General Linear Models (GLM)

NCSS Statistical Software NCSS.
com
Chapter 212
General Linear
Models (GLM)
Introduction
This procedure performs an analysis of variance or analysis of covariance on up to ten factors using the general
linear models approach. The experimental design may include up to two nested terms, making possible various
repeated measures and split-plot analyses.
Because the program allows you to control which interactions are included and which are omitted, it can analyze
designs with confounding such as Latin squares and fractional factorials.
Kinds of Research Questions

A large amount of research consists of studying the influence of a set of independent variables on a response
(dependent) variable. Many experiments are designed to look at the influence of a single independent variable
(factor) while holding other factors constant. These experiments are called single-factor experiments and are
analyzed with the one-way analysis of variance (ANOVA). A second type of design considers the impact of one
factor across several values of other factors. This experimental design is called the factorial design.
The factorial design is popular among researchers because it not only lets you study the individual effects of
several factors in a single experiment, but it also lets you study their interaction. Interaction is present when the
response variable fails to behave the same at values of one factor when a second factor is varied. Since factors
seldom work independently, the study of their interaction becomes very important.
This procedure will also analyze repeated-measures and split-plot designs. These designs are popular in many
disciplines in which experiments are needed that take several measurements on an individual through time.
Examples are pre-post type tests administered to various groups of individuals.
Analysis of covariance (ANCOVA) is another design that may be analyzed using this procedure. ANCOVA is
useful when you want to improve precision by removing various extraneous sources of variation from your study.
The Linear Model

We begin with an infinite population of individuals with many measurable characteristics. These individuals are
(mentally) separated into two or more treatment populations based on one or more of these characteristics. A
random sample of the individuals in each population is drawn. A treatment is applied to each individual in the
sample and an outcome is measured. The data so obtained are analyzed using an analysis of variance table that
produces an F-test.
212-1
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com
General Linear Models (GLM)
A mathematical model may be formulated that underlies each analysis of variance. This model expresses the
response variable as the sum of parameters of the population. For example, a linear mathematical model for a two-
factor experiment is
Yijk = m + a i + b j + ( ab) ij + eijk
where i=1,2,...,I; j=1,2,...,J; and k=1,2,...,K. This model expresses the value of the response variable, Y, as the sum
of five components:
m the mean.
ai the contribution of the ith level of a factor A.
bj the contribution of the jth level of a factor B.
(ab)ij the combined contribution of the ith level of a factor A and the jth level of a factor B.
eijk the contribution of the kth individual. This is often called the “error.”
Note that this model is the sum of various constants. This type of model is called a linear model. It becomes the
mathematical basis for our discussion of the analysis of variance. Also note that this serves only as an example.
Many linear models could be formulated for the two-factor experiment.
Assumptions
The following assumptions are made when using the F-test.
1. The response variable is continuous.
2. The eijk follow the normal probability distribution with mean equal to zero.
3. The variances of the eijk are equal for all values of i, j, and k.
4. The individuals are independent.
Limitations
There are few limitations when using these tests. Sample sizes may range from a few to several hundred. If your
data are discrete with at least five unique values, you can assume that you have met the continuous variable
assumption. Perhaps the greatest restriction is that your data comes from a random sample of the population. If
you do not have a random sample, the F-test will not work.
When missing cells occur in your design, you must take special care to be sure that appropriate interaction terms
are removed from the ANOVA model.
Special restrictions apply when you are running an analysis with nested terms, as in repeated measures designs.
First of all, you cannot have covariates with nested terms. Second, although the sample sizes of groups (the
“between” factor) may be unequal, all data must be present for each nested factor. For example, if you are running
a pre-post design, you must have both pre- and post- scores for each individual. You cannot include individuals
that have only one or the other.
Multiple Comparison Procedures

The multiple comparison procedures are discussed in the One-Way Analysis of Variance chapter. Note that if
covariates are used, no multiple comparisons are shown.
212-2
Data Structure
The data must be entered in a format that puts the response in one variable and the values of each of the factors in
other variables. An example of the data for a randomized-block design is shown next.
Randomized Block dataset
Block Treatment Response
1 1 123
2 1 230
3 1 279
1 2 245
2 2 283
3 2 245
1 3 182
2 3 252
3 3 280
1 4 203
2 4 204
3 4 227
Procedure Options
This section describes the options available in this procedure.
Variables Tab
These panels specify the variables used in the analysis and the model.
Response Variables
Response Variable(s)
Specifies the response (dependent) variable to be analyzed. If you specify more than one variable here, a separate
analysis is run for each variable.
Covariate Specification
Covariate(s)
One or more covariates may be specified, causing an analysis of covariance (ANCOVA) to be run. Note that you
cannot specify covariates if any of your factors are of the nested type.
Factor Specification
Factor Variable
At least one factor variable must be specified. This variable’s values indicate how the values of the response
variable should be categorized. Examples of factor variables are gender, age groups, “yes” or “no” responses, etc.
Note that the values in the variable may be either numeric or text. The treatment of text variables is specified for
each variable by the Data Type option on the database.
212-3
Type
This option specifies whether the factor is fixed, random, or nested.
A fixed factor includes all possible levels, like male and female for gender, includes representative values across
the possible range of values, like low, medium, and high temperatures, or includes a set of values to which
inferences will be limited, like New York, California, and Maryland.
A random factor is one in which the chosen levels represent a random sample from the population of values. For
example, you might select four classes from the hundreds in your state or you might select ten batches from an
industrial process. The key is that a random sample is chosen. In NCSS, a random factor is “crossed” with other
random and fixed factors. Two factors are crossed when each level of one includes all levels of the other.
A nested factor is a special type of random factor whose levels (values) are not repeated for all combinations of
the factors before it. That is, if factor B is nested in factor A, each level of factor A has its own set of values for
factor B.
For example, suppose that factor A represents three fourth-grade classrooms of twenty students in a particular
state. Further suppose that factor B represents the sixty children in these classrooms. If factors A and B were
crossed, then all sixty children would somehow simultaneously be attending all three classrooms. However, if
each classroom has a mutually exclusive set of twenty children, we say that children are nested within classrooms
or B is nested within A. Notice that nesting occurs when each level of the first factor (the classrooms) contains
separate levels of the second factor (the children).
Note that nested factors should be numbered consecutively, just like random and fixed factors. In the preceding
example, you would number the children from one to sixty. You cannot have two individuals with the same
identification number.
Comparisons
Comparisons are only valid for fixed factors. This option lets you specify comparisons that you want to run on
this factor. A comparison is formulated in terms of the means as follows:
J
Ci = ∑w m
j =1
ij j
In this equation, there are J levels in the factor, the means for each level of the factor are denoted mi, and wij
represents a set of J weight values for the ith comparison. The comparison value, Ci, is tested using a t-test. Note
that if the wij sum to zero across j, the comparison is called a “contrast” of the means.
Comparisons may be specified by simply listing the weights. For example, suppose a factor has three levels
(unique values). Further suppose that the first level represents a control group, the second a treatment at one dose,
and the third a treatment at a higher dose. Three comparisons come to mind: compare each of the treatment groups
to the control group and compare the two treatment groups to each other. These three comparisons would be
Control vs. Treatment 1 -1,1,0
Control vs. Treatment 2 -1,0,1
Treatment 1 vs. Treatment 2 0,-1,1
You might also be interested in comparing the control group with the average of both treatment groups. The
weights for this comparison would be -2,1,1.
When a factor is quantitative, it might be of interest to divide the response pattern into linear, quadratic, cubic, or
other components. If the sample sizes are equal and the factor levels are equally spaced, these so-called
components of trend may be studied by the use of simple contrasts. For example, suppose a quantitative factor has
three levels: 5, 10, and 15. Contrasts to test the linear and quadratic trend components would be
Linear trend -1,0,1
Quadratic trend 1,-2,1
212-4
If the sample sizes for the groups are unequal (the design is unbalanced), adjustments must be made for the
differing sample sizes.
NCSS will automatically generate some of the more common sets of contrasts, or it will let you specify up to
three custom contrasts yourself. The following common sets are designated by this option.
• None
No comparisons are generated.
• Standard Set
This option generates a standard set of contrasts in which the mean of the first level is compared to the
average of the rest, the mean of the second group is compared to the average of those remaining, and so on.
The following example displays the type of contrast generated by this option. Suppose there are four levels
(groups) in the factor. The contrasts generated by this option are:
-3,1,1,1 Compare the first-level mean with the average of the rest.
0,-2,1,1 Compare the second-level mean with the average of the rest.
0,0,-1,1 Compare the third-level mean with the fourth-level mean.
• Polynomial
This option generates a set of orthogonal contrasts that allow you to test various trend components from linear
up to sixth order. These contrasts are appropriate even if the levels are unequally spaced or the group sample
sizes are unequal. Of course, these contrasts are only appropriate for data that are at least ordinal. Usually,
you would augment the analysis of this type of data with a multiple regression analysis.
The following example displays the type of contrasts generated by this option. Suppose there are four equally
spaced levels in the factor and each group has two observations. The contrasts generated by this option are
(scaled to whole numbers):
-3,-1,1,3 Linear component.
1,-1,-1,1 Quadratic component.
-1,3,-3,1 Cubic component.
• Linear Trend
This option generates a set of orthogonal contrasts and retains only the linear component. This contrast is
appropriate even if the levels are unequally spaced and the group sample sizes are unequal. See Orthogonal
Polynomials above for more detail.
• Linear-Quadratic Trend
This option generates the complete set of orthogonal polynomials, but only the results for the first two (the
linear and quadratic) are reported.
• Linear-Cubic Trend
This option generates the complete set of orthogonal polynomials, but only the results for the first three are
reported.
• Linear-Quartic Trend
This option generates the complete set of orthogonal polynomials, but only the results for the first four are
reported.
212-5
• Each with First

This option generates a set of nonorthogonal contrasts appropriate for comparing each of the remaining levels
with the first level. The following example displays the type of contrast generated by this option. Suppose
there are four levels (groups) in the factor. The contrasts generated by this option are:
-1,1,0,0 Compare the first- and second-level means.
-1,0,1,0 Compare the first- and third-level means.
-1,0,0,1 Compare the first- and fourth-level means.
• Each with Last

This option generates a set of nonorthogonal contrasts appropriate for comparing each of the remaining levels
with the last level. The following example displays the type of contrast generated by this option. Suppose
there are four levels (groups) in the factor. The contrasts generated by this option are:
-1,0,0,1 Compare the first- and fourth-level means.
0,-1,0,1 Compare the second- and fourth-level means.
0,0,-1,1 Compare the third- and fourth-level means.
• Custom
This option indicates that the contrasts listed in the corresponding three boxes of the Comparison panel should
be used.
Model Specification
This section specifies the experimental design model.
Which Model Terms
A design in which main effect and interaction terms are included is called a saturated model. Often, it is useful to
omit various interaction terms from the model. This option lets you specify which interactions to keep very easily.
If the selection provided here is not flexible enough for your needs, you can specify custom here and enter the
model directly.
The options included here are as follows.
• Full Model
The complete, saturated model is analyzed. This option requires that you have no missing cells, although you
can have an unbalanced design. Hence, you cannot use this option with Latin square or fractional factorial
designs.
• Up to 1-Way
A main-effects only model is run. All interactions are omitted.
• Up to 2-Way
All main-effects and two-way interactions are included in the model.
• Up to 3-Way
All main-effects, two-way, and three-way interactions are included in the model.
• Up to 4-Way
All main-effects, two-way, three-way, and four-way interactions are included in the model.
212-6
• Custom
This option indicates that you want the Custom Model (given in the next box) to be used.
• Write Model in ‘Custom Model’ Field

When this option is checked, no analysis is performed when the procedure is run. Instead, a copy of the full
model is stored in the Custom Model box. You can then delete selected terms from the model without having
to enter all the terms you want to keep.
Custom Model
When Custom Model (see Which Model Terms above) is selected, the model itself is entered here. If all main
effects and interactions are desired, you can enter the word “ALL” here. For complicated designs, it is usually
easier to check the next option, Write Model in ‘Custom Model’ Field, and run the procedure. The appropriate
model will be generated and placed in this box. You can then edit it as you desire.
The model is entered using letters separated by the plus sign. For example, a three-factor factorial in which only
two-way interactions are needed would be entered as follows:
A+B+AB+C+AC+BC.
A simple repeated-measures design would look like this:
A+B(A)+C+AC+BC(A).
Custom Comparisons Tab

This panel is used when the Comparison option of one or more factors is set to Custom. The Custom option
means that the contrast coefficients are to be entered by the user. The boxes on this panel contain the user-
supplied contrast coefficients. The first row is for factor one, the second row for factor two, and so on.
Custom Comparisons
The following options are only used if Comparisons is set to 'Custom' on the Variables tab.
Custom (1-3)
This option lets you write a user-specified comparison by specifying the weights of that comparison. Note that
there are no numerical restrictions on these coefficients. They do not even have to sum to zero. However, this is
recommended. If the coefficients do sum to zero, the comparison is called a contrast. The significance tests
anticipate that only one or two of these comparisons are to be run. If you run several, you should make some type
of Bonferroni adjustment to your alpha value.
When you put in your own contrasts, you must be careful that you specify the appropriate number of weights. For
example, if the factor has four levels, four weights must be specified, separated by commas. Extra weights are
ignored. If too few weights are specified, the missing weights are set to zero.
These comparison coefficients designate weighted averages of the level-means that are to be statistically tested.
The null hypothesis is that the weighted average is zero. The alternative hypothesis is that the weighted average is
nonzero. The weights (comparison coefficients) are specified here.
As an example, suppose you want to compare the average of the first two levels with the average of the last two
levels in a six-level factor. You would enter “-1,-1,0,0,1,1.”
As a second example, suppose you want to compare the average of the first two levels with the average of the last
three levels in a six-level factor. The contrast would be
-3,-3,0,2,2,2.
Note that in each case, we have used weights that sum to zero. This is why we could not use ones in the second
example.
212-7
Reports Tab
The following options control which reports are displayed.
Select Reports
EMS Report ... Means Report
Specify whether to display the indicated reports.
Test Alpha
The value of alpha for the statistical tests and power analysis. Usually, this number will range from 0.10 to 0.001.
A common choice for alpha is 0.05, but this value is a legacy from the age before computers when only printed
tables were available. You should determine a value appropriate for your particular study.
Multiple Comparison Tests

Note that if covariates are used, no multiple comparison tests are displayed.
Bonferroni Test (All-Pairs) ... Tukey-Kramer Confidence Intervals
These options specify which MC tests and confidence intervals to display.
Tests for Two-Factor Interactions
This option specifies whether multiple comparison tests are generated for two-factor interaction terms. When
checked, the means of two-factor interactions will be tested by each active multiple comparison test. The multiple
comparison test will treat the means as if they came from a single
factor. For example, suppose factor A as two levels and factor B has three levels. The AB interaction would then
have six levels. The active multiple comparison tests would be run on these six means.
Care must be used when interpreting multiple comparison tests on interaction means. Remember that the these
means contain not only the effects of the interaction, but also the main effects of the two factors. Hence these
means contain the combined effects of factor A, factor B, and the AB interaction. You cannot interpret the results
as representing only the AB interaction.
Multiple Comparison Tests – Options

MC Alpha
Specifies the alpha value used by the multiple-comparison tests.
MC Decimals
Specify how many decimals to display in the multiple comparison sections.
Report Options
Precision
Specify the precision of numbers in the report. Single precision will display seven-place accuracy, while the
double precision will display thirteen-place accuracy.
Variable Names
Indicate whether to display the variable names or the variable labels.
Value Labels
Indicate whether to display the data values or their labels.
212-8
Plots Tab
These options specify the plots of group means.
Select Plots
Means Plot(s)
Specify whether to display the indicated plots. Click the plot format button to change the plot settings.
Y-Axis Scaling
Specify the method for calculating the minimum and maximum along the vertical axis. Separate means that each
plot is scaled independently. Uniform means that all plots use the overall minimum and maximum of the data.
This option is ignored if a minimum or maximum is specified.
Example 1 – Running a GLM ANOVA

This section presents an example of how to run an analysis of the data presented above. These data are contained
in the Randomized Block dataset.
You may follow along here by making the appropriate entries or load the completed template Example 1 by
clicking on Open Example Template from the File menu of the General Linear Models (GLM) window.
1 Open the Randomized Block dataset.

• From the File menu of the NCSS Data window, select Open Example Data.
• Click on the file Randomized Block.NCSS.
• Click Open.
2 Open the General Linear Models (GLM) window.

• On the menus, select Analysis, then ANOVA, then General Linear Models (GLM). The General Linear
Models (GLM) procedure will be displayed.
• On the menus, select File, then New Template. This will fill the procedure with the default template.
3 Specify the variables.

• Select the Variables tab.
• Double-click in the Response Variables box. This will bring up the variable selection window.
• Select Response from the list of variables and then click Ok.
• Double-click in the Factor 1 Variable box. This will bring up the variable selection window.
• Select Block from the list of variables and then click Ok.
• Select Random in the Type box for Factor 1.
• Double-click in the Factor 2 Variable box. This will bring up the variable selection window.
• Select Treatment from the list of variables and then click Ok.
• Select Fixed in the Type box for Factor 2.
• Select Linear in the Comparisons box for Factor 2.
4 Specify the reports.

• Select the Reports tab.
• Check the Tukey-Kramer Test option of the Multiple Comparison Tests.
5 Run the procedure.

• From the Run menu, select Run Procedure. Alternatively, just click the green Run button.
212-9
We will now document this output, one section at a time.
Expected Mean Squares Section

Source Term Denominator Expected

Term DF Fixed? Term Mean Square
A (Block) 2 No S(AB) S+bsA
B (Treatment) 3 Yes AB S+sAB+asB
AB 6 No S(AB) S+sAB
S(AB) 0 No S
Note: Expected Mean Squares are for the balanced cell-frequency case.
The expected mean square expressions are provided to show the appropriate error term for each factor. The
correct error term for a factor is that term that is identical except for the factor being tested.
Source Term
The source of variation or term in the model.
DF
The degrees of freedom, which is the number of observations used by this term.
Term Fixed?
Indicates whether the term is fixed or random.
Denominator Term
Indicates the term used as the denominator in the F-ratio.
Expected Mean Square
This expression represents the expected value of the corresponding mean square if the design was completely
balanced. S represents the expected value of the mean square error (sigma). The uppercase letters represent either
the adjusted sum of squared treatment means if the factor is fixed, or the variance component if the factor is
random. The lowercase letter represents the number of levels for that factor, and s represents the number of
replications of the experimental layout.
These EMS expressions are provided to determine the appropriate error term for each factor. The correct error
term for a factor is that term whose EMS is identical except for the factor being tested.
In this example, the appropriate error term for factor B is the AB interaction. The appropriate error term for AB is
S (mean square error). Since there are zero degrees of freedom for S, the terms A and AB cannot be tested.
Analysis of Variance Table Section

Analysis of Variance Table
Source Sum of Mean Prob Power

Term DF Squares Square F-Ratio Level (Alpha=0.05)
A (Blocks) 2 10648.67 5324.333
B (Treatment) 3 4650.917 1550.306 1.09 0.421359 0.177941
AB 6 8507.333 1417.889
S 0 0
Total (Adjusted) 11 23806.92
Total 12
* Term significant at alpha = 0.05
212-10
Source Term
The source of variation, which is the term in the model.
DF
The degrees of freedom, which is the number of observations used by the corresponding model term.
Sum of Squares
This is the sum of squares for this term. It is usually included in the ANOVA table for completeness, not for direct
interpretation.
Mean Square
An estimate of the variation accounted for by this term; it is the sum of squares divided by the degrees of
freedom.
F-Ratio
The ratio of the mean square for this term and the mean square of its corresponding error term. This is also called
the F-test value.
Prob Level
The significance level of the above F-ratio, or the probability of an F-ratio larger than that obtained by this
analysis. For example, to test at an alpha of 0.05, this probability would have to be less than 0.05 to make the F-
ratio significant. Note that if the value is significant at the specified value of alpha, a star is placed to the right of
the F-Ratio.
Power (Alpha=0.05)
Power is the probability of rejecting the hypothesis that the means are equal when they are in fact not equal.
Power is one minus the probability of type II error (β). The power of the test depends on the sample size, the
magnitudes of the variances, the alpha level, and the actual differences among the population means.
The power value calculated here assumes that the population standard deviation is equal to the observed standard
deviation and that the differences among the population means are exactly equal to the differences among the
sample means.
High power is desirable. High power means that there is a high probability of rejecting the null hypothesis when
the null hypothesis is false. This is a critical measure of precision in hypothesis testing.
Generally, you would consider the power of the test when you accept the null hypothesis. The power will give
you some idea of what actions you might take to make your results significant. If you accept the null hypothesis
with high power, there is not much left to do. At least you know that the means are not different. However, if you
accept the null hypothesis with low power, you can take one or more of the following actions:
1. Increase your alpha level. Perhaps you should be testing at alpha = 0.05 instead of alpha = 0.01.
Increasing the alpha level will increase the power.
2. Increasing your sample size will increase the power of your test if you have low power. If you have high
power, an increase in sample size will have little effect.
3. Decrease the magnitude of the variance. Perhaps you can redesign your study so that measurements are
more precise and extraneous sources of variation are removed.
212-11
Means and Standard Errors Section, Plots Sections

Means and Standard Errors , and Plots Sections
Standard
Term Count Mean Error
All 12 229.4167
A: Blocks
1 4 188.25 0
2 4 242.25 0
3 4 257.75 0
B: Treatment
1 3 210.6667 21.74005
2 3 257.6667 21.74005
. . . .
. . . .
Term
The label for this line of the report.
Count
The number of observations in the mean.
Mean
The value of the sample mean.
Standard Error
The standard error of the mean. Note that these standard errors are the square root of the mean square of the error
term for this term divided by the count. These standard errors are not the same as the simple standard errors
calculated separately for each group. The standard errors reported here are those appropriate for testing multiple
comparisons.
Note that the standard errors for the means of Block are zero since there is no error term for this factor. This may
be seen by looking at the Expected Mean Squares Report above.
212-12
Plot of Means
These plots display the means for each factor and two-way interactions. Note how easily you can see patterns in
the plots.
Multiple-Comparison Sections
Tukey-Kramer Multiple-Comparison Test
Response: Response
Term B: Treatment
Alpha=0.050 Error Term=AB DF=6 MSE=1417.889 Critical Value=4.8956
Different
Group Count Mean From Groups
1 3 210.6667
2 3 257.6667
3 3 238
4 3 211.3333
These sections present the results of the multiple-comparison procedures selected. These reports all use a uniform
format that will be described by considering Tukey-Kramer Multiple-Comparison Test. The reports for the other
procedures are similar. For more information on the interpretation of the various multiple-comparison procedures,
turn to the section by that name in the One-Way ANOVA chapter.
Alpha
The level of significance that you selected.
Error Term
The term in the ANOVA model that is used as the error term.
DF
The degrees of freedom for the error term.
MSE
The value of the mean square error.
Critical Value
The value of the test statistic that is “just significant” at the given value of alpha. This value depends on which
multiple-comparison procedure you are using. It is based on the t-distribution or the studentized range
distribution. It is the value of t, F, or q in the corresponding formulas.
Group
The label for this group.
Count
Mean
Different from Groups
A list of those groups that are significantly different from this group according to this multiple-comparison
procedure. All groups not listed are not significantly different from this group.
212-13
Planned-Comparison Section
This section presents the results of any planned comparisons that were selected.
Planned Comparison: B Linear Trend
Response: Response
Term B: Treatment
Alpha=0.050 Error Term=AB DF=6 MSE=1417.889
Comparison Value=-3.950387 T-Value=0.1817 Prob>|T|=0.861794 Decision(0.05)=Do Not Reject

Comparison Standard Error=21.74005 Comparison Confidence Interval = -57.14637 to 49.24559
Comparison
Group Coefficient Count Mean
1 -0.6708204 3 210.6667
2 -0.2236068 3 257.6667
3 0.2236068 3 238
4 0.6708204 3 211.3333
Alpha
The level of significance that you selected.
Error Term
The term in the ANOVA model that is used as the error term.
DF
The degrees of freedom of the error term.
MSE
The value of the mean square error.
Comparison Value
The value of the comparison. This is formed by multiplying the Comparison Coefficient times the Mean for each
group and summing.
T-Value
The t-test used to test whether the above Comparison Value is significantly different from zero.
k
∑c M
i =1
i i
tf =
k
ci2
MSE ∑ i =1
ni
where MSE is the mean square error, f is the degrees of freedom associated with MSE, k is the number of
groups, ci is the comparison coefficient for the ith group, Mi is the mean of the ith group, and ni is the sample size
of the ith group.
Prob>|T|
The significance level of the above T-Value. The Comparison is statistically significant if this value is less than
the specified alpha.
Decision(0.05)
The decision based on the specified value of the multiple comparison alpha.
212-14
Comparison Standard Error

This is the standard error of the estimated comparison value. It is the denominator of the T-Value (above).
Group
The label for this group.
Comparison Coefficient
The coefficient (weight) used for this group. Note that for our example, the weights are appropriate for the linear-
trend component of a set of orthogonal polynomials.
Count
Mean
GLM ANOVA Checklist

This checklist, prepared by a professional statistician, is a flowchart of the steps you should complete to conduct a
valid analysis. Since this topic is vast, this flowchart will give only a brief summary. You should consult
appropriate statistical books in your field for further details. We recommend Winer (1990) and Keppel (1991) as
good books to use, but there are many others available that are equally useful.
Step 1 – Data Preparation
Introduction
This step involves scanning your data for anomalies, keypunch errors, typos, and so on. You would be surprised
how often we hear of people completing an analysis, only to find that they had mistakenly selected the wrong
variables.
Sample Size
The sample size (number of nonmissing rows) has a lot of ramifications. The analysis of variance was originally
developed under the assumption that the sample sizes of each treatment combination are equal. In practice this
seldom happens, but the closer you can get to equal sample sizes the better.
Missing Values
The number and pattern of missing values are always issues to consider. Usually, we assume that missing values
occur at random throughout your data. If this is not true, your results will be biased since a particular segment of
the population is underrepresented.
If you have missing values, it will be important to identify the degree of unbalance in your design. You should
also check to see if there are any missing cells. If there are, you cannot run a full model. You will have to assume
some interactions are zero and remove them from the ANOVA model.
Type of Data
The mathematical basis of the F-test assumes that the data are continuous. Because of the rounding that occurs
when data are recorded, all data are technically discrete. The validity of assuming the continuity of the data then
comes down to determining when we have too much rounding. For example, most statisticians would not worry
about human-age data that was rounded to the nearest year. However, if these data were rounded to the nearest ten
212-15
years or further to only three groups (young, adolescent, and adult), most statisticians question the validity of the
probability statements. Some studies have shown that the F-test is reasonably accurate when the data have only
five possible values (most would call this discrete data). If your data contain less than five unique values, any
probability statements made are tenuous.
Also, you should double-check to ensure that you are going to use the appropriate design. Our experience is that
many researchers use a factorial design when they should be using a repeated measures design. Consider again the
examples of each type of design and make sure you are using the correct one.
Outliers
Generally, outliers cause distortion in most popular statistical tests. You must scan your data for outliers (the box
plot is an excellent tool for doing this). If you have outliers, you have to decide if they are one-time occurrences
or if they would occur in another sample. If they are one-time occurrences, you can remove them and proceed. If
you know they represent a certain segment of the population, you have to decide between biasing your results (by
removing them) or leaving them in and invalidating the normality assumption.
Step 2 – Setup and Run the GLM ANOVA Panel
Introduction
Now comes the fun part: running the program. NCSS is designed to be simple to operate, but it can still seem
complicated. When you go to run a procedure such as this for the first time, take a few minutes to read through
the chapter again and familiarize yourself with the issues involved.
Enter Variables
The templates are set with ready-to-run defaults. About all you have to do is select the appropriate variables
(columns of data).
Select All Plots

As a rule, you should select the means plots. They add a great deal to your ability to interpret the data.
Specify Alpha
Most beginners at statistics forget this important step and let the alpha value default to the standard 0.05. You
should consciously decide what value of alpha is appropriate for your study. The 0.05 default came about when
people had to rely on printed probability tables and there were only two values available: 0.05 or 0.01. Now you
can set the value to whatever is appropriate.
A special note on setting the Multiple Comparison alpha. You will often want to reset this value to 0.10 so that the
individual tests are made at a more reasonable significance level.
Step 3 – Check Assumptions
Introduction
Testing the assumptions of normality and equal variance is often difficult in a multi-way analysis of variance. We
suggest that you make several passes through your data using our one-way ANOVA program, studying each
factor separately. We suggest this because the one-way ANOVA program displays extensive diagnostic
information for checking equal variance and normality. Although this method does not account for the
interactions among the factors, it is often the best you can do to assess the validity of your assumptions.
212-16
Sometimes, the ANOVA model can be recoded so that you can run it through our regression program. When this
is possible, you can analyze the residuals to assess normality and equal variance.
Random Sample
These statistical procedures were designed with the assumption that the sample population was selected randomly.
The validity of this assumption depends on the method used to select the sample. If you have not used valid
sampling techniques, the F-test will not work.
Check Descriptive Statistics

You should check the Means and Standard Errors Section first to determine if the Counts and the Means are
reasonable. If you have selected the wrong variable, these values will alert you.
Step 4 – Interpret Findings
Introduction
You are now ready to conduct your tests. The basic plan of attack for analyzing your output is as follows:
1. Glance through the reports, checking the means, the F-tests, and so forth for obvious problems.
2. Look at the power of the nonsignificant tests. Could the lack of significance be the result of a small
sample size?
3. Determine which main effects and interactions are significant.
4. Use care in interpreting a main effect when its interaction with another term is significant.
5. Use planned comparisons, paired comparisons, and plots of means to view the experimental results and
discuss what they reveal.
Examples of Various Experimental Designs

We will now present examples of how to run various popular types of experimental designs.
Randomized-Block Design
The randomized-block design is a very popular experimental design. The focus of the analysis is on a set of two
or more treatments. A blocking variable is used to account for extraneous factors. Each block receives all
treatments. These treatments are randomly assigned within the block.
The data in the Randomized Block dataset show how to enter the data for this type of design. You should
designate the block term as random and the treatment term as fixed. Set the Which Model Terms option to Up to
1-Way (removing the interaction term). In a typical randomized-block design, the interaction term becomes the
error term, so it does not have to be fit separately. Doing this will reduce the amount of time needed to complete
the calculations.
Single-Factor Repeated-Measures Design

The single-factor repeated-measures design is similar to the randomized-block design. In this design the
individuals (analogous to the blocks) are measured over time. Unlike the randomized-block design, however, the
treatments are not applied in random order. Instead, the treatments are always applied in the same order. For
212-17
example, you might conduct a pre-test, apply some treatment to the individuals, and conduct a post-test. You
cannot apply the post-test first.
The data in the Randomized Block dataset show how to enter the data for this type of design if you think of
blocks as the individuals and treatments as time of measurement.
It turns out that even though the randomization method is different, the analysis of this design is identical to that
described above for the randomized-block design. The individuals become the blocks. This variable is designated
random. The repeated-measures variable (the variable representing time) becomes the treatment. This variable is
designated as fixed. Set the Which Model Terms option (Model Tab) to Up to 1-Way to omit the interaction term.
Example 2 – Latin-Square Design

Fractional-rep designs are known for their ability to provide insight about several factors with a minimum number
of observations. This efficiency comes from an experimental setup that ignores many interaction terms. The Latin
square is one such design. It may be analyzed with NCSS. The following table shows a set of Latin-square data
from page 313 of Snedecor and Cochran (1972).
Latin-Square Data from Snedecor and Cochran
Column
Row 1 2 3 4 5
1 B:257 E:230 A:279 C:287 D:202
2 D:245 A:283 E:245 B:280 C:260
3 E:182 B:252 C:280 D:246 A:250
4 A:203 C:204 D:227 E:193 B:259
5 C:231 D:271 B:266 A:334 E:338
The following table shows the data as it would be entered for analysis in NCSS. The Custom Model statement
“A+B+C” would be used since many of the interactions cannot be estimated. The factors would be designated as
fixed or random depending on the experimental situation.
Latin Square dataset

Rows Columns Letters Yield
1 1 B 257
1 2 E 230
1 3 A 279
1 4 C 287
1 5 D 202
2 1 D 245
2 2 A 283
2 3 E 245
2 4 B 280
2 5 C 260
3 1 E 182
3 2 B 252
3 3 C 280
3 4 D 246
3 5 A 250
4 1 A 203
4 2 C 204
4 3 D 227
4 4 E 193
4 5 B 259
212-18
Latin Square dataset (continued)

Rows Columns Letters Yield
5 1 C 231
5 2 D 271
5 3 B 266
5 4 A 334
5 5 E 338
Example 3 – Repeated-Measures Design

A Repeated Measures ANOVA is a particular type of three-factor design that uses two error terms. In this design,
treatments are applied to experimental units of different sizes. For example, in an educational study, one treatment
might be applied to whole classrooms. A second treatment might consist of the students’ responses to a pre-test
and a post-test. Such a design employs two error terms. One error term is the between-classes error for testing the
first factor. The other error term is the within-student error for testing the second factor.
This procedure analyzes data from an experimental design represented by the following mathematical model:
Yijkl = µ + Ai + Sij + Bk + ABik + eijkl
In this model, A is the between-group treatment, Sij is the between-group error, B is the within-subject treatment,
and eijkl is the within-subject error.
This is a specialized technique with strict assumptions. An advanced statistical text dealing with the topic should
be consulted before the technique is employed.
The data below illustrate how the data should be set up. An experiment was conducted to study the effects of
exercise on heart rate. The subjects were randomly divided into three groups of six. The first group did not have a
regular exercise plan. The second group exercised once a week. The third group exercised daily. Each subject’s
heart rate was recorded when the experiment began and again at the end of ten weeks. These data are stored in a
dataset called Heart. You might want to open this dataset and run the analysis yourself.
Heart dataset
Exercise Subject Time Heart Rate

None 1 0 87
None 1 10 89
None 2 0 67
None 2 10 65
None 3 0 55
None 3 10 58
None 4 0 66
None 4 10 68
None 5 0 88
None 5 10 90
None 6 0 75
None 6 10 73
Weekly 7 0 84
Weekly 7 10 78
Weekly 8 0 78
Weekly 8 10 72
Weekly 9 0 64
Weekly 9 10 53
Weekly 10 0 73
212-19
Heart dataset (continued)
Exercise Subject Time Heart Rate

Weekly 10 10 65
Weekly 11 0 84
Weekly 11 10 82
Weekly 12 0 55
Weekly 12 10 53
Daily 13 0 72
Daily 13 10 55
Daily 14 0 83
Daily 14 10 72
Daily 15 0 75
Daily 15 10 63
Daily 16 0 55
Daily 16 10 49
Daily 17 0 83
Daily 17 10 68
Daily 18 0 63
Daily 18 10 54
To run this analysis, you specify Heart Rate as the Response Variable, Exercise as Factor 1 (designate it as fixed),
Subject as Factor 2 (designate it as nested), and Time as Factor 3 (designate it as fixed). Select the full model.
When the analysis is complete, the following output is displayed.
Repeated-Measure ANOVA Report for Heart Rate Data

A: Exercise 2 331.1667 165.5833 0.61 0.555761 0.133134
B(A): Subject 15 4064.833 270.9889
C: Time 1 277.7778 277.7778 50.51 0.000004* 0.999998
AC 2 234.7222 117.3611 21.34 0.000041* 0.999793
BC(A) 15 82.5 5.5
S 0 0
Total (Adjusted) 35 4991
Total 36
Means and Standard Error Section

Standard
Term Count Mean Error
All 36 69.83334
A: Exercise
Daily 12 66 4.752095
None 12 73.41666 4.752095
Weekly 12 70.08334 4.752095
C: Time
0 18 72.61111 .5527708
10 18 67.05556 .5527708
AC: Exercise,Time
Daily,0 6 71.83334 .9574271
Daily,10 6 60.16667 .9574271
None,0 6 73 .9574271
None,10 6 73.83334 .9574271
Weekly,0 6 73 .9574271
Weekly,10 6 67.16666 .9574271
212-20
Example 4 – Analysis of Covariance

The analysis of covariance uses features from both analysis of variance and multiple regression. The usual one-
way classification model in analysis of variance is
Yij = µi + e1ij
where Yij is the j observation in the i group, µi represents the true mean of the ith group, and eij are the residuals
th th
or errors in the above model (usually assumed to be normally distributed). Suppose you have measured a second
variable with values Xij that is linearly related to Y. Further suppose that the slope of the relationship between Y
and X is constant from group to group. You could then write the analysis of covariance model
Yij = µi + ß(Xij-X..) + e2ij
where X.. represents the overall mean of X. If X and Y are closely related, you would expect that the errors, e2ij,
would be much smaller than the errors, e1ij, giving you more precise results.
The analysis of covariance is useful for many reasons, but it does have the (highly) restrictive assumption that the
slope is constant over all the groups. This assumption is often violated, which limits the technique’s usefulness.
You will want to study more about this technique in statistical texts before you use it.
Running an analysis of covariance is easy in NCSS. You fill out the procedure template as usual for an ANOVA.
To change your ANOVA into an ANCOVA, you simply specify one or more covariates. We will now take you
through an extended example showing how to run an Ancova as well as how to test the assumption of equal
slopes. The following data give the home state, age, and IQ of thirty teenagers. The variables X1-X4 are for use in
testing the Ancova assumption of equal slopes and they will be explained later.
Suppose we wish to test for differences in IQ among the three states while controlling for age (the covariate).
These data are contained in the ANCOVA database. You should open this database now if you want to follow
along.
ANCOVA dataset
State Age IQ X1 X2 X3 X4
Iowa 12 100 -1 -1 -12 -12
Iowa 13 102 -1 -1 -13 -13
Iowa 12 97 -1 -1 -12 -12
Iowa 14 96 -1 -1 -14 -14
Iowa 15 105 -1 -1 -15 -15
Iowa 18 106 -1 -1 -18 -18
Iowa 12 105 -1 -1 -12 -12
Iowa 14 103 -1 -1 -14 -14
Iowa 12 99 -1 -1 -12 -12
Iowa 10 98 -1 -1 -10 -10
Utah 14 104 0 2 0 28
Utah 11 105 0 2 0 22
Utah 12 106 0 2 0 24
Utah 15 103 0 2 0 30
Utah 17 102 0 2 0 34
Utah 18 99 0 2 0 36
Utah 19 107 0 2 0 38
Utah 16 105 0 2 0 32
Utah 15 103 0 2 0 30
Utah 14 103 0 2 0 28
Texas 15 105 1 -1 15 -15
Texas 16 106 1 -1 16 -16
Texas 12 103 1 -1 12 -12
212-21
ANCOVA dataset (continued)
State Age IQ X1 X2 X3 X4
Texas 13 99 1 -1 13 -13
Texas 14 93 1 -1 14 -14
Texas 11 104 1 -1 11 -11
Texas 18 103 1 -1 18 -18
Texas 19 100 1 -1 19 -19
Texas 18 101 1 -1 18 -18
Texas 16 104 1 -1 16 -16
We begin by loading the ANCOVA dataset and the GLM ANOVA options panel. We specify IQ as the Response
Variable, Age as the Covariate, and State as Factor 1. We run the procedure and the analysis of covariance table
is displayed.
Analysis of Covariance Report


X(Age) 1 5.239314 5.239314 0.47 0.497230 0.101741
A: State 2 28.38448 14.19224 1.28 0.293886 0.253298
S 26 287.3607 11.05233
Total 30
Notice that now, in addition to the test for factor A, we also have a test for the covariate. This test, the one along
the line labeled “X(Age),” tests the significance of the covariate. If it is not significant (as is the case in this
example), analysis of covariance should not be used. However, if it is significant, you may proceed to the next F-
test, the one dealing with factor A (State). This is the test that is usually desired in the analysis of covariance. It
tests whether the adjusted means of the three states are different. The means are adjusted as if all three states had
the same age. That is, the means for each state are adjusted to the average value of age. These adjusted
means are shown in the Means and Effects report. If you run the analysis without the covariate, you'll notice that
these means are different.
Since the covariate (Age) is not significant, you should stop here. However, for the sake of instruction, we will
assume that the covariate is significant and proceed to test whether the slopes between IQ and Age are the same in
the three states. The following steps will lead you through this test:
1. Construct a new contrast variable for each degree of freedom of the factor. In our current example, the
three levels (states) of factor A yield two degrees of freedom, so we must create two contrast variables.
These are shown as X1 and X2.
2. Multiply each of these new variables by the covariate variable. In our example, X3=(X1)(Age) and
X4=(X2)(Age).
3. Run another ANCOVA, using the same setup as before except now you fit the three covariates Age, X3,
and X4. Call these the Model 2 results, and call the previous results with just the single covariate the
Model 1 results.
212-22
Second Analysis of Covariance Report


X(Age) 1 9.740934 9.740934 0.94 0.341886 0.153727
X(X3) 1 22.27164 22.27164 2.15 0.155572 0.290793
X(X4) 1 21.07455 21.07455 2.03 0.166672 0.277862
A: State 2 46.57466 23.28733 2.25 0.127408 0.412350
S 24 248.6402 10.36001
Total 30
4. Finally, create the F-test for equality of slopes as follows. The formula is
( SSE1 − SSE2 ) / k
Fk ,m =
MSE2
where k is the degrees of freedom of the factor (in our example, this is 2), m is the degrees of freedom of
the mean square for error in model2, SSE1 and SSE2 are the sums of squares error for model1 and model2,
and MSE2 is the mean square for error in model2.
The calculations for this example proceed as follows:
F2,24 = [(287.3607-248.6402)/2]/10.36001 = 1.86875.
This F-ratio would then be compared against a tabulated 0.05 F-value, 3.403, which you could find in the
probability calculator or in a statistics book. Since 1.86875 < 3.403, we would not reject the equality of
slopes assumption in this case.
One final note, you should generate a scatter plot, which shows the response variable on the vertical axis, the
covariate on the horizontal axis, and uses different symbols for each group. The least squares trend line can also
be displayed. This plot will let you visually assess the validity of the assumption of equal slope.
Example 5 – Hierarchical-Classification Design

Snedecor and Cochran (1967), page 286, present an example of a hierarchical-classification design. In this
example, four plants were selected at random, and three leaves were randomly selected from each plant. Two
samples were taken from each leaf, and the amount of calcium in the sample was recorded. The data are displayed
below. The data are stored in a database called Plant.
Plant dataset
Row Plant Leaf Calcium

1 1 1 3.28
2 1 1 3.09
3 1 2 3.52
4 1 2 3.48
5 1 3 2.88
6 1 3 2.80
7 2 4 2.46
8 2 4 2.44
9 2 5 1.87
10 2 5 1.92
11 2 6 2.19
12 2 6 2.19
13 3 7 2.77
14 3 7 2.66
212-23
Plant dataset (continued)
Row Plant Leaf Calcium

15 3 8 3.74
16 3 8 3.44
17 3 9 2.55
18 3 9 2.55
19 4 10 3.78
20 4 10 3.87
21 4 11 4.07
22 4 11 4.12
23 4 12 3.31
24 4 12 3.31
To run this analysis, you specify Calcium as the Response Variable, Plant as Factor 1 (designate it as random),
and Leaf as Factor 2 (designate it as nested). Select the Full Model. When the analysis is complete, the following
output is displayed.
Plant Data Example Report


A: Plant 3 No B(A) S+sB+bsA
B(A): Leaf 8 No S(AB) S+sB
S(AB) 12 No S
Note: Expected Mean Squares are for the balanced cell-frequency case.

A: Plant 3 7.560346 2.520115 7.67 0.009725*
B(A): Leaf 8 2.6302 0.328775 49.41 0.000000*
S 12 0.07985 0.006654167
Total 24
212-24
Example 6 – Split-Split-Plot Design

Gomez and Gomez (1984), page 142-153, present an example of a split-split-plot design. This design is an
extension of the split-plot design in which the plots are divided into subplots. Thus there are two nesting factors
and three levels of precision. There are three plot sizes: the main plot, the subplot, and the sub-subplot.
Their example is of a 5 x 3 x 3 factorial experiment with three replications (blocks). The treatments are five levels
of nitrogen on a main plot, three types of management applied to the subplot, and three varieties of rice applied to
sub-subplots.
The data are too lengthy to be presented here, but they are contained in the Grain Yield dataset.
You may follow along here by making the appropriate entries or load the completed template Example 6 by
clicking on Open Example Template from the File menu of the General Linear Models (GLM) window.
1 Open the Grain Yield dataset.

• From the File menu of the NCSS Data window, select Open Example Data.
• Click on the file Grain Yield.NCSS.
• Click Open.
2 Open the General Linear Models (GLM) window.

• On the menus, select Analysis, then ANOVA, then General Linear Models (GLM). The General Linear
Models (GLM) procedure will be displayed.
• On the menus, select File, then New Template. This will fill the procedure with the default template.
3 Specify the variables and model.

• Select the Variables tab.
• Set the Response Variable to Yield.
• Set Factor Variable 1 to Rep.

• Set its Type to Random.
• Set Factor Variable 2 to Nitrogen.

• Set its Type to Fixed.
• Set Factor Variable 3 to Plot.

• Set its Type to Nested.
• Set Factor Variable 4 to Management.

• Set Factor Variable 5 to Subplot.

• Set its Type to Nested.
• Set Factor Variable 6 to Variety.

• Set Which Model Terms to Custom Model.

• Set the Custom Model to A+B+C(AB)+D+BD+CD(AB)+F+BF+DF+BDF.
4 Specify the reports.

• Select the Reports tab.
• Check the EMS Report and the ANOVA Report, and the Means Report.
212-25
5 Run the procedure.

• From the Run menu, select Run Procedure. Alternatively, just click the green Run button.
The following reports will appear.


A: Rep 2 No C(AB) S+defsC+bcdefsA
B: Nitrogen 4 Yes C(AB) S+defsC+acdefsB
C(AB): Plot 8 No S(ABCDEF) S+defsC
D: Management 2 Yes CD(AB) S+efsCD+abcefsD
BD 8 Yes CD(AB) S+efsCD+acefsBD
CD(AB) 20 No S(ABCDEF) S+efsCD
F: Variety 2 Yes S(ABCDEF) S+abcdesF
BF 8 Yes S(ABCDEF) S+acdesBF
DF 4 Yes S(ABCDEF) S+abcesDF
BDF 16 Yes S(ABCDEF) S+acesBDF
S(ABCDEF) 60 No S
Analysis of Variance Table Section


A: Rep 2 0.7319945 0.3659973 0.66 0.543910
B: Nitrogen 4 61.64082 15.41021 27.70 0.000097* 0.999999
C(AB): Plot 8 4.451351 0.5564188 1.12 0.361290
D: Management 2 42.93611 21.46805 82.00 0.000000* 1.000000
BD 8 1.102973 0.1378717 0.53 0.822648 0.182126
CD(AB) 20 5.236335 0.2618167 0.53 0.942667
F: Variety 2 206.0132 103.0066 207.87 0.000000* 1.000000
BF 8 14.14451 1.768063 3.57 0.001916* 0.969208
DF 4 3.851769 0.9629423 1.94 0.114899 0.552586
BDF 16 3.699232 0.231202 0.47 0.953759 0.264199
S 60 29.73249 0.4955415
Total 135
A look at Gomez and Gomez (1984) page 153 will show that these are the exact same answers that were given
there.
212-26

General Linear Models (GLM)

Uploaded by

Copyright:

Available Formats

General Linear Models (GLM)

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

General Linear Models (GLM)

Uploaded by

Copyright:

Available Formats

NCSS Statistical Software NCSS.

Kinds of Research Questions

The Linear Model

Multiple Comparison Procedures

• Each with First

• Each with Last

• Write Model in ‘Custom Model’ Field

Custom Comparisons Tab

Multiple Comparison Tests

Multiple Comparison Tests – Options

Example 1 – Running a GLM ANOVA

1 Open the Randomized Block dataset.

2 Open the General Linear Models (GLM) window.

3 Specify the variables.

4 Specify the reports.

5 Run the procedure.

We will now document this output, one section at a time.

Expected Mean Squares Section

Source Term Denominator Expected

Analysis of Variance Table Section

Source Sum of Mean Prob Power

* Term significant at alpha = 0.05

Means and Standard Errors Section, Plots Sections

Alpha=0.050 Error Term=AB DF=6 MSE=1417.889 Critical Value=4.8956

Planned Comparison: B Linear Trend

Alpha=0.050 Error Term=AB DF=6 MSE=1417.889

Comparison Value=-3.950387 T-Value=0.1817 Prob>|T|=0.861794 Decision(0.05)=Do Not Reject

Comparison Standard Error

GLM ANOVA Checklist

Step 1 – Data Preparation

Step 2 – Setup and Run the GLM ANOVA Panel

Select All Plots

Step 3 – Check Assumptions

Check Descriptive Statistics

Step 4 – Interpret Findings

Examples of Various Experimental Designs

Single-Factor Repeated-Measures Design

Example 2 – Latin-Square Design

Latin Square dataset

Latin Square dataset (continued)

Example 3 – Repeated-Measures Design

Exercise Subject Time Heart Rate

Heart dataset (continued)

Exercise Subject Time Heart Rate

Repeated-Measure ANOVA Report for Heart Rate Data

Means and Standard Error Section

Example 4 – Analysis of Covariance

ANCOVA dataset (continued)

Analysis of Covariance Report

Source Sum of Mean Prob Power

Second Analysis of Covariance Report

Source Sum of Mean Prob Power

Example 5 – Hierarchical-Classification Design

Row Plant Leaf Calcium

Plant dataset (continued)

Row Plant Leaf Calcium

Plant Data Example Report

Source Term Denominator Expected

Analysis of Variance Table

Source Sum of Mean Prob Power