7.

5: Choosing the Right Test

Learning Objectives
This table is designed to help you decide which statistical test or descriptive statistic is appropriate for your
experiment. In order to use it, you must be able to identify all the variables in the data set and tell what kind of
variables they are.

measurement
test nominal variables ranked variables purpose notes example
variables

count the number

of red, pink and
test fit of
white flowers in a
observed use for small
Exact test for genetic cross, test
1 – – frequencies to sample sizes (less
goodness-of-fit fit to expected
expected than 1000)
1 : 2 : 1 ratio,
frequencies
total sample
< 1000

count the number

of red, pink and
test fit of
use for large white flowers in a
observed
Chi-square test of sample sizes genetic cross, test
1 – – frequencies to
goodness-of-fit (greater than fit to expected
expected
1000) 1 : 2 : 1 ratio,
frequencies
total sample
> 1000

count the number

of red, pink and
test fit of
used for large white flowers in a
observed
G–test of sample sizes genetic cross, test
1 – – frequencies to
goodness-of-fit (greater than fit to expected
expected
1000) 1 : 2 : 1 ratio,
frequencies
total sample
> 1000

test fit of count the number

observed of red, pink and
frequencies to white flowers in a
Repeated G–tests
2 – – expected - genetic cross, test
of goodness-of-fit
frequencies in fit to expected
multiple 1 : 2 : 1 ratio, do

experiments multiple crosses

measurement
test nominal variables ranked variables purpose notes example
variables

count the number

of live and dead
patients after
treatment with
test hypothesis drug or placebo,
use for small
that proportions test the hypothesis
Fisher's exact test 2 – – sample sizes (less
are the same in that the proportion
than 1000)
different groups of live and dead is
the same in the
two treatments,
total sample
< 1000

John H. McDonald 7.5.1 1/14/2022 https://stats.libretexts.org/@go/page/1764

Chi-square test of 2 – – test hypothesis use for large count the number
independence that proportions sample sizes of live and dead
are the same in (greater than patients after
different groups 1000) treatment with
drug or placebo,
test the hypothesis
that the proportion
of live and dead is
the same in the
two treatments,
total sample
> 1000

count the number

of live and dead
patients after
treatment with
test hypothesis drug or placebo,
large sample sizes
G–test of that proportions test the hypothesis
2 – – (greater than
independence are the same in that the proportion
1000)
different groups of live and dead is
the same in the
two treatments,
total sample
> 1000

count the number

of live and dead
patients after
treatment with
test hypothesis alternate drug or placebo,
that proportions hypothesis is a test the hypothesis
Cochran-Mantel-
3 – – are the same in consistent that the proportion
Haenszel test
repeated pairings direction of of live and dead is
of two groups difference the same in the
two treatments,
repeat this
experiment at
different hospitals
measurement
test nominal variables ranked variables purpose notes example
variables

description of
Arithmetic mean – 1 – central tendency - -
of data
median height of
trees in forest, if
most trees are
description of more useful than
short seedlings
Median – 1 – central tendency mean for very
and the mean
of data skewed data
would be skewed
by a few very tall
trees
used more in
description of everyday life than
Range – 1 – -
dispersion of data in scientific
statistics
Variance – 1 – description of forms the basis of -
dispersion of data many statistical

John H. McDonald 7.5.2 1/14/2022 https://stats.libretexts.org/@go/page/1764

 tests; in squared
units, so not very
understandable
in same units as
original data, so
Standard description of
– 1 – more -
deviation dispersion of data
understandable
than variance
description of
Standard error of accuracy of an
– 1 – - -
the mean estimate of a
mean
description of
Confidence accuracy of an
– 1 – - -
interval estimate of a
mean
measurement
test nominal variables ranked variables purpose notes example
variables

test the hypothesis

that the mean blindfold people,
value of the ask them to hold
One-sample t–test – 1 – measurement - arm at 45 angle,
∘

variable equals a see if mean angle

theoretical is equal to 45 ∘

expectation
test the hypothesis
that the mean just another name compare mean
values of the for one-way heavy metal
Two-sample t–test 1 1 – measurement anova when there content in mussels
variable are the are only two from Nova Scotia
same in two groups and New Jersey
groups
compare mean
test the hypothesis heavy metal
that the mean content in mussels
values of the from Nova Scotia,
One-way anova 1 1 – measurement - Maine,
variable are the Massachusetts,
same in different Connecticut, New
groups York and New
Jersey
compare mean
heavy metal
after a significant content in mussels
one-way anova, from Nova Scotia
Tukey-Kramer test for significant vs. Maine, Nova
1 1 – -
test differences Scotia vs.
between all pairs Massachusetts,
of groups Maine vs.
Massachusetts,
etc.
Bartlett's test 1 1 – test the hypothesis usually used to compare standard
that the standard see whether data deviation of heavy
deviation of a fit one of the metal content in
measurement mussels from

