Sample Size

Calculation with
R
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 R a free,
open-source package
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
 • In designing an experiment, a key question is:
The Why of How many animals/subjects do I need for my
Sample Size experiment?
• Too small of a sample size can under detect the
Calculations 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 𝑀𝑒𝑎𝑛𝐴 − 𝑀𝑒𝑎𝑛0
𝐸𝑓𝑓𝑒𝑐𝑡 𝑆𝑖𝑧𝑒 ≈
How to estimate Effect Size: 𝑆𝑡𝑑. 𝑑𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛
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, then 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 Calculation within R
• As opposed to GPower, which allows you to enter details such as means and standard
deviations into the program and it will calculate effect size for you, that is not the
case for R
• Most R functions for sample size only allow you to enter effect size
• If you want to estimate effect size from background information, you’ll need to
calculate it yourself first
• Throughout this Module, I will provide an equation to calculated effect size for each
of the statistical tests

❖Disclaimer: Most of the examples and practice problems are the same as an earlier GPower
Module. However, it was not always clear how effect size was calculated in GPower or in R,
so sometimes the sample size calculated was different between the two. When in doubt, I
would go with the result that gives the higher sample size to avoid undersampling.
 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 R: Basics
• This module assumes the user is familiar with R
• For an introduction or refresher, please check out the following material
• https://cran.r-project.org/doc/manuals/r-release/R-intro.pdf
• http://www.r-tutor.com/r-introduction
• https://www.statmethods.net/

• R can be downloaded here: https://cran.r-project.org/

• I strongly suggest also getting RStudio, an integrated development
environment: https://rstudio.com/
Organization of tests
• As opposed to the earlier GPower Module, which organized tests
taxonomically based on types of variables, this module will follow a
different order
• The order will be based on the packages available in R
• We will start with basic statistical tests that are easily calculated
• For each test:
• Introduction slide: description, example, R code, and effect size calculation
• Result slide: shows R code and results for the example question
• Practice: 2-3 questions to practice on your own
• Answers: parameters, R-code, and resulting sample size for practice questions
 # Name of Test in R? Package Function
1 One Mean T-test Yes pwr pwr.t.test
2 Two Means T-test Yes pwr pwr.t.test
3 Paired T-test Yes pwr pwr.t.test
4 One-way ANOVA Yes pwr pwr.anova.test
5 Single Proportion Test Yes pwr pwr.p.test
6 Two Proportions Test Yes pwr pwr.2p.test
7 Chi-Squared Test Yes pwr pwr.chisq.test
8 Simple Linear Regression Yes pwr pwr.f2.test
9 Multiple Linear Regression Yes pwr pwr.f2.test
10 Correlation Yes pwr pwr.r.test
11 One Mean Wilcoxon Test Yes* pwr pwer.t.test + 15%
12 Mann-Whitney Test Yes* pwr pwer.t.test + 15%
13 Paired Wilcoxon Test Yes* pwr pwer.t.test + 15%
14 Kruskal Wallace Test Yes* pwr pwr.anova.test + 15%
15 Repeated Measures ANOVA Yes WebPower wp.rmanova
16 Multi-way ANOVA (1 Category of interest) Yes WebPower wp.kanova
17 Multi-way ANOVA (>1 Category of interest) Yes WebPower wp.kanova
18 Non-Parametric Regression (Logistic) Yes WebPower wp.logistic
19 Non-Parametric Regression (Poisson) Yes WebPower wp.poisson
20 Multilevel modeling: CRT Yes WebPower wp.crt2arm/wp.crt3arm
21 Multilevel modeling: MRT Yes WebPower wp.mrt2arm/wp.mrt3arm
22 GLMM Yes^ Simr & lme4 n/a
*-parametric test with non-parametric correction
^-detailed in future Module
One Mean T-Test
Description: This tests if a sample mean is any different R Code: pwer -> pwr.t.test
from a set value for a normally distributed variable.
pwr.t.test(d = , sig.level = , power = , type = c("two.sample",
"one.sample", "paired"))

• d=effect size
• sig.level=significant level
• power=power of test
Numeric. Cat. Var(s) Cat. Var Cat Var. # of Parametric Paired • type=type of test
Var(s) Group # Interest

1 0 0 0 Yes N/A

Example: Is the average body temperature of college

students any different from 98.6°F?
• H0=98.6°F, H1≠98.6°F
• We will guess that the effect sizes will be medium
• For t-tests: Effect size calculation
0.2=small, 0.5=medium, and 0.8 large effect sizes • Cohen’s D = (M2-M1)/SD
• Selected Two-tailed, because we were asking if temp • M2=Mean 2
differed, not whether it was simply lower or higher • M1=Mean 1
• SD =Standard deviation
One Mean T-Test
Results:

> #sample number

> pwr.t.test(d=0.50, sig.level=0.05, power=0.80, type="one.sample", alternative="two.sided")

One-sample t test power calculation

n = 33.36713 → Round up to 34
d = 0.5
sig.level = 0.05
power = 0.8
alternative = two.sided
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).
Sleep
-0.55 0.16 2.6 0.65 -0.23 0.21 -4.3 2 -1.7 1.9
Change

1. You are interested in determining if the average weight change in a year for
college freshman is greater than zero.
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).
• Effect size = (MeanH1-MeanH0)/SD= (14,500-20,000)/6000 = -0.917
• One-tailed test
• pwr.t.test(d=-0.917, sig.level=0.05, power=0.80, type="one.sample", alternative=“less")
• n = 8.871645 -> 9 samples
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).
Sleep Change -0.55 0.16 2.6 0.65 -0.23 0.21 -4.3 2 -1.7 1.9

• Effect size =(MeanH1-MeanH0)/SD =(-0.446-0)/1.96 = -0.228

• Two-tailed test
• pwr.t.test(d=-0.228, sig.level=0.05, power=0.80, type="one.sample", alternative=“two.sided")
• n = 152.91 -> 153 samples
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), and used one-tailed test
• pwr.t.test(d=0.80, sig.level=0.05, power=0.80, type="one.sample", alternative=“greater")
• n = 11.14 -> 12 samples
Two Means T-test
Description: this tests if a mean from one group is different R Code: pwer -> pwr.t.test
from the mean of another group for a normally distributed
variable. AKA, testing to see if the difference in means is pwr.t.test(d = , sig.level = , power = , type = c("two.sample",
different from zero. "one.sample", "paired"))

• d=effect size
Numeric. Cat. Var(s) Cat. Var Cat Var. # of Parametric Paired • sig.level=significant level
Var(s) Group # Interest
• power=power of test
1 1 2 1 Yes No • type=type of test

