Sample Size And Gpower Module

Sample Size Calculation
with GPower
Dr. Mark Williamson, Statistician
Biostatistics, Epidemiology, and Research Design Core
DaCCoTA

Purpose
◦ This Module was created to provide
instruction and examples on sample size
calculations for a variety of statistical tests
on behalf of BERDC
◦ The software used is GPower, the
premiere free software for sample size
calculation that can be used in Mac or
Windows
https://dl2.macupdate.com/images/icons256/24037.png

Background
◦ The Biostatistics, Epidemiology, and Research Design
Core (BERDC) is a component of the DaCCoTA
program
◦ Dakota Cancer Collaborative on Translational Activity
has as its goal to bring together researchers and
clinicians with diverse experience from across the
region to develop unique and innovative means of
combating cancer in North and South Dakota
◦ If you use this Module for research, please reference
the DaCCoTA project

The Why of Sample Size Calculation
◦ In designing an experiment, a key question is:
How many animals/subjects do I need for my experiment?
◦ Too small of a sample size can under-detect the effect of interest in
your experiment
◦ Too large of a sample size may lead to unnecessary wasting of
resources and animals
◦ Like Goldilocks, we want our sample size to be ‘just right’
◦ The answer: Sample Size Calculation
◦ Goal: We strive to have enough samples to reasonably detect an
effect if it really is there without wasting limited resources on too
many samples.
https://upload.wikimedia.org/wikipedia/commons/thumb/e/ef/The_Three_Bears_-
_Project_Gutenberg_eText_17034.jpg/1200px-The_Three_Bears_-_Project_Gutenberg_eText_17034.jpg

Key Bits of Sample Size Calculation
Effect size: magnitude of the effect under the alternative hypothesis
• The larger the effect size, the easier it is to detect an effect and require
fewer samples
Power: probability of correctly rejecting the null hypothesis if it is false
• AKA, probability of detecting a true difference when it exists
• Power = 1-β, where β is the probability of a Type II error (false negative)
• The higher the power, the more likely it is to detect an effect if it is
present and the more samples needed
• Standard setting for power is 0.80
Significance level (α): probability of falsely rejecting the null hypothesis even
though it is true
• AKA, probability of a Type I error (false positive)
• The lower the significance level, the more likely it is to avoid a false
positive and the more samples needed
• Standard setting for α is 0.05
• Given those three bits, and other information based on the specific design,
you can calculate sample size for most statistical tests
https://images-na.ssl-images-amazon.com/images/I/61YIBfLPPuL._SX355_.jpg

Effect Size in detail
◦ While Power and Significance level are usually set irrespective of
the data, the effect size is a property of the sample data
◦ It is essentially a function of the difference between the means of
the null and alternative hypotheses over the variation (standard
deviation) in the data
How to estimate Effect Size:
A. Use background information in the form of preliminary/trial data to
get means and variation, then calculate effect size directly
B. Use background information in the form of similar studies to get
means and variation, then calculate effect size directly
C. With no prior information, make an estimated guess on the effect size
expected, or use an effect size that corresponds to the size of the effect
◦ Broad effect sizes categories are small, medium, and large
◦ Different statistical tests will have different values of effect size for each
category
𝐸𝑓𝑓𝑒𝑐𝑡 𝑆𝑖𝑧𝑒 ≈
𝑀𝑒𝑎𝑛1 − 𝑀𝑒𝑎𝑛2
𝑆𝑡𝑑. 𝑑𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛

Statistical Rules of the Game
Here are a few pieces of terminology to refresh yourself with before embarking on calculating sample size:
◦ Null Hypothesis (H0): default or ‘boring’ state; your statistical test is run to either Reject or Fail to Reject the Null
◦ Alternative Hypothesis (H1): alternative state; usually what your experiment is interested in retaining over the Null
◦ One-Tailed Test: looking for a deviation from the H0 in only one direction (ex: Is variable X larger than 0?)
◦ Two-tailed Test: looking for a deviation from the H0 in either direction (ex: Is variable Y different from 0?)
◦ Parametric data: approximately fits a normal distribution; needed for many statistical tests
◦ Non-parametric data: does not fit a normal distribution; alternative and less powerful tests available
◦ Paired (dependent) data: categories are related to one another (often result of before/after situations)
◦ Un-paired (independent) data: categories are not related to one another
◦ Dependent Variable: Depends on other variables; the variable the experimenter cares about; also known as the Y or
response variable
◦ Independent Variable: Does not depend on other variables; usually set by the experimenter; also known as the X or
predictor variable

Using GPower: Basics
◦ Download for Mac or PC
◦ http://www.psychologie.hhu.de/arbeitsgruppen/allgemeine-psychologie-
und-arbeitspsychologie/gpower.html
◦ Three basic steps:
◦ Select appropriate test:
◦ Input parameters
◦ Determine effect size (can use background info or guess)
◦ For situations when you have some idea of parameters such as mean,
standard deviation, etc., I will refer to this as having Background
Information
◦ If not, I will refer to this as Naïve

Using GPower: Graphics
◦ Central and noncentral distributions
◦ Shows the distribution of the null
hypothesis (red) and the alternative
(blue)
◦ Also has the critical values
◦ X-Y plot for a range of values
◦ Can generate plots of one of the
parameters α, effect size, power and
sample size, depending on a range of
values of the remaining parameters.

Taxonomy of Designs Covered
◦ 1 Numerical
◦ Parametric: One mean T-test
◦ Non-parametric: One mean Wilcoxon Test
◦ 1 Numerical + 1 Categorical
◦ Categorical groups=2:
◦ Independent (non-paired):
◦ Parametric: Two means T-test
◦ Non-parametric: Mann-Whitney Test
◦ Dependent (paired):
◦ Parametric: Paired T-test
◦ Non-Parametric: Paired Wilcoxon Test
◦ Categorical groups>2:
◦ Independent (non-paired):
◦ Parametric: One-way ANOVA
◦ Non-Parametric: Kruskal Wallace Test
◦ Dependent (paired):
◦ Parametric: Repeated Measures ANOVA
◦ Non-Parametric: Friedman Test
◦ 1 Numerical + 2+ Categorical
◦ Single Category of Interest: Multi-Way ANOVA
Blocked ANOVA
Nested ANOVA
Split-Plot ANOVA
◦ Multiple Categories of Interest: Multi-Way ANOVA
◦ 1 Categorical: Proportion Test
◦ 1 Categorical + 1 Categorical
◦ Independent (non-paired): Fisher’s Exact Test
◦ Dependent (paired): McNamar’s Test
◦ 1 Categorical + 1+ Categorical
◦ Categorical groups≥2: Goodness-of-Fit Test
◦ 1 Numerical + 1 Numerical
◦ Parametric: Simple Linear Regression
◦ Non-parametric: Spearman Rank-order Regression
◦ 1 Numerical + 2+ Numerical
◦ Parametric: Multiple Linear Regression
◦ Non-Parametric: Logistic and Poisson Regression
◦ 1 Numerical + 1+ Numerical + 1+ Categorical:
◦ Only 1 Category of Interest: ANCOVA
◦ Multiple Categories of Interest: {GLMM}

# Name of Test Numeric. Var(s) Cat. Var(s) Cat. Var Group
#
Cat Var. # of
Interest
Parametric Paired
1 One Mean T-test 1 0 0 0 Yes N/A
2 One Mean Wilcoxon Test 1 0 0 0 No N/A
3 Two Means T-test 1 1 2 1 Yes No
4 Mann-Whitney Test 1 1 2 1 No No
5 Paired T-test 1 1 2 1 Yes Yes
6 Paired Wilcoxon Test 1 1 2 1 No Yes
7 One-way ANOVA 1 1 >2 1 Yes No
8 Kruskal Wallace Test 1 1 >2 1 No No
9 Repeated Measures ANOVA 1 1 >2 1 Yes Yes
10 Friedman Test 1 1 >2 1 No Yes
11 Multi-way ANOVA (1 Category of interest) 1 ≥2 ≥2 1 Yes No
12 Multi-way ANOVA (>1 Category of interest) 1 ≥2 ≥2 >1 Yes No
13 Proportions Test 0 1 2 1 N/A N/A
14 Fisher’s Exact Test 0 2 2 2 N/A No
15 McNemar’s Test 0 2 2 2 N/A Yes
16 Goodness-of-Fit Test 0 ≥1 ≥2 1 N/A No
17 Simple Linear Regression 2 0 N/A N/A Yes N/A
18 Multiple Linear Regression >2 0 N/A N/A Yes N/A
19 Pearson’s Correlation 2 1 N/A N/A Yes No
20 Non-Parametric Regression (Logistic) ≥2 0 N/A N/A No N/A
21 Non-Parametric Regression (Poisson) ≥2 0 N/A N/A No N/A
22 ANCOVA >1 ≥1 >1 ≥1 Yes N/A

Format for each test
Overview
Example
{Parameter Calculations}
Practice
Answers

One Mean T-Test: Overview
Description: this tests if a sample
mean is any different from a set
value for a normally distributed
variable.
Example:
◦ Is the average body temperature of college
students any different from 98.6°F?
◦ H0=98.6°F, H1≠98.6°F
GPower:
◦ Select t tests from Test family
◦ Select Means: difference from constant (one sample case) from Statistical test
◦ Select A priori from Type of power analysis
◦ Background Info:
a) Select One or Two from the Tail(s), depending on type
b) Enter 0.05 in α err prob box – or a specific α you want for your study
c) Enter 0.80 in Power (1-β err prob) box - or a specific power you want for your study
d) Hit Determine =>
e) Enter in the Mean H0, Mean HI, and SD, then hit Calculate and transfer to main window
(this will calculate effect size and add it to the Input Parameters)
f) Hit Calculate on the main window
g) Find Total sample size in the Output Parameters
◦ Naïve:
a) Run a-c as above
b) Enter Effect size guess in the Effect size d box (small=0.2, medium=0.5, large=0.8)
c) Hit Calculate on the main window
d) Find Total sample size in the Output Parameters
Numeric.
Var(s)
Cat. Var(s) Cat. Var
Group #
Cat Var. #
of Interest
Parametric Paired
1 0 0 0 Yes N/A

