the matrix X
and A from the two listings.
8.1.1 Matrix Arithmetic
Matrices have their own special kind of arithmetic that you may not be aware of, so here
is a very short course. For the following examples, I will be using the matrices X
1
, Y,
and Z as dened by the R commands:
> X < matrix ( 1: 9 , nrow=3,byrow=TRUE)
> X
[ , 1] [ , 2] [ , 3]
[ 1 , ] 1 2 3
[ 2 , ] 4 5 6
1
For matrices I will use upper case bold letters for variable names in the text to make it easier to distin-
guish them from non-matrix variables as you read along. Obviously, this is not possible in R itself but for
the text hopefully this will make it easier to follow.
Biological Data Analysis Using R
8.1. MATRICES IN R 123
[ 3 , ] 7 8 9
> Y < matrix ( 9: 1 , nrow=3)
> Y
[ , 1] [ , 2] [ , 3]
[ 1 , ] 9 6 3
[ 2 , ] 8 5 2
[ 3 , ] 7 4 1
> Z < matrix( 1: 12 ,nrow=4)
> Z
[ , 1] [ , 2] [ , 3]
[ 1 , ] 1 5 9
[ 2 , ] 2 6 10
[ 3 , ] 3 7 11
[ 4 , ] 4 8 12
One of the main things you have to pay attention to when dealing with matrices is the
number of rows and columns in the matrices. In these example matrices, X and X are
square matrices (e.g., they have the same number of rows and columns whereas X is
not square as it has 4 rows and 3 columns of data. To access the number of rows and
columns in a matrix you must use the function dim().
Scalar Addition & Subtraction
Matrices may be shifted by the addition or subtraction of a constant scalar value (e.g.,
2 + X). Scalar addition and subtraction take the value of the scalar and add it to every
element in the matrix.
> X
[ , 1] [ , 2] [ , 3]
[ 1 , ] 1 2 3
[ 2 , ] 4 5 6
[ 3 , ] 7 8 9
> X + 3
[ , 1] [ , 2] [ , 3]
[ 1 , ] 4 5 6
[ 2 , ] 7 8 9
[ 3 , ] 10 11 12
Matrix Addition & Subtraction
For both addition and subtraction of matrices, the numbers of rows and columns must
be identical. If they are, the addition and/or subtraction operation results in the elemente-
wise addition of each matrix. In R you can use the normal addition (+) and subtraction
(-) operators as demonstrated below.
> X+Y
[ , 1] [ , 2] [ , 3]
[ 1 , ] 10 8 6
[ 2 , ] 12 10 8
[ 3 , ] 14 12 10
But when they are not the same size, R will barf up an error message to you telling you
they are not amenable to this operation.
Biological Data Analysis Using R
124 CHAPTER 8. MATRIX ANALYSIS
> X+Z
Error in X + Z : nonconformable arrays
Scalar Multiplication
The values within a matrix may be scaled by the multiplication of a scalar value (e.g., 0.5
X). Scalar multiplication results in every single element in the matrix being multiplied
by the scalar value. For example:
> X
[ , 1] [ , 2] [ , 3]
[ 1 , ] 1 2 3
[ 2 , ] 4 5 6
[ 3 , ] 7 8 9
> X 2
[ , 1] [ , 2] [ , 3]
[ 1 , ] 2 4 6
[ 2 , ] 8 10 12
[ 3 , ] 14 16 18
Element-wise Multiplication
It is possible to multiply two matrices where what you are wanting is a new matrix that
is the element-wise product of each of the original matrices. This is sometimes called
the Hadamard product or the Schur product. In R this operation is conducted using
the regular multiplication character,
*
, between the two matrices. The result of this
operation is a new matrix, the same dimensions as the two original ones.
> X
[ , 1] [ , 2] [ , 3]
[ 1 , ] 1 2 3
[ 2 , ] 4 5 6
[ 3 , ] 7 8 9
> Y
[ , 1] [ , 2] [ , 3]
[ 1 , ] 9 6 3
[ 2 , ] 8 5 2
[ 3 , ] 7 4 1
> X Y
[ , 1] [ , 2] [ , 3]
[ 1 , ] 9 12 9
[ 2 , ] 32 25 12
[ 3 , ] 49 32 9
Multiplication
Matrix multiplication is slightly more complicated than multiplication among scalars or
multiplying a scalar by a matrix. For example, in matrix multiplication, AB = BA.
This is because of the way that matrices are multiplied. Moreover, there are several
restrictions to which sets of matrices can be multiplied together.
Biological Data Analysis Using R
8.1. MATRICES IN R 125
For example, consider the operation A = XY where the matrix X has r
X
rows and c
X
columns of data and the matrix Y has r
Y
rows and c
Y
columns of data. For this operation
to be dened, the number of columns in X, c
X
, must equal the number of rows in Y (e.g.,
c
X
= r
Y
). If these are not equal, then you cannot perform the multiplication. Moreover,
the resulting matrix A will have r
X
rows and c
Y
columns. This is because the matrix
multiplication is conducted as:
A
ij
=
N

k=1
X
i,k
Y
k,j
Essentially every row of X is multiplied against the corresponding column of Y.
In R matrix multiplication uses a unique operator that you probably havent seen yet. To
indicate that you want two matrices to be multiplied (and not the Hadamard product as
above) you use the compound operator % %. That is right, it is a pair of percent signs
surrounding the normal multiplication character (a.k.a. the asterisk). Two examples
using the matrices X and Y are given below. Notice how XY = YX.
> X
[ , 1] [ , 2] [ , 3]
[ 1 , ] 1 2 3
[ 2 , ] 4 5 6
[ 3 , ] 7 8 9
> Y
[ , 1] [ , 2] [ , 3]
[ 1 , ] 9 6 3
[ 2 , ] 8 5 2
[ 3 , ] 7 4 1
> X %% Y
[ , 1] [ , 2] [ , 3]
[ 1 , ] 46 28 10
[ 2 , ] 118 73 28
[ 3 , ] 190 118 46
> Y %% X
[ , 1] [ , 2] [ , 3]
[ 1 , ] 54 72 90
[ 2 , ] 42 57 72
[ 3 , ] 30 42 54
> X %% I
[ , 1] [ , 2] [ , 3]
[ 1 , ] 1 2 3
[ 2 , ] 4 5 6
[ 3 , ] 7 8 9
> X (X %% I )
[ , 1] [ , 2] [ , 3]
[ 1 , ] 0 0 0
[ 2 , ] 0 0 0
[ 3 , ] 0 0 0
> I %% X
[ , 1] [ , 2] [ , 3]
[ 1 , ] 1 2 3
[ 2 , ] 4 5 6
[ 3 , ] 7 8 9
>
Here both X and Y are both square and have the same number of rows and columns
(e.g., the simplest case because we dont have to make sure the correct rows and columns
match). The identity matrix, I dened in the section above is shown here with its groovy
Biological Data Analysis Using R
126 CHAPTER 8. MATRIX ANALYSIS
properties. Matrix multiplication by the identity matrix is transitive and will result in
the original matrix. A kind of matrix version of the scalar multiplying by one.
2
Here is an example using the matrices X and Z, who have different dimensions.
> Z
[ , 1] [ , 2] [ , 3]
[ 1 , ] 1 5 9
[ 2 , ] 2 6 10
[ 3 , ] 3 7 11
[ 4 , ] 4 8 12
> Z %% X
[ , 1] [ , 2] [ , 3]
[ 1 , ] 84 99 114
[ 2 , ] 96 114 132
[ 3 , ] 108 129 150
[ 4 , ] 120 144 168
> X %% Z
Error in X %% Z : nonconformable arguments
In the rst case, Z %%X is dened and provides a result because the number of columns
in Z match the number of rows in X. The reverse of this multiplication, X %%Z, is
undened and R tells you so.
8.1.2 Matrix Operations
There are several other operations that can be conducted on matrices that you will
probably run across as you begin playing with matrices. Here are a smattering of a
few.
The Diagonal
It is often necessary to interact with the diagonal, dened as the elements in the matrix
whose row index are equal to the column index, of a matrix. For example, in a covariance
matrix, the diagonal elements are the variance estimates. In R you can get access to
the diagonal of a matrix by using the diag(). Some examples using the diag() function
include:
> X
[ , 1] [ , 2] [ , 3]
[ 1 , ] 1 2 3
[ 2 , ] 4 5 6
[ 3 , ] 7 8 9
> diag (X)
[ 1] 1 5 9
> Z
[ , 1] [ , 2] [ , 3]
[ 1 , ] 1 5 9
[ 2 , ] 2 6 10
[ 3 , ] 3 7 11
[ 4 , ] 4 8 12
> diag ( Z)
[ 1] 1 6 11
2
There are other matrices that have this property that are not as simple as this one and if you take some
multivariate statistics, it will blow your mind how cool they are...
Biological Data Analysis Using R
8.1. MATRICES IN R 127
Notice how even for non-square matrices the diagonal is dened. You can also extract
and insert particular values for the diagonal as demonstrated below:
> X
[ , 1] [ , 2] [ , 3]
[ 1 , ] 1 2 3
[ 2 , ] 4 5 6
[ 3 , ] 7 8 9
> origDiag < diag (X)
> origDiag
[ 1] 1 5 9
> diag (X) < c(42,23,4)
> X
[ , 1] [ , 2] [ , 3]
[ 1 , ] 42 2 3
[ 2 , ] 4 23 6
[ 3 , ] 7 8 4
> diag (X) < origDiag
> X
[ , 1] [ , 2] [ , 3]
[ 1 , ] 1 2 3
[ 2 , ] 4 5 6
[ 3 , ] 7 8 9
A commonly used matrix that can easily be constructed using the diag() function is the
Identity Matrix, whose symbol is I. This matrix has the zeros everywhere except on the
diagonal
> I < matrix ( 0 , nrow=3, ncol =3)
> diag ( I ) < 1
> I
[ , 1] [ , 2] [ , 3]
[ 1 , ] 1 0 0
[ 2 , ] 0 1 0
[ 3 , ] 0 0 1
Finally, there is an operator called the trace of a matrix that is typically written as tr(A),
which is the sum of the diagonal elements. If A is a variance, covariance matrix as is
commonly found in multivariate statistics, then its trace is the overall variance. In R we
can nd the trace using a combination of the sum() and diag() functions as:
> X
[ , 1] [ , 2] [ , 3]
[ 1 , ] 1 2 3
[ 2 , ] 4 5 6
[ 3 , ] 7 8 9
> sum( diag ( X ) )
[ 1] 15
Matrix Determinant
The determinant of a matrix is scalar factor of a matrix. The calcuation of the determi-
nant is somewhat complicated when we get to matrices that have more than two rows
and columns and Ill let you go nd a linear algebra book to look into it if you so desire.
For small matrices, the determinant of a matrix, denoted as |A| is given as:
Biological Data Analysis Using R
128 CHAPTER 8. MATRIX ANALYSIS
|A| =

a
11
a
12
a
21
a
22

= a
11
a
22
a
12
a
21
In R the function det() is used to estimate the determinant of a matrix.
> X < matrix ( c( 1 , 6 , 3 , 4) , nrow=2)
> X
[ , 1] [ , 2]
[ 1 , ] 1 3
[ 2 , ] 6 4
> det (X)
[ 1] 14
Matrix Transpose
The transpose of a matrix is an operation that exchanges the row and column indices
of the elements. This will change the dimensions of the matrix if it is not square. No-
tationally, you will see several different ways to represent a transpose such as A

or
A
T
.
In R the transpose operation is performed with the t () function.
> Z
[ , 1] [ , 2] [ , 3]
[ 1 , ] 1 5 9
[ 2 , ] 2 6 10
[ 3 , ] 3 7 11
[ 4 , ] 4 8 12
> t ( Z)
[ , 1] [ , 2] [ , 3] [ , 4]
[ 1 , ] 1 2 3 4
[ 2 , ] 5 6 7 8
[ 3 , ] 9 10 11 12
> t ( t ( Z) )
[ , 1] [ , 2] [ , 3]
[ 1 , ] 1 5 9
[ 2 , ] 2 6 10
[ 3 , ] 3 7 11
[ 4 , ] 4 8 12
Notice that the transpose of a transpose is equal to the original variable.
Matrix Inversion
For scalars, the inverse is dened as x
1
=
1
x
but for matrices it is slightly more com-
plicated. There are even large groups of matrices that cannot be inverted. One property
that prevents inversion is if the matrix is singular (think black hole of mathematics or
matrices that have a zero determinant).
A common use for matrix inversion is in estimation of regression coefcients by least
squares. In 6.2, we used the lm() function to estimate the intercept and slope coefcients.
This can be done using matrix algebra and the inversion function ginv() found in the MASS
library. A one column matrix of slope coefcients B is estimated from the formula:
Biological Data Analysis Using R
8.1. MATRICES IN R 129
B = (X

X)
1
X

Y
Where the matrix Y matrix is the normal matrix of response variables and the X matrix
has the rst column of all ones (1) for the intercept and the remaining columns as the
predictor variables.
> X < matrix ( c ( rep( 1 , 10) , 1: 10) , ncol=2 )
> X
[ , 1] [ , 2]
[ 1 , ] 1 1
[ 2 , ] 1 2
[ 3 , ] 1 3
[ 4 , ] 1 4
[ 5 , ] 1 5
[ 6 , ] 1 6
[ 7 , ] 1 7
[ 8 , ] 1 8
[ 9 , ] 1 9
[ 10 , ] 1 10
> Y < matrix ( c(19,25,14,15,24,17,19,27,29,25))
> Y
[ , 1]
[ 1 , ] 19
[ 2 , ] 25
[ 3 , ] 14
[ 4 , ] 15
[ 5 , ] 24
[ 6 , ] 17
[ 7 , ] 19
[ 8 , ] 27
[ 9 , ] 29
[ 10 , ] 25
> l i brary (MASS)
> ginv ( t (X) %% X ) %% ( t (X) %% Y )
[ , 1]
[ 1 , ] 16.3333333
[ 2 , ] 0.9212121
> lm( Y c ( 1: 10) )
Cal l :
lm( formula = Y c ( 1: 10) )
Coef f i ci ent s :
( I ntercept ) c ( 1: 10)
16.3333 0.9212
You can see from the comparison, both lm() and the matrix multiplication/inversion
method produce the same estimates for the intercept and the slope coefcient. If you
were to make Z <Y mean(Y) (e.g., standardize it for mean zero), you could have the X ma-
trix without the column for the interscept (
0
= 0) and you could get the same estimate
for the slope coefcient,
1
.
Eigen Decompositions
An eigenvalue/eigenvector decomposition is a magical property of matrices that can
only be appreciated by some experience in matrix algebra. However, we will be using
them in the next section so it seems there is a need to introduce them here. Start by
Biological Data Analysis Using R
130 CHAPTER 8. MATRIX ANALYSIS
considering the square (kxk) matrix X and the identity matrix (I) in the characteristic
equation |AI| = 0.
Using the matrix:
> A < matrix ( c( 1 , 6 , 3 , 4) , nrow=2)
> A
[ , 1] [ , 2]
[ 1 , ] 1 3
[ 2 , ] 6 4
The eigenvalues for the matrix are given by solving the characteristic formula:
0 = |AI| (8.1)
=