Example: Is the average body temperature higher in women

than in men?
• H0=0°F, H1>0°F
• We will guess that the effect sizes will be medium
• For t-tests: Effect size calculation
0.2=small, 0.5=medium, and 0.8 large effect sizes • Cohen’s D = (M2-M1)/SDpooled
• Selected greater, because we only cared to test if • M2=Mean 2
women’s temp was higher, not lower (group 1 is • M1=Mean 1
women, group 2 is men) • SDpooled =Pooled standard deviation
• SDpooled=√((SD12+ SD22)/2)
Two Means T-test
Results:

> #sample number

> pwr.t.test(d=0.5, sig.level=0.05, power=0.80, type="two.sample", alternative="greater")

Two-sample t test power calculation

n = 50.1508 → Round up to 51, per group

d = 0.5
sig.level = 0.05
power = 0.8
alternative = greater

NOTE: n is number in *each* group

Two Means 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 daily caloric intake different
between men and women. You collected trial data and found the average
caloric intake for males to be 2350.2 (SD=258), 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).
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

3. You are interested in determining if the average glucose level in blood is lower
in men than women
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 males to be 2350.2 (SD=258), while females had
intake of 1872.4 (SD=420).
• Effect size = (MeanH1-MeanH0)/ SDpooled =(2350.2-1872.4)/ √((2582+ 4202)/2) = 477.8/348.54 = 1.37
• two-tailed test
• pwr.t.test(d=1.37, sig.level=0.05, power=0.80, type=“two.sample", alternative=“two-sided")
• n = 9.43 -> 10 samples per group
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).
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
• Effect size = (MeanH1-MeanH0)/ SDpooled =(4.59-4.98)/ √((2.582+ 2.882)/2) = -0.14
• two-tailed test
• pwr.t.test(d=-0.14, sig.level=0.05, power=0.80, type=“two.sample", alternative=“two-sided")
• n = 801.87 -> 802 samples per group
3. You are interested in determining if the average glucose level in blood is lower in men than women
• Guessed a small effect (0.20), then used a one-tailed test
• pwr.t.test(d=-0.20, sig.level=0.05, power=0.80, type=“two.sample", alternative=“less")
• n = 309.8 -> 310 samples per group
Paired T-test
Description: this tests if a mean from one group is different R Code: pwer -> pwr.t.test
from the mean of another group, where the groups are
dependent (not independent) for a normally distributed pwr.t.test(d = , sig.level = , power = , type = c("two.sample",
variable. Pairing can be leaves on same branch, siblings, the "one.sample", "paired"))
same individual before and after a trial, etc. • d=effect size
Numeric. Cat. Var(s) Cat. Var Cat Var. # of Parametric Paired
• sig.level=significant level
Var(s) Group # Interest • power=power of test
• type=type of test
1 1 2 1 Yes Yes

Example: Is heart rate higher in patients after a run

compared to before a run?
• H0;bpm (after) – bpm (before) ≤ 0
• H1; bpm (after) – bpm (before) > 0
• We will guess that the effect sizes will be large Effect size calculation
• For t-tests: • Cohen’s D = (M2-M1)/SDpooled
0.2=small, 0.5=medium, and 0.8 large effect sizes • M2=Mean 2
• Selected One-tailed, because we only cared if bpm • M1=Mean 1
was higher after a run • SDpooled =Pooled standard deviation
• Group 1 is after the run, while group 2 is before the • SDpooled=√((SD12+ SD22)/2)
run
Paired T-test
Results:
> #sample number
> pwr.t.test(d=0.8, sig.level=0.05, power=0.80, type="paired", alternative="greater")

Paired t test power calculation

n = 11.14424 → Round up to 12 pairs

d = 0.8
sig.level = 0.05
power = 0.8
alternative = greater

NOTE: n is number of *pairs*

Paired 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 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.
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
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.
Paired T-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 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.
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
• Effect size = (MeanH1-MeanH0)/ SDpooled =(98.1-85.4)/ √((26.82+ 27.22)/2) =12.7/27 = 0.47
• one-tailed test
• pwr.t.test(d=0.47, sig.level=0.05, power=0.80, type=“paired", alternative=“greater")
• n = 29.39 -> 30 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).
• Effect size = (MeanH1-MeanH0)/ SD =(0.73)/ 2.9 = 0.25
• two-tailed test
• pwr.t.test(d=0.25, sig.level=0.05, power=0.80, type=“paired", alternative=“two.sided")
• n = 127.52 -> 128 pairs
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 {used a negative effect to match the ‘less’ alternative}
• pwr.t.test(d=-0.20, sig.level=0.05, power=0.80, type=“paired", alternative=“less")
• n = 155.92-> 156 pairs
One-Way ANOVA
Description: this tests if at least one mean is different among R Code: pwer -> pwr.anova.test
groups, where the groups are larger than two, for a normally
distributed variable. ANOVA is the extension of the Two pwr.anova.test(k =, f = , sig.level = , power = )
Means T-test for more than two groups.
• k=number of groups
• f=effect size
• sig.level=significant level
• power=power of test
Numeric. Cat. Var(s) Cat. Var Cat Var. # of Parametric Paired
Var(s) Group # Interest

1 1 >2 1 Yes No

Example: Is there a difference in new car interest rates across

6 different cities?
• H0=0%, H1≠0% Effect size calculation
• There are a total of 6 groups (cities) • η2 = SStreat / SStotal
• We will guess that the effect sizes will be small • SStreat =treatment sum of squares
• For f-tests: • SStotal =total sum of squares
0.1=small, 0.25=medium, and 0.4 large effect sizes • f = √((η2 /(1- η2)
• No Tails in ANOVA
• Groups assumed to be the same size
One-Way ANOVA
Results:

> pwr.anova.test(k =6 , f =0.1 , sig.level=0.05 , power =0.80 )

Balanced one-way analysis of variance power calculation

k=6
n = 214.7178 → Round up to 215 samples per group
f = 0.1
sig.level = 0.05
power = 0.8