One Mean T-Test: Example
Is the average body temperature of
college students any different from
98.6°F?
◦ H0=98.6°F, H1≠98.6°F
◦ From a trial study, you found the
mean temperature to be 98.2° with a
standard deviation of 0.733.
◦ Selected Two-tailed, because we were
asking if temp differed, not whether
it was simply lower or higher
Results:
◦ Total number of samples needed is
29.
Dropdown menu items you specified
Values you entered
Value(s) GPower calculated
Sample size calculation

One Mean T-Test: Practice
Calculate the sample size for the following scenarios (with α=0.05, and power=0.80):
1. You are interested in determining if the average income of college freshman is less than $20,000. You
collect trial data and find that the mean income was $14,500 (SD=6000).
2. You are interested in determining if the average sleep time change in a year for college freshman is
different from zero. You collect the following data of sleep change (in hours).
3. You are interested in determining if the average weight change in a year for college freshman is greater
than zero.
Sleep Change -0.55 0.16 2.6 0.65 -0.23 0.21 -4.3 2 -1.7 1.9

One Mean T-Test: Answers
1. You are interested in determining if the average income of college freshman is less than $20,000. You collect trial
data and find that the mean income was $14,500 (SD=6000).
◦ Found an effect size of 0.91, then used a one-tailed test to get a total sample size of 9
2. You are interested in determining if the average sleep time change in a year for college freshman is different from
zero. You collect the following data of sleep change (in hours).
◦ Mean H0=0, Mean H1=-0.446 with SD=1.96; found an effect size of 0.228 then used a two-tailed test to get a total
sample size of 154
3. You are interested in determining if the average weight change in a year for college freshman is greater than zero.
◦ Guessed a large effect size (0.8), then used a one-tailed test to get a total sample size of 12
Sleep Change -0.55 0.16 -2.6 0.65 -0.23 0.21 -4.3 2 -1.7 1.9

One Mean Wilcoxon: Overview
Description: this tests if a sample
mean is any different from a set value
for a non-normally distributed
variable
Example:
◦ Is the average number of children in Grand
Forks families greater than 1?
◦ H0=1 child, H1>1 child
GPower:
◦ Select Means: Wilcoxon signed-rank test (one sample case) from Statistical test
b) Select Parent Distribution (Laplace, Logistic, or min ARE) depending on variable (min
ARE is good default if you don’t know for sure)
c) Enter 0.05 in α err prob box – or a specific α you want for your study
d) Enter 0.80 in Power (1-β err prob) box - or a specific power you want for your study
e) Hit Determine =>
f) Enter in the Mean H0, Mean HI, and SD, then hit Calculate and transfer to main window
g) Hit Calculate on the main window
h) Find Total sample size in the Output Parameters
◦ Naïve:
a) Run a-d as above
b) Enter Effect size guess in the Effect size d box
Numeric.
Var(s)
Group #
Cat Var. #
of Interest
Parametric Paired
1 0 0 0 No N/A

One Mean Wilcoxon: Example
Is the average number of children in
Grand Forks families greater than 1?
◦ H0=1 child, H1>1 child
◦ You have no background information
and you don’t think it is normally
distributed
◦ Default distribution: for non-normal,
use min ARE – weakest, but least
assumptions
◦ Try looking at a large effect
◦ Selected One-tailed, because we only
cared if the number is greater than the
null (1 child)
Results:
◦ Total number of samples needed is 13.
Values you entered

One Mean Wilcoxon: Practice
1. You are interested in determining if the average number of insurance claims for Grand Forks families
different from 3. You collect trial data and find that the claim number was 2.8 (SD=0.53, with a
Laplace distribution).
2. You are interested in determining if the average number of pets in Grand Forks families is greater
than 1. You collect the following trial data for pet number.
3. You are interested in determining if the average number of yearly trips to the emergency room for
Grand Forks families is different from 5?
Pet Number 1 1 1 3 2 1 0 0 0 4

One Mean Wilcoxon: Answers
1. You are interested in determining if the average number of insurance claims for Grand Forks families different from 3.
You collect trial data and find that the claim number was 2.8 (SD=0.53, with a Laplace distribution).
◦ Found an effect size of 0.38, then used a two-tailed test with Laplace distribution to get a total sample size of 39
2. You are interested in determining if the average number of pets in Grand Forks families is greater than 1. You collect
the following trial data for pet number.
◦ Mean H0=1, Mean H1=1.3 with SD=1.34;
◦ Found an effect size of 0.224
◦ Because I didn’t know the distribution, went conservatively with min ARE, then used a one-tailed test to get a total sample
size of 145
3. You are interested in determining if the average number of yearly trips to the emergency room for Grand Forks families
is different from 5?
◦ Guessed a medium effect (0.5) and because I didn’t know the distribution went conservatively with min ARE, then used a
two-tailed test to get a total sample size of 39
Pet Number 1 1 1 3 2 1 0 0 0 4

Two Means T-Test: Overview
Description: this tests if a mean from one
group is different from the mean of another
group for a normally distributed variable.
AKA, testing to see if the difference in means
is different from zero.
Example:
◦ Is the average body temperature higher in
men than in women?
◦ H0=0°F, H1>0°F
Numeric.
Var(s)
Group #
Cat Var. #
of Interest
Parametric Paired
1 1 2 1 Yes No
GPower:
◦ Select Means: Difference between two independent means (two groups) from Statistical
test
d) Enter 1.00 in the Allocation ratio N2/N1 box – unless your group sizes are not equal, then
enter the ratio
e) Hit Determine =>
f) Enter in the Mean and SD for each group, then hit Calculate and transfer to main window
◦ Naïve:
a) Run a-d as above

Two Means T-test : Example
Is the average body temperature
higher in men than in women?
mean temperature to be 98.1° for
men with a standard deviation (SD)
of 0.699 and 98.4 ° for women with
a SD of 0.743.
◦ Selected one-tailed, because we only
cared to test if men’s temp was
higher than women’s, not lower
◦ Group 1 is men, group 2 is women
Results:
◦ Need a sample size of 73 per gender,
for a total of 146.
Values you entered

Two Means T-Test: Practice
1. You are interested in determining if the average daily caloric intake different between men and
women. You collected trial data and found the average caloric intake for 10 males to be 2350.2
(SD=258), 5 while females had intake of 1872.4 (SD=420).
2. You are interested in determining if the average protein level in blood different between men and
women. You collected the following trial data on protein level (grams/deciliter).
3. You are interested in determining if the average glucose level in blood is lower in men than women.
Male Protein 1.8 5.8 7.1 4.6 5.5 2.4 8.3 1.2
Female Protein 9.5 2.6 3.7 4.7 6.4 8.4 3.1 1.4

Two Means T-Test: Answers
1. You are interested in determining if the average daily caloric intake different between men and women. You collected
trial data and found the average caloric intake for 10 males to be 2350.2 (SD=258), while 5 females had intake of 1872.4
(SD=420).
◦ Found an effect size of 1.37, then used a two-tailed test with Allocation Ratio of 0.5 (5 women/10 men) to get a total sample
size of 22 (15 men, 7 women)
2. You are interested in determining if the average protein level in blood different between men and women. You collected
the following trial data on protein level (grams/deciliter).
◦ Mean group 1 (men)=4.5875 with SD=2.575, mean group 2 (women)=4.975 with SD=2.875
◦ Found an effect size of 0.142, then used a two-tailed test with Allocation Ratio of 1.0 (equal group sizes) to get a total sample
size of 1560 (780 each for men and women)
1. You are interested in determining if the average glucose level in blood is lower in men than women in a predominately
female college.
◦ Guessed a small effect (0.2), then used a two-tailed test with Allocation Ratio of 3.0 (3 women: 1 man) to get a total sample
size of 826 (207 men, 619 women)
Male Protein 1.8 5.8 7.1 4.6 5.5 2.4 8.3 1.2
Female Protein 9.5 2.6 3.7 4.7 6.4 8.4 3.1 1.4

Mann-Whitney Test: Overview
group for a non-normally distributed variable.
AKA, testing to see if the difference in means
is different from zero.
Example:
◦ Does the average number of snacks per day
for individuals on a diet differ between
young and old persons?
◦ H0=0 difference in snack number,
H1≠0 difference in snack number
Numeric.
Var(s)
Group #
Cat Var. #
of Interest
Parametric Paired
1 1 2 1 No No
GPower:
◦ Select Means: Wilcoxon-Mann-Whitney test (two groups) from Statistical test
b) Select Parent Distribution (Laplace, Logistic, or min ARE) depending on variable (min ARE is
good default if you don’t know for sure)
e) Enter 1.00 in the Allocation ratio N2/N1 box – unless your group sizes are not equal, then enter
the ratio
f) Hit Determine =>
g) Enter in the Mean and SD for each group, then hit Calculate and transfer to main window (this will
calculate effect size and add it to the Input Parameters)
h) Hit Calculate on the main window
i) Find Total sample size in the Output Parameters
◦ Naïve:
a) Run a-e as above

Mann-Whitney Test: Example
Does the average number of snacks
per day for individuals on a diet
differ between young and old
persons?
◦ H0=0 difference in snack number,
H1≠0 difference in snack number
and you think it might be a small effect
(0.20)
◦ Selected two-tailed, because we only
care if there is a difference in the two
groups
Hovering over the Effect size d box
will show a bubble of size conventions
Results:
Need a sample size of 456 for each age
for a total of 912.
Values you entered

Mann-Whitney Test: Practice
1. You are interested in determining if the average number of walks per week for individuals on a diet is
lower in younger people than older. You collected trial data and found mean walk number for young
dieters to be 1.4 (SD=0.18) and 2.5 for older (SD=0.47).
2. You are interested in determining if the number of meals per day for individuals on a diet is higher in
younger people than older. You collected trial data on meals per day.
3. You are interested in determining if the average number of weight loss websites visited per day for
individuals on a diet is different in younger people than older.
Young meals 1 2 2 3 3 3 3 4
Older meals 1 1 1 2 2 2 3 3