_
1 3
6 4
_

_
1 0
0 1
_

_
1 3
6 4
_

= (1 )(4 ) 18
=
2
5 14
If we solve for we see that possible values are 7 and 2. These are called the eigenvalues
of the matrix A.
Each eigenvalue has an associated eigenvector such that:
Ax = x
Where x is a vector (e.g, a matrix with only one column) that is matched to each of
the k eigenvalues. The equation above is called the characteristic equation for the right
eigenvector and a left eigenvector exists and has the form xA = x. From both of these,
we need to solve for x. Starting with the largest eigenvalue,
1
= 7, we have:
_
1 3
6 4
_ _
e
1
e
2
_
=
1
_
e
1
e
2
_
(8.2)
If we multiply these out, we get the following equations:
1e
1
+ 3e
2
= 7e
1
6e
1
+ 4e
2
= 7e
2
And here we have two equations in two variables and can easily solve for the values
of e
1
and e
2
and these values dene the eigenvector v
1
= [e
1
, e
2
] that is linked to the
eigenvalue
1
. We can do the same for the second vector (which I will let you play with in
those boring weekend hours where you are wishing that you had some really cool math
problem to solve).
Biological Data Analysis Using R
8.1. MATRICES IN R 131
It is important to point out here that the values for v
1
can be scaled. As you look at the
equations above we can solve for the components and nd that e
1
=
e
2
2
. There are a lot
of values for e
1
and e
2
that make this statement true. And if we think about the vector
v
1
= [e
1
, e
2
] as a project away from the origin a distance of e
1
on one axis and e
2
on a
second orthogonal axis it may make a bit more sense. There are several vectors that will
point in a direction that will intersect the point (e
1
, e
2
) all of which are the same except
for a scaling factor. This is graphically shown in Figure 8.1.2 with two vectors pointing
in the same direction but with different lengths.
Figure 8.1: Image depicting two vectors v
red
= [4, 2] and v
blue
= [2, 1] that are projecting in the
same direction but have different magnitudes.
The reason I bring this up is that it is common for routines that calculate vectors, such
as we are doing here for the eigenvector decomposition, to scale the vectors such that
their lengths are set to some normalizing constant such as 1. As a result, if you solve for
v
1
and then check it below with the eigen() function you may not get the same values but
if you were to plot the vectors, the lines away from the origin would be pointing in the
same direction.
There are some interesting properties of eigenvalues and eigenvectors.
If the original matrix is symmetric (actually non-negative semi-denite but whose
Biological Data Analysis Using R
132 CHAPTER 8. MATRIX ANALYSIS
watching), the original matrix A =

k
i=1

i
e
i
e

i
. This is called the spectral decompo-
sition of the matrix A.
The product of the eigenvalues is equal to the determinant of the original matrix
(e.g.,

k
i=1

i
= |A|).
The sum of the eigenvalues is equal to the trace of the matrix (e.g.,

k
i=1
n
i

i
= tr(A)
where n
i
is a
If it is possible to invert A then the eigenvalues of A
1
will be the inverse of the
eigenvalues of A (e.g,. they will be
1
i
.
The eigenvectors of A and A
1
are identical.
R has a eigen() function that takes a square matrix and returns the eigen values and
eigenvectors as a list. Here is an example using our little friend the A matrix we touched
on above.
> A
[ , 1] [ , 2]
[ 1 , ] 1 3
[ 2 , ] 6 4
> rootsOfA < eigen ( A)
> rootsOfA
$values
[ 1] 7 2
$vectors
[ , 1] [ , 2]
[ 1 , ] 0.4472136 0.7071068
[ 2 , ] 0.8944272 0.7071068
Baring the possibility that I actually just copied and pasted the results from eigen() into
the discussion above on v
i
= [e
1
, e
2
], the answer looks like it should.
8.2 Stage-Classied Matrix Models
Stage-classied matrix models are concerned with understanding the processes that in-
uence the persistence of populations. These models tacitly assume that the continuum
of life histories for a species can be partitioned into discrete stages and that a census
of individuals in a population can be performed wherein we can tally the number of
individuals in each of these discrete stages. Some species lend themselves to stage-
classication better than others and the distinctions on how to go about dening stages
is best left to another course. Here we are going to introduce the notation of a matrix
model in R and then perform some analyses on these models. This Chapter is intended
to only whet your appetite a bit on matrix models and for those that are interested, you
should seek out another course or at least read a good text such as Caswell (2001).
8.2.1 Transition Matrices & Census Vectors
For the sake of discussion, lets assume that we are working with a plant, Grenus growii,
that has the following four different distinct life stages. Moreover, from our vast knowl-
Biological Data Analysis Using R
8.2. STAGE-CLASSIFIED MATRIX MODELS 133
edge of this organism, we have the accompanying information about the way this species
proceeds through life stages.
Seed The seed stage lasts a single time step (e.g., there is no persistent seed bank) and
only 50% of the seeds actually germinate, the others are either eaten or rot.
Seedling The seedling stage is a non-reproductive stage and herbivory removes 20%
of the individuals that get into this stage and the remaining individuals become
juveniles.
Juvenile The juvenile stage is the rst reproductive stage and on average each juvenile
produces 1.3 offspring. Depending upon the habitat the juvenile is located in, half
move on to the next stage and a quarter stay as a juvenile. The remining ones are
eaten.
Adult The nal adult stage is where most of the reproduction happens with each indi-
vidual producing an average of 3.1 offspring. Half of the adults persist in the adult
stage from one time step to the next.
A diagram of this ctions species is shown in Figure 8.2.
Figure 8.2: The A graphical depiction of the life history stages in the ctitious plant Grenus growii
Here each of the spheres in this image represent a stage. The arrows between the stages
depict either fertility estimates (labeled f
X
) when they point back to the seed stage, or
transitions (labeled p
XY
signifying the probability that an individual proceeds to stage
X from stage Y . From the description we have above, we can associate values with
this particular life history diagram with particular parameters. In Table 8.1 I show the
parameters for each of the variables listed.
These parameters can now be put into a transition matrix
3
, A, that has a particularly
strict form.
A =
_

_
f
1
f
2
f
3
f
4
p
21
p
22
p
23
p
24
p
31
p
32
p
33
p
34
p
41
p
42
p
43
p
44
_

_
(8.3)
3
Actually this is not a transition matrix as it does not sum to 1 rather it is a Leslie matrix but I think I
can get away with generalizing the term a bit here.
Biological Data Analysis Using R
134 CHAPTER 8. MATRIX ANALYSIS
Table 8.1: Table of life history values separated into A Fertility estimates (the fX items) and B
transition probabilities depicting the movement between stages and within stages.
A. Fertility Estimates
Stage Parameter Value
Seed f
1
0
Seeding f
2
0
Juvenile f
3
1.3
Adult f
4
3.1
B. Transition probabilities.
Transition Parameter Value
Seed Seedling p
21
0.5
Seedling Juvenile p
32
0.8
Juvenile Adult p
43
0.5
Juvenile Juvenile p
33
0.25
Adult Adult p
44
0.5
The items in the matrix are partitioned into two components, the top row records the
fecundity values, f
X
, and the second and remaining rows depict the probabilities of
transition, p
XY
. Inserting the observed values into this matrix gives us:
A =
_

_
0 0 1.3 3.1
0.5 0 0 0
0 0.8 0.25 0
0 0 0.5 0.5
_

_
(8.4)
In R we can create this matrix using the following code:
> A < matrix ( 0 , nrow=4, ncol =4)
> A[ 1 , 3] < 1.3
> A[ 1 , 4] < 3.1
> A[ 2 , 1] < 0.5
> A[ 3 , 2] < 0.8
> A[ 3 , 3] < 0.25
> A[ 4 , 3] < 0.5
> A[ 4 , 4] < 0.5
> A
[ , 1] [ , 2] [ , 3] [ , 4]
[ 1 , ] 0.0 0.0 1.30 3.1
[ 2 , ] 0.5 0.0 0.00 0.0
[ 3 , ] 0.0 0.8 0.25 0.0
[ 4 , ] 0.0 0.0 0.50 0.5
The entries in this matrix have some rather special properties if we put the values into
it as directed.
Biological Data Analysis Using R
8.2. STAGE-CLASSIFIED MATRIX MODELS 135
Intrinsic Growth Rate
The Euler-Lotkas integral equation for the instantaneous grow rate, r, is well known to
most biologists (...) and has the form:
1 =
_

0
l(x)m(x)e
rx
dx
where the term l(x) is the fraction of reproductive individuals surviving to x, m(x) is the
fertility rate of individuals at x, and r is the growth. The r component here is the part
that we are interested in looking at because:
r =
_
_
_
< 1 : Populationsizedecayingexponentially
= 1 : Stablesizethroughtime
> 1 : Populationsizeincreasingexponentially
We can provide an estimate of r using an eigenvalue decomposition of the transition
matrix A. Due to the way the matrix is set up, the largest non-imaginary eigenvalue of
the matrix (
1
as dened in 8.1.2) is equal to r. So, once the matrix A is entered into R
, we can nd the growth parameter as:
> A
[ , 1] [ , 2] [ , 3] [ , 4]
[ 1 , ] 0.0 0.0 1.30 3.1
[ 2 , ] 0.5 0.0 0.00 0.0
[ 3 , ] 0.0 0.8 0.25 0.0
[ 4 , ] 0.0 0.0 0.50 0.5
> eigen ( A)
$values
[ 1] 1.2075472+0.0000000i 0.0067844+0.8194141i 0.00678440.8194141i
[ 4] 0.4439783+0.0000000i
$vectors
[ , 1] [ , 2] [ , 3] [ , 4]
[ 1 , ] 0.8603823+0i 0.7490103+0.0000000i 0.7490103+0.0000000i 0.4753001+0i
[ 2 , ] 0.3562521+0i 0.00378390.4570089i 0.0037839+0.4570089i 0.5352740+0i
[ 3 , ] 0.2976372+0i 0.4052283+0.1306829i 0.40522830.1306829i 0.6170499+0i
[ 4 , ] 0.2103303+0i 0.1682952+0.1431813i 0.16829520.1431813i 0.3268348+0i
here we can see that
1
is not a complex number (the +0.0000000i part tells us that) even
though there are some complex eigenvalues (roots) of this matrix. Moreover, it suggests
that the overall behavior of this transition matrix is to increase overall population size
with an instantaneous rate of r 1.2.
The particular values of will determine the overall long term behavior of the population.
Essentially as time increases t : 0 , the impact of is determined by raising it to
higher and higher powers. Figure 8.3 shows the projected impact on population growth
rate as a function to two values for
red
= 0.8 and
blue
= 1.2.
Biological Data Analysis Using R
136 CHAPTER 8. MATRIX ANALYSIS
Figure 8.3: Effects of the instantaneous growth rate as a function of time for both exponential
growth (
blue
= 1.2) and exponential decay (
red
= 0.8).
Stable Stage Distribution
The values in A also contain information on the relative proportion of individuals that
will be in each stage class as the population stabilizes into a steady state (either growth,
stable, or declining). This information is contained in the eigenvector that is associated
with
1
. From the output above we see that:
> ssd < as . numeric ( eigen ( A) $vectors [ , 1] )
> ssd
[ 1] 0.8603823 0.3562521 0.2976372 0.2103303
> sum( ssd)
[ 1] 1.724602
> ssd < ssd / sum( ssd)
> ssd
[ 1] 0.4988875 0.2065706 0.1725831 0.1219587
> sum( ssd)
[ 1] 1
Here you see that the eigenvalues are scaled to unit size (e.g., t(e i ) %%e i = 1) as mentioned
above which results in a total sum of the vector of sum(ssd) = 1.724602. If we are interested in
Biological Data Analysis Using R
8.2. STAGE-CLASSIFIED MATRIX MODELS 137
nding the proportion of the population that is in each stage then we need to standardize
the vector so that the sum(ssd) = 1 and this is done by dividing every element by the total.
As a result, ssd suggests that at equilibrium there should be 49% of the individuals as
seeds, 21% as seedlings, 17% as juveniles and 12% as adults.
We will return to these numbers and the estimate for r in the next subsection when we
iterate the data manually.
Bar Plots
As in the previous example, we determined the stable age distribution to estimate the
proportion of the total population that is in each group. Graphically, this material could
be depicted as a bargraph and since we havent covered how to make bar graphs yet,
this is as good a time as any...
There is an option in the normal plot () function, type="h" that will kind of plot bars of your
data to a gure. Actually, these are high density lines and not real bar plots. This is
what I used to make Figure 4.2 and at that time it got the job done correctly, but a true
bar plot is something that looks a bit different than those lines.
R provides the function barplot() that takes a vector of heights and produces a general
barplot for you. Without modications, the function barplot() does not produce a very
interesting plot in my opinion. However, there are several optional arguments that can
be used to create a more informative graphic. They include:
names.arg a vector of names that you can have placed on the xaxis below the bars
width controls the width of the bars.
space controls the amount of area between the bars with a value of zero having the
bars touch and positive numbers equal to that number of bar width (e.g., space=2
plots a bar and then 2 bar widths before the next bar shows up).
horiz is a logical ag that will plot the bars horizontally instead of vertically.
col can pass as a single color or a vector of colors which are used to color the bars.
ylim can adjust the limit of the yaxis as in normal plotting routines.
xlab \& ylab Labels for the x and yaxes.
Using the data from
1
in the previous section, we can plot the data as (shown in Figure
8.4.
> ssd
[ 1] 0.4988875 0.2065706 0.1725831 0.1219587
> barplot ( ssd)
> barplot ( ssd , ylim=c ( 0 , 1) , xlab="Stage" , ylab="Proportion of Individuals" ,
+ names. arg=c ( "Seed" ,"Seedling" ,"Juvenile" ,"Adult") , col =c ( "red" ,"blue" ,"green" ,"yellow" ) )
The barplot() function can also be used to create stacked graphs 8.5
To create this example, I used the following code which as t
Biological Data Analysis Using R
138 CHAPTER 8. MATRIX ANALYSIS
Figure 8.4: Examples of two different calls to the plotting function barplot(). The parameters used
to create these plots is given in the R code.
> x < matrix ( runi f ( 9) , nrow=3)
> x
[ , 1] [ , 2] [ , 3]
[ 1 , ] 0.2355922 0.396869276 0.5674993
[ 2 , ] 0.7247734 0.001881527 0.9215767
[ 3 , ] 0.4625868 0.767329832 0.6408461
> barplot ( x, names. arg=c ( "Control" ,"A" ,"B") , xlab="Treatments" , ylab="Value" ,
+ legend=c ( "Category A" ,"Category B" ,"Category C" ) )
These stacked plots treat every column of data as a single bar and the order in which the
rows are presented is the order in which the stacking occurs. You can standardize the
plot to all have the same height by dividing each column by that columns sum providing
a proportional barplot.
Biological Data Analysis Using R
8.2. STAGE-CLASSIFIED MATRIX MODELS 139
Figure 8.5: Example of a stacked bar plot with multiple categories represented in each Treatment.
8.2.2 Projecting Stage Sizes
In this matrix model we have been playing with, the census count of individuals in
each of the four stages can be represented by the vector n and in R as a matrix whose
dimensions are (4x1). Assuming that I start with 12 seeds, 34 seedlings, 21 juveniles, and
12 adults, the vector can be depicted as:
> n < matrix ( c(12,34,21,12))
> n
[ , 1]
[ 1 , ] 12
[ 2 , ] 34
[ 3 , ] 21
[ 4 , ] 12
Using this notation, we can predict what the number of individuals in the next time slice
will be given A and n as:
n
t+1
= An
t
Biological Data Analysis Using R
140 CHAPTER 8. MATRIX ANALYSIS
> A
[ , 1] [ , 2] [ , 3] [ , 4]
[ 1 , ] 0.0 0.0 1.30 3.1
[ 2 , ] 0.5 0.0 0.00 0.0
[ 3 , ] 0.0 0.8 0.25 0.0
[ 4 , ] 0.0 0.0 0.50 0.5
> n
[ , 1]
[ 1 , ] 12
[ 2 , ] 34
[ 3 , ] 21
[ 4 , ] 12
> A %% n
[ , 1]
[ 1 , ] 64.50
[ 2 , ] 6.00
[ 3 , ] 32.45
[ 4 , ] 16.50
So after one generation, we can see that the number of seeds, juveniles, and adults all
increased but the number of seedlings decreased. If we look at the next time step, we
see that:
n
t+2
= An
t+1
= AAn
t+1
= A
2
n
t
And in general the vector of stage sizes at any arbitrary time step can be written as:
n
t
= A
t
n
0
(8.5)
Lets make a matrix of n values for time 1 11 in R and calculate the number of individ-
uals in each stage for each time step. I use 11 here because the matrix starts counting
at column 1 which will correspond to our time t = 0 so when t = 10 the column will be
11. Lets also set the rst column (our t = 0) equal to the census population size we were
using above.
> N < matrix ( 0 , nrow=4, ncol =11)
> N
[ , 1] [ , 2] [ , 3] [ , 4] [ , 5] [ , 6] [ , 7] [ , 8] [ , 9] [ , 10] [ , 11]
[ 1 , ] 0 0 0 0 0 0 0 0 0 0 0
[ 2 , ] 0 0 0 0 0 0 0 0 0 0 0
[ 3 , ] 0 0 0 0 0 0 0 0 0 0 0
[ 4 , ] 0 0 0 0 0 0 0 0 0 0 0
> N[ , 1] < n
> N
[ , 1] [ , 2] [ , 3] [ , 4] [ , 5] [ , 6] [ , 7] [ , 8] [ , 9] [ , 10] [ , 11]
[ 1 , ] 12 0 0 0 0 0 0 0 0 0 0
[ 2 , ] 34 0 0 0 0 0 0 0 0 0 0
[ 3 , ] 21 0 0 0 0 0 0 0 0 0 0
[ 4 , ] 12 0 0 0 0 0 0 0 0 0 0
Now, for time steps 1 10 (and in the matrix N columns 2 11) we will use the equation
8.5 to calculate the number of individuals in each group.
Biological Data Analysis Using R
8.2. STAGE-CLASSIFIED MATRIX MODELS 141
> t < 1
> N[ , ( t +1) ] < A %% N[ , t ]
> t < t + 1
> N
[ , 1] [ , 2] [ , 3] [ , 4] [ , 5] [ , 6] [ , 7] [ , 8] [ , 9] [ , 10] [ , 11]
[ 1 , ] 12 64.50 0 0 0 0 0 0 0 0 0
[ 2 , ] 34 6.00 0 0 0 0 0 0 0 0 0
[ 3 , ] 21 32.45 0 0 0 0 0 0 0 0 0
[ 4 , ] 12 16.50 0 0 0 0 0 0 0 0 0
> t
[ 1] 2
OK, here I am going to do something that saves some typing (you can use the up cursor
key to repeat the last entry you typed in the R interpreter and I will use this to make my
life a bit easier). I have dened the variable t such that it will be used to indicate which
column of the matrix to use (the ( t+1) part) as well as the exponent to the matrix A. Then
I will increment the variable t by one and redo it again and again until Ive lled up the
columns of N.
In the following code examples, I show that you can use a semicolon (;) to put more than
one command on a line. Again, I combine the assignment of counts to the appropriate
column of N and then update the counter variable t each time through until all eleven
columns are full. In Chapter 11 you will learn how to use a loop to do this much easier
but until then using the up cursor key in the R interpreter is good enough.
> N[ , ( t +1) ] < A %% N[ , t ] ; t < t + 1
> N
[ , 1] [ , 2] [ , 3] [ , 4] [ , 5] [ , 6] [ , 7] [ , 8] [ , 9] [ , 10] [ , 11]
[ 1 , ] 12 64.50 93.3350 0 0 0 0 0 0 0 0
[ 2 , ] 34 6.00 32.2500 0 0 0 0 0 0 0 0
[ 3 , ] 21 32.45 12.9125 0 0 0 0 0 0 0 0
[ 4 , ] 12 16.50 24.4750 0 0 0 0 0 0 0 0
> N[ , ( t +1) ] < A %% N[ , t ] ; t < t + 1
> N[ , ( t +1) ] < A %% N[ , t ] ; t < t + 1
> N[ , ( t +1) ] < A %% N[ , t ] ; t < t + 1
> N[ , ( t +1) ] < A %% N[ , t ] ; t < t + 1
> N[ , ( t +1) ] < A %% N[ , t ] ; t < t + 1
> N[ , ( t +1) ] < A %% N[ , t ] ; t < t + 1
> N[ , ( t +1) ] < A %% N[ , t ] ; t < t + 1
> N[ , ( t +1) ] < A %% N[ , t ] ; t < t + 1
> N
[ , 1] [ , 2] [ , 3] [ , 4] [ , 5] [ , 6] [ , 7] [ , 8]
[ 1 , ] 12 64.50 93.3350 92.65875 95.68719 131.93725 168.77519 193.20372
[ 2 , ] 34 6.00 32.2500 46.66750 46.32937 47.84359 65.96862 84.38759
[ 3 , ] 21 32.45 12.9125 29.02813 44.59103 48.21126 50.32769 65.35682
[ 4 , ] 12 16.50 24.4750 18.69375 23.86094 34.22598 41.21862 45.77316
[ , 9] [ , 10] [ , 11]
[ 1 , ] 226.86065 281.25553 343.80907
[ 2 , ] 96.60186 113.43032 140.62776
[ 3 , ] 83.84928 98.24381 115.30521
[ 4 , ] 55.56499 69.70713 83.97547
So this is a large number of values here so lets plot this out to see what the stages do
as we go through 10 time steps. The code used to produce the image in Figure 8.2.2
is:
> pl ot ( 1: 11 ,N[ 1 , ] , xlab="" , ylab="" , axes=F, bty="n" , col ="red" , ylim=ylim, type="l" , lwd=2)
> par ( new=T)
> pl ot ( 1: 11 ,N[ 2 , ] , xlab="" , ylab="" , axes=F, bty="n" , col ="blue" , ylim=ylim, type="l" , lwd=2)
Biological Data Analysis Using R
142 CHAPTER 8. MATRIX ANALYSIS
> par ( new=T)
> pl ot ( 1: 11 ,N[ 3 , ] , xlab="" , ylab="" , axes=F, bty="n" , col ="green" , ylim=ylim, type="l" , lwd=2)
> par ( new=T)
> pl ot ( 1: 11 ,N[ 4 , ] , xlab="t" , ylab="Number of Individuals" , axes=T, bty="n" , col ="pink" ,
+ ylim=ylim, type="l" , lwd=2)
> legend(2,350, c ( "Seed" ,"Seedling" ,"Juvenile" ,"Adult") , col =c ( "red" ,"blue" ,"green" ,"pink") ,
+ lwd=2, bty="n")
I use the par(new=T) to overlay the lines on a single graph (see 4.1.1 for more on this). I
also turn off the labels and axes for the rst three plots because if you plot them over
and over again, they look too dark on the graphic (think printing the same line on top of
itself numerous times). On the last one, I set the labels for the axes and the turn on the
axes. Also included is the code I used to add the legend to the image. See ?legend for a
complete discussion of the options that you can provide to this function.
Figure 8.6: Size of the four stage classes through time.
We can check some of the values that we estimated directly from A using the eigen
decomposition by looking at the numbers in the matrix N. First, the growth rate we
estimated from the rst eigenvalue
1
1.2 looks pretty close to that estimated from the
raw counts.
> eigen ( A) $values [ 1]
Biological Data Analysis Using R
8.2. STAGE-CLASSIFIED MATRIX MODELS 143
[ 1] 1.207547+0i
> sum(N[ , 11] ) / sum(N[ , 10] )
[ 1] 1.215202
And the proportion of individuals in each class was estimated by standardizing the rst
eigenvalue v
1
= v
1
/