NOTE: n is number in each group

One-way ANOVA: Practice
Calculate the sample size for the following scenarios (with Option 1 Option 2 Option 3 Option 4
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 6.3 9.9 5.1 1.0
the following trial data of weight lost in pounds (shown on
right)
2. You are interested in determining if there is a difference in 2.8 4.1 2.9 2.8
white blood cell counts between 5 different medication
regimes.
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 Option 1 Option 2 Option 3 Option 4
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 6.3 9.9 5.1 1.0
the following trial data of weight lost in pounds (shown on
right)
• η2 = SStreat / SStotal =31.47/(31.47+62.87) = 0.33
2.8 4.1 2.9 2.8
• f = √((0.33/(1- 0.33) = 0.7
• 4 groups
• pwr.anova.test(k =4 , f =0.7 , sig.level=0.05 , power =0.80 )
• n = 6.63 -> 7 samples per group (28 total) 7.8 3.9 3.6 4.8
2. You are interested in determining if there is a difference in
white blood cell counts between 5 different medication
regimes. 7.9 6.3 5.7 3.9
• Guessed a medium effect size (0.25)
• 5 groups
• pwr.anova.test(k =5 , f =0.25 , sig.level=0.05 , power =0.80 )
4.9 6.9 4.5 1.6
• n = 39.15 -> 40 samples per group (200 total)
Single Proportion Test
Description: this tests when you only have a single proportion R Code: pwer -> pwr.p.test
and you want to know if the proportions of certain values
differ from some constant proportion. pwr.p.test(h = , sig.level =, power =, alternative="two.sided",
"less", or "greater" )
Numeric. Cat. Var(s) Cat. Var Cat Var. # of Parametric Paired
Var(s) Group # Interest • h=effect size
0 1 2 1 N/A N/A • sig.level=significant level
• power=power of test
Example: Is there a significance difference in cancer • alternative=type of tail
prevalence of middle-aged women who have a sister with
breast cancer (5%) compared to the general population
prevalence (2%)?
• H0=0, H1≠0
• You don’t have background info, so you guess that
Effect size calculation
there is a small effect size
• h= 2*asin(sqrt(p1))-2*asin(sqrt(p2))
• For h-tests:
• p1=proportion 1
0.2=small, 0. 5=medium, and 0.8 large effect sizes
• p2=proportion 2
• Selected Two-sided, because we don’t care about
directionality
Single Proportion Test
Results:

> #sample number

> pwr.p.test(h=0.2, sig.level=0.05, power=0.80, alternative="two.sided")

proportion power calculation for binomial distribution (arcsine transformation)

h = 0.2
n = 196.2215 → Round up to 197
sig.level = 0.05
power = 0.8
alternative = two.sided
Single Proportion: 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 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).
Single Proportion: Answers
Calculate the sample size for the following scenarios (with α=0.05, and 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.
• h= 2*asin(sqrt(0.00495))-2*asin(sqrt(0.00490))=0.0007
• pwr.p.test(h=0.0007, sig.level=0.05, power=0.80, alternative=“greater")
• n = 12617464 -> 12,617,464 samples
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).
• Guess a very low effect size (0.001)
• pwr.p.test(h=-0.001, sig.level=0.05, power=0.80, alternative=“less")
• n = 6182557 -> 6,182,557 samples
Two Proportions Test
Description: this tests when you only have two groups and R Code: pwer -> pwr.2p.test
you want to know if the proportions of each group are
different from one another. pwr.2p.test(h = , sig.level =, power =,
alternative="two.sided", "less", or "greater" )
Numeric. Cat. Var(s) Cat. Var Cat Var. # of Parametric Paired
Var(s) Group # Interest
• h=effect size
0 2 2 2 N/A No • sig.level=significant level
• power=power of test
• alternative=type of tail
Example: Is the expected proportion of students passing a
stats course taught by psychology teachers different from
the observed proportion of students passing the same stats
class taught by mathematics teachers?
• H0=0, H1≠0 Effect size calculation
• You don’t have background info, so you guess that • h= 2*asin(sqrt(p1))-2*asin(sqrt(p2))
there is a small effect size • p1=proportion 1
• For h-tests: • p2=proportion 2
0.2=small, 0. 5=medium, and 0.8 large effect sizes
• Selected Two-sided, because we don’t care about
directionality
Two Proportions Test
Results:

> #sample number

> pwr.2p.test(h=0.2, sig.level=0.05, power=.80, alternative="two.sided")

Difference of proportion power calculation for binomial distribution (arcsine transformation)

h = 0.2
n = 392.443 → Round up to 393
sig.level = 0.05
power = 0.8
alternative = two.sided

NOTE: same sample sizes

Two Proportions: 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 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.
Psychology Yes Yes Yes No No Yes Yes Yes Yes No
Biology No No Yes Yes Yes No Yes No Yes Yes

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.
Two Proportions: Answers
Calculate the sample size for the following scenarios (with α=0.05, and power=0.80):
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.
• P1=7/10=0.70, P2=6/10=0.60
• h= 2*asin(sqrt(0.60))-2*asin(sqrt(0.70))=-0.21
• pwr.2p.test(h=-0.21, sig.level=0.05, power=0.80, alternative=“two.sided")
• n = 355.96 -> 356 samples
Psychology Yes Yes Yes No No Yes Yes Yes Yes No
Biology No No Yes Yes Yes No Yes No Yes Yes
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.
• Guess that the expected proportion (P1) =0.85
• h= 2*asin(sqrt(0.85))-2*asin(sqrt(0.75))=0.25
• pwr.2p.test(h=0.25, sig.level=0.05, power=0.80, alternative=“greater")
• n = 197.84 -> 198 samples
Chi-Squared Test
Description: Extension of proportions test, which asks if table R Code: pwer -> pwr.chisq.test
of observed values are any different from a table of expected
ones. Also called Goodness-of-fit test. pwr.chisq.test(w =, df = , sig.level =, power = )

• w=effect size
Numeric. Cat. Var(s) Cat. Var Cat Var. # of Parametric Paired • df=degrees of freedom
Var(s) Group # Interest • sig.level=significant level
0 ≥1 ≥2 1 N/A No • power=power of test

Example: Does the observed proportions of phenotypes

from a genetics experiment different from the expected
9:3:3:1?
• H0=0, H1≠0
• You don’t have background info, so you guess that Effect size calculation
there is a medium effect size • w = √(Χ2 /(n*df))
• For w-tests: • X2= Chi-squared = ∑(O-E)2/E
0.1=small, 0.3=medium, and 0.5 large effect sizes • O=observed
• Degrees of freedoms is the number of proportions • E=expected
minus 1 • n=number of samples
4 (phenotypes) – 1 = 3 • df= degrees of freedom
Chi-Squared Test
Results:

> #sample number

> pwr.chisq.test(w=0.3, df=3, sig.level=0.05, power=0.80)

Chi squared power calculation

w = 0.3
N = 121.1396 → Round up to 122
df = 3
sig.level = 0.05
power = 0.8

NOTE: N is the number of observations

Chi-Squared: 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 ethnic ratios in a company differ by gender. You
collect the following trial data from 200 employees.
Gender White Black Am. Indian Asian
Male 0.60 0.25 0.01 0.14
Female 0.65 0.21 0.11 0.03