Mann-Whitney Test: Answers
1. You are interested in determining if the average number of walks per week for individuals on a diet is lower in younger
people than older. You collected trial data and found mean walk number for young dieters to be 1.4 (SD=0.18) and 2.5
for older (SD=0.47).
◦ Found an effect size of 3.09, then used a one-tailed test with a min ARE distribution and Allocation Ratio of 1.0 to get a
total sample size of 6 (3 each for young and old)
2. You are interested in determining if the number of meals per day for individuals on a diet is higher in younger people
than older. You collected trial data on meals per day
◦ Mean group 1 (older)=1.875 with SD=0.834, mean group 2 (younger)=2.625 with SD=0.916
◦ Found an effect size of 0.53, then used a one-tailed test with a Logistic distribution and Allocation Ratio of 1.0 to get a total
sample size of 82 (41 each for young and old)
3. You are interested in determining if the average number of weight loss websites visited per day for individuals on a diet
is different in younger people than older.
◦ Guessed a medium effect (0.5) and Logistic Regression, then used a two-tailed test and Allocation Ratio of 1.0 to get a total
sample size of 118 (59 each for young and old)
Young meals 1 2 2 3 3 3 3 4
Older meals 1 1 1 2 2 2 3 3

Paired T-test: Overview
group, where the groups are dependent (not
independent) for a normally distributed variable.
Pairing can be leaves on same branch, siblings,
the same individual before and after a trial, etc.
Example:
◦ Is heart rate higher in patients after a run
compared to before a run?
◦ H0=0 bpm, H1>0 bpm
GPower:
◦ Select Means: Difference between two dependent means (matched) from Statistical test
d) Hit Determine =>
e) Enter in the Mean and SD for each group, as well as the Correlation between groups, then
hit Calculate and transfer to main window (this will calculate effect size and add it to the
Input Parameters)
◦ Naïve:
a) Run a-c as above
Numeric.
Var(s)
Group #
Cat Var. #
of Interest
Parametric Paired
1 1 2 1 Yes Yes

Paired T-test: Example
Is heart rate higher in patients after a run
compared to before a run?
◦ H0=0 bpm, H1>0 bpm
◦ From a trial study, you found the mean beats
per minute (bpm) before the run to be 86.7
with SD=11.28 and 144.7 with SD=29.12
after the run. The correlation between
measurements before and after is 0.34.
◦ Selected One-tailed, because we only cared if
bpm was higher after a run
◦ Group 1 is after the run, while group 2 is
before the run
◦ If you only have the mean difference, can use
that in the From differences option
◦ If you don’t have correlation data, can guess it
(0.5 is good standard if you have no idea)
Results:
◦ Need a sample size of 4 individuals (each with
a before and after run measurement).
◦ Because the difference in means was so large,
the effect size was huge, so it does not take
many samples to test this question
Values you entered

Paired T-Test: Practice
1. You are interested in determining if heart rate is higher in patients after a doctor’s visit compared to
before a visit. You collected the following trial data and found mean heart rate before and after a visit.
2. You are interested in determining if metabolic rate in patients after surgery is different from before
surgery. You collected trial data and found a mean difference of 0.73 (SD=2.9).
3. You are interested in determining if glucose levels in patients after surgery are lower compared to
before surgery.
BPM before 126 88 53.1 98.5 88.3 82.5 105 41.9
BPM after 138.6 110.1 58.44 110.2 89.61 98.6 115.3 64.3

Paired T-Test : Answers
1. You are interested in determining if heart rate is higher in patients after a doctor’s visit compared to before a visit.
You collected the following trial data and found mean heart rate before and after a visit.
◦ Mean 1 (before)=85.4 with SD=27.2; Mean 2 (after)=98.1 with SD=26.8; correlation between groups=0.96
◦ Found an effect size of 1.66, then used a one-tailed test to get a total sample size of 4 pairs
2. You are interested in determining if metabolic rate in patients after surgery is different from before surgery. You
collected trial data and found a mean difference of 0.73 (SD=2.9).
◦ Found an effect size of 0.25, then used a two-tailed test to get a total sample size of 126
3. You are interested in determining if glucose levels in patients after surgery are lower compared to before surgery.
◦ Guessed a small effect (0.20), then used a one-tail test to get a total sample size of 156
BPM before 126 88 53.1 98.5 88.3 82.5 105 41.9
BPM after 138.6 110.1 58.44 110.2 89.61 98.6 115.3 64.3

Paired Wilcoxon: Overview
group, where the groups are dependent (not
independent) for a non-normally distributed
variable.
Example:
◦ Are genome methylation patterns different
between identical twins?
◦ H0=0% methylation, H1≠0% methylation
GPower:
◦ Select Means: Wilcoxon signed-rank test (matched pairs) from Statistical test
b) Select Parent Distribution (Laplace, Logistic, or min ARE) depending on variable (min
ARE is good default if you don’t know for sure)
e) Hit Determine =>
f) Enter in the Mean and SD for each group, as well as the Correlation between groups, then
hit Calculate and transfer to main window (this will calculate effect size and add it to the Input
Parameters)
◦ Naïve:
a) Run a-d as above
Numeric.
Var(s)
Group #
Cat Var. #
of Interest
Parametric Paired
1 1 2 1 No Yes

Paired Wilcoxon: Example
Are genome methylation patterns
different between identical twins?
◦ H0=0% methylation,
H1>0% methylation
on this, so assume there is going to be a
small effect (0.2).
◦ Selected one-tailed, because can’t have
negative methylation
Results:
◦ Need a sample size of 181 pairs of
twins (362 individuals total).
Values you entered

Paired Wilcoxon: Practice
1. You are interested in determining if genome methylation patterns are different between fraternal
twins. You collected trial data and found mean methylation levels in twin A at 43.2 (SD=20.9) and at
65.7 (SD=28.1) for twin B.
2. You are interested in determining if genome methylation patterns are higher in the first fraternal twin
born compared to the second. You collected the following trial data on methylation level difference
(in percentage).
3. You are interested in determining if the cancer risk rate is lower in the first identical twin born
compared to the second.
Methy. Diff (%) 5.96 5.63 1.25 1.17 3.59 1.64 1.6 1.4

Paired Wilcoxon: Answers
1. You are interested in determining if genome methylation patterns are different between fraternal twins. You collected trial
data and found mean methylation levels in twin A at 43.2 (SD=20.9) and at 65.7 (SD=28.1) for twin B.
◦ With a default correlation of 0.5, found an effect size of 0.89, then used a two-tailed test with a min ARE distribution to get a
total sample size of 14 pairs
2. You are interested in determining if genome methylation patterns are higher in the first fraternal twin born compared to
the second. You collected the following trial data on methylation level difference (in percentage).
◦ Mean of difference=2.505 with SD=1.87
◦ Found an effect size of 1.33, then used a one-tailed test with a min ARE distribution to get a total sample size of 6 pairs
3. You are interested in determining if the cancer risk rate is lower in the first identical twin born compared to the second.
◦ Guessed a small effect (0.2) and because I didn’t know the distribution went conservatively with min ARE, then used a one-
tailed test to get a total sample size of 181 pairs
Methy. Diff (%) 5.96 5.63 1.25 1.17 3.59 1.64 1.6 1.4

One-way ANOVA: Overview
Description: this tests if at least one mean is
different among groups, where the groups are
larger than two, for a normally distributed
variable. ANOVA is the extension of the
Two Means T-test for more than two groups.
Example:
◦ Is there a difference in new car interest
rates across 6 different cities?
◦ H0=0, H1≠0
GPower:
◦ Select F tests from Test family
◦ Select ANOVA: Fixed effects, omnibus, one-way from Statistical test
a) Enter 0.05 in α err prob box – or a specific α you want for your study
b) Enter 0.80 in Power (1-β err prob) box - or a specific power you want for your study
c) Enter the number of groups
d) Hit Determine =>
e) Enter in the Mean and Size for each group, as well as the SD within each group, then hit
Calculate and transfer to main window (this will calculate effect size and add it to the Input
Parameters)
◦ Naïve:
a) Run a-c as above
Numeric.
Var(s)
Group #
Cat Var. #
of Interest
Parametric Paired
1 1 >2 1 Yes No

One-way ANOVA: Example
Is there a difference in new car
interest rates across 6 different
cities?
◦ H0=0%, H1≠0%
following information for means
◦ With a SD of 0.786 and group sizes
of 9
◦ No Tails in ANOVA
◦ If groups are equal size, can enter the
size in the Purple box, then click
Equal n
Results:
◦ A total number of 48 samples are
needed (8 per group).
Values you entered
City 1 2 3 4 5 6
mean 13.19444 12.61111 13.30667 13.24444 13.48333 12.2

One-way ANOVA: Practice
Calculate the sample size for the following scenarios
(with α=0.05, and power=0.80):
1. You are interested in determining there is a
difference in weight lost between 4 different surgery
options. You collect the following trial data of
weight lost in pounds (shown on right)
2. You are interested in determining if there is a
difference in white blood cell counts between 5
different medication regimes.
Option 1 Option 2 Option 3 Option 4
6.3 9.9 5.1 1.0
2.8 4.1 2.9 2.8
7.8 3.9 3.6 4.8
7.9 6.3 5.7 3.9
4.9 6.9 4.5 1.6

One-way ANOVA: Answers
Calculate the sample size for the following scenarios
(with α=0.05, and power=0.80):
1. You are interested in determining there is a
difference in weight lost between 4 different surgery
options. You collect the following trial data of
weight lost in pounds (shown on right)
◦ 4 groups with group size of 5; means of 5.94, 6.22, 4.36, and
2.82 with SD=2.23
◦ Found an effect size of 0.61 for a total sample size of 36
difference in white blood cell counts between 5
different medication regimes.
◦ 5 groups; guessed a medium effect size (0.25) for a total
sample size of 200
Option 1 Option 2 Option 3 Option 4
6.3 9.9 5.1 1.0
2.8 4.1 2.9 2.8
7.8 3.9 3.6 4.8
7.9 6.3 5.7 3.9
4.9 6.9 4.5 1.6