4
i=1
v
1i
is pretty close to what we see in N (and I throw in the rst
census so that you dont think I put values in there that were already pretty close).
> N[ , 1] / sum(N[ , 1] )
[ 1] 0.1518987 0.4303797 0.2658228 0.1518987
> N[ , 11] / sum(N[ , 11] )
[ 1] 0.5028525 0.2056811 0.1686445 0.1228219
> ssd
[ 1] 0.4988875 0.2065706 0.1725831 0.1219587
If we were to iterate this a bit longer you would see that the brute force method of
getting the population growth rate and the stable age distributions converge towards
what was estimated. In fact, Figure 8.2.2 shows the mean absolute deviation (MAD)
representing the differences between the distribution of individuals in each stage from
the predicted stable stage distribution (ssd) we calculated earlier. As you can see, it
approaches the expected values pretty quickly.
Biological Data Analysis Using R
144 CHAPTER 8. MATRIX ANALYSIS
Figure 8.7: Differences in estimated proportions of individuals in each stage from what was
expected through time.
8.3 Useful Functions
The following functions were introduced in this chapter and you will be required to use
them for the exercises. To get more information on any of these functions, use the R
help system.
%
*
% Binary operator to performmatrix multiplication. An example would be X \%\\% Y.
as.matrix(x) Coerces the variable x into the data type matrix.
barplot(x) Creates a barplot of the values in x.
det(x) Calculates, if possible, the determinant of the matrix in x.
diag(x) Returns the diagonal (e.g., those entries whose row and column indices are
equal) of the matrix in x.
dim(x) Returns the dimensions of the matrix x (e.g., the number of rows and columns).
Biological Data Analysis Using R
8.3. USEFUL FUNCTIONS 145
eigen(x) Returns the eigenvalue/eigenvector pairs for the matrix in x as a list. Values
are sorted in descending numerical order and vectors are scaled to unit length.
ginv(x) Attempts to calculate the generalized inverse of x.
legend(x,y,c) Creates a legend for the plot at the coordinates (x, y) with the entries
in c.
matrix(x) Creates a new instance of the matrix data type of the values in x. You will
probably need to specify nrow and ncol to set the proper size for the matrices.
read.table(x) Reads the le x into memory. See ?read.table for the copious amounts of
additional parameters that may be needed as well as Chapter 3.
t(x) Returns the transpose of the matrix in x (e.g., reverses the row and column
indices)
Biological Data Analysis Using R
146 CHAPTER 8. MATRIX ANALYSIS
8.4 Exercises
The following exercises are meant to help you understand the items presented in this
Chapter.
1. In considering the instantaneous growth rate r, it was mentioned that
1
> 0 and this
is what you will nd in most cases. However, it is possible to get values of < 0.
For the following values of make a graph of t vs.
t
as shown in Figure 8.3 and
describe the behavior of the population if these were the real values of r.
(a) 1 <
1
< 0.
(b)
1
< 1.
2. Create a matrix of random numbers using the runif () function and make a barplot of
the values. What happens when you pass the optional argument beside=T?
3. Standardize the columns of data in the matrix from the previous example so that the
sum of each column is equal to 1. Replot this with using the function barplot() as done
for Figure 8.5 with the beside=F option. How does standardizing each row inuence
the display of the plot?
Biological Data Analysis Using R
Chapter 9
Working With Strings
While the majority of biological data is numeric in nature there are still several important
reasons to be able to manipulate character-based information. For example, you may be
downloading all the references from a online database such as WebOfScience and want
to mine the abstracts for metadata. You may also be interested in working with sequence
data which consists of mostly text information. In this relatively short chapter we will
learn about how we can work with string in data in R and look at a few examples using
genetic sequences.
In this chapter, you will focus on the following topics:
Learn how to work with string data to perform tasks such as parsing, searching,
and replacement.
Learn how to access sequence based data and pre-process it for importation into R
Learn how to create genetic distance matrices.
Construct Neighbor-Joining trees and display them in R
9.1 Parsing Text Data
At a most basic level you need to understand that character data in R is treated as a
single token in the same way that integer and numeric data is treated. For example,
consider the following code:
> x < c ( "bob" ,"mary" ,"johnathan")
> length ( x )
[ 1] 3
> x < "George Stephen Sr."
> length ( x )
[ 1] 1
> x < c( 1 , 2 , 3)
> length ( x )
[ 1] 3
> x < 3
> length ( x )
[ 1] 1
147
148 CHAPTER 9. WORKING WITH STRINGS
9.1.1 Finding Lengths of Character Sequences
So R treats a character data type, independent of the length of the items in the variable,
as a single entry. Once we understand this then the rest of this Chapter really begins to
take shape and make sense.
So, if R thinks that the everything between a pair of quotes is a single instance of a
character data type then how do we gure out how many letters are contained between
the quotes? The answer here is the function nchar().
> x < "George Stephen Sr."
> nchar ( x )
[ 1] 18
Another commonly used function for dealing with strings is the strsplit () function. This
function takes the string of characters that you are interested in splitting as well as the
character you want to split it on and returns the chunks as a list. This returning-as-a-
list behavior is kind of a pain in the butt so at the same time I introduce this function I
will also show the unlist() function at the same time.
1
> partsOfName < unl i st ( st r spl i t ( x, " ") )
> partsOfName
[ 1] "George" "Stephen" "Sr."
> nchar ( partsOfName )
[ 1] 6 7 3
Here is another example as to how we may go about cycling through a set of words in
a phrase and doing some operation on them. The rst sentence from the rst chapter
of Darwins The Origin Of Species is, WHEN we look to the individuals of the same
variety or sub-variety of our older cultivated plants and animals, one of the rst points
which strikes us, is, that they generally differ much more from each other, than do
the individuals of any one species or variety in a state of nature. While this is a very
interesting sentence, we are going to use it to show you how to break down the sentence
into an array of words and then tally the number of times each word is used.
We begin by making the sentence all lowercase and without punctuation because the
simple matching procedure would consider When different than when and the strsplit ()
function will cut up the string on the spaces (that I what I will tell it to do)
> phrase < "when we look to the individuals of the same variety or sub-variety of our older "
+ "cultivated plants and animals one of the first points which strikes us is that they "
+ "generally differ much more from each other than do the individuals of any one species or "
+ "variety in a state of nature"
> wordList < unl i st ( st r spl i t ( phrase , " " ) )
> tabl e ( wordList )
wordList
a and animals any cul ti vated di f f er
1 1 1 1 1 1
do each f i r s t from general l y in
1 1 1 1 1 1
i ndi vi dual s i s look more much nature
2 1 1 1 1 1
of ol der one or other our
5 1 2 2 1 1
1
This function takes a list and turns the items in it into a vector which is easier to work with.
Biological Data Analysis Using R
9.1. PARSING TEXT DATA 149
plants points same species state stri kes
1 1 1 1 1 1
subvari et y than that the they to
1 1 1 4 1 1
us vari et y we when which
1 2 1 1 1
9.1.2 Extracting Substrings
It is not possible to use the normal subscripting approaches to access the individual
characters within strings because R treats the entire sequence of characters between
the quotation marks as a single item. However, you can extract internal components of
a string by using the substring() function.
> phrase < "A Goat, that was sitting next to the gentleman in white, shut his eyes and said
+ in a loud voice, She ought to know her way to the ticket-office, even if she doesnt know
+ her alphabet! "
> substring ( phrase , 34, 70)
[ 1] "the gentleman in white, shut his eyes"
> substring ( phrase , 98)
[ 1] "She ought to know her way to the ticket-office, even if she doesnt know her alphabet! "
The function takes the string to be searched and the starting and ending locations in
the string and returns the characters in between. If you do not provide an ending
number, it will return all the characters up to the end. This is a shorthand way of saying
substring( phrase, x, nchar(phrase) ).
It is also possible to use vector notation in pulling out substrings by passing vectors to
the start and end arguments.
> startPosi ti ons < c(34,3,58,172,67)
> endPositions < c(36,6,61,174,70)
> substring ( phrase , startPosi ti ons , endPositions )
[ 1] "the" "Goat" "shut" "her" "eyes"
9.1.3 Concatenating Strings
Vectors of character data can be concatenated to form a single long string. This is very
helpful in creating labels for graphs that have to include the value of a variable and
for times when you need to open a lot of data les that have a predictable le naming
scheme. In R string concatenation is accomplished using the paste() function.
> stri ngVector < substring ( phrase , startPosi ti ons , endPositions )
> stri ngVector
[ 1] "the" "Goat" "shut" "her" "eyes"
> paste ( stringVector , col l apse=" ")
[ 1] "the Goat shut her eyes"
> paste ( stringVector , col l apse="|")
[ 1] "the|Goat|shut|her|eyes"
Biological Data Analysis Using R
150 CHAPTER 9. WORKING WITH STRINGS
9.1.4 Matching & Substitution
The nal tasks we will look into in this section on string operations are matching and
substitutions. There are a lot of times when the ability to see if a particular set of
strings has a specic substring within it. This is the realm of matching and is primarily
accomplished by the functions grep() and regexpr(). This last function allows you to use
what are called Regular Expressions (RE) to scan through string. While this is a very
powerful method for pattern matching and is something that if you are going to do any
extensive work with strings should know, I am not going to cover it in this Chapter. In
fact, it probably needs its own chapter and perhaps in a future version of this text I will
include it. For those of you who work with string data on a regular basis, look up the
regexpr function and have at it, it will make your life easier. For the rest of us, lets dig into
grep for a little light matching exercises.
The grep function takes a pattern that you are looking for and a string that you want to
look into. A simple example would be:
> x < "The quick brown fox jumped over the candle stick"
> grep ( "fox" , x )
[ 1] 1
> any( grep ( "fox" , x ) )
[ 1] TRUE
> any( grep ( "o" , x ) )
[ 1] TRUE
> any( grep ( "dog" , x ) )
[ 1] FALSE
In general, the grep function returns an integer indicating that the string either has or
does not have a copy of the pattern in it. I wrapped the grep function here inside the
any() function because it will take either a single argument or a vector of arguments and
return a logical value.
It is also possible to substitute values in a string with new items. There are two functions
that perform string substitutions, sub and gsub. Both of these functions take at least three
arguments;
1. A pattern to match,
2. The string to replace the matched pattern with, and
3. The string to search within.
The sub function replaces the rst occurrence of the pattern whereas gsub replaces all of
them (the g stands for global).
> x < "The quick brown fox jumped over the candle stick with all the kings men."
> sub( "the" ,"THE" , x )
[ 1] "The quick brown fox jumped over THE candle stick with all the kings men."
> gsub( "the" ,"THE" , x )
[ 1] "The quick brown fox jumped over THE candle stick with all THE kings men."
> gsub( "the" ,"THE" , x, ignore . case=T)
[ 1] "THE quick brown fox jumped over THE candle stick with all THE kings men."
Both of these functions have optional arguments, the most common one of which is
ignore.case option that allows the searching and replacing to either take into consideration
Biological Data Analysis Using R
9.1. PARSING TEXT DATA 151
the case of the letters when matching or not.
9.1.5 Slightly More In Depth Examples: Genetic Sequence Analyses
Genetic sequences are essentially long character strings and R has a few different li-
braries available to you for the analysis of sequence data. I am not going to get into what
a genetic sequence is, if you do not already know about it then you probably should not
be calling yourself a biologist... In this section, we will:
1. Briey discuss how we go about getting DNA sequence data
2. Learn how to align sequences
3. Import sequence aligned sequence data into R
4. Create a distance matrix from the sequences
5. Use R to estimate a Neighbor-Joining tree from the sequence data
Getting DNA Sequence Data
The mother of all sequence repositories that you can access (without actually doing the
sequencing yourself) is the NCBI web database located at http://www.ncbi.nlm.nih.gov/
Here you can run database queries based upon taxa, genes, groups, or whatever. The
basic results of a search are given as an annotation (just below). This annotation has
three parts,
1. The meta data in the top section that contains the locus denition, size, who found
it, references and a the taxonomy of the organism.
2. The FEATURES of the record that describe what is in the sequences (coding and
non-coding regions if known), some geographical and taxonomic information that
has been standardized (good for data mining and putting on a map) as well as the
translation of genetic sequence into amino acids if appropriate.
3. The ORIGIN which contains the raw sequence information.
An example of a record is given below
LOCUS FJ347583 278 bp DNA l i near INV 01JUL2009
DEFINITION Araptus attenuatus haplotype 5 muscle protein 20 (MP20) gene ,
part i al sequence .
ACCESSION FJ347583
VERSION FJ347583.1 GI:227345175
KEYWORDS .
SOURCE Araptus attenuatus
ORGANISM Araptus attenuatus
Eukaryota; Metazoa; Arthropoda ; Hexapoda; Insecta ; Pterygota ;
Neoptera ; Endopterygota ; Coleoptera ; Polyphaga ; Cucujiformia ;
Curculionidae ; Scolytinae ; Araptus .
REFERENCE 1 ( bases 1 to 278)
AUTHORS Garrick ,R.C. , Meadows,C. A. , Nason, J.D. , Cognato , A. I . and Dyer ,R. J.
TITLE Variable nuclear markers f or a Sonoran Desert bark beetl e , Araptus
attenuatus Wood ( Curculionidae : Scolytinae ) , with appl i cati ons to
rel ated genera
Biological Data Analysis Using R
152 CHAPTER 9. WORKING WITH STRINGS
JOURNAL Conserv . Genet . 10 ( 4) , 11771179 (2009)
REFERENCE 2 ( bases 1 to 278)
AUTHORS Garrick ,R.C. , Meadows,C. A. , Nason, J.D. , Cognato , A. I . and Dyer ,R. J.
TITLE Di rect Submission
JOURNAL Submitted (26SEP2008) Department of Biology , Vi rgi ni a
Commonwealth University , 1000 West Cary Street , Richmond, VA 23284,
USA
FEATURES Location/Qual i f i ers
source 1..278
/organism="Araptus attenuatus"
/mol type="genomic DNA"
/db xref ="taxon:634056"
/haplotype="5"
gene <1..>278
/gene="MP20"
/note="muscle protein 20; coding region not determined"
ORIGIN
1 ctaaaatcaa cacttccgga ggacaattta aattcatgga aaacatcaac aagtaagaaa
61 aaaataattt gacatgtaaa taatgtagag aaaattcata aacattccta t t t t t t at t g
121 at t t gt caat at t t agt t t g gaactaaact ctgacaatca attatacagg gtgacaattc
181 taat tacat t tccattcaat gccaactaga aatttcgtga aaaaaaaatt gt t t ct at gc
241 caaacatact gt t t t at aag at t t aat t cc agaaattt
//
Sequence Formats & Aligning Genetic Sequences
The format of the sequence data like this is a bit verbose but very informative. When
we work with sequence data we will use an abbreviated le format, the FASTA format,
to work with sequences. This format is very compact and as a result, it is rather easy to