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.
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
Chi-Squared: Answers
Calculate the sample size for the following scenarios (with α=0.05, and power=0.80):
1. You are interested in determining if the ethnic ratios in a company differ by gender. You collect the following
trial data from 200 employees. Gender White Black Am. Indian Asian
Male 60 25 1 14
Female 65 21 11 3
• If they were equal the expected ratios should be the same as the overall ethnic ratios (62.5, 23.0, 6.0, 8.5)
• Will just focus on males
• Χ2 (Chi-squared)= ∑(O-E)2/E = (60-62.5)2/62.5 + (25-23)2/23 + (1-6)2/6 + (14-8.5)2/8.5
• =0.10 + 0.17 + 4.17 + 3.56 = 8
• w = √(Χ2 /(n*df))= √(8/(200*3))=0.115
• pwr.chisq.test(w=0.115, df=3, sig.level=0.05, power=0.80)
• n = 824.39 -> 825 samples
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.
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

• Χ2 (Chi-squared) = ∑(O-E)2/E = (7-5)2/5 + (5-5)2/5 + (5-5)2/5 + (3-5)2/5 = 0.8 + 0 + 0 + 0.8 = 1.6

• w = √(Χ2 /(n*df))= √(1.6/(20*3))=0.163
• pwr.chisq.test(w=0.163, df=3, sig.level=0.05, power=0.80)
• n = 410.34 -> 411 samples
Simple Linear Regression
Description: this test determines if there is a significant R Code: pwer -> pwr.f2.test
relationship between two normally distributed numerical
variables. The predictor variable is used to try to predict the pwr.f2.test(u =, v= , f2=, sig.level =, power = )
response variable.
• u=numerator degrees of freedom
Numeric. Cat. Var(s) Cat. Var Cat Var. # of Parametric Paired
• v=denominator degrees of freedom
Var(s) Group # Interest • f2=effect size
2 0 N/A N/A Yes N/A • sig.level=significant level
• power=power of test
Example: Is there a relationship between height and
weight in college males?
• H0=0, H1≠0
• You don’t have background info, so you guess that there is a
large effect size Effect size calculation
• For f2-tests: • f2=R=√(R2)
0.02=small, 0.15=medium, and 0.35 large effect sizes
• R=correlation coefficient
• For simple regression (only one predictor variable) =
numerator df=1 • R2 =goodness-of-fit
• Output will be denominator degrees of freedom rather than • Use adjusted R2
sample size; will need to round up and add 2 to get sample
size
Simple Linear Regression
Results:

> #sample number

> pwr.f2.test(u=1, f2=0.35, sig.level=0.05, power=0.80)

Multiple regression power calculation

u=1
v = 22.50313
f2 = 0.35
sig.level = 0.05
power = 0.8

> #denominator df to sample size

> round(22.5031,0)+2
[1] 25 → Sample size
Simple Linear Regression: Practice
Calculate the sample size for the following scenarios (with α=0.05, and power=0.80):
1. You are interested in determining if height (meters) in plants can predict yield (grams of
berries). You collect the following trial data.
Yield 46.8 48.7 48.4 53.7 56.7
Height 14.6 19.6 18.6 25.5 20.4

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).
Simple Linear Regression: Answers
Calculate the sample size for the following scenarios (with α=0.05, and power=0.80):
1. You are interested in determining if height (meters) in plants can predict yield (grams of
berries). You collect the following trial data. Yield 46.8 48.7 48.4 53.7 56.7
Height 14.6 19.6 18.6 25.5 20.4
• Created variables in R
• yield<-c(46.8, 48.7, 48.4, 53.7, 56.7)
• height<-c(14.6, 19.6, 18.6, 25.5, 20.4)
• Ran linear model to find R-squared
• linearMod <- lm(height~yield)
• summary(linearMod) -> adj R2=0.2784
• f2=R=√(adj R2)= √(0.4588)=0.53
• pwr.f2.test(u=1, f2=0.53, sig.level=0.05, power=0.80)
• v=14.96 -> 15+ 2(variables) ->17 samples
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 predictor so 1 df
• pwr.f2.test(u=1, f2=0.35, sig.level=0.05, power=0.80)
• v=22.5 -> 23+ 2(variables) ->25 samples
Multiple Linear Regression
Description: The extension of simple linear regression. The R Code: pwer -> pwr.f2.test
first major change is there are more predictor variables. The
second change is that interaction effects can be used. Finally, pwr.f2.test(u =, v= , f2=, sig.level =, power = )
the results typically can’t be plotted.
• u=numerator degrees of freedom
Numeric. Cat. Var(s) Cat. Var Cat Var. # of Parametric Paired
Var(s) Group # Interest • v=denominator degrees of freedom
>2 0 N/A N/A Yes N/A • f2=effect size
• sig.level=significant level
• power=power of test
Example: Can height, age, and time spent at the gym, predict
weight in adult males?
• H0=0, H1≠0
• You don’t have background info, so you guess that there is a
medium effect size
• For f2-tests: Effect size calculation
0.02=small, 0.15=medium, and 0.35 large effect sizes • f2=R=√(R2)
• Numerator degrees of freedom is the number of predictor • R=correlation coefficient
variables (3) • R2 =goodness-of-fit
• Output will be denominator degrees of freedom rather than • Use adjusted R2
sample size; will need to round up and add the total number
of variables (4)
Multiple Linear Regression
Results:

> #sample number

> pwr.f2.test(u=3, f2=0.15, sig.level=0.05, power=0.80)

Multiple regression power calculation

u=3
v = 72.70583
f2 = 0.15
sig.level = 0.05
power = 0.8

> #denominator df to sample size

> round(72.70583,0)+4
[1] 77 → Sample Size
Multiple Linear Regression: Practice
Calculate the sample size for the following scenarios (with α=0.05, and power=0.80):
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.
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
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).
Multiple Linear Regression: Answers
Calculate the sample size for the following scenarios (with α=0.05, and power=0.80):
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.
• Created variables in R Yield 46.8 48.7 48.4 53.7 56.7
• yield<-c(46.8, 48.7, 48.4, 53.7, 56.7) Height 14.6 19.6 18.6 25.5 20.4
• height<-c(14.6, 19.6, 18.6, 25.5, 20.4) Weight 95.3 99.5 94.1 110 103
• weight<-c(95.3, 99.5, 94.1, 110, 103) Fertilizer 2.1 3.2 4.3 1.1 4.3
• Fert<-c(2.1, 3.2, 4.3, 1.1, 4.3)
• Ran linear model to find R-squared
• linearMod2 <-lm(height~yield + weight + Fert)
• summary(linearMod2) -> Adj R2 = 0.6765
• f2=R=√(adj R2)= = √(0.6765)=0.822
• pwr.f2.test(u=3, f2=0.822, sig.level=0.05, power=0.80)
• v=13.7 -> 14+ 4(variables) ->18 samples
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 variables (df=3)
• pwr.f2.test(u=3, f2=0.35, sig.level=0.05, power=0.80)
• v=31.31 -> 32+ 4(variables) ->36 samples
Correlation
Description: this test determines if there is a difference R Code: pwr -> pwer.r.test
between two numerical values. It is like simple regression,
but is not identical. pwr.r.test(r = , sig.level = , power = )