Kruskal Wallace Test: Overview
Description: this tests if at least one mean is different
among groups, where the groups are larger than two
for a non-normally distributed variable. There really
isn’t a standard way of calculating sample size in
GPower, but you can use a rule of thumb:
1. Run Parametric Test
2. Add 15% to total sample size
(https://www.graphpad.com/guides/prism/7/statistics/in
dex.htm?stat_sample_size_for_nonparametric_.htm)
Example:
◦ Is there a difference in draft rank across 3
different months?
◦ H0=0, H1≠0
GPower:
◦ Select ANOVA: Fixed effects, omnibus, one-way from Statistical test
c) Enter the number of groups
d) Hit Determine =>
e) Enter in the Mean and Size for each group, as well as the SD within each group, then hit
Calculate and transfer to main window (this will calculate effect size and add it to the Input
Parameters)
h) Add 15% to size (Total + Total*0.15)
◦ Naïve:
a) Run a-c as above
e) Add 15% to size (Total + Total*0.15)
Numeric.
Var(s)
Group #
Cat Var. #
of Interest
Parametric Paired
1 1 >2 1 No No

Kruskal Wallace Test: Example
Is there a difference in draft rank
across 3 different months?
◦ H0=0, H1≠0
◦ You don’t have background info, so
you guess that there is a medium
effect size (0.25)
Results:
needed (53 per group) for the
parametric (ANOVA)
◦ For the non-parametric:
◦ 159 + 159*0.15 = 182.85
◦ Round up
◦ Total of 183 samples (61 per group)
are needed.
◦ Notice that non-parametric is weaker
Values you entered

Kruskal Wallace Test: Practice
Calculate the sample size for the following scenarios (with
α=0.05, and power=0.80):
1. You are interested in determining there is a difference in
hours worked across 3 different groups (faculty, staff, and
hourly workers). You collect the following trial data of
weekly hours (shown on right).
assistant professor salaries across 25 different departments.
Faculty Staff Hourly
42 46 29
45 45 42
46 37 33
55 42 50
42 40 23

Kruskal Wallace Test: Answers
α=0.05, and power=0.80):
1. You are interested in determining there is a difference in hours
worked across 3 different groups (faculty, staff, and hourly
workers). You collect the following trial data of weekly hours
(shown on right).
◦ 3 groups with group size of 5; means of 46, 42 and 35.4 with
SD=8.07
◦ Found an effect size of 0.54 for a parametric sample size of 39
◦ 39*1.15=44.85 -> round up to total sample size of 45 (15 per
group)
assistant professor salaries across 25 different departments.
◦ 25 groups; guessed a small effect size (0.10) for a parametric
sample size of 2275
◦ 2275*1.15=2616.26 -> round up to total sample size of 2625 (105
per group)
Faculty Staff Hourly
42 46 29
45 45 42
46 37 33
55 42 50
42 40 23

Repeated Measures ANOVA: Overview
repeated measures (more than two) for a
normally distributed variable. Repeated
Measures ANOVA is the extension of the
Paired T-test for more than two groups.
Example:
◦ Is there a difference in blood pressure at 1,
2, 3, and 4 months post-treatment?
◦ H0=0 bpm, H1≠0 bpm
GPower:
◦ Select ANOVA: Repeated measures, within factors from Statistical test
◦ Background Info or Naïve:
c) Enter the number of groups and measurements (see next page)
d) Enter the Corr among rep measures and the Nonsphericity correction ε (see next page)
e) Hit Determine =>
f) Enter calculated Partial η2 (eta squared) (see next page)
◦ for Background, enter calculated eta squared and select as in SPSS for the Effect size
specification in the Options bar
◦ for Naïve, enter guess eta squared (based on convention) and keep the Effect size
specification on as in Gpower 3.0 and will also have to enter a guess for the Corr among
rep measures parameter
g) Hit Calculate and transfer to main window
Numeric.
Var(s)
Group #
Cat Var. #
of Interest
Parametric Paired
1 1 >2 1 Yes Yes

Repeated Measures ANOVA: parameters and how to calculate them
◦ Number of groups: how many different groups are being subjected to the repeated measurements
◦ In the simple case, it is one (college students)
◦ In more complex designs, it may be more than one (college freshman, sophomores, juniors, and seniors)
◦ Number of measurements: how many repeats of a measurement
◦ Ex. number of times blood pressure is measured
◦ Corr among rep measures: Correlation
◦ No easy way to get a single correlation value
◦ Can average all the values from a correlation table
◦ Or default to 0.5 unless you have reason to believe it higher or lower
◦ Nonsphericity correction ε: The assumption of sphericity,
◦ Can be estimated with background info
◦ Otherwise, if the data is assumed to be spherical, enter 1
◦ Spherical: variances of the differences between all possible pairs of the within subjects variable should be equivalent
◦ Partial η2 (eta squared): proportion of the total variance in a dependent variable that is associated with the membership of different groups
defined by an independent variable
◦ The larger the eta, the larger the effect size
◦ Can be estimated with background information (Equation is η2 = SSeff / (SSeff + SSerr) where SSeff is the sum of squares between groups and
the SSerr is the sum of squares within groups
◦ Otherwise, enter guessed value; convention is small=0.02, medium=0.06, large=0.14

Repeated Measures ANOVA: Example
Is there a difference in blood
pressure at 1, 2, 3, and 4 months
post-treatment?
◦ H0=0, H1≠0
◦ 1 group, 4 measurements
◦ Have no background info
◦ Assume 0.5 correlation and
Nonsphericity correction of 1.0
◦ Assume small partial eta-squared
(0.02)
Results:
needed (each getting 4
measurements).
Values you entered

Repeated Measures ANOVA: Practice
α=0.05, and power=0.80):
1. You are interested in determining if there is a difference in
blood serum levels at 6, 12, 18, and 24 months post-
treatment. You collect the following trial data of blood
serum in mg/dL (shown on right).
antibody levels at 1, 2, and 3 months post-treatment.
◦ Info: no background info, but expect nonsphericity correction of
1, correlation of 0.5, and medium eta squared (remember to
select as in GPower 3.0 in options)
6 months 12 months 18 months 24 months
38 38 46 52
13 44 15 29
32 35 53 60
35 48 51 44
21 27 29 36

Repeated Measures ANOVA: Answers
Calculate the sample size for the following scenarios (with α=0.05,
and power=0.80):
1. You are interested in determining if there is a difference in blood serum
levels at 6, 12, 18, and 24 months post-treatment? You collect the
following trial data of blood serum in mg/dL (shown on right).
◦ Ran Repeated measures in SPSS with data to get η2 = SSeff / (SSeff + SSerr)
◦ η2 = 19531.2 / (19531.2 + 1789.5) = 0.92 (selected as in SPSS in options)
◦ Sphericity was non-significant (p=0.712), so nonsphericity correction is 1.0
◦ Got effect size of 3.39 for a total sample size of 3
2. You are interested in determining if there is a difference in antibody
levels at 1, 2, and 3 months post-treatment?
◦ Guessed a medium eta squared (0.06), selected as in GPower 3.0 in options
◦ Set correlation to default of 0.5 and nonsphericity correction to 1.0
6 months 12 months 18 months 24 months
38 38 46 52
13 44 15 29
32 35 53 60
35 48 51 44
21 27 29 36

Friedman Test: Overview
repeated measures (more than two) for a non-
normally distributed variable. The Friedman test
is the extension of the Two Means Wilcoxon
test for more than two groups.
Example:
◦ Is there a difference in taste preference
across three different desserts for a group
of students?
◦ H0=0, H1≠0
GPower:
◦ Select ANOVA: Repeated measures, within factors from Statistical test
c) Enter the number of groups and measurements
d) Enter the Nonsphericity correction ε
e) Hit Determine =>
f) Enter Partial η2 (eta squared)
◦ for Background, enter calculated eta squared and select as in SPSS for the Effect size
specification in the Options bar
◦ for Naïve, enter guess eta squared (based on convention) and keep the Effect size
specification on as in Gpower 3.0 and will also have to enter a guess for the Corr among
rep measures parameter
j) Add 15% to size (Total + Total*0.15)
Numeric.
Var(s)
Group #
Cat Var. #
of Interest
Parametric Paired
1 1 >2 1 No Yes

Friedman Test: Example
Is there a difference in taste preference
across three different desserts for a
group of students?
◦ H0=0, H1≠0
◦ From a trial study, you found a sphericity
of 0.888
◦ The eta-squared was
◦ SSeff / (SSeff + SSerr)
◦ 18 / (18 + 40) = 0.31
◦ Make sure to select the correction Option
Results:
◦ A total number of 15 samples are needed
for parametric
◦ Non-parametric:
◦ 15 + 15*0.15 =17.25 (round up) -> 18
◦ A total of 18 samples are needed (each
getting 3 measurements).
Values you entered

Friedman Test: Practice
α=0.05, and power=0.80):
taste preference across four different main courses for a
group of students. You collect the following trial data of
food preferences(shown on right).
movie preference across five different genres for a group of
students.
steak lobster burgers noodles
9 1 6 6
8 8 9 4
9 8 10 7
5 9 5 2
8 2 7 3

Friedman Test: Answers
Calculate the sample size for the following scenarios (with α=0.05, and
power=0.80):
1. You are interested in determining if there is a difference in taste
preference across four different main courses for a group of
students. You collect the following trial data of food
preferences(shown on right).
◦ Ran Repeated measures in SPSS with data to get η2 = SSeff / (SSeff + SSerr)
◦ η2 = 793.8 / (793.8 + 36.7) = 0.95 (selected as in SPSS in options)
◦ Sphericity was non-significant (p=0.165), so nonsphericity correction is 1.0
◦ Got effect size of 4.35 for a parametric sample size of 3
◦ 3*1.15=3.45 -> round up to total sample size of 4
2. You are interested in determining if there is a difference in movie
preference across five different genres for a group of students.
◦ Guessed a large eta squared (0.14), selected as in GPower 3.0 in options
◦ Set correlation to default of 0.5 and nonsphericity correction to 1.0
◦ Got effect size of 0.40 for a parametric sample size of 9
◦ 9*1.15=10.35 -> round up to total sample size of 11
steak lobster burgers noodles
9 1 6 6
8 8 9 4
9 8 10 7
5 9 5 2
8 2 7 3