use. In general, FASTA les are simple text les that have blocks of information for each
sequence. Each block contains a summary line that must begin with the greater than
character (>) and can be anything you like. It is common to put the accession numbers,
locus identier, taxonomy and other information into this line. The lines following the
summary line is the raw sequence. If you want to have more than a single taxon in a
le, you just put the next taxon block blow the previous one and continue. In general
they look like this (this is an excerpt from an example data set that you have in the class
folder):
>Pinus caribaea var . hondurensis
GGTTCAAGTCCCTCTATCCCCACCCAGGTTCGGTCCCGAAAGGAYTGATCTATCTTCTCCAATTCCATTG
GTTCGAATCCATTCTAATTTCTCGATTCTTTTACCTCGCTATTTTTTTTTTTTCATGAAGAGAAGAAATT
AGAACATGAATCTTTTCATCCATCTTATGACAAGTTGAGTTGATCTGTTAATAAGTTGATCATATGATCA
ATTTATTTTGTGATATATGATCTACATAGAATAGATTAGATCNTTTTTAAATTATTCAATTGCAGTCCAT
TTTTATCATATTAGTGACTTCCAGATCGAAAATAATAAAGATCATTCTAAAAACTAGTAAAAATACCTTT
TTACTTCTTTTTAGTTGACACAAGTTAAAACCCTGTACCAGGATGATCCACAGGGAAGAGCCGGGGATAG
CTCATTTGGTAAACCAAAGGACTGAAAATCCTCGTGTCACCAGTTCAAAT
>Pinus echinata
ACCCAGGTTCGTTCCCGAACGGATTGATCTATCTTCTCCAATTCCATTGGTTCGAATCCATTCTAATTTC
TCGATTCTTTTACCTCGCTATTTTTTTTTTTCATGAAGAGAAGAAATTAGAACATGAATCTTTTCATCCA
TCTTATGACAAGTTGAGTTGATCTGTTAATAAGTTGATCATATGATCAATTTATTTTGTGATATATGATC
TACATAGAATAGATTAGATCATTTTTAAATTATTCAATTGCAGTCCATTTTTATCATATTAGTGACTTCC
AGATCGAAAATAATAAAGATCATTCTAAAAACTAGTAAAAATACCTTTTTACTTCTTTTTAGTTGACACA
AGTTAAAACCCTGTACCAGGATGATCCACAGGGAAGAGCCGGGATAGCTCAGTTGGTAGAGCAGAGGACT
GAAAATC
When conducting analyses of genetic sequence data, it is important that you are con-
dent that all the sequences you have are of homologous portions of the genome. For the
Biological Data Analysis Using R
9.1. PARSING TEXT DATA 153
example I used here, I downloaded some genetic sequence data for a handful of conifers
in the family Pinaceae from the NCBI website. The sequences I was looking for is a com-
mon inter-genic spacer region between the genes encoding for tRNA-trnL and tRNA-trnF.
These sequences were between 390-470 base pairs in length and are in the le named
confiers.fasta in the folder for this chapter. I cleaned up the summary lines in this le so
it only has the genus and species names rather than all the other stuff. This makes it a
bit easier for you in the future when you interact with the data.
Before I played with these sequences, I ran an alignment on them to make sure we were
dealing with the matching sequences across taxa. There are many ways to do this and I
just used the online ClustalW server at http://align.genome.jp to align the sequences for
me. This is not something you want to do by hand and it is much better to let a computer
do some of the work for you. This algorithm aligns all the sequences and returns the
le in a clustal format. This is another text le but this time all the species have been
displayed in blocks with homologous sequence locations in the same text column. An
example of this is shown below with gaps (insertions/deletions) indicated as the dash
character ().
Pinus caribaea var . hondurensi CCCACCCAGGTTCGGTCCCGAAAGGAYTGATCTATCTTCTCCAATT
Pinus taeda ACCCAGGTTCGTTCCCGAACGGATTGATCTATCTTCTCCAATT
Pinus ponderosa ACCCAGGTTCGTTCCCGAACGGATTGATCTATCTTCTCCAATT
Pinus echinata ACCCAGGTTCGTTCCCGAACGGATTGATCTATCTTCTCCAATT
This le is also located in the folder for this chapter and is called conifers.aln and this is
the le we will be working with.
Getting Aligned Sequences Into R
R does not by default recognize sequence data as anything more elegant than a sequence
of characters. As a result, several people have developed libraries for you to use that have
a lot of general functionality to them. In this section, I am going to use the library ape. If
you do not have this library installed on your machine, see Appendix B for an overview
of the process.
I am assuming that you currently have the data le in a location that you can reach it
easily from within R . To load the aligned sequences into R type the following:
> l i brary ( ape )
> seqs < read . dna( "confiers.aln" , format="clustal")
> cl ass ( seqs )
[ 1] "DNAbin"
> summary( seqs )
23 DNA sequences in binary format stored in a matrix .
Al l sequences of same length : 526
Labels : Abies alba Abies kawakamii Abies vei t chi i Abies homolepis Larix potani ni i Cedrus atl anti ca . . .
Base composition:
a c g t
0.310 0.187 0.160 0.343
Biological Data Analysis Using R
154 CHAPTER 9. WORKING WITH STRINGS
There are several things that you can do with these aligned sequences. You can look for
motifs, examine CG content, etc. I will leave these options for you to play with later in
the exercises.
Constructing A Neighbor Joining Tree
To construct a Neighbor Joining (NJ) tree, we rst need to create a distance matrix that
estimates the distances between pairs of sequences that we have in our le. There are
several different kinds of distance metrics that you can use in the calculation of this
distance matrix (see ?dist.dna for more information on these). We will use the default value
which is Kimuras 2-parameter model called K90.
> D < di st . dna( seqs )
> cl ass (D)
[ 1] "dist"
> summary(D)
Min. 1st Qu. Median Mean 3rd Qu. Max.
0.00000 0.07252 0.09310 0.26890 0.15720 1.45700
The function dist.dna() takes as an argument a set of sequences that you have read in (the
must be of class DNAbin as shown above) and spits out the distance matrix. The distance
matrix, D, is a particular kind of matrix that holds the lower triangle of the pair-wise
distance calculations. If you print it out, you will get a whole lot of output as it prints
the taxa names for row and column headers.
Since D is a general distance matrix, we can look at the values in it. Figure 9.1 shows
a histogram of the distance values that have been estimated in D. From this we see that
there are several values that are low meaning that the sequences are very similar to each
other and then there are some that are 2-3 peaks that are larger suggesting some degree
of sequence divergence.
To create a NJ tree from these distances, we use the function nj () .
> njTree < nj (D)
> cl ass ( njTree )
[ 1] "phylo"
> summary( njTree )
Phylogenetic tree : njTree
Number of t i ps : 23
Number of nodes : 21
Branch lengths :
mean: 0.03838704
variance : 0.01999758
di stri buti on summary:
Min. 1st Qu. Median 3rd Qu. Max.
0.0009736 0.0000000 0.0004898 0.0150700 0.8610000
No root edge .
Fi rst ten t i p l abel s : Abies alba
Abies kawakamii
Abies vei t chi i
Abies homolepis
Larix potani ni i
Cedrus atl anti ca
Larix decidua
Cedrus deodara
Biological Data Analysis Using R
9.1. PARSING TEXT DATA 155
Figure 9.1: Histogram of distance estimates among all sequences using the K90 model of sub-
stitutions
Larix l ari ci na
Pinus roxburghii
No node l abel s .
This function take a distance matrix and returns a tree that is of the class phylo. We
can see that internally the variable njTree has some internal information that may be of
interest (e.g., branch lengths, etc) but the real way we can understand it is by looking at
a graphic of the tree that is produced. To do this, we use the plot () command and pass it
the njTree variable as plot(njTree).
2
The topology of the tree (Figure 9.2) is easy to interpret and it is quite obvious where
those very large distances shown in Figure 9.1 come from. From this topology we can
see that:
1. The Pinus species are generally together forming a polytomy that connects to the
2
You may be surprised by the utility of the plot function as it seems to know how to plot everything. Well
in actuality this function is simply a wrapper that takes whatever you pass to it and determines if the class
of the object you passed has its own plot command. For the tree, the native command is plot.phylo() and you
have to look up that command to see the available options for it.
Biological Data Analysis Using R
156 CHAPTER 9. WORKING WITH STRINGS
Figure 9.2: Neighbor joining tree based upon the trnL-trnF intergenic spacer sequences and the
K90 model of sequence evolution.
other genera in the family.
2. The Larix, Abies, and Cedrus for generally self contained groups.
3. The most divergent groups are the Picea and Keteleeria samples.
There is quite a bit more that can be done here but I think that is enough to get you on
the right track if you are interested in using R for some basic sequence analysis.
9.2 Producing Formatted Output
Often in the use of R there is a need to produce a particular kind of output from an
analysis of to display the contents of a particular variable. R does a pretty good job
itself, but it has some limitations. For example, you may want to print out a matrix of
values but only have 2 decimal places printed for each entry. Or you may want to export
a table of values as HTML so that I can copy and paste it into another program
Biological Data Analysis Using R
9.2. PRODUCING FORMATTED OUTPUT 157
9.2.1 Formatting Strings For Printing
format ( x, trim = FALSE, di gi t s = NULL, nsmall = 0,
j ust i f y = c ( "left" , "right" , "centre" , "none") ,
width = NULL, na. encode = TRUE, sci ent i f i c = NA,
big . mark = "" , big . i nt erval = 3,
small . mark = "" , small . i nt erval = 5,
decimal . mark = "." , zero . pri nt = NULL, drop0trai l i ng = FALSE, . . . )
9.2.2 Formatting Tables
A common type of format to be output to another format is tabular data. Tables are
common features of statistical analysis and as such you will nd it necessary to cut
a table out of R and paste it into a document in the same way that graphics can be
exported from R to be used in your manuscripts and reports.
For these examples, I will just created a matrix of values and add row and column names
using the functions rownames and colnames.
> x < matrix ( rnorm( 12) , nrow=3)
> x
[ , 1] [ , 2] [ , 3] [ , 4]
[ 1 , ] 0.1678067 0.8856766 0.3955881 0.7677516
[ 2 , ] 1.0302831 0.7392326 0.8333904 0.3235135
[ 3 , ] 0.4396607 1.7622323 0.8763023 0.6091688
> colnames ( x ) < c ( "Header A" , "Header B" , "Header C" , "Header D")
> rownames( x ) < c ( "Row 1" , "Row 2" , "Row 3")
> x
Header A Header B Header C Header D
Row 1 0.1678067 0.8856766 0.3955881 0.7677516
Row 2 1.0302831 0.7392326 0.8333904 0.3235135
Row 3 0.4396607 1.7622323 0.8763023 0.6091688
> theMatrixTable < xtabl e ( x, caption="Caption For Table" , al i gn="l|cccc")
The variable theMatrixTable now is a xtable object. What we do with it at this point depends
upon how you want to interact with it.
Getting L
A
T
E
XOutput
If you print it out as is, it will display the contents in L
A
T
E
X, a typesetting language that
is used to create very nice looking manuscripts and books (this entire book has been
written in it). If you use L
A
T
E
Xto write your manuscripts then you are set and the listing
that follows show the formatting that results and the Table 9.1 that follows is what it
looks like when it is inserted into a L
A
T
E
Xdocument.
% l atex tabl e generated in R 2.8.0 by xtabl e 1.54 package
% Wed Dec 31 14:22:46 2008
\begin{tabl e }[ ht ]
\begin{center}
\caption{Caption For Table}
\begin{tabular}{ l | cccc}
\hl i ne
& Header A & Header B & Header C & Header D \\
\hl i ne
Biological Data Analysis Using R
158 CHAPTER 9. WORKING WITH STRINGS
Row 1 & 0.17 & 0.89 & 0.40 & 0.77 \\
Row 2 & 1.03 & 0.74 & 0.83 & 0.32 \\
Row 3 & 0.44 & 1.76 & 0.88 & 0.61 \\
\hl i ne
\end{tabular}
\end{center}
\end{tabl e}
Table 9.1: Caption For Table
Header A Header B Header C Header D
Row 1 0.17 0.89 -0.40 0.77
Row 2 -1.03 0.74 -0.83 -0.32
Row 3 0.44 1.76 -0.88 0.61
You can also print the table to a le by calling the function print(theMatrixTable,le=theleName.tex).
There are several other options available to you with the print function, see ?print.xtable for
more information.
Exporting In HTML for Web or Word
If you do not use L
A
T
E
Xand are a biologist that does a lot of mathematical, programming,
or scientic work then you should be. That being said there are many people for which a
general overpriced and under powered word processor (which shall remain nameless but
is buggy and prone to viruses and screwing up your manuscripts, you know which one I
mean) is the best you can expect to master. The xtable can be exported into a format you
can open up in said program by rst exporting the le as type="html". To export it as such
call the command > print(theMatrixTable,type=html,le=MyHTMLizedTable.html) and the table will be
saved. You can then open it up in your favorite word processor and it will turn the html
table into a normal table that you can manipulate in your documents. An example of the
html markup that this function produces is given below and an image of it is presented
in Figure 9.3.
<! html tabl e generated in R 2.8.0 by xtabl e 1.54 package >
<! Wed Dec 31 14:22:51 2008 >
<TABLE border=1>
<CAPTION ALIGN="bottom"> Caption For Table </CAPTION>
<TR>
<TH> </TH>
<TH> Header A </TH>
<TH> Header B </TH>
<TH> Header C </TH>
<TH> Header D </TH>
</TR>
<TR>
<TD> Row 1 </TD>
<TD al i gn="center"> 0.17 </TD>
<TD al i gn="center"> 0.89 </TD>
<TD al i gn="center"> 0.40 </TD>
<TD al i gn="center"> 0.77 </TD>
</TR>
<TR>
<TD> Row 2 </TD>
<TD al i gn="center"> 1.03 </TD>
<TD al i gn="center"> 0.74 </TD>
<TD al i gn="center"> 0.83 </TD>
Biological Data Analysis Using R
9.3. PLOTTING SPECIAL CHARACTERS 159
<TD al i gn="center"> 0.32 </TD>
</TR>
<TR>
<TD> Row 3 </TD>
<TD al i gn="center"> 0.44 </TD>
<TD al i gn="center"> 1.76 </TD>
<TD al i gn="center"> 0.88 </TD>
<TD al i gn="center"> 0.61 </TD>
</TR>
</TABLE>
The HTML above produces a table that when imported into Firefox looks like that pre-
sented in Figure 9.3.
Figure 9.3: The html printout of a xtable as interpreted in Firefox. You can also import tables
saved as html into popular word processors and use them as normal table items in the creation of
your documents.
There are several other options available to you with the print function, see ?print.xtable for
more information.
9.3 Plotting Special Characters
There are some special characters that you should be aware of when trying to get your
data output into a readable format. These characters are not necessarily ones that you
specically type on the keyboard rather they are ones that are available as their own
buttons on the keyboard, namely the tab character, the newline character, and the bell
character.
All the characters on your keyboard (assuming that you are using an en US keyboard)
are specied in as single variables in ASCII (ASCII stands for the American Standard Code
for Information Interchange). Obviously, since the rst A stands for American, there are
a lot of characters that you see on a computer screen that you cannot type directly on
a keyboard such as letters with accents, Greek and Latin characters (, , ), and then
there are all those non-US English characters and hieroglyphs. Your terminal that you
are running R from cannot handle these characters but you can get them into plots that
you make.
R has the nice ability to produce slightly complicated output for the axes of your plots
as well as for putting into most graphics you produce. Items such as subscripts, su-
perscripts, and mathematical symbols are easily produced using just a few different
functions.
Biological Data Analysis Using R
160 CHAPTER 9. WORKING WITH STRINGS
The primary way for producing formatted text for a graphics output is through the use
of the expression function. And the best method for looking at the ability of R to provide
nice mathy like output is to look at its own demo. So, start R and type:
> demo( plotmath)
This command will show you a short number of tables in a gure window that have
examples of the different kinds of math plotting that R handles. Associated with each
table, when R sources the demo script it passes the optional echo=TRUE parameter so that
all the commands that are used to produce the output are also shown in the R command
interface. This way you can see how each of the cells in the displayed tables is being
encoded. An example of some of the copious output is:
> draw. plotmath. cel l ( expression ( i t al i c ( x ) ) , i , nr ) ; i < i + 1
> draw. plotmath. cel l ( expression ( bold ( x ) ) , i , nr ) ; i < i + 1
> draw. plotmath. cel l ( expression ( bol di t al i c ( x ) ) , i , nr ) ; i < i + 1
The demo script itself dened the function draw.plotmath.cell() so dont worry about that part.
The part you should focus on is the (expression(bold(x)) parts. There are several options that
you can pass to the expression function and it is not quite worth listing them all here since
you see them in the R demo itself. However, I will show some of the more common
methods in the plot shown in Figure 9.4.
> x < rnorm(100)
> y < 23 + 1.4x + 2rnorm(100)
> pl ot ( x, y , bty="n" , ylab=expression (X[ st uf f ] ) , xlab=expression ( chi 2) , col ="red")
For both the x and y-axes, I use the expression function to create labels with subscripts
and superscripts. If you like, you can dene these values as individual variables prior to
plotting if you like to keep the plot command a bit cleaner, there is really no difference in
the speed at which R would evaluate them. Here is another example:
> xl abel < expression ( bold ( x [ i ] ) )
> yl abel < expression ( i t al i c ( x [ i ] 2) )
> pl ot ( x, y , bty="n" , xlim=c( 0 , 20) , type="l" , lwd=2, col ="blue" , xlab=xlabel , ylab=yl abel )
Look at the demo(plotmath) output to see the diversity of plotting approaches.
Biological Data Analysis Using R
9.4. USEFUL FUNCTIONS 161
Figure 9.4: Example of using the expression function to annotate a graphic.
9.4 Useful Functions
The following functions were introduced in this chapter and you will be required to use
them for the exercises. To get more information on any of these functions, use the R
help system.
any(x,y) Returns a logical response to x having any instance of y in it.
cat(x) Concatenates the objects in x and dumps them out to the interface.
format(x) Formats the object x for rigid (some say pretty) printing.
substring(x,s,f) This returns takes the string in x and returns the substring starting
at position s and nishing at position f.
strsplit(x,c)functions!strsplit Splits the string x on the character (or characters in c).
nchar(x) Returns the number of characters in the string x.
expressionx This function takes the variables in x and turns them into a string ex-
pression to be plotted in a function.
Biological Data Analysis Using R
162 CHAPTER 9. WORKING WITH STRINGS
nj(x) This function performs the neighbor joining function on the distance matrix
x.
unlist(x) Takes the list x and returns it as a vector.
Biological Data Analysis Using R
9.5. EXERCISES 163
9.5 Exercises
The following exercises are meant to help you understand the items presented in this
Chapter.
1. Create a table fromthe data data <matrix( rnorm(9), nrow=3 ) and label the rows as c("Richmond",
"Petersburg", "Varina") and the columns as c("PPM(A)", "PPM(B)", "PPM(C)". Use the xtable
library to export this table as HTML and then import it into your answers. (This
is a very helpful methodology for getting formatted data out of R and into your
manuscripts).
2. Create a table of the different words found on the rst page of the Chapter entitled
Preface in this text.
3. Using the strsplit function to break apart the raw text of the rst four paragraphs of
the Chapter entitled Preliminaries into sentences (HINT: use the
. as the character to break apart the string on and you can copy and paste it from
the pdf). Then use the grep command to nd the sentences that have the word are in
them.
4. Show how you would use the sub command to x the sentence, Dr. Dyer is a loser?
(And when I say x I mean make it say that I am not...)
5. How many characters are in the rst paragraph of this Chapter?
6. Create a density plot of the
2
distribution and make the main label say using the
expression() function (hint: this character is called chi).
7. In the previous graph, plot a dotted vertical line to indicate where the mean value of
the distribution is and put the character symbol next to it.
8. Using the aligned sequences to create a few different distance matrices by changing
the model type that you pass to the function dist.dna(). Do alternate distance models
have different densities of values? (Hint: plot a density plot for each distance matrix
on the same graph similar to what is shown in Figure 9.1).
9. Do these different distance models produce different tree topologies when using the
nj () function? If so, show the trees and describe the differences you see in the trees.
10. Do the functions nj () , fastme.bal(), and bionj () produce the same looking topologies? You
should read the functions to see what they are as you probably havent worked with
them yet. Explain.
Biological Data Analysis Using R
164 CHAPTER 9. WORKING WITH STRINGS
Biological Data Analysis Using R
Part III
Extending R
165
Chapter 10
Basic Scripts
When I use the term script here, what I am referring to is a set of R commands that you
put into a text le and have R evaluate. Learning how to write scripts will help you out
in the following ways:
1. In general we are all lazy. It seems to be a monumental task to type the same thing
into R over and over again. Scripts allow you to put your commands into a text le
and have R run them for you.
2. Keeping your analyses and data sets together is a great way for you to not loose
a record of what you have done. At a later date, you can come back and pick up
where you left off. If you have more data or another angle at the analysis, having a
record of how the previous analyses were performed is a huge benet.
3. There are times when you have to do the same thing over and over again, say make
graphs of a large number of variables or transform a lot of different data sets using
the same algorithm. If you put the commands in a script, and later when we get
into programming (Chapter 11) and functions (Chapter 12) you can run it over and
over again with ease (remember the lazy thing?).
So in essence, scripts are enablers for our laziness.
In this chapter, you will focus on the following topics:
Learn about basic script writing
Understand differences between code evaluated from a script and that same code
typed into the interactive R command line
Execute scripts in R
10.1 Writing Scripts
A script is nothing more than a series of commands that R recognizes and evaluates.
Within a script, you can dene data (Chapter 2), functions (Chapter 12), or other oper-
167
168 CHAPTER 10. BASIC SCRIPTS
ations. It is convenient to have a record of the commands that you use in R to produce
output.
10.1.1 Knowing Directories
A script must be in text and it must reside in a location where you can tell R it is located.
When you start an interactive session in R , it notes the current directory that you are
using. This is what is called the cwd or current working directory. Now if you are using R
from a GUI-ish installation such as on Windows , you have to tell R which directory to OS
use as a starting place. You can change the cwd from the Change dir... command in the
File menu. If you are staring R from a terminal (in OSX or some Unix variant), then
the directory where you started will be the cwd.
Here are a few tips that I nd helpful when I work with R :
It is a pretty good idea to keep your data sets and the scripts that you use to analyze
these data in the same directory. Use your descriptive skills in naming your data
and scripts such that you know what is contained in the le without looking at
it (e.g., perhaps a data set named DogwoodGerminationRates27.csv and the R
script as AnalysisOfDogwoodGermination.R; just makes it easier).
It is also a good idea to make sure that you separate your directories of data and
associated scripts such that it is easy for you to nd the right directory. Keeping
it all mashed together into a single directory can cause problems with data sets
having the same name (e.g., the infamous data.txt).
Always provide labels for each column of data. At some time in the future you will
need to look at the data set and gure out what that column of data represent.
In your scripts, provide a lot of comments. Lines that start with the hash character
(#) are ignored by R and you can use them for adding comments about what the
script, program, functions, or variables actually mean. I cannot emphasize this
enough. You will leave this class and at some point in the future look back on
some script you wrote and want to gure out how it works and without copious
comments you will fail and have a small sense of being genuine looser. You have
been warned.
10.1.2 The Editor
You can write a script in any basic text editor. For some installations of R , there is a
pseudo-GUI associated with it (e.g., Windows) because there is no real command line
terminal in the OS. This interface to R often has an integrated editor built into it and
if it is there you should probably use it unless you have another editor of choice that
you feel more comfortable with.
1
If you do not want to use the supplied editor or do not
have one available, you may want to check out TEXTMATE or TEXTWRANGLER on OSX,
1
There have literally been decades of wars fought over the choice of the real editor. If you are interested
in cultural aspects of programming and programmers (e.g., nerds like myself) re up a google search for vi
vs. emacs and sit back and enjoy.
Biological Data Analysis Using R
10.2. EVALUATING SCRIPTS 169
E or CRIMSON EDITOR (or the million others that are on this pedestrian platform) on
Windows, and for Unix/Linux you can use GEDIT, KATE, EMACS or VI (n.b. If you learn
one these last two you will never need another editor on any platform).
The important component of the editor that you are looking for is one that understands
R (or SPlus) and can provide you with syntax highlighting, parenthesis matching, and
automatic indentation. These are things that just make your life easier. After all, if you
are going to be spending a lot of time in front of your computer, you may as well have
tools that help instead of get in the way. Speaking of getting in the way, you should
never, under any circumstance, even think of using Word to do any of this.
OK, so open your editor and we will make a very small script that does something entire
useless. There is a data set named ScriptExampleData1.txt in the class folder. Make sure
you script is saved in the same directory as the data le. In R type the following code
and see what happens.
> theData < read . tabl e ( "ScriptExampleData1.txt" , header=T, sep=",")
> summary( theData )
Population Height Sex
A:5 Min. :23.40 Female:5
B:4 1st Qu.:27.70 Male :4
Median :29.70
Mean :30.04
3rd Qu.:32.70
Max. :38.20
> range ( theData$Height )
[ 1] 23.4 38.2
> l evel s ( theData$Population )
[ 1] "A" "B"
It should have loaded theData and provided a summary of it as shown. If not, you are
probably not in the correct directory. Change to the right directory and redo.
Now, take the same code and put it into your script le. Obviously, you do not want to
copy the responses that the R engine had provided to you, just the commands that you
typed. Save the script as AnalysisOfScriptData.R (note you must have the .R sufx on the
script le). Congratulations, you have written your rst script. In the next section we
will evaluate the script and note a few differences.
10.2 Evaluating Scripts
The R engine can load and evaluate scripts relatively easily. Take a look at the docu-
mentation for the source() command by typing ?source into R and give it a read. OK, ready?
In R type source("AnalysisOfScriptData.R") and see what happens... Nothing. Why is this?
The same commands produced lots of output when typed directly into R ...
The issue is that when you are typing commands into R you are doing so in an interactive
mode. You say do this and it says OK. However, when you are executing the contents
of a script, it is not entirely clear where output should go, another le, to the screen,
some other place. As a result, if you want to get a response from stuff in a script you
need to tell R to print the results. So for example, if you change your script to look
like:
Biological Data Analysis Using R
170 CHAPTER 10. BASIC SCRIPTS
theData < read . tabl e ( "ScriptExampleData1.txt" , header=T, sep=",")
pri nt ( summary( theData ) )
pri nt ( range ( theData$Height ) )
pri nt ( l evel s ( theData$Population ) )
and from R source it youll get:
> source ( "AnalysisOfScriptData.R")
Population Height Sex
A:5 Min. :23.40 Female:5
B:4 1st Qu.:27.70 Male :4
Median :29.70
Mean :30.04
3rd Qu.:32.70
Max. :38.20
[ 1] 23.4 38.2
[ 1] "A" "B"
Again, notice that here the output was only the response of the commands, the com-
mands themselves were not echoed to the R environment. You can get R to echo each
command and then provide the results when it is in a script by adding the optional
echo=TRUE option to the source() function as shown in the output below:
> source ( "AnalysisOfScriptData.R" , echo=TRUE)
> theData < read . tabl e ( "ScriptExampleData1.txt" , header=T, sep=",")
> pri nt ( summary( theData ) )
Population Height Sex
A:5 Min. :23.40 Female:5
B:4 1st Qu.:27.70 Male :4
Median :29.70
Mean :30.04
3rd Qu.:32.70
Max. :38.20
> pri nt ( range ( theData$Height ) )
[ 1] 23.4 38.2
> pri nt ( l evel s ( theData$Population ) )
[ 1] "A" "B"
This is helpful if you are debugging a script (e.g., guring out why it is crashing or giving
you the wrong answers).
So, in a script, things wont be printed out to the R terminal unless you tell it to. And
it is relatively appropriate to ask why you are wanting some things printed out as the
script is executing. The variables in a script are available in the main R memory so if
you dene a new variable in the script, after the rst time you source() it, you will have
access to it. However, because you can add variables to the main memory of R from a
script, I typically erase all variables from memory at the beginning of each script using
the command rm( list=ls() ). This way it is easy to see that the variable x you are working
with is the real one and not another x you had used two hours ago. This is a very
important point. Again, we are thinking about the future here and we need to make sure
that the things that we do in our analyses are reproducible at some point in the future.
Relying on variables that are outside our script and are only memory because we did
something before running our scripts will lead to frustration (bet on it!).
Biological Data Analysis Using R
10.3. ADDING COMMENTS TO YOUR CODE 171
In Chapter 9 there was a more complete discussion of how you can format your data for
printing. As you begin writing scripts right now, just focus on writing the routines that
you need to use to get an answer and later you can focus on making it look pretty.
10.3 Adding Comments To Your Code
Speaking of looking pretty, you must add comments to your code so that you remember
what is going on inside that le. To comment code in R you put a hash character at
the beginning of the section that you want to be commented. This will comment the line
from that point to the right. Everything to the left of the hash character is considered
code that will be evaluated.
x < 20 # thi s comment wi l l l et the assignment happen
# thi s i s a comment that spans multiple l i nes and won t
# be evaluated even i f i t has l ogi cal R code in i t
# x < 21
pri nt ( x )
Empty lines are also a nice feature to sprinkle through your scripts so that logical par-
titions can be identied. The R interpreter ignores all commented material and all lines
that do not have anything on them, so you are not penalized for not having it there.
Biological Data Analysis Using R
172 CHAPTER 10. BASIC SCRIPTS
10.4 Useful Functions
The following functions were introduced in this chapter and you will be required to use
them for the exercises. To get more information on any of these functions, use the R
help system.
# Indicates the start of a comment. The R interpreter ignores everything to the right
of this symbol.
rm(x) This function removes the variable x (or if x is a list of variable names all of
them) from memory.
source(x) This function causes R to look for the script named x and evaluate its
contents from start to nish. This works just as if you had typed in the lines of the
script with the exception of how variables are printed out to the terminal.
cat(x) This function dumps the contents of x to the GUI output as a single entity.
print(x) Send the contents of the variable x to the terminal output.
summary(x) Provides a summary of the variable x.
Biological Data Analysis Using R
10.5. EXERCISES 173
10.5 Exercises
The following exercises are meant to help you understand the items presented in this
Chapter.
1. How do you remove all variables from memory in the current workspace?
2. What happens when you set the optional argument verbose=TRUE when calling source?
3. Are you lazy?
4. Can R evaluate scripts that are written in Word or Excel?
5. How do you change the current working directory in R ?
6. How does the optional argument echo=TRUE change in the output of sourcing a script in
R ?
7. How would you print the summary of a data frame from within a script?
8. What character is used to indicate a comment?
9. How would you comment out several lines of code in a script?
10. Why is it important to comment your code?
Biological Data Analysis Using R
174 CHAPTER 10. BASIC SCRIPTS
Biological Data Analysis Using R
Chapter 11
Programming
Programming is the art of making a computer, who understands that it has to do only
exactly what you tell it to do, to get it to do the things you want to do. The language that R
uses for programming is derived from S-Plus and will be familiar looking to anyone who
has programmed in another language or seen other programming languages before.
In general, the majority of programming in R will be very linear. While it is possible to
program in an object-orientated fashion (and indeed it is not that bad of an implemen-
tation in my opinion), I wont be covering that in this book. The programs that Ill help
you build will have a start, proceed through a set of operations, have some conditional
statement, perhaps some loops, print out some stuff or save it to a le, and then exit. If
you have never programmed before you need to think about programming as a kind of
recipe, a very precise one. You need to think about the problem that you are going to
solve by writing a program. And then you need to think about the exact steps that you
will need to do to accomplish what you are attempting to do.
In this chapter, we will tackle a rather easy problem as a test case to show off how to
construct a very simple program. The problem that we are going to deal with is how to
measure canopy light from a hemispheric photo. An example of a Hemispheric photo is
given in Figure 11.1. This photo was taken by S.B. Weiss from the winter roosting habitat
of the monarch buttery in the Monarch Biosphere Reserve, Mexico. In this image, it is
easy to see the amount of canopy closure when taken from the hemispherical lens.
What we are going to do in this chapter is determine how much of that image is open
sky as a surrogate to measure available light in these forests. In the next few sections,
I will show some basic programming tools that we will use to write this program. Then
we will walk through the loading of an image and discuss how we get information from
and manipulate image data. Finally we will set out to write the program in a step-wise
fashion and nish with the completed program.
In this chapter, you will focus on the following topics:
Be introduced to some basic programming logic and the corresponding R grammar.
Develop a detailed pseudocode for a given program.
In a step-wise fashion, develop and test the program.
175
176 CHAPTER 11. PROGRAMMING
Figure 11.1: Hemispherical photograph of winter roosting habitat at Monarch Biosphere Reserve,
Mexico. Photo by S.B. Weiss made available by the Creative Commons Atribution 2.5
11.1 Looping
As mentioned in Chapter 2, R is primarily a vector language. The consequence of this
is that if you are looking for a language to do fast loops through a data set, R is not it.
In fact, Perl or Python would actually be faster to do looping-like algorithms. That being
said, there are reasons we occasionally need to use loops in R and here is a general
overview.
A loop, when referred to in a programming language, is a sequence of statements that
are repeated over and over again until some condition is reached. The items inside the
loop are typically contained within curly brackets (e.g.,{}).
11.1.1 The While
The while looping metric is a good one to use if you have a particular condition which you
want to check over and over again and perform some operations as long as the condition
is in one state. The while loop has the form while(COND){ <code goes here> }. The COND term
Biological Data Analysis Using R
11.2. CONDITIONAL STATEMENTS 177
in the parenthesis is evaluated as a logical statement each time you go through the loop
and will continue as long as COND=TRUE. When COND=FALSE, the loop exits and R starts to
evaluate statements after the closing curly bracket. There can be a lot of code between
the brackets.
The following example loops as long as x < 10 and prints out the value of x each time
through the loop.
> x < 0
> while ( x < 10 ) {
+ x < x + 1
+ cat ( x, " ")
+ }
1 2 3 4 5 6 7 8 9 10
When you start looping here, x = 0 and at each time through the loop, the variable x is
incremented and printed out on the console.
11.1.2 The For
Another common loop is one that actually focuses on the value of a counting variable
(e.g., the index in the loop). What this looping metric does is combine the initialization
of the condition variable (a counter) as a numeric value, increment the counter each
through the loop, and exits the loop when some condition on the counter is correct. The
general form of the for statement is for( COND ){ <code goes here> }. The COND can be one
of many different constructs that sets up a counting variable. Here are some examples
using the variable x.
> f or ( i in seq ( 0 , 9) ) {
+ cat ( i )
+ }
0123456789
> f or ( i in 0: 9) {
+ cat ( i )
+ }
0123456789
> x < seq( 0 , 9)
> f or ( i in seq ( length ( x ) ) ) {
+ cat ( x [ i ] )
+ }
0123456789
> f or ( i in x) {
+ cat ( i )
+ }
0123456789
For the COND the variable i is used as the counting variable along with the keyword in.
11.2 Conditional Statements
The next tool in your R programming toolbox is the conditional statement. Conditional
statements control the ow of logic through the a script or program. There are many
Biological Data Analysis Using R
178 CHAPTER 11. PROGRAMMING
cases where you would like to run some command or sets of commands if some condition
is true. For example,
if( CONDITION ) then RESPONSE
else if( OTHER_CONDITION ) then OTHER_RESPONSE
else FINAL_RESPONSE
Here the logic asks about the state of CONDITION, and OTHER CONDITION. If CONDITION is TRUE then
RESPONSE is done and none of the other conditions are evaluated nor are their responses
performed. The R interpreter just skips everything until the end of the set of condition-
als. If CONDITION is not TRUE but OTHER CONDITION is, then the only response to be performed
is OTHER CONDITION. If neither CONDITION nor OTHER CONDITION are true then FINAL RESPONSE is per-
formed. Note, only one response is ever performed each time.
In the example below, I set up a vector of boolean (TRUE|FALSE) variables and then loop
through them one at a time and see what they
> observations < as . l ogi cal ( c (TRUE, FALSE, FALSE, TRUE, TRUE) )
> observations
[ 1] TRUE FALSE FALSE TRUE TRUE
> f or ( obs in observations )
+ pri nt ( obs )
[ 1] TRUE
[ 1] FALSE
[ 1] FALSE
[ 1] TRUE
[ 1] TRUE
> f or ( obs in observations ) {
+ i f ( obs == TRUE )
+ cat ( obs , "it is true \n")
+ el se
+ cat ( "not\n")
+ }
TRUE i t i s true
not
not
TRUE i t i s true
TRUE i t i s true
We can also use conditional operators as a CONDITION in a if statement. In the example
below, we cycle through the numbers 1 through 10. And for each of them, we determine if
they are odd or even using the modulus operator %%. This operator returns the remainder
after a division.
> f or ( i in 1:10){
+ i f ( i %% 2 )
+ cat ( i , " is odd\n")
+ el se
+ cat ( i , " is even\n")
+ }
1 i s odd
2 i s even
3 i s odd
4 i s even
5 i s odd
6 i s even
7 i s odd
Biological Data Analysis Using R
11.2. CONDITIONAL STATEMENTS 179
8 i s even
9 i s odd
10 i s even
Each time through, the remainder of i %% 2 is evaluated. Possible values for this are
1 and 0 which when evaluated as.logical () , turn out to be either TRUE or FALSE printing the
appropriate message.
11.2.1 Bracketing
There is a little bit of bracket magic going on here and I should take the time to make
a few comments. Notice in the previous listing, there were brackets {} surrounding the
content inside the for loop. These brackets are essential because there is more than
one line of code inside the for loop. If there were only one line (see previous code listing
where print(obs) is the only code inside the for loop) then the enclosing brackets are
optional.
As a general rule, after any conditional (e.g., the if/else if/else) or loop (e.g., while/for) if
there is only one line of code then you do not need to use brackets if you do not want to.
Examples include:
> i f ( rnorm( 1) > 0.5 )
+ pri nt ( "greater")
> while ( TRUE )
+ pri nt ( "this will last forever")
This rule is recursive in that the one line of code is any line that is not a conditional
or a loop. In the next example, I loop through the numbers 1-10 and look for those
even numbers that are not divisible by 4 (n.b. I could have used a compound condi-
tional statement such as if( !(i%%2) && (i%%4)) but that would have really screwed up my
example).
> f or ( i in 1:10)
+ i f ( ! ( i %% 2 ) )
+ i f ( i %% 4 )
+ cat ( "the value=" , i , "\n")
the value= 2
the value= 6
the value= 10
In some sense, you can think of these kinds of one-liners as just extensions as one-
offs. There is nothing wrong with using brackets even in these cases. In fact, it may
open up your code a bit and make it a bit easier to read in the future. You just do not
have to use them.
However, where you want more than one statement to be executed after a loop or condi-
tional statement then you must use brackets. T
Biological Data Analysis Using R
180 CHAPTER 11. PROGRAMMING
11.3 Outlining A Program
The most difcult part of programming is understanding where to start. Writing a pro-
gram, on the surface, appears to be a daunting task in intself. However, when I write
programs I tend to think of them not as a single large program but as a series of smaller
steps. The key to doing this is to understand the sequence of steps that we need to
accomplish so that the program can do what is required.
So, rst things rst. State what you want the program to do in specic terms. For
this Chapter we will be working on developing a program that calculates the amount of
canopy openness from a hemispheric image (Figure 11.1). If you havent already done
so, I recommend that you look at Chapter 7 to refresh yourself on how we work with the
internals of an image.
Next, we need to get out a sheet of paper and write down, exactly, how the program is
going to work. It is important that we include all the steps necessary and in the order in
which they are to be performed. An example of this would be:
1. Load image into memory
2. Determine what parts of image are open canopy
3. Determine total area of image
4. Print out the proportion of canopy that is open.
So, each of these steps is a relatively easy one by itself and we will create the overall
program by breaking it up into manageable parts.
11.4 Creating A Program
It is often necessary to incrementally build a program. Using the outline in the previous
section, we can open a new le and create a script that does each of these items in suc-
cession. Typically, I nd it helpful to work on the R command line to test out particular
sets of commands and when I have it exactly like I like it then I move it to a script.
11.4.1 Step 1: Loading An Image Into Memory
In Chapter 7, we examined how to load images into memory, translate them into vari-
ous formats, and get into their knickers, so to speak. So to begin with, the image as I
retrieved it from Wikipedia is a JPEG image. I will begin by turning it into a PPM formatted
image as discussed in Chapter 7 using the program GIMP (http://www.gimp.org), al-
though you could use any image manipulation program and there are several free ones
available for you on the internets. The PPM le is what you have access to in the class
folder for Chapter 11.
> l i brary ( pixmap)
> img < read .pnm( f i l e ="Hemiphoto monarch habitat1.ppm")
Read 637563 items
Biological Data Analysis Using R
11.4. CREATING A PROGRAM 181
Figure 11.2: The blue channel of the
canopy picture displayed as a greyscale im-
age.
Figure 11.3: A histogram of values in the
blue channel (Figure 11.2).
> pl ot ( img)
Now we have the image loaded and a plot that is identical to that displayed in Figure
11.1 and we must gure out how to have it represented.
11.4.2 Step 2: What Is Open Canopy
The variable img has the following components and here we need to gure out what parts
of the image are the sky parts.
> names( attri butes ( img ) )
[ 1] "size" "cellres" "bbox" "bbcent" "channels" "red" "green"
[ 8] "blue" "class"
Remembering that there are three different channels in a PPM le, one for red, one for
green, and one for blue, perhaps we should look there rst. You can plot each of the
channels as an image by creating a pixmapGrey() image and see the intensity of each color
channel.
> pl ot ( pixmapGrey( img@blue ) )
> pl ot ( pixmapGrey( img@red ) )
> pl ot ( pixmapGrey( img@green ) )
And from this you will see that the different channels look pretty much the same when
evaluating the area that is considered the sky in this image. For our purposes, I will
we will only use the blue channel as displayed in Figure 11.2.
Biological Data Analysis Using R
182 CHAPTER 11. PROGRAMMING
So if that is the component of the image that we are going to use, we now need to deter-
mine which values to look for. To do this, you can easily make a histogram composed
of the values in the blue channel of the image using the command hist( img@blue ). We can
see from Figure 11.3 that there is a tremendous amount of values in this channel at the
low end, a peak at around 0.2 and another at the top end close to 1.0.
We can get a bit more specic with this image and plot the intensity of a particular row
of values in the blue channel to double check that we think values close to 1.0 should
represent light values and those near 0.0 are the dark regions. The following commands
create the image displayed in Figure 11.4 where the raw values along the 230
th
row of
pixels (indicated by the red dashed line) are shown in blue. It is easy to see that the
value in the blue channel gets larger as the dashed line crosses the image.
Figure 11.4: Intensity of blue channel values in the image as taken through a slice of the image
(at pixel row 230 as indicated by red dashed line).
> pl ot ( img, axes=T, bty="n" , xlab="Image Width" , ylab="Image Height")
> par ( new=T)
> abline (230,0, col ="red" , lwd=2, l t y=2)> par ( new=T)
> par ( new=T)
> pl ot ( img@blue[ 230 , ] , bty="n" , type="l" , xlab="" , ylab="" , col ="blue" ,
+ lwd=3, axes=F, ylim=c( 10,10))
So, at this point, we need to make a value judgement. We are fairly condent that values
close to one in the blue channel (and others you can go check yourself) represent areas
in the image where it is pretty light. But, we need to make a cut-off such that if we look
at a pixel, we can put it into the light or not-light category. For the purposes of this
exercise, I will assume that values that are 0.98 are to be considered as sky and I will
also make the restriction that I need the pixels in each channel to meet or exceed this
cut-off.
Now, to nd out how much of the image is sky (using this denition), we must:
Biological Data Analysis Using R
11.4. CREATING A PROGRAM 183
1. Loop through every matrix and the items in each matrix.
2. Evaluate if the value should be considered as sky or not.
3. Use a variable to keep track of all the pixels that meet the criteria
So to our script, we will add the following lines of code
> numRows < img@size [ 1]
> numCols < img@size [ 2]
> f or ( row in 1:numRows ) {
+ f or ( col in 1:numCols ) {
+ i f ( img@red[ row, col ] >= 0.98 &
+ img@green[ row, col ] >= 0.98 &
+ img@blue [ row, col ] >= 0.98 )
+ numSky < numSky + 1
+ }
+ }
> numSky
[ 1] 9624
So, in the image across all three color channels, we nd a total of 9, 624 pixels that can
be considered to represent the sky.
1
11.4.3 Step 3: Determine The Total Area Of The Image
OK, nally we are almost nished. We need to now determine what the total number of
pixels there are in the image so that we can get a standardized percent of open canopy.
We could use the total number of pixels 461
2
= 212, 521 but the image taken with the
sh-eye lens is not square, rather it is a circle that ts in a square whose side has 461
pixels. So, we need to gure out the area of this circle as:
> r < 461/2
> total Area < pi r 2
> total Area
[ 1] 166913.6
> (4612total Area ) /total Area
[ 1] 0.2732395
As a side note, the last expression in the code listing shows what percentage of area that
we would bias our estimation by if we just used the total number of pixels in the image,
27.3% is a reasonable sized bias!
11.4.4 Step 4: Print Out The Proportion Of Canopy That Is Sky
This part is fairly easy and doesnt require much.
1
While this part of the exercise was excellent at showing some of the programming paradigms and how
they can be combined to give an answer, it is also true that Step 2 can be accomplished in R using the
one-liner sum( img@blue >= 0.98 &img@green >= 0.98 &img@red >= 0.98 ). Here the three conditionals return a
vector of logical variables, which the function sum() coerces into integers. While it would have been much
shorter to do it this way, it would have negated all the quality teaching experiences that I was laying on
you...
Biological Data Analysis Using R
184 CHAPTER 11. PROGRAMMING
> numSky / total Area
[ 1] 0.05765857
11.4.5 The Complete Program
The complete program is listed below with comments. There are a few changes in the
program that I made to make it a bit easier to work with. Comments should be self
explanatory and are indicated by lines that start with the hash character (#).
# removes al l vari abl es from memory at st art of scri pt
rm( l i s t =l s ( ) )
# load the pixmap l i brary to open the image
l i brary ( pixmap)
# I put the f i l e name i nto a vari abl e so
# i t could be changed easi l y at the top
# of the f i l e i f necessary
fileName = "Hemiphoto monarch habitat1.ppm"
# I also put the cr i t er i a i nto a vari abl e
# so we can change i t in one place to see
# how the resul ts di f f er
skyCri teri a < 0.98
# Read in the image and f i nd the number of
# rows and columns in i t
img < read .pnm( f i l e =fileName )
numRows < img@size [ 1]
numCols < img@size [ 2]
# Loop through each row
f or ( row in 1:numRows ) {
# Loop through each column
f or ( col in 1:numCols ) {
# Evaluate the cel l in each f or
# sky cri t eri a
i f ( img@red[ row, col ] >= 0.98 &
img@green[ row, col ] >= 0.98 &
img@blue [ row, col ] >= 0.98 )
numSky < numSky + 1
}
}
# Find t ot al are of f i sheye ci r cl e
r < numRows/2
total Area < pi r 2
# Pri nt out the percent ca
percentCanopyOpen = numSky/total Area
cat ( Canopy Opening: , percentCanopyOpen, \n ) ;
11.5 Synopsis
This has been a very simple little program that we made. Despite it being simplistic, it
does show you how to go about creating a simple analysis program. R is not a general
Biological Data Analysis Using R
11.5. SYNOPSIS 185
programming language and you are not going to make large programs with it. The key
to R is knowing how to get something put together, take it a step at a time, and break
the components into reasonably sized, easy to accomplish pieces. This is where you
start.
In Chapter 12 we will build upon what has been done here when we discuss Functions.
We can encapsulate code into functions and make our lives much easier. For now, play
around with the program and the exercises and get comfortable with typing code.
Biological Data Analysis Using R
186 CHAPTER 11. PROGRAMMING
11.6 Useful Functions
The following functions were introduced in this chapter and you will be required to use
them for the exercises. To get more information on any of these functions, use the R
help system.
x %% y The modulus operator. This returns the remainder of the division x/y.
as.logical(x) Coerces x into a logical variable if possible. See 2.4.7 for more infor-
mation on logical variables.
rm(x) This function removes the variable x from memory
abline(a,b) This function plots a line with intercept of a and a slope of b in the current
graphics window.
for(INDEX SEQUENCE ) A main looping construct that specically uses the counter IN-
DEX that is contained in SEQUENCE.
while(COND) A looping construct that continues to loop until some condition is met.
As long as COND==TRUE the loop will continue.
if(COND) The evaluation of the condition COND. If it is TRUE then the next line following
the if statement is executed. If it is FALSE then the next line is skipped. You can
include several lines to be evaluated after this and other evaluation statements by
enclosing the code in curly brackets {}.
else if(OTHER COND) The second evaluation of a condition. This must not be the rst
conditional (e.g., there is an else here that implies a previous if or else if statement
that this is following).
else The last of a conditional, if all the previous ones did not turn out to be true,
then whatever follows the else will be evaluated. It is not necessary that you have
one of these at the end, you may want to not do anything unless some specic
conditions occur.
Biological Data Analysis Using R
11.7. EXERCISES 187
11.7 Exercises
The following exercises are meant to help you understand the items presented in this
Chapter.
1. Write a short program that lists all numbers from 1 to 100 and determines if they are
divisible by 2 and 3.
2. Write a program using the for () loop that prints the numbers from 42 down to 27, one
on each line.
3. List some of the assumptions that are included in how the variable numSky is deter-
mined.
4. Using the program we created in this Chapter, make a graph of percent canopy with
different cut-off values. In your opinion what would be the most biologically mean-
ingful cutoff?
5. Change the program to use a cutoff value based on the sum of the individual color
channel values rather than the current requirement that they all be simultaneously
over some threshold.
6. How many lines of output do you expect to get from the following code? HINT: Think
before you try to run this program.
while ( 1) {
cat ( "All work and no play makes Dr. D a dull boy.\n" ) ;
}
7. Create an outline of the steps that would nd the number of values in a matrix that
is equal to or greater than 20.
8. Implement the program you outlined using the matrix M <matrix( runif(25,10,30), nrow=5) as
your input. Make sure to comment your code appropriately.
9. What is the proper syntax for conditions passed to an if statement requires x to be
greater than 23 and y to be equal to or less than 4?
10. How many else statements can you have after an if statement?
Biological Data Analysis Using R
188 CHAPTER 11. PROGRAMMING
Biological Data Analysis Using R
Chapter 12
Functions
Throughout this book, weve used both built-in function such as sqrt() and sum() as well as
some that are located in external libraries that we had to load (such as skewness() in 4.3.1
and read.pnm() in 7.2). These functions have been really helpful in making you scripts
look clean and readable and have made you life rather easy as you performed some
basic statistical analysis. Think what a pain it would have been if you had to write code
every time you wanted to calculate a sqrt() of a number... (Im not even sure how it is
done).
Writing your own functions in R is a very useful way to save a lot of typing. You can
consider a function a small self-contained bundle of instructions that you can call when
every you need to. Say you are picky about the way your graphics look, or that there
is a particular set of routines that you use to make translations of your data from one
format to another. Putting this code into function and putting that function in a location
where you can get access to when every you need it is a real treat.
In this Chapter you will learn the following skills:
Learn the syntax required to write your own functions.
Understand the scope of a variable and why you should care.
Create a basic library of routines that you can use in the future.
12.1 Function Syntax
The format of a function basically has the following three parts:
1. The name of the function. The creation of a name for a function is just as important
as for a variable. I nd it helpful to try to make the name tell me what the function
does (Im funny that way), which means it typically starts with a verb such as
convertMissingData(), removeLameExcuses(), or makeTheGraphTheWayILikeIt().
2. The assignment to the name. Right after the name you will have the assignment of
the generic function() function to the variable (see the syntax below). This tells R that
the name is not a variable but will actually be the name of a function.
189
190 CHAPTER 12. FUNCTIONS
3. The function contents. This is the part that you get to write. Here is where you put
all the stuff together to do whatever it needs to do.
In general, these three parts are put together to look like:
doMyBidding < function ( ) {
# Function Contents
}
Now this is fairly boring function here, it takes no arguments and doesnt return any-
thing to you. It is kind of like saying, R go to your special place and do something
but dont tell me what it is. As you write functions, they will be considerably more
complicated (and hopefully useful).
In this Chapter I will post in the raw code for the function itself followed by the output
of R from the command line. The straight posting of the function syntax allows you to
cut-and-paste them into the R interpreter (even though you will learn it better by typing
it).
Also, functions that you have dened are available in the local memory of the interpreter
in the same way as local variables are. If you use the ls () command to list the items in
memory it will show your function names along side your variable names.
> l s ( )
[ 1] "doMyBidding" "x"
12.1.1 Returning Values From A Function
Most likely you are calling some function because you are interested in getting a re-
sponse to it. It is not common to write functions that do no give you something back in
return.
To return a value from a function, R has you put the name of the variable on the last
line of the function. An example of this is the following function that returns a single
number.
gimmeANumber < function ( ) {
42
}
> gimmeANumber ( )
[ 1] 42
> gimmeANumber ( )
[ 1] 42
And a slightly better function here that actually returns a random number:
gimmeAnotherNumber < function ( ) {
x < runi f (1,1,100)
x
}
> gimmeAnotherNumber ( )
[ 1] 87.3278
> gimmeAnotherNumber ( )
[ 1] 64.97312
Biological Data Analysis Using R
12.1. FUNCTION SYNTAX 191
You can also use the return return() to exit the function and potentially return a value.
Here is an example that checks to see if the passed argument is the right kind, if it is
not it prints an error and returns, otherwise it performs a calculation and then returns
the result.
gimmeHalf < function ( theValue ) {
# check to see i f i t i s a numeric value
# i f i t i s the return hal f
i f ( i s . numeric ( theValue ) ) {
return ( theValue / 2. 0)
}
# i f i t isn t then complain
el se {
cat ( "The value" , theValue , "is not a number, try again.\n")
return ( )
}
}
> gimmeHalf ( 12 )
[ 1] 6
> gimmeHalf ( "Hello partner! " )
The value Hello partner ! i s not a number, try again.
NULL
Notice here that when the function left the else section of the function by calling the
return() without any arguments then the function actually returned the NULL value. If you
are not interested in having a function return NULL, something that signals to you that
the value passed to the function may be incorrect then you can remove the last return()
statement and have the function not return anything. Here is what that function would
look like.
gimmeHalf < function ( theValue ) {
# check to see i f i t i s a numeric value
# i f i t i s the return hal f
i f ( i s . numeric ( theValue ) )
return ( theValue / 2. 0)
# i f i t isn t then complain
el se
cat ( "The value" , theValue , "is not a number, try again.\n")
}
> gimmeHalf ( 14)
[ 1] 7
> gimmeHalf ( "bob")
The value bob i s not a number, try again.
Vector Arguments
By default you function above can work on vectors of values just as easy as single
numbers. This is because a vector of numbers will return TRUE when asked if it is.numeric()
(see 2.4.8 for more on this). Here is an example,
> x < seq(2,20, by=3)
> i s . numeric ( x )
[ 1] TRUE
Biological Data Analysis Using R
192 CHAPTER 12. FUNCTIONS
> i s . vector ( x )
[ 1] TRUE
> x
[ 1] 2 5 8 11 14 17 20
> gimmeHalf ( x )
[ 1] 1.0 2.5 4.0 5.5 7.0 8.5 10.0
So by default, you can work with vectors of your values just as easy as single numbers.
This is pretty cool and you should try to remember the love that R has for vector oper-
ations because it is much faster to call your gimmeHalf() function by passing it vector of
value than using a loop to go through the vector and calling gimmeHalf() for each individual
value...
Here is a slightly longer example of a function. Notice that inside the function, I have
added some comments. This is a very good idea because it allows you to document what
you are doing inside the function. In fact, I typically write functions by:
1. Write the signature of the function, the funcName <function(){ } part.
2. Using comments, write the sequence of events that have to occur inside the function
so I can see what needs to be done (breaking large problems into small ones here)
3. Fill in the code to allow R to do my bidding.
So lets walk through these steps and make a function. The purpose of this function is
to get a little encouragement for my programming endeavors by having R return some
nice praise for me.
Step 1: Create signature The signature for this function will be:
giveMeSomeMomLove < function ( ) {}
Step 2: Using comments create logic of function: The overall goal of this function is to
return a random statement from my mother so I will have to set up some statements,
nd a random one,, and then return it.
giveMeSomeMomLove < function ( ) {
# set up a vector of l ovi ng mother sayings
# pick a random number to use as index f or responses
# I f you put the name vector and the index on the l ast l i ne
}
Step 3: Fill in the R logic: Now that I have the comments set out, it is fairly easy for me
to use them as a guide in laying out the logic of function. You do not have to document
every line of code in your functions, but if you put in enough so that it is obvious what is
going to happen next, you will nd yourself being happy with your past self more often
than hating what you had forgotten to do (?).
giveMeSomeMomLove < function ( ) {
# set up a vector of l ovi ng mother sayings
momSayings < c ( "Honey, your dad and I think you are doing just fine." ,
"Come home this weekend, I made your favorite dessert." ,
"We think you are the BEST student at VCU." ,
"You know I took calculus back in college, maybe I can help." ,
Biological Data Analysis Using R
12.1. FUNCTION SYNTAX 193
"I just know youll be able to find a good job after college.")
# pick a random number to use as index f or responses
resp = round( runi f ( 1 , 1, length ( momSayings) ) )
# I f you put the name vector and the index on the l ast l i ne
momSayings[ resp ]
}
> giveMeSomeMomLove ( )
[ 1] "We think you are the BEST student at VCU."
> giveMeSomeMomLove ( )
[ 1] "Honey, your dad and I think you are doing just fine."
> giveMeSomeMomLove ( )
[ 1] "You know I took calculus back in college, maybe I can help."
Feel free to add some of your own mother sayings here
12.1.2 Passing Values To A Function
The most common way you will interact with a function is probably by giving it some
variables and expecting to get something back.
getI denti tyMatri x < function ( numRows ) {
# make a square matrix with al l zeros
I < matrix ( 0, nrow=numRows, ncol=numRows )
# make the diagonal al l ones
diag ( I ) < 1
# return i t to the cal l er
I
}
> getI denti tyMatri x ( 2)
[ , 1] [ , 2]
[ 1 , ] 1 0
[ 2 , ] 0 1
> getI denti tyMatri x ( 5)
[ , 1] [ , 2] [ , 3] [ , 4] [ , 5]
[ 1 , ] 1 0 0 0 0
[ 2 , ] 0 1 0 0 0
[ 3 , ] 0 0 1 0 0
[ 4 , ] 0 0 0 1 0
[ 5 , ] 0 0 0 0 1
Default Values
Functions can have default values associated with variables that are passed to them.
Weve seen this many times so far as youve looked up and seen the function signatures
of built in variables. This is a very convenient feature for you and your users. In general,
when you think of writing functions you should not try to make them so specic that
you have a lot of different functions that do almost the same thing, rather you should
make them robust and if you can combine a few functions into a single one whose
values change depending upon a parameter you pass to it, it is better overall form. For
example, the function getIdentityMatrix() returns a square matrix with ones down the
diagonal. This matrix is a pretty special one (see ??) in matrix analysis and probably
Biological Data Analysis Using R
194 CHAPTER 12. FUNCTIONS
should have its own function just because of its status. However, there are a number of
reasons why you may need a square matrix with a single value down the diagonal and
perhaps it would be more robust to create a function such as:
getDiagonalMatrix < function ( si ze , value=1 ) {
theMat < matrix ( 0 , nrow=si ze , ncol=si ze )
diag ( theMat ) < value
theMat
}
> getDiagonalMatrix ( 3)
[ , 1] [ , 2] [ , 3]
[ 1 , ] 1 0 0
[ 2 , ] 0 1 0
[ 3 , ] 0 0 1
> getDiagonalMatrix ( 3 , 42)
[ , 1] [ , 2] [ , 3]
[ 1 , ] 42 0 0
[ 2 , ] 0 42 0
[ 3 , ] 0 0 42
Now this function has a default value to set the diagonal values to (e.g., 1) producing the
Identity matrix I by default, however, it can also produce any diagonal matrix when you
pass an additional parameter to the function. If you do not pass it to the function, it
is assigned in the signature for you by default. This makes the function perhaps more
robust and useful. Of course, this is all up to you, you are the programmer here and you
get to make the decisions. After all, there are several different ways to get the correct
result when programming and as Biologists, we should focus on the biology and use
tools like R as simple tools.
12.2 Scope
The scope of a variable determines the value that it has depending upon where it is
located. This topic is a pretty important one and can be a bit tricky at times.
myFunc < function ( x) {
x < 42
cat ( "x inside function is" , x, "\n")
}
> x < 21
> x
[ 1] 21
> myFunc( x )
x i nsi de i s 42
> x
[ 1] 21
myFunc < function ( a) {
x < 42
cat ( "other x inside function is" , x, "\n")
}
> x < 23
> myFunc( x )
other x i nsi de function i s 42
> x
[ 1] 23
Biological Data Analysis Using R
12.3. USEFUL FUNCTIONS 195
12.3 Useful Functions
The following functions were introduced in this chapter and you will be required to use
them for the exercises. To get more information on any of these functions, use the R
help system.
function(args)code Creates a function that has the code inside code requiring the ar-
guments args.
return(x) Returns the value x from the function which means it is immediately exited
and no more code is executed in the function.
Biological Data Analysis Using R
196 CHAPTER 12. FUNCTIONS
12.4 Exercises
The following exercises are meant to help you understand the items presented in this
Chapter.
1. Create a function that allows you to pass it a regression model and it will return a
string that contains the formula for the model as you would like to have it displayed
on a graph.
2. Create a function that takes a single vector of values and creates a histogram and
density line from that data in a new graphics window.
3. Explain scope and how it pertains to the values assigned to variables.
4. Create a function that takes an ANOVA or Regression model and saves the ANOVA
table to a le. You should probably allow the user to pass a le name to the function.
5. How do you set default values for a function when you write it?
6. Explain how you get your functions to accept vector arguments.
7. Create a function that returns random numbers but allow the user to set an optional
argument that will only return even numbers.
8. How would you remove a function from the memory of R ?
9. Lets assume that you have a folder full of data les named Data1, Data2, Data3, . . .,
Data40. Write a function that creates these le names dynamically. You will want to
allow the user to specify the base name of the les (e.g., Data) as well as the starting
and ending numbers (e.g., 1 and 40) but set the starting number to default to 0.
10. How do you make sure that the arguments that are passed to your functions are the
right kind of variables? For example, what if I passed the variable x <"this is the end"
to a function that expects a number.
Biological Data Analysis Using R
Appendix A
Answers to Exercises
In this section you will nd answers to the odd numbered Exercises presented in each
Chapter. These answers are meant to help you start on the exercises facilitating your
completion of the remaining questions. It is my recommendation that you look at the
answers only after you have completed them just to make sure that what you thought
you were doing is the correct thing. Dont look ahead....
Ansers to Chapter 2.
197
198 APPENDIX A. ANSWERS TO EXERCISES
Biological Data Analysis Using R
Appendix B
Installing Additional Libraries
The R statistical computing environment is made more robust by the addition of external
libraries. Libraries can be written in R , C, or FORTRAN by you or other people who want to
expand the functionality and utility of R .
B.1 Library Availability
There is a list of libraries available at http://cran.r-project.org. As of the time of this
writing, there are currently 1621 different packages in the repository. All are available
for you to install and use at your discression. Each should also come with a set of
documentation covering all the functions that are included in the library, descriptions of
the data sets, and some overall discussions on the library along with the library.
B.2 Installing Libraries
B.2.1 Using install.packages() As A GUI
The easiest way for you to install a libarary is to do so from within R itself. To do this,
your machine must be connected to the internet. R knows how to nd, download, and
install binary versions of packages using a tck/tk interface GUI interface.
If you conduct the installation as a normal user that does not have administrative priv-
ilages on your computer, the libraries will be installed in a location that is in your own
home directory. Depending upon which operating system you have, this will be in dif-
ferent places. The main thing to worry about here is that when you install libraries into
your own directory they will only be available to that user and will not be available for
any other users on that machine. If two people use the same machine then they will have
to install it twice, once in each home directory. Conversely, if you have administrative
privelages on the machine you are using, you can install the libraries into a location that
everyone that uses that machine can access.
To start the installation process, issue the command:
199
200 APPENDIX B. INSTALLING ADDITIONAL LIBRARIES
> i nst al l . packages ( )
And this will bring up a window (using tck/tk so it wont look quite like the normal
window on your operating system) that allows you to select which mirror you would
like to use for downloading. An example window is shown in Figure B.2.1. In general,
you should select a location that is geographically proximite to your current location.
All of these mirror servers are kept up-to-date pretty well and you shouldnt nd any
differences among the packages on any of them.
Once you have selected your preferred mirror server, another window will be presented
(resembling Figure B.2.1) that lists all the packages that are available to be installed.
Be careful here, this simple interface does not check to see which packages you already
have installed, it only lists all the packages that are at your disposal. So just because
there is a package on that list doesnt mean that you do not already have it installed on
your machine.
Select the package, or packages, that you want to install from the list. To select more
than one, click on more than one... To deselect a package, click on it a second time and it
will be deslected. Once you hit the OK button on this window, the install .packages() function
will look to see what dependencies the selected packages have (e.g., PackageA requires
PackageB but you didnt know that and didnt select it). Packages will be downloaded and
installed in the correct location. After they are installed, you should be able to use them
immediately (e.g., without restarting R ).
B.2.2 Using install.packages() For Specic Libraries
If you know the name of the package that you are interested in installing you can use
the install .packages() function directly by passing it a name, or list of names, of the packages
you are interesed in. This will skip the Package Selection Window step shown in Figure
B.2.1. The syntax for this would be:
> i nst al l . packages ( "theNameOfTheLibraryNeeded")
Libraries have also be partitioned into different Task Views. These are meta-packages
that contain several different packages under a particular theme. Below are a list of the
views that are available as of January 2009 (these categories and desriptions are lifted
directly from the website.
Bayesian Bayesian Inference
ChemPhys Chemometrics and Computational Physics
Cluster Cluster Analysis & Finite Mixture Models
Distributions Probability Distributions
Econometrics Computational Econometrics
Environmetrics Analysis of Ecological and Environmental Data
ExperimentalDesign Design of Experiments (DoE) & Analysis of Experimental Data
Biological Data Analysis Using R
B.2. INSTALLING LIBRARIES 201
Figure B.1: Example of CRAN mirror window as viewed on Linux
Biological Data Analysis Using R
202 APPENDIX B. INSTALLING ADDITIONAL LIBRARIES
Figure B.2: All packages that can be installed from the selected mirror server on my machine.
Biological Data Analysis Using R
B.2. INSTALLING LIBRARIES 203
Finance Empirical Finance
Genetics Statistical Genetics
Graphics Graphic Displays & Dynamic Graphics & Graphic Devices & Visualization
gR gRaphical Models in R
MachineLearning Machine Learning & Statistical Learning
Multivariate Multivariate Statistics
NaturalLanguageProcessing Natural Language Processing
Optimization Optimization and Mathematical Programming
Pharmacokinetics Analysis of Pharmacokinetic Data
Psychometrics Psychometric Models and Methods
Robust Robust Statistical Methods
SocialSciences Statistics for the Social Sciences
Spatial Analysis of Spatial Data
Survival Survival Analysis
TimeSeries Time Series Analysis
You can install all the libraries in these particular views by invoking the command:
> i nst al l . packages ( "ViewName")
You will still have to specify the mirror server to use and once you do, R will take it
from there. This could be a lengthy process as it may require numerous packages to be
downloaded and installed. Be patient.
B.2.3 From the Command Line
Finally, there is one other method that I typically use on my machines. This is because I
typically download the source packages rather than the pre-compiled binaries. However,
this method also works with binaries. You can download the package from the CRAN
site directly and then open a command-line Terminal and change to the directory where
the package is located. From there issue the command:
R CMD INSTALL ThePackageYouDownloaded.tar.gz
and R will install it for you. If you do this as the root or administrator person, it will
install it in a globally accessable location so any user on that machine will have access
to it.
Biological Data Analysis Using R
204 APPENDIX B. INSTALLING ADDITIONAL LIBRARIES
Biological Data Analysis Using R
Bibliography
Caswell, H. (2001). Matrix population Models: Construction, Analysis, and Interpretation.
Sinauer Associates, Sunderland, Mass., 2nd edition edition.
205
Index
class, 115
clustal le, 153
coercion, 9
comment character (#), 172
data types, 8
character, 10
complex, 11
constant, 11
data frame, 18, 26
factors, 16
integer, 8
list, 17
logical, 13
matrix, 14
NULL, 31
numeric, 9
raw, 12
vector, 13
distributions
dchisq, 43, 68
df, 43, 68
dnorm, 43
pchisq, 43, 54
pf, 43
pnorm, 43
qchisq, 43, 44, 54
qf, 43, 46
qnorm, 43
qt, 46
rchisq, 43
rf, 43
rnorm, 43, 54, 58, 68
rpois, 63
runif, 65, 107
fasta le, 152
gure
axis labels, 56
title, 56
functions, 6
%%, 186
abline, 186
any, 150, 161
as.factor, 72, 86
as.index, 186
as.matrix, 86, 144
as.matrix(), 121
attributes, 33
barf, 123
barplot, 137
binom.test, 73
c, 86
cat, 118, 161, 172
cbind, 29, 40, 86, 107
class, 20, 33
colnames, 86, 157
components, 6
cov, 64
density, 57, 58
det, 128
diag(), 126
dim, 123
dist.dna, 154
eigen, 132
else, 186
else if, 186
expression, 161
for, 186
format, 161
function, 195
gimeMeSomeMomLove, 192
ginv, 128
grep, 150
grey, 118
gsub, 150
image, 118
index, 127, 186
kurtosis, 60
length, 20, 86
206
INDEX 207
levels, 17
lm, 128
load, 32, 40
log, 7
ls, 32
matrix, 121, 145
max, 61, 118
mean, 58, 129
merge, 39, 40
min, 61
names, 33
nchar, 148, 161
nj, 154
par, 49
paste, 11, 20, 95, 149
plot, 47, 155
print, 172
q, 31
qchisq, 45
range, 50, 61, 81, 86
rbind, 28, 40
read.dna, 153
read.table, 27, 28, 86, 145
read.table(), 121
rep, 14, 20
return, 191, 195
rexp, 54
rm, 32, 40, 172, 186
rnorm, 54, 56, 118
round, 107
row.names, 33
rownames, 86, 157
rpois, 52, 54
save, 32, 40
sd, 58
seq, 14, 20
skewness, 59
source, 172
strsplit, 148
sub, 150
subset, 35, 36, 40
substring, 149, 161
sum, 127
summary, 17, 20, 86, 93, 107,
172
t, 128
table, 17, 68, 72
unlist, 148, 162
var, 58
while, 186
genetic distance, 154
grahics
pdf, 51
graphics
abline, 94, 107
barplot, 137, 144
bg, 48
bmp, 51
bty, 48
bxp, 85
cairo pdf, 51
cex, 48
col, 48
density plot, 57
dev.copy, 52, 53
dev.off, 52, 53
fg, 48
hist, 52, 55
jpeg, 51, 53
legend, 142, 145
line plot, 47
lty, 48
lwd, 48
main, 48
mfrow, 48, 61
optional parameters, 48
overlaid, 49
par, 48
pch, 48, 107
pictex, 51
plot, 46, 68, 85
png, 51
postscript, 51
quartz, 51
rug, 104
scatter plot, 46, 47
sub, 48
text, 107
tiff, 51
topo.colors, 52
type, 48
x11, 51
xlab, 48
xlim, 49
ylab, 48
Biological Data Analysis Using R
208 INDEX
ylim, 49
matrix
%*%, 144
addition, 123
det, 144
diag, 144
diagonal, 126
dim, 144
eigen, 145
element-wise multiplication, 124
ginv, 145
Hadamard product, 124
multiplication, 124
scalar addition, 123
scalar multiplication, 124
scalar subtraction, 123
Schur product, 124
subtraction, 123
t, 145
trace, 127
Neighbor Joining, 154
operator
assignment, 18
logical, 19
numerical, 18
operator order, 18
Pinaceae, 153
stats
anova, 93, 107
aov, 107
binom.test, 86
chisq.test, 72, 76, 86
cor.test, 67, 79, 86
interaction formula, 99
Kruskal-Wallis Test, 82, 83
kruskal.test, 86
lm, 92, 107
Mann-Whitney, 80
mean, 58, 68, 81, 86
median, 63
nj, 162
no intercept, 100
quantile, 63
sd, 68
step, 107
t.test, 107
TukeyHSD, 107
var, 58, 68
Wilcoxon, 80
Wilcoxon Test, 81
variable, 7
Biological Data Analysis Using R