• r=correlation
Numeric. Cat. Var(s) Cat. Var Cat Var. # of Parametric Paired • sig.level=significant level
Var(s) Group # Interest
• power=power of test
2 0 N/A N/A Yes No

Example: Is there a correlation between hours studied and

test score?
• H0=0, H1≠0
• You don’t have background info, so you guess that
there is a large correlation Effect size calculation
• For correlation levels (r): • r=correlation coefficient
0.1=small, 0.3=medium, and 0.5 large correlations
Correlation
Results:

> #sample number

> pwr.r.test(r=0.5, sig.level=0.05, power=0.80)

approximate correlation power calculation (arctangh transformation)

n = 28.24841 → Round up to 29
r = 0.5
sig.level = 0.05
power = 0.8
alternative = two.sided
Correlation: Practice
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 correlation between height and
weight in men
Height 178 166 172 186 182
Males
Weight 165 139 257 225 196

2. You are interested in determining if, in lab mice, the correlation between
longevity (in months) and average protein intake (grams).
Correlation: 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 correlation between height and
weight in men 178
Males
166 172
Height 186 182

• Created variables in R and ran correlation test Weight 165 139 257 225 196

• MH <-c(178,166,172,186,182)
• MW <-c(165,139,257,225,196)
• cor(MH, MW) -> 0.37
• pwr.r.test(r=0.37, sig.level=0.05, power=0.80)
• n = 54.19 -> 55 samples
2. You are interested in determining if, in lab mice, the correlation between longevity (in
months) and average protein intake (grams).
• Guessed large (0.5) correlation
• pwr.r.test(r=0.5, sig.level=0.05, power=0.80)
• n = 28.24 -> 29 samples
Non-Parametric T-tests
Description: versions of the t-tests for non-parametric Examples: (for t-tests, 0.2=small, 0.5=medium, and 0.8 large effect
sizes)
data. One Mean Wilcoxon:
• One Mean Wilcoxon: sample mean against set value Is the average number of children in Grand Forks families different
• Mann-Whitney: two sample means (unpaired) than 1?
• Paired Wilcoxon: two sample means (paired) • H0=1 child
• H1>1 child
• You don’t have background info, so you guess that there is a medium
Name Numeric. Cat. Var(s) Cat. Var Cat Var. # of Parametric Paired
Var(s) Group # Interest effect size
• Select one-tailed (greater)
One Mean 1 0 0 0 No N/A
Wilcoxon Mann-Whitney:
Mann-Whitney
Does the average number of snacks per day for individuals on a diet
1 1 2 1 No No
differ between young and old persons?
Paired Wilcoxon • H0=0 difference in snack number,
1 1 2 1 No Yes • H1≠0 difference in snack number
• You don’t have background info, so you guess that there is a small
• There aren’t any R packages that had useful non- •
effect size
Select two-sided
parametric t-tests Paired Wilcoxon:
• I suggest using the parametric + 15% approach Is genome methylation patterns different between identical twins?
• H0=0% methylation
• H1≠0% methylation
Effect size calculation • You don’t have background info, so you guess that there is a large
• Cohen’s D: (M2-M1)/SD; (M2-M1)/Sdpooled; (Meandiff)/ SDdiff effect size
• Select one-tailed (greater)
Non-parametric Tests
Results:
>#One Mean Wilcoxon >#Paired Wilcoxon
> pwr.t.test(d=0.5, sig.level=0.05, power=0.80, type="one.sample", alternative="greater") > pwr.t.test(d=0.8, sig.level=0.05, power=0.80, type="paired",
One-sample t test power calculation alternative="greater")
n = 26.13753 Paired t test power calculation
d = 0.5 n = 11.14424
sig.level = 0.05 d = 0.8
power = 0.8 sig.level = 0.05
alternative = greater power = 0.8
> #Non-parametric correction alternative = greater
> round(26.13753*1.15,0) NOTE: n is number of *pairs*
[1] 30 → Total sample size > #Non-parametric correction
> round(11.14424*1.15,0)
>#Mann-Whitney [1] 13 → Total number of pairs
> pwr.t.test(d=0.2, sig.level=0.05, power=0.80, type=“two.sample",
alternative="two.sided")
Two-sample t test power calculation
n = 198.1508
d = 0.2
sig.level = 0.05
power = 0.8
alternative = two.sided
> #Non-parametric correction
> round(198.1508*1.15,0)
[1] 228 → Total sample size
Non-Parametric T-tests: 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 number of pets in Grand Forks families is
greater than 1. You collect the following trial data for pet number.
Pets 1 1 1 3 2 1 0 0 0 4
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.
Young meals 1 2 2 3 3 3 3 4
Older meals 1 1 1 2 2 2 3 3

3. 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).
Methy. Diff (%) 5.96 5.63 1.25 1.17 3.59 1.64 1.6 1.4
Non-Parametric T-tests: Answers
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 number of pets in Grand Forks families is greater than 1. You
collect the following trial data for pet number.
• Effect size = (MeanH1-MeanH0)/SD= (1.3-1.0)/1.34 =0.224 Pets 1 1 1 3 2 1 0 0 0 4
• One-tailed test
• pwr.t.test(d=0.224, sig.level=0.05, power=0.80, type="one.sample", alternative=“greater")
• n =124.58*1.15 (then round up)-> 143 samples
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.
Young meals 1 2 2 3 3 3 3 4
Older meals 1 1 1 2 2 2 3 3
• Effect size = (MeanH1-MeanH0)/SDpooled =(2.625-1.875)/ √((0.922+ 0.832)/2) = 0.856
• One-tailed test
• pwr.t.test(d=0.856, sig.level=0.05, power=0.80, type=“two.sample", alternative=“greater")
• n = 17.59*1.15 (then round up)-> 20 samples per group
3. You are interested in determining if genome methylation patterns are different in the first fraternal twin born
compared to the second. You collected the following trial data on methylation level difference (in percentage).
• Effect size = (Meandiff)/ SDdiff =(2.78)/ 2.01 = 1.38
• Two-tailed test
Methy. Diff (%) 5.96 5.63 1.25 1.17 3.59 1.64 1.6 1.4
• pwr.t.test(d=1.38, sig.level=0.05, power=0.80, type=“paired", alternative=“two.sided")
• n = 6.29*1.15 (then round up) -> 7 pairs
Kruskal Wallace Test
Description: this tests if at least one mean is different among R Code: pwer -> pwr.anova.test
groups, where the groups are larger than two for a non-normally
distributed variable. (AKA, non-parametric ANOVA). There really isn’t pwr.anova.test(k =, f = , sig.level = , power = )
a good way of calculating sample size in R, but you can use a rule of
thumb: • k=number of groups
1. Run Parametric Test • f=effect size
2. Add 15% to total sample size • sig.level=significant level
• power=power of test
Numeric. Cat. Var(s) Cat. Var Cat Var. # of Parametric Paired
Var(s) Group # Interest

1 1 >2 1 No No

Example: Is there a difference in draft rank across 3 different

months?
• H0=0, H1≠0 Effect size calculation
• There will be a total of 3 groups (months) • η2 = SStreat / SStotal
• You don’t have background info, so you guess that • SStreat =treatment sum of squares
there is a medium effect size • SStotal =total sum of squares
• For f-tests: • f = √((η2 /(1- η2)
0.1=small, 0.25=medium, and 0.4 large effect sizes
• No Tails in ANOVA
• Groups assumed to be the same size
Kruskal Wallace Test
Results:

> #sample number of ANOVA

> pwr.anova.test(k =3 , f =0.25 , sig.level=0.05 , power =0.80 )

Balanced one-way analysis of variance power calculation

k=3
n = 52.3966
f = 0.25
sig.level = 0.05
power = 0.8

NOTE: n is number in each group

> #15% correction factor

> 52.3996 * 1.15
[1] 60.25954 → Round up to 61 samples per group
Kruskal Wallace Test: Practice
Calculate the sample size for the following scenarios Faculty Staff Hourly
(with α=0.05, and power=0.80):
1. You are interested in determining there is a
42 46 29
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). 45 45 42
2. You are interested in determining there is a
difference in assistant professor salaries across 25
different departments. 46 37 33

55 42 50

42 40 23
Kruskal Wallace Test: Answers
Calculate the sample size for the following scenarios Faculty Staff Hourly
(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 42 46 29
(faculty, staff, and hourly workers). You collect the
following trial data of weekly hours (shown on right).
• η2 = SStreat / SStotal =286.5/(286.5+625.2) = 0.314 45 45 42
• f = √((0.314/(1- 0.314) = 0.677
• 3 groups
• pwr.anova.test(k =3, f =0.677, sig.level=0.05, power =0.80) 46 37 33
• n =8.09*1.15 (then round up)-> 10 samples per group
2. You are interested in determining there is a
difference in assistant professor salaries across 25 55 42 50
different departments.
• Guess small effect size (0.10)
• 25 groups 42 40 23
• pwr.anova.test(k =25, f =0.10, sig.level=0.05, power =0.80)
• n =90.67*1.15 (then round up)-> 105 samples per group
Repeated Measures ANOVA
Description: this tests if at least one mean is different among R Code: WebPower -> wp.rmanova
groups, where the groups are repeated measures (more than
two) for a normally distributed variable. Repeated Measures wp.rmanova(ng = NULL, nm = NULL, f = NULL, nscor = 1, alpha =
ANOVA is the extension of the Paired T-test for more than two 0.05, power = NULL, type = 0)
• ng=number of groups
groups. • nm=number of measurements
• f=effect size
Numeric. Cat. Var(s) Cat. Var Cat Var. # of Parametric Paired • nscor=nonsphericity correction coefficient
Var(s) Group # Interest • alpha=significant level of test
1 1 >2 1 Yes Yes • power=statistical power
• type=(0,1,2) The value "0" is for between-effect; "1" is for
Example: Is there a difference in blood pressure at 1, 2, 3, within-effect; and "2" is for interaction effect
and 4 months post-treatment?
• H0=0, H1≠0 Effect size calculation NOTE:
𝜎
• 1 group, 4 measurements • 𝑓= 𝑚 • Within-effects: variability
𝜎
• of a particular value for
You don’t have background info, so you guess that there is a • σm = standard deviation of group
individuals in a sample
small effect size means • Between-effects:
• For f-tests: σ𝐾 2 examines differences
𝑘=1(𝑚𝑘 −𝑚)
0.1=small, 0.25=medium, and 0.4 large effect sizes • 𝜎𝑚 = between individuals
𝑘
• For the nonsphericity correction coefficient, 1 means • mk = group mean
sphericity is met. There are methods to estimate this but • m = overall mean
will go with 1 for this example. • k=number of groups,
• Type will be 1, as we want within-effect • σ =overall standard deviation
Repeated Measures ANOVA
Results:

> #sample size

> wp.rmanova(n=NULL, ng=1, nm=4, f=0.1, nscor=1,
+ alpha=0.05, power=0.80, type=1)
Repeated-measures ANOVA analysis

n f ng nm nscor alpha power

1091.559 0.1 1 4 1 0.05 0.8
→ Round up to 1092 samples total
NOTE: Power analysis for within-effect test
URL: http://psychstat.org/rmanova
Repeated Measures ANOVA: Practice
Calculate the sample size for the following scenarios 6 months 12 months 18 months 24 months
(with α=0.05, and power=0.80):
1. You are interested in determining if there is a
38 38 46 52
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).
13 44 15 29
2. You are interested in determining if there is a
difference in antibody levels at 1, 2, and 3 months
post-treatment. 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 6 months 12 months 18 months 24 months
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).
38 38 46 52
𝜎𝑚 σ𝐾
𝑘=1(𝑚𝑘 −𝑚)
2
• 𝑓= = /σ
𝜎 𝑘

(27.8−37.3)2 +(38.4−37.3)2 +(38.8−37.3)2 +(25.2−37.3)2

• f= / 12.74 = 0.608
4
13 44 15 29
• To get sphericity, ran ANOVA
• library(ez)
• anova3 <- ezANOVA(ex3, dv=Serum, wid=Patient, within=.(Month),detailed=TRUE)
• print(anova3$ANOVA)
32 35 53 60
• Sphericity was non-significant (0.43), so coefficient of 1
• One group, four measurements, within-effects so type 1
• wp.rmanova(n=NULL, ng=1, nm=4, f=0.608, nscor=1, alpha=0.05, power=0.80, type=1)
• n =30.81-> 31 samples total
35 48 51 44
2. You are interested in determining if there is a difference in antibody levels
at 1, 2, and 3 months post-treatment.
• Guess a nonsphericity correction of of 1 and medium effect 0.25
• One group, three measurements, type 1
• wp.rmanova(n=NULL, ng=1, nm=3, f=0.25, nscor=1, alpha=0.05, power=0.80, type=1) 21 27 29 36
• n =155.66-> 156 samples total
Multi-Way ANOVA (1 Category of Interest)
Description: this test is an extension of ANOVA, where there R Code: WebPower -> wp.kanova
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.). wp.kanova(ndf = NULL, f = NULL, ng = NULL, alpha = 0.05, power =
NULL)
Numeric. Cat. Var(s) Cat. Var Cat Var. # of Parametric Paired
Var(s) Group # Interest
• ndf=numerator degrees of freedom
1 ≥2 ≥2 1 Yes No • f=effect size
• ng=number of groups
Example: Is there difference in treatment (Drug A, B, and C) • alpha=significance level
from a series of four different hospital sections (Block 1, 2, 3, • power=statistical power
and 4)?
• H0=0, H1≠0 Effect size calculation
• Category of interest: Treatment • 𝑓= 𝑏
𝜎
𝜎𝑤
• Want to control for the Sections (Blocking) • 𝜎𝑏 =standard deviation of blocking sections
• Numerator df (Treatment) = 3-1=2 σ𝐾 2
𝑘=1(𝜇𝑘 −𝜇)
• Number of groups (Treatment *Section)=3*4=12 • 𝜎𝑏 = , where 𝜇𝑘 =mean of section, 𝜇=overall mean, and
𝐾
• You don’t have background info, so you guess that there is a 𝐾=number of sections
medium effect size • 𝜎𝑤 =standard deviation of all groups (treatment*section)
• For f-tests: σ𝐽𝑗=1 σ𝐾
2
𝑘 (𝜇𝑗𝑘 −𝜇)
0.1=small, 0.25=medium, and 0.4 large effect sizes • 𝜎𝑤 = , where 𝜇𝑗𝑘 =mean of group, 𝜇=overall mean,
𝐽𝐾

𝐽=number of treatments,
• and 𝐾=number of sections
Multi-Way ANOVA (1 Category of Interest)
Results:

> #sample size

> wp.kanova(ndf=2, f=0.25, ng=12, alpha=0.05, power=0.80)
Multiple way ANOVA analysis

n ndf ddf f ng alpha power

157.3764 2 145.3764 0.25 12 0.05 0.8
→ Round up to 158 total samples
NOTE: Sample size is the total sample size
URL: http://psychstat.org/kanova
Multi-Way ANOVA (>1 Category of Interest)
Description: this test is an extension of ANOVA, where there R Code: WebPower -> wp.kanova
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. wp.kanova(ndf = NULL, f = NULL, ng = NULL, alpha = 0.05, power =
NULL)

• ndf=numerator degrees of freedom

• f=effect size
Numeric. Cat. Var(s) Cat. Var Cat Var. # of Parametric Paired • ng=number of groups
Var(s) Group # Interest • alpha=significance level
1 ≥2 ≥2 >1 Yes No • power=statistical power

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 Cancer Stage Effect size calculation (sort of)
• η2 = 𝜎𝑚 2 / 𝜎2
• Numerator df = Treat DF * Age DF * Stage DF = (3-1)*(3- 𝑡
2
1)*(5-1)=2*2*4=16 • 𝜎𝑚 =between-group variance
• Number of groups = Treat*Age*Stage = 3*3*5=45 • 𝜎𝑡2 =total variance
• You don’t have background info, so you guess that there is a
• f = η2 Τ(1 − η2 )
small effect size
• For f-tests:
0.1=small, 0.25=medium, and 0.4 large effect sizes
Multi-Way ANOVA (>1 Category of Interest)
Results:

> #sample size

> wp.kanova(ndf=16, f=0.10, ng=45, alpha=0.05, power=0.80)
Multiple way ANOVA analysis

n ndf ddf f ng alpha power

1940.159 16 1895.159 0.1 45 0.05 0.8
→ Round up to 1941 total samples
NOTE: Sample size is the total sample size
URL: http://psychstat.org/kanova
Multi-Way ANOVA: Practice
Calculate the sample size for the following scenarios Drug A Drug B Drug C
(with α=0.05, and power=0.80):
c a e c a e c a e
1. You are interested in determining if there is a
-6.4 8.7 -3.1 1.3 -6.0 6.8 -2.0 -4.3 -1.2
difference in treatment (Drug A, B, and C), while
controlling for age (child=c, adult=a, elder=e). You
collect the following trial data for treatment (shown -8.2 -6.3 -6.5 3.6 1.3 2.4 1.5 1.3 1.1

on right).
7.9 -1 -1.5 3.9 -1.9 1.3 2.5 -8.2 -9.7
2. You are interested in determining if there is a
difference in treatment (Drug A, B, and C) across age
(child, adult, elder) and cancer stage (I, II, III, IV, V).
You collect trial data and find that the between-
group variance is 27.3, while the total variance is
85.2.
Multi-Way ANOVA: Answers
Calculate the sample size for the following scenarios (with α=0.05, and Drug A Drug B Drug C
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=c, adult=a, elder=e). You collect the c a e c a e c a e
following trial data for treatment (shown on right).
• Only care about Drug, so focus on treatment only -6.4 8.7 -3.1 1.3 -6.0 6.8 -2.0 -4.3 -1.2
2
σ𝐾 2 σ𝐽𝑗=1 σ𝐾
𝑘 (𝜇𝑗𝑘 −𝜇)
𝜎𝑏 𝑘=1(𝜇𝑘 −𝜇)
• 𝑓= = ൘
𝜎𝑤 𝐾 𝐽𝐾

• J= drug groups, K=age groups

-8.2 -6.3 -6.5 3.6 1.3 2.4 1.5 1.3 1.1
• 𝑓= ((−1.82 − −0.84 )2 +((1.41 − −0.84 )2 +((−2.11 − −0.84 )2 )Τ3 /
(−2.23 − (−0.84))2 +(0.47 − (−0.84))2 +(−3.7 − (−0.84))2 +(2.93 − (−0.84))2 +
ൗ3 ∗ 3
(−2.2 − (−0.84))2 +(3.5 − (−0.84))2 +(0.67 − (−0.84))2 +(−3.73 − (−0.84))2 +(−3.27 − (−0.84))2

• 𝑓 = 1.74Τ2.65 = 0.657 7.9 -1 -1.5 3.9 -1.9 1.3 2.5 -8.2 -9.7
• Numerator df = 3 (Drug treatments) -1 = 2
• Number of groups = 3*3 = 9
• wp.kanova(ndf=2, f=0.657, ng=9, alpha=0.05, power=0.80)
• n =26.6-> 27 samples total (3 per group)
2. You are interested in determining if there is a difference in treatment (Drug A,
B, and C) across age (child, adult, elder) and cancer stage (I, II, III, IV, V). You
collect trial data and find that the between-group variance is 27.3, while the total variance
is 85.2.
• 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
• 2
η2 = 𝜎𝑚 / 𝜎𝑡2 = 27.3/85.2 =0.32
• f = η2Τ(1 − η2 ) = 0.32Τ(1 − 0.32) = 0.686
• wp.kanova(ndf=16, f=0.686, ng=45, alpha=0.05, power=0.80)
• n 67.03-> 68 samples, need 90 samples to have even groups (2 per group)
Logistic Regression
Description: Tests whether a predictor variable is a significant R Code: WebPower -> wp.logistic
predictor of a binary outcome, with or without other
covariates. It is a type of non-parametric regression:
numerical variables are not normally distributed. In Logistic wp.logistic(n = NULL, p0 = NULL, p1 = NULL, alpha = 0.05,
power = NULL, alternative = c("two.sided", "less", "greater"),
regression, the response variable (Y) is binary (0/1). family = c("Bernoulli", "exponential", "lognormal", "normal",
"Poisson", "uniform"), parameter = NULL)

Numeric. Cat. Var(s) Cat. Var Cat Var. # of Parametric Paired • p0= Prob(Y=1|X=0): the probability of observing 1 for the outcome
Var(s) Group # Interest variable Y when the predictor X equals 0
• p1= Prob(Y=1|X=1): the probability of observing 1 for the outcome
≥2 0 N/A N/A No N/A variable Y when the predictor X equals 1
• alpha= significance level
Example: Does body mass index (BMI) influences mortality • power= statistical power
(yes 1, no 0)? • alternative= direction of the alternative hypothesis ("two.sided" or
• H0=0, H1≠0 "less" or "greater")
• You must have at least some background (or good guess) on • family= distribution of the predictor ("Bernoulli","exponential",
"lognormal", "normal","Poisson", "uniform"). The default is
the p0 and p1 probabilities; let’s use 0.15 and 0.25
"Bernoulli"
• Will use ‘two-sided’ because we don’t care about direction • parameter= corresponding parameter for the predictor’s distribution.
• BMI seems normally distributed, so will go with normal for The default is 0.5 for "Bernoulli", 1 for "exponential", (0,1) for
the family (but should confirm the distribution for whatever "lognormal" or "normal", 1 for "Poisson", and (0,1) for "uniform"
predictor variable you use)
Effect size calculation
• Can leave the parameter empty at the default of mean=0,
SD=1 • N/A, uses probability information instead
Logistic Regression
Results:

> #sample size

> wp.logistic(p0=0.15, p1=0.25, alpha=0.05, power=0.80, alternative="two.sided", family="normal")
Power for logistic regression

p0 p1 beta0 beta1 n alpha power

0.15 0.25 -1.734601 0.6359888 165.3687 0.05 0.8
→ Round up to 166 total samples
URL: http://psychstat.org/logistic
Poisson Regression
Description: Tests whether a predictor variable influences the R Code: WebPower -> wp.poisson
rate of events over a set period, with or without other
covariates. It is a type of non-parametric regression: wp.poisson(n = NULL, exp0 = NULL, exp1 = NULL, alpha = 0.05,
numerical variables are not normally distributed. In Poisson power = NULL, alternative = c("two.sided", "less", "greater"), family
= c("Bernoulli", "exponential", "lognormal", "normal", "Poisson",
regression, the events within the rate are assumed to be "uniform"), parameter = NULL)
independent. Subjects can have multiple events, as long as
they are independent. • exp0= the base rate under the null hyp.(must be positive value)
Numeric. Cat. Var(s) Cat. Var Cat Var. # of Parametric Paired • exp1= the relative increase of the event rate. It is used for calculation
Var(s) Group # Interest of the effect size
• alpha= significance level
≥2 0 N/A N/A No N/A • power= statistical power
• alternative= direction of the alternative hypothesis ("two.sided" or
Example: Does a change in drug dose decrease the rate of "less" or "greater")
adverse affects? • family= distribution of the predictor ("Bernoulli","exponential",
• H0=0, H1≠0 "lognormal", "normal","Poisson", "uniform"). The default is
• You must have at least some background (or good guess) on the "Bernoulli"
exp0 and exp1 rate; let’s use 1.0 and 0.80 • parameter= corresponding parameter for the predictor’s distribution.
The default is 0.5 for "Bernoulli", 1 for "exponential", (0,1) for
• Will use ‘less’ because we’re asking if the alternative hypothesis
"lognormal" or "normal", 1 for "Poisson", and (0,1) for "uniform"
has a lower rate than the null
• Because I have no idea about the distribution of drug dosage, I
will go with uniform (but should confirm the distribution for
whatever predictor variable you use) Effect size calculation
• Can leave the parameter empty at the default of mean=0, SD=1 • N/A, uses rate information instead
Poisson Regression
Results:

> #sample size

> wp.poisson(exp0=1.0, exp1=0.80, alpha=0.05, power=0.80, alternative ="less", family="uniform")
Power for Poisson regression

n power alpha exp0 exp1 beta0 beta1

1666.539 0.8 0.05 1 0.8 0 -0.2231436
→ Round up to 1667 total samples
URL: http://psychstat.org/poisson
Logistic/Poisson Regression: Practice
Calculate the sample size for the following scenarios (with α=0.05, and power=0.80):
1. You are interested in determining if body temperature influences sleep disorder prevalence
(yes 1, no 0). You collect the following trial data.
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

2. You are interested in determining if the rate of lung cancer incidence changes with a drug
treatment.
Logistic/Poisson Regression: Answers
Calculate the sample size for the following scenarios (with α=0.05, and power=0.80):
1. You are interested in determining if body temperature influences sleep disorder prevalence
(yes 1, no 0). You collect the following trial data. 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
• Logistic Regression (two.sided)
• Mean temp is 98.4 (SD=0.436) -> range of one SD=(97.964 --98.836)
• P0=0.33 (as only one had sleep disorder at ranges outside one SD); P1=0.67
• Temperature is normally distributed
• wp.logistic(p0=0.33, p1=0.67, alpha=0.05, power=0.80, alternative="two.sided", family="normal")
• n =40.80-> 41 samples total
2. You are interested in determining if the rate of lung cancer incidence changes with a drug
treatment.
• Poisson Regression (two.sided)
• Expect the base rate (intercept) for male lung cancer is 57.8 (per 100,000), so exp0 = exp(57.8/100000) = 1.0005
• Expect the relative increase of the event rate (slope) to be -1.02, so exp1 = exp(-1.02) = 0.36
• Go with default distribution of Bernoulli
• wp.poisson(exp0=1.0005, exp1=0.36, alpha=0.05, power=0.80, alternative =“two.sided", family=“Bernoulli")
• n =56.8-> 59 samples total
Multilevel Modeling: Cluster Randomized Trials
Description: Multilevel models are used when data are R Code: WebPower -> wp.crt2arm
clustered within a hierarchical structure that will make them
wp.crt2arm(n=NULL, f = NULL, J = NULL, icc = NULL, power = NULL, alpha =
non-independent. Also