Multi-Way ANOVA (1 Category of interest): Overview
Description: this test is an extension of
ANOVA, where there is more than one
category, but only one category is of interest.
The other category/categories are things that
need to be controlled for
(blocking/nesting/random effects/etc.).
Example:
◦ Is there difference in treatment (Drug A, B,
and C) from a series of four different
hospital sections (Block 1, 2, 3, and 4)?
◦ H0=0, H1≠0
GPower:
◦ Select ANOVA: Fixed effects, special, main effects and interactions from Statistical test
c) Enter the number of groups and Numerator df (df for the category of interest only)
d) Hit Determine =>
e) Enter calculated Partial η2 (eta squared)
f) Hit Calculate and transfer to main window
◦ Naïve:
a) Follow steps a-c as above
b) Estimate effect size f
Numeric.
Var(s)
Group #
Cat Var. #
of Interest
Parametric Paired
1 ≥2 ≥2 1 Yes No

Multi-Way ANOVA: Parameters and how to calculate them
◦ Numerator df = degrees of freedom for the effect you want to test
◦ Degrees of freedom found by taken the number of categories for the predictor variable or interaction and subtracting one
◦ For interactions between variables, the degrees of freedom need to be multiplied together
◦ For example, in a design with three seating locations and two genders:
◦ For the predictor variable seating location, enter 3 (locations) – 1 = 2 df
◦ For the predictor variable gender, enter 2 (genders) – 1 = 1 df
◦ For the interaction, enter (3 – 1) * (2 – 1) = 2 df
◦ Since we only want to look at the effect of a single category: just use that category’s DF
◦ If in the sample above, we only care about seating (we included gender to control for the difference b/t genders) = 2df
◦ Number of groups = found by multiplying the number of levels in both predictor variables. In this example
there are 2 genders and 3 seating locations, so 2 x 3 = 6

Multi-Way ANOVA: Categories to
control for
◦ When you are running a multi-way ANOVA for only 1 category of
interest, that means you only care about one category (say drug
type), but there are other factors that might interfere with your ability
to detect that category because those categories have variation
among the groups
◦ Examples of interfering factors: gender, location, age, etc.
◦ Below are common classes of effects that are controlled for in
ANOVA
◦ Blocking: samples come from different blocks (locations) –ex. Nutrient
treatment types across different fields
◦ Nesting: samples are located within higher order variables –ex. Leaves are
nested within branch
◦ Split-Plot: More complicated version of blocking with two levels of
experimental units – ex. Pesticide treatment in greenhouse trays across
whole and subplots
https://i.ytimg.com/vi/gKkMOaANugI/maxresdefault.jpg
https://www.statisticshowto.datasciencecentral.com/wp-content/uploads/2015/07/nested-anova.png
https://blog.minitab.com/hubfs/Imported_Blog_Media/split_plot_3.jpg

Multi-Way ANOVA (1 Category of interest): Example
Is there difference in treatment
(Drug A, B, and C) from a series of
four different hospital sections
(Block 1, 2, 3, and 4)?
◦ H0=0, H1≠0
◦ Category of interest: Treatment
◦ Want to control for the Sections
(Blocking)
◦ No background information
◦ Assume medium effect size
◦ Numerator df (Treatment ) = 3-1=2
◦ Number of groups (Treatment *
Sections) = 3*4 =12
Results:
needed, so 13.17 per group, rounding
up to 14 per group (168 total).
Values you entered

Multi-Way ANOVA (>1 Category of interest): Overview
Description: this test is an extension of
ANOVA, where there is more than one
category, and each category is of interest. If
there is two categories, it is 2-way ANOVA;
three categories, 3-way ANOVA, etc.
Example:
◦ Is there difference in treatment (Drug
A, B, and C) across age (child, adult,
elder) and cancer stage (I, II, III, IV,
V)?
◦ H0=0, H1≠0
GPower:
◦ Select ANOVA: Fixed effects, specia, main effects and interactions from Statistical test
c) Enter the number of groups and Numerator df (df for all categories of interest)
d) Hit Determine =>
e) Enter calculated Partial η2 (eta squared)
◦ Naïve:
a) Follow steps a-c as above
b) Estimate effect size f
Numeric.
Var(s)
Group #
Cat Var. #
of Interest
Parametric Paired
1 ≥2 ≥2 >1 Yes No

Multi-Way ANOVA (>1 Category of interest):
Example
Is there difference in treatment
(Drug A, B, and C) across age (child,
adult, elder) and cancer stage (I, II,
III, IV, V)?
◦ H0=0, H1≠0
◦ Categories of interest: Treatment, Age,
and Caner Stage
◦ Based on trial study, the partial eta-
squared=0.0042
◦ Numerator df = Treat DF * Age DF *
Stage DF = (3-1)*(3-1)*(5-
1)=2*2*4=16
◦ Number of groups = Treat*Age*Stage
= 3*3*5=45
Results:
needed, so 101.82 per group, rounding
up to 102 (4590 total).
Values you entered

Multi-Way ANOVA: Non-Parametric
◦ There really isn’t a non-parametric option like in the one-way or repeated measures ANOVA
◦ Two options:
◦ Turn to a more sophisticated model for your statistical test, like a Generalized Linear Mixed Model (explained later)
◦ Run a parametric test, then add the extra 15% at the end like we’ve done before for simpler non-parametric ANOVA
paralogs
Non-Parametric Reanalysis
Test Question Parametric
Sample size
Calculations Non-parametric
Sample size
Multi-Way, 1
category of
interest
Is there difference in treatment (Drug A, B,
and C) from a series of four different
hospital sections (Block 1, 2, 3, and 4)?
251 251+251*0.15=288.65
288.65/12(groups)=24.05 -> 25
25*12(groups)=300
300
Multi-Way, 1
category of
interest
Is there difference in treatment (Drug A, B,
and C) across age (child, adult, elder) and
cancer stage (I, II, III, IV, V)?
4582 4582+4582*0.15=5269.3
5269.3/45(groups)=117.09 ->118
118*45(groups) = 5310
5310

Multi-Way ANOVA: Practice
α=0.05, and power=0.80):
treatment (Drug A, B, and C), while controlling for age
(child, adult, elder). You collect the following trial data for
treatment (shown on right).
difference in treatment (Drug A, B, and C) across age
(child, adult, elder) and cancer stage (I, II, III, IV, V).
Drug A Drug B Drug C
child adult elder child adult elder child adult elder
-6.4 8.7 -3.1 1.3 -6.0 6.8 -2.0 -4.3 -1.2
-8.2 -6.3 -6.5 3.6 1.3 2.4 1.5 1.3 1.1
7.9 -1 -1.5 3.9 -1.9 1.3 2.48 -8.2 -9.7

Multi-Way ANOVA: Answers
power=0.80):
1. You are interested in determining if there is a difference in treatment
(Drug A, B, and C), while controlling for age (child, adult, elder). You
collect the following trial data for treatment (shown on right).
◦ Only care about treatment so numerator df is 3-1=2
◦ Number of groups is 3*3=9
◦ Ran Univariate ANOVA in SPSS with data to get η2 = SSeff / (SSeff +
SSerr)
◦ η2 = 68.8 / (68.8 + 96.3) = 0.416
treatment (Drug A, B, and C) across age (child, adult, elder) and
cancer stage (I, II, III, IV, V).
◦ Care about treatment, age, and cancer stage
◦ Numerator df = (3-1)*(3-1)*(5-1)=2*2*4=16
◦ Number of groups is 3*3*5=45
◦ Guessed a medium effect size (0.25) for a total sample size of 323
Drug A Drug B Drug C
child adult elder child adult elder child adult elder
-6.4 8.7 -3.1 1.3 -6.0 6.8 -2.0 -4.3 -1.2
-8.2 -6.3 -6.5 3.6 1.3 2.4 1.5 1.3 1.1
7.9 -1 -1.5 3.9 -1.9 1.3 2.48 -8.2 -9.7

Proportion Test: Overview
Description: this tests when you only have a
single categorical value with only two groups,
and you want to know if the proportions of
certain values differ from some constant
proportion. (binomial, AKA=Yes/No, 1/0, etc.)
Example:
◦ Is there a significance difference in cancer
prevalence of middle-aged women who
have a sister with breast cancer compared
to the general population prevalence
(proportion=0.02, or 2%)?
◦ H0=0, H1≠0
GPower:
◦ Select Exact from Test family
◦ Select Proportion: Difference from constant (binomial test, one sample case) from
Statistical test
a) Select One or Two Tail(s) as appropriate to your question
d) Enter the constant proportion (whatever constant proportion you are measuring your
proportion of interest against)
e) Hit Determine =>
f) Select Difference P2-P1 from the Calc P2 from option (doesn’t really matter, all give you
about the same effect size)
g) Put the constant proportion in the P1 box
h) Put the alternative proportion in the P2 box:
a) For Background: enter the value from your background information
b) For Naïve: guess the value and enter it in
◦ 0.2=small, 0. 5=medium, and 0.8 large effect sizes
i) Hit Sync Values, then Calculate and transfer to main window
j) Hit Calculate on the main window
k) Find Total sample size in the Output Parameters
Numeric.
Var(s)
Group #
Cat Var. #
of Interest
Parametric Paired
0 1 2 1 N/A N/A

Proportion Test: Example
Is there a significance difference in
cancer prevalence of middle-aged
women who have a sister with
breast cancer compared to the
general population prevalence
(proportion=0.02, or 2%)?
◦ H0=0, H1≠0
◦ Based on background information,
we found the prevalence (proportion)
of women w/ sister who has breast
cancer to get breast cancer was 0.05.
◦ Want to use two-tailed, because only
care if there is a difference, not
directionality
Results:
needed.
Values you entered

Proportion Test: Practice
power=0.80):
1. You are interested in determining if the male incidence rate proportion of cancer in
North Dakota is higher than the US average (prop=0.00490). You find trial data
cancer prevalence of 0.00495.
2. You are interested in determining if the female incidence rate proportion of cancer in
North Dakota is lower than the US average (prop=0.00420).

Proportion Test: Answers
1. You are interested in determining if the male incidence rate (proportion per 100,000) of cancer in
North Dakota is higher than the US average (prop=0.00490). You find trial data cancer prevalence of
0.00495.
◦ The difference in P2-P1 is 0.00005 (constant proportion is 0.0049)
◦ Got effect size of 0.00005 and used a one-tailed test for a total sample size of 12,113,157
2. You are interested in determining if the female incidence rate proportion of cancer in North Dakota
is lower than the US average (prop=0.00420).
◦ Guessed a very small effect size (0.0001) and entered a constant proportion of 0.00420
◦ Used a one-tailed test for a total sample size of 2,490,591

Fisher’s Exact Test: Overview
Description: this test when you only have 2
categorical variables and you want to know if the
proportions are different between groups, where the
groups are not related. Essentially testing between an
observed proportion and an expected one. Also called
independent Chi-Squared test.
Example:
◦ Is the expected proportion of students
passing a stats course taught by psychology
teachers of 0.85 different from the
observed proportion of students passing
the same stats class taught by mathematics
teachers of 0.95
◦ H0=0, H1≠0
GPower:
◦ Select Proportion: Inequality, two independent groups (Fisher’s exact test) from
Statistical test
b) Enter values for Proportion p1 and p2
a) P1 is the ‘expected value’
b) P2 is the ‘observed value’ – if you don’t have background info, you will have to
guess on this
e) Enter 1.0 in Allocation ration N2/N1 – unless the group sizes are different
Numeric.
Var(s)
Group #
Cat Var. #
of Interest
Parametric Paired
0 2 2 2 N/A No

Fisher’s Exact Test: Example
Is the expected proportion of
students passing a stats course
taught by psychology teachers of
0.85 different from the observed
proportion of students passing the
same stats class taught by
mathematics teachers of 0.95?
◦ H0=0, H1≠0
◦ P1 is the expected value of 0.85
◦ P2 is the observed value of 0.95
directionality
Results:
needed.
Values you entered

Fisher’s Exact Test: Practice
1. You are interested in determining if the expected proportion (P1) of students passing a stats course taught by
psychology teachers is different than the observed proportion (P2) of students passing the same stats class taught
by biology teachers. You collected the following data of passed tests. You also know that twice as many students
take the psychology class than the biology one.
1. You are interested in determining of the expected proportion (P1) of female students who selected YES on a
question was higher than the observed proportion (P2) of male students who selected YES. The observed
proportion of males who selected yes was 0.75.
Psychology Yes Yes Yes No No Yes Yes Yes Yes No
Biology No No Yes Yes Yes No Yes No Yes Yes

Fisher’s Exact Test: Practice
1. You are interested in determining if the expected proportion (P1) of students passing a stats course taught by
psychology teachers is different than the observed proportion (P2) of students passing the same stats class taught
by biology teachers. You collected the following data of passed tests. You also know that twice as many students
take the psychology class than the biology one.
◦ Psychology proportion is 7/10=0.70; biology proportion is 6/10=0.60; allocation ratio is 0.5
◦ For a two-tailed test, the total sample size is 812 (541 for psychology class, 271 for biology class)
2. You are interested in determining of the expected proportion (P1) of female students who selected YES on a
question was higher than the observed proportion (P2) of male students who selected YES. The observed
proportion of males who selected yes was 0.75.
◦ Don’t have any info on the female students, but will guess that it is 0.10 higher (P1=0.85)
◦ For a one-tailed test, the total sample size is 430 (215 each male and female)
Psychology Yes Yes Yes No No Yes Yes Yes Yes No
Biology No No Yes Yes Yes No Yes No Yes Yes

McNamar’s Test: Overview
Description: this test when you only have 2
categorical variables and you want to know if
the proportions are different between groups,
where the groups are related (paired). Still
testing between an observed proportion and
an expected one. Also called dependent Chi-
Squared test.
Example:
◦ Is there a difference in innate immune
response (good/poor) for patients taking
Vitamin D or not (yes/no)?
◦ H0=0, H1≠0
GPower:
◦ Select Proportion: Inequality, two dependent groups (McNemar) from Statistical
test
b) Enter value for Odds ratio: (see next page)
e) Enter value for Prop discordant pairs: (see next page)
f) Hit Calculate on main window
Numeric.
Var(s)
Group #
Cat Var. #
of Interest
Parametric Paired
0 2 2 2 N/A Yes

McNamar’s Test: Proportion table
◦ To calculate the Odds Ratio and Prop discordant pairs, it is important to
understand what the data looks like
◦ To the right is trial data from our example: the number are proportions for
each category pair
◦ Note: the proportions must all sum to 1.0
◦ Discordant pairs are the proportions for the groups that don’t match up
◦ Good Immune Response for Vitamin D and Poor response with No Vitamin D
◦ Poor Immune Response for Vitamin D and Good response with No Vitamin D
◦ The stronger the two categories are related to each other, the smaller the discordant pair
proportions should be
◦ The Odds Ratio is the quotient of the discordant pairs
◦ 0.02/0.13= 0.1538 (Note: The opposite quotient 0.13 / 0.02 = 6.5 produces the same
result in GPower)
◦ The Proportion of discordant pairs is the sum of the discordant pairs
◦ 0.02+0.13=0.15
Vitamin D
good poor
No Vitamin D good 0.70 0.02
poor 0.13 0.15
discordant pairs

McNamar’s Test: Example
Is there a difference in innate
immune response (good/poor) for
patients taking Vitamin D or not
(yes/no)?
◦ H0=0, H1≠0
◦ Based on background information
from the previous page,
◦ Odds Ratio=0.02/0.13=0.1538
◦ Prop discordant pairs=0.02+0.13=0.15
directionality
Results:
needed.
Values you entered

McNamar’s Test: Practice
1. You are interested in determining if there is a difference in overnight stay (yes/no) for patients having surgery
with anesthesia or not (yes/no). You collected the following trial information.
1. You are interested in determining if there is a difference in snacking (yes/no) for students in the afternoon
whether they had school or not (yes/no). You collected the following trial information.
Anesthesia
yes no
No Anesthesia yes 0.30 0.10
no 0.08 0.52
Student 1 2 3 4 5 6 7 8 9 10
Snack Yes No Yes Yes No Yes No Yes No Yes
School No No Yes Yes Yes Yes Yes Yes Yes No

McNamar’s Test: Answers
1. You are interested in determining if there is a difference in overnight stay (yes/no) for patients having surgery
with anesthesia or not (yes/no). You collected the following trial information.
◦ Odds ratio is 0.08/0.10=0.80; proportion of discordant pairs is 0.08+0.10=0.18
◦ For a two-tailed test, total sample size is 3562
2. You are interested in determining if there is a difference in snacking (yes/no) for students in the afternoon
whether they had school or not (yes/no). You collected the following trial information.
◦ Odds ratio is 0.2/0.3=0.667; proportion of discordant pairs is 0.2+0.3=0.5
◦ For a two-tailed test, total sample size is 398
Anesthesia
yes no
No Anesthesia yes 0.30 0.10
no 0.08 0.52
Student 1 2 3 4 5 6 7 8 9 10
Snack Yes No Yes Yes No Yes No Yes No Yes
School No No Yes Yes Yes Yes Yes Yes Yes No
School
yes no
No School yes 4/10 3/10
no 2/10 1/10

Goodness-of-Fit Test: Overview
Description: Extension of Chi-squared
tests, which asks if table of observed
values are any different from a table of
expected ones. Also generically called
Chi-Squared.
Example:
◦ Does the observed proportions of
phenotypes from a genetics experiment
different from the expected 9:3:3:1?
◦ H0=0, H1≠0
GPower:
◦ Select X2 from Test family
◦ Select Goodness-of-fit tests: Contingency tables from Statistical test
c) Enter X in DF (degrees of freedom), which = number of categories -1
d) Hit Determine =>
e) Add the number of cells that equals the number of proportions
f) Enter the expected proportions in the p(H0) column
g) Enter in the observed proportions in the p(H1) column
◦ If you have counts rather than proportions, you can enter them in and then click on Normalize
p(H0) and Normalize p(H1)
h) Hit Calculate and transfer to main window
i) Hit Calculate on the main window
j) Find Total sample size in the Output Parameters
◦ Naïve:
a) Run steps a-c above
b) Estimate an effect size w
Numeric.
Var(s)
Group #
Cat Var. #
of Interest
Parametric Paired
0 ≥1 ≥2 1 N/A No

Goodness-of-Fit Test: Example
Does the observed proportions of
phenotypes from a genetics
experiment different from the
expected 9:3:3:1?
◦ H0=0, H1≠0
◦ The observed ratios we’d like to test
are 9:3:4:0
◦ DF=4-1=3
◦ Notice the change between the
counts and the normalized
proportions
Results:
needed.
Values you entered

Goodness-of-Fit Test: Practice
1. You are interested in determining if the ethnic ratios in a company differ by gender. You collect the following trial
data.
2. You are interested in determining if the proportions of student by year (Freshman, Sophomore, Junior, Senior) is
any different from 1:1:1:1. You collect the following trial data.
Gender White Black Am. Indian Asian
Male 0.60 0.25 0.01 0.14
Female 0.65 0.21 0.11 0.03
Student 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Grade Frs Frs Frs Frs Frs Frs Frs Soph Soph Soph Soph Soph Jun Jun Jun Jun Jun Sen Sen Sen

Goodness-of-Fit Test: Answers
1. You are interested in determining if the ethnic ratios in a company differ by gender. You collect the following trial
data.
◦ Got an effect size of 1.04 and degrees of freedom (4-1) of 3; total sample size is 10
2. You are interested in determining if the proportions of student by year (Freshman, Sophomore, Junior, Senior) is
any different from 1:1:1:1. You collect the following trial data.
◦ Proportions were: Freshman=7/20=0.35, Sophomore=5/20=0.25, Junior=5/20=0.25, Senior=3/20=0.15
◦ Got an effect size of 0.28 and degrees of freedom (4-1) of 3; total sample size is 137
Gender White Black Am. Indian Asian
Male 0.60 0.25 0.01 0.14
Female 0.65 0.21 0.11 0.03
Student 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Grade Frs Frs Frs Frs Frs Frs Frs Soph Soph Soph Soph Soph Jun Jun Jun Jun Jun Sen Sen Sen

Simple Linear Regression: Overview
Description: this test determines if there is a
significant relationship between two normally
distributed numerical variables. The predictor
variable is used to try to predict the response
variable.
Example:
◦ Can height predict weight in college males?
◦ H0=0, H1≠0
GPower:
◦ Select Linear multiple regression: Fixed model, R2 deviation from zero from Statistical
test- simple linear regression is just the limit case of multiple regression where there is only 1
predictor variable
c) Enter 1 in Number of predictors
d) Hit Determine =>
e) Selected the From correlation coefficient option and add squared multiple correlation
◦ Add the correlation coefficient from background information after squaring it
◦ Naïve:
b) Estimate an effect size f2
Numeric.
Var(s)
Group #
Cat Var. #
of Interest
Parametric Paired
2 0 N/A N/A Yes N/A

Simple Linear Regression: Example
Is there a relationship between
height and weight in college
males?
◦ H0=0, H1≠0
◦ No background, but expect the
correlation to be strong between
the two, so go with large effect
size
Results:
◦ A total number of 25 samples
are needed.
Values you entered

Simple Linear Regression: Practice
1. You are interested in determining if height (meters) in plants can predict yield (grams of berries). You
collect the following trial data.
2. You are interested in determining if the size of a city (in square miles) can predict the population of
the city (in # of individuals).
Yield 46.8 48.7 48.4 53.7 56.7
Height 14.6 19.6 18.6 25.5 20.4

Simple Linear Regression: Answers
Calculate the sample size for the following questions:
1. You are interested in determining if height (meters) in plants can predict yield (grams of berries). You
◦ Running a regression found a squared multiple correlation coefficient (R-squared)=0.459
◦ Got an effect size of 0.85; for 1 predictor, found total sample size of 12
2. You are interested in determining if the size of a city (in square miles) can predict the population of
the city (in # of individuals).
◦ Guessed a large effect size (0.35); for 1 predictors, found total sample size of 25
Yield 46.8 48.7 48.4 53.7 56.7
Height 14.6 19.6 18.6 25.5 20.4

Multiple Linear Regression: Overview
Description: The extension of simple linear
regression. The first major change is there
are more predictor variables. The second
change is that interaction effects can be used.
Finally, the results typically can’t be plotted.
Example:
◦ Can height, age, and time spent at the gym,
predict weight in adult males?
◦ H0=0, H1≠0
GPower:
◦ Select Linear multiple regression: Fixed model, R2 deviation from zero from Statistical test
c) Enter the number of predictor variables in Number of predictors
d) Hit Determine =>
e) Enter correlation coefficient:
a) From correlation coefficient: enter squared multiple correlation coefficient
b) From predictor correlations: add number of predictors, then click Specify matrices, from
there add the correlation coefficients from a correlation matrix of the outcome (response) with
all the predictor variables, and finally Accept values
◦ Naïve:
b) Estimate an effect size f2
Numeric.
Var(s)
Group #
Cat Var. #
of Interest
Parametric Paired
>2 0 N/A N/A Yes N/A

Multiple Linear Regression: Example
Can height, age, and time spent at
the gym, predict weight in adult
males?
◦ H0=0, H1≠0
◦ From trial data, we obtained the
following correlation matrix
◦ Only need about highlighted values
for the matrix
Results:
needed.
Values you entered
Weight Height Age GymTime
Weight 1 0.317 0.025 0.304
Height 0.317 1 -0.466 0.800
Age 0.025 -0.466 1 -0.485
GymTime 0.304 0.800 -0.485 1

Multiple Linear Regression: Practice
1. You are interested in determining if height (meters), weight (grams), and fertilizer added (grams) in
plants can predict yield (grams of berries). You collect the following trial data.
2. You are interested in determining if the size of a city (in square miles), number of houses, number of
apartments, and number of jobs can predict the population of the city (in # of individuals).
Yield 46.8 48.7 48.4 53.7 56.7
Height 14.6 19.6 18.6 25.5 20.4
Weight 95.3 99.5 94.1 110 103
Fertilizer 2.1 3.2 4.3 1.1 4.3

Multiple Linear Regression: Answers
Calculate the sample size for the following questions:
1. You are interested in determining if height (meters), weight (grams), and fertilizer added (grams) in
plants can predict yield (grams of berries). You collect the following trial data.
◦ Running a regression found a squared multiple correlation coefficient (R-squared)=0.944
◦ Got an effect size of 16.85; for 3 predictors, found total sample size of 6
2. You are interested in determining if the size of a city (in square miles), number of houses, number of
apartments, and number of jobs can predict the population of the city (in # of individuals).
◦ Guessed a large effect size (0.35); for 4 predictors, found total sample size of 40
Yield 46.8 48.7 48.4 53.7 56.7
Height 14.6 19.6 18.6 25.5 20.4
Weight 95.3 99.5 94.1 110 103
Fertilizer 2.1 3.2 4.3 1.1 4.3

Pearson’s Correlation: Overview
Description: this test determines if there is a
difference between two correlation values.
The categorical variable is a grouping variable
that is binomial.
Example:
◦ Is the correlation between hours studied
and test score for group A statistically
different than the correlation between
hours studied and test score for group B?
◦ H0=0, H1≠0
GPower:
◦ Select z tests from Test family
◦ Select Correlations: Two Independent Person r’s from Statistical test
d) Enter 1.0 in Allocation ratio N2/N1 – unless the group sizes are different between the
categories
e) Hit Determine =>
f) Add the correlation coefficients 1 and 2
◦ Naïve:
a) Run steps a-d above
b) Estimate an effect size q
Numeric.
Var(s)
Group #
Cat Var. #
of Interest
Parametric Paired
2 1 N/A N/A Yes No

Pearson’s Correlation: Example
Is the correlation between hours
studied and test score for group A
statistically different than the
correlation between hours studied
and test score for group B?
◦ H0=0, H1≠0
◦ Don’t have data, but expect that the
two groups will be quite similar so
expect effect size to be small
◦ Two tails because we only care about
difference, not direction
Results:
needed, 1573 for each of the two
groups.
Values you entered

Pearson’s Correlation: Practice
1. You are interested in determining if the correlation between height and weight in men is different from the
correlation between height and weight in women. You collect the following trial data.
2. You are interested in determining if, in lab mice, the correlation between longevity (in months) and average
protein intake (grams) is higher from the correlation between longevity and average fat intake (grams).
Males
Height 178 166 172 186 182
Weight 165 139 257 225 196
Females
Height 174 175 157 168 166
Weight 187 182 149 132 143

Pearson’s Correlation: Answers
1. You are interested in determining if the correlation between height and weight in men is different from the
correlation between height and weight in women. You collect the following trial data.
◦ Correlation coefficient for men=0.37; coefficient for women=0.66
◦ Got effect size of -0.40, and with an allocation ratio of 1.0, total sample size is 198 (99 per group)
◦ NOTE: same result if effect size is 0.40, sign doesn’t matter here
2. You are interested in determining if, in lab mice, the correlation between longevity (in months) and average
protein intake (grams) is higher from the correlation between longevity and average fat intake (grams).
◦ Guess a medium effect (0.3), and with an allocation ratio of 1.0, total sample size is 282 (141 per group)
Males
Height 178 166 172 186 182
Weight 165 139 257 225 196
Females
Height 174 175 157 168 166
Weight 187 182 149 132 143

Non-Parametric Regression (Logistic): Overview
Description:
Non-parametric regression is any regression where the
numerical variables are not normally distributed. The
two that can be calculate in GPower are Logistic and
Poisson.
In Logistic regression, the response variable (Y) is
binary (0/1). It tests whether a continuous predictor is
a significant predictor of a binary outcome, with or
without other covariates. (Can also run with a
dichotomous predictor but will only show continuous
here).
Example:
◦ Does body mass index (BMI) influences
mortality (yes 1, no 0)?
◦ H0=0, H1≠0
GPower:
◦ Select Logistic Regression from Statistical test
◦ Select Options button and set the Input effect size as .. Two probabilities
d) Enter value in Pr(Y=1|Y=1) H0 box-(explained in next slide)
e) Enter value in Pr(Y=1|Y=1) H1 box-(explained in next slide)
f) Enter value in R2 other X box-(explained in next slide)
g) Select X distribution – will have to examine or predict distribution of predictor (X)
variable-(explained in next slide)
h) Enter parameters in the X parameter box(es) – will depend on the distribution
Numeric.
Var(s)
Group #
Cat Var. #
of Interest
Parametric Paired
≥2 0 N/A N/A No N/A

Logistic Regression: parameters and how to calculate them
◦ Pr(Y=1|X=1) H0 – probability of Y variable (response) =1, when main predictor is one standard deviation
above its mean
◦ Based on background information, or make informed guess
◦ Pr(Y=1|X=1) H1 - probability of Y variable (response) =1, when main predictor is at its mean
◦ Based on background information, or make informed guess
◦ R2 other X – Expected R-squared between main predictor variable and over covariates; amount of variability
in main predictor that is accounted for by covariates
◦ If there are no covariates (as in the simplest case of a single predictor), enter 0
◦ Otherwise, calculate with background data by regressing main predictor onto data for all other covariates
◦ Rule of thumb for naïve estimation: low association=0.04, moderate association=0.25, strong association=0.81
◦ X distribution – will have to examine or predict distribution of predictor (X) variable
◦ Select normal unless you think the main predictor is distributed differently
◦ X parameter box(es) – will depend on the distribution
◦ For normal, the μ (mu) is the z-score population mean of main predictor, while sigma (σ) is that predictor’s z-
score population Standard Deviation

Non-Parametric Regression (Logistic): Example
Does body mass index (BMI)
influences mortality (yes 1, no 0)?
◦ H0=0, H1≠0
◦ Based on background data, expect
Pr(Y=1|X=1) H0 to be low at 0.15
and Pr(Y=1|X=1) H1 to be higher,
at 0.25
◦ Since BMI is only predictor, the R2
other X is 0
◦ Expect BMI to be normally
distributed (mean=0, SD=1)
◦ Two tails because we only care about
difference, not direction
◦ Make sure to Choose Options
Results:
needed.
Values you entered

Non-Parametric Regression (Poisson): Overview
Description:
In Poisson regression, the response variable (Y) is a
rate. It tests whether a continuous predictor variable
influences the rate of events over a set period, with or
without covariates (can also do dichotomous predictor
but will only show continuous here).
Note that Poisson regression assumes independence
of observations which is that the occurrence or
absence of a previous event should not influence
whether another event will occur. Subjects can have
multiple events, as long as they are independent.
Example:
◦ Does a change in drug dose decrease the
rate of adverse affects?
◦ H0=0, H1≠0
GPower:
◦ Select Poisson Regression from Statistical test
b) Enter value in Exp(β1) box – (explained in next slide)
e) Enter value in Base rate exp(β) box-(explained in next slide)
f) Enter value in Mean exposure box –(explained in next slide)
g) Enter value in R2 other X box-same as logistic; enter 0 if no covariates
h) Select X distribution – will have to examine or predict distribution of predictor (X)
variable
i) Enter parameters in the X parameter box(es) – will depend on the distribution
j) Hit Calculate on the main window
k) Find Total sample size in the Output Parameters
Poisson Only:
Numeric.
Var(s)
Group #
Cat Var. #
of Interest
Parametric Paired
≥2 0 N/A N/A No N/A

Poisson Regression: parameters and how to calculate them
◦ Exp(β1) –change in response rate with a 1-unit increase in the predictor variable
◦ If we expect response rate to go UP by 25% per 1-unit predictor increase, the Exp(β1) would be 1.25
◦ If we expect response rate to go DOWN by 30% per 1-unit predictor increase, the Exp(β1) would be 0.70
◦ Base rate exp(β) –The response rate we expect if there is no intervention
◦ Need to know your unit of exposure, what rate is being counted-could be time, distance, sample size, volume, etc.
◦ For that unit of exposure, the base rate is the number of events expected per length of exposure
◦ Ex.: 13 events/30 days= base rate of 0.43; 1 event/20 miles= base rate of 0.05; 5 events/10 volunteers= base rate of 0.5
◦ Mean exposure – length of exposure (how long you want the study to last in terms of your unit of exposure)
◦ R2 other X – Expected R-squared between main predictor variable and over covariates; amount of variability in main predictor that is
accounted for by covariates
◦ If there are no covariates (as in the simplest case of a single predictor), enter 0
◦ Otherwise, calculate with background data by regressing main predictor onto data for all other covariates
◦ Rule of thumb for naïve estimation: low association=0.04, moderate association=0.25, strong association=0.81
◦ X distribution – will have to examine or predict distribution of predictor (X) variable
◦ Select normal unless you think the main predictor is distributed differently
◦ X parameter box(es) – will depend on the distribution
◦ For normal, the μ (mu) is the z-score population mean of main predictor, while sigma (σ) is that predictor’s z-score population
Standard Deviation

Non-Parametric Regression (Poisson): Example
Does a change in drug dose decrease the
rate of adverse affects?
◦ H0=0, H1≠0
◦ Expect a moderate decrease in rate with
increase in drug dose, so will set Exp(β1) to
0.80
◦ Our unit of exposure will be days and evens
will be adverse effects, so let’s set the Base rate
exp(β0) to be:
◦ 1 adverse effect/1 day = 1
◦ We want to try this for a month, so let's have
Mean exposure be 30 days
◦ Since Drug dose is only predictor, the R2 other
X is 0
◦ Not sure what rate of adverse effects'
distribution is, so to be safe, went with uniform
◦ One tail, because we are specifically interested
in a decrease in rate
Results:
◦ A total number of 98 samples are needed.
Values you entered

Non-Parametric Regression: Practice
1. You are interested in determining if body temperature influences sleep disorder prevalence (yes 1, no 0). You
2. You are interested in determining if the rate of lung cancer incidence changes with a drug treatment.
Temperature 98.6 98.5 99.0 97.5 98.8 98.2 98.5 98.4 98.1
Sleep Disorder? No No Yes No Yes No No Yes No

Non-Parametric Regression: Answers
1. You are interested in determining if body temperature influences sleep disorder prevalence (yes 1, no 0). You collect the
following trial data.
◦ Want logistic regression with two-tails; temperature mean is 98.4 with SD=0.436 -> (97.964 --98.836)
◦ Pr(Y=1|X=1) H0 = 0.33 (as only one had sleep disorder at ranges outside one SD); Pr(Y=1|X=1) H1= 0.67
◦ No other Xs, so R-squared is 0
◦ Temperature is normally distributed, but set the mean to 0 and SD=1 (want it standardized rather than actual)
◦ Got a total sample size of 195
2. You are interested in determining if the rate of lung cancer incidence changes with an environmental factor over a 3
years.
◦ Want poisson regression with two-tails
◦ Expect an Exp(β1) of 1.02 (up by 2% per every unit increase)
◦ Researching rate of lung cancer in men per year found rate of 57.8 (per 100,000)
◦ Mean exposure is 3, as we want to try this for 3 years; No other Xs, so R-squared is 0
◦ Try normal distribution of X with 0,1
◦ Got a total sample size of 116
Temperature 98.6 98.5 99.0 97.5 98.8 98.2 98.5 98.4 98.1
Sleep Disorder? No No Yes No Yes No No Yes No

ANCOVA: Overview
Description: Blend of ANOVA and regression.
Tests whether the means of a dependent
variable (DV) are equal across levels of a
categorical independent variable (IV) often
called a treatment, while statistically controlling
for the effects of other continuous variables that
are not of primary interest, known as covariates.
Example:
◦ Is there a difference in recovery time across
two conditions and three diagnostic groups
with baseline-scores as the co-variate?
◦ H0=0, H1≠0
GPower:
◦ Select ANCOVA: Fixed effects, main effects and interactions from Statistical test
c) Fill out boxes for Numerator df and Number of groups-same as Multi-way ANOVA
d) Fill out box for Number of covariates-depends on number you are including in design
e) Hit Determine =>
f) Select Direct
g) Add partial eta-squared-same as ANOVA
h) Hit Calculate and transfer to main window
◦ Naïve:
b) Estimate an effect size f
Numeric.
Var(s)
Group #
Cat Var. #
of Interest
Parametric Paired
>1 ≥1 >1 ≥1 Yes N/A

ANCOVA: Example
Is there a difference in recovery
time across two conditions and
three diagnostic groups with
baseline-scores as the co-variate?
◦ H0=0, H1≠0
◦ Expect a medium effect
◦ Numerator df = (2-1)*(3-2)=2
◦ Number of groups=2*3=6
◦ Only one covariate
Results:
◦ A total of 158 samples are needed.
Values you entered

ANCOVA: Practice
α=0.05, and power=0.80):
weight loss due to treatment (Drug X, Y, and Z), while
controlling for body weight. You collect the following trial
data for treatment (shown on right).
weight loss due to a diet style (1, 2, 3, or 4), while
controlling for starting body weight and body height.
Drug X Drug Y Drug Z
Starting
weight
Weight
lost
Starting
weight
Weight
lost
Starting
weight
Weight
lost
120 1.9 120 5.7 239 1.1
144 6.6 236 9 332 4.6
270 5.9 204 4.4 268 12.6
169 9.6 127 7.5 207 12.5
189 3.9 134 3.7 183 18.5

ANCOVA: Answers
power=0.80):
1. You are interested in determining if there is a difference in weight loss
due to treatment (Drug X, Y, and Z), while controlling for body weight.
You collect the following trial data for treatment (shown on right).
◦ Numerator df =3-1=2
◦ 3 groups (3 treatments) and 1 covariate (weight)
◦ Ran general linear model in SPSS with data to get η2 = SSeff / (SSeff +
SSerr)
◦ η2 = 64 / (64 +235.3)=0.213
2. You are interested in determining if there is a difference in weight loss
due to a diet style (1, 2, 3, or 4), while controlling for starting body
weight and body height.
◦ Numerator df =4-1=3
◦ 4 groups (4 diet styles) and 2 covariates (weight, and height)
◦ Guessed medium effect size (0.25) for a total sample size of 179
Drug X Drug Y Drug Z
Starting
weight
Weight
lost
Starting
weight
Weight
lost
Starting
weight
Weight
lost
120 1.9 120 5.7 239 1.1
144 6.6 236 9 332 4.6
270 5.9 204 4.4 268 12.6
169 9.6 127 7.5 207 12.5
189 3.9 134 3.7 183 18.5

Generalized Linear Mixed Models, a postlude
Description: The most expansive model
for statistical analysis. Can used fixed
effects, random effects, categorical
predictor variables, numerical predictor
variables, and a variety of distributions
(Not just normal). Often abbreviated as
GLMMs.
Example:
◦ Can height, weight, rate of driving speed,
gender, age, ethnicity, and number of weekly
drinks predict the probability of car crashes?
◦ H0=0, H1≠0
◦ Unfortunately, GLMMS are beyond the
scope of GPower
◦ Stay tuned for approaches in other software
programs such as R, SAS, and Excel
Numeric.
Var(s)
Group #
Cat Var. #
of Interest
Parametric Paired
≥1 ≥0 ≥0 ≥0 Y/N N/A

Common Clinical Study Designs
Name Description Typical Statistical Method
Randomized
Control Trial
A controlled clinical trial that randomly (by chance) assigns participants to two or more groups. There
are various methods to randomize study participants to their groups
ANOVA
Double Blind
Method
A type of randomized controlled clinical trial/study in which neither medical staff/physician nor the
patient knows which of several possible treatments/therapies the patient is receiving.
ANOVA
Cohort Study A clinical research study in which people who presently have a certain condition or receive a particular
treatment are followed over time and compared with another group of people who are not affected by
the condition.
Regression
Case Control Case-control studies begin with the outcomes and do not follow people over time. Researchers choose
people with a particular result (the cases) and interview the groups or check their records to ascertain
what different experiences they had. They compare the odds of having an experience with the outcome
to the odds of having an experience without the outcome.
Regression
Cross-sectional
Study
The observation of a defined population at a single point in time or time interval. Exposure and
outcome are determined simultaneously
Regression
Info from https://research.library.gsu.edu/c.php?g=115595&p=755213 and https://hsl.lib.umn.edu/biomed/help/understanding-research-study-designs

Acknowledgements
◦ The DaCCoTA is supported by the National Institute of
General Medical Sciences of the National Institutes of
Health under Award Number U54GM128729.
◦ For the labs that use the Biostatistics, Epidemiology, and
Research Design Core in any way, including this Module,
please acknowledge us for publications. "Research
reported in this publication was supported by
DaCCoTA (the National Institute of General Medical
Sciences of the National Institutes of Health under
Award Number U54GM128729).

Sample Size And Gpower Module

More Related Content

What's hot (20)

Similar to Sample Size And Gpower Module (20)

Recently uploaded (20)

Sample Size And Gpower Module