John H. McDonald 7.5.3 1/14/2022 https://stats.libretexts.org/@go/page/1764

 variable is the assumptions of an Nova Scotia,
same in different anova Maine,
groups Massachusetts,
Connecticut, New
York and New
Jersey
measurement
test nominal variables ranked variables purpose notes example
variables

compare mean
heavy metal
test hypothesis content in mussels
that the mean from Nova Scotia,
values of the Maine,
measurement Massachusetts,
subgroups must
variable are the Connecticut, New
Nested anova 2+ 1 – be arbitrary
same in different York and New
(model II)
groups, when Jersey; several
each group is mussels from each
divided into location, with
subgroups several metal
measurements
from each mussel
compare
test the hypothesis
cholesterol levels
that different
in blood of male
groups, classified
vegetarians,
Two-way anova 2 1 – two ways, have -
female
the same means of
vegetarians, male
the measurement
carnivores, and
variable
female carnivores
test the hypothesis just another name compare the
that the means of for two-way cholesterol level
the continuous anova when one in blood of people
Paired t–test 2 1 –
variable are the nominal variable before vs. after
same in paired represents pairs of switching to a
data observations vegetarian diet
compare the
test the hypothesis cholesterol level
that the means of used when the in blood of people
Wilcoxon signed- the measurement differences of before vs. after
2 1 –
rank test variable are the pairs are severely switching to a
same in paired non-normal vegetarian diet,
data when differences
are non-normal
measurement
test nominal variables ranked variables purpose notes example
variables

Linear regression – 2 – see whether - measure chirping

variation in an speed in crickets
independent at different
variable causes temperatures, test
some of the whether variation
variation in a in temperature
dependent causes variation in
variable; estimate chirping speed; or
the value of one use the estimated
unmeasured relationship to

John H. McDonald 7.5.4 1/14/2022 https://stats.libretexts.org/@go/page/1764

 variable estimate
corresponding to temperature from
a measured chirping speed
variable when no
thermometer is
available
measure salt
intake and fat
intake in different
see whether two
Correlation – 2 – - people's diets, to
variables covary
see if people who
eat a lot of fat also
eat a lot of salt
test the hypothesis
that an equation
with X , X , etc.
2 3

Polynomial
– 2 – fits the Y variable - -
regression
significantly
better than a
linear regression
measure chirping
speed vs.
first test the temperature in
test the hypothesis homogeneity of four species of
Analysis of that different slopes; if they are crickets, see if
covariance 1 2 – groups have the not significantly there is significant
(ancova) same regression different, test the variation among
lines homogeneity of the species in the
the Y -intercepts slope or Y -
intercept of the
relationships
measurement
test nominal variables ranked variables purpose notes example
variables

measure air
temperature,
fit an equation
humidity, body
Multiple relating several X
– 3+ – - mass, leg length,
regression variables to a
see how they
single Y variable
relate to chirping
speed in crickets
fit an equation
give different
relating an
doses of a drug
independent
(the measurement
measurement
Simple logistic variable), record
1 1 – variable to the -
regression who lives or dies
probability of a
in the next year
value of a
(the nominal
dependent
variable)
nominal variable
Multiple logistic 1 2+ – fit an equation - record height,
regression relating more than weight, blood
one independent pressure, age of
measurement multiple people,
variable to the see who lives or
probability of a dies in the next
value of a year

John H. McDonald 7.5.5 1/14/2022 https://stats.libretexts.org/@go/page/1764

 dependent
nominal variable
measurement
test nominal variables ranked variables purpose notes example
variables

compare the
cholesterol level
in blood of people
test randomness before vs. after
of direction of switching to a
Sign test 2 – 1 -
difference in vegetarian diet,
paired data only record
whether it is
higher or lower
after the switch
40 ears of corn (8
from each of 5
test the hypothesis often used as a varieties) are
Kruskal–Wallis that rankings are non-parametric ranked for
1 – 1
test the same in alternative to one- tastiness, and the
different groups way anova mean rank is
compared among
varieties
40 ears of corn are
often used as a
ranked for
see whether the non-parametric
Spearman rank tastiness and
– – 2 ranks of two alternative to
correlation prettiness, see
variables covary regression or
whether prettier
correlation
corn is also tastier

Contributor
John H. McDonald (University of Delaware)

John H. McDonald 7.5.6 1/14/2022 https://stats.libretexts.org/@go/page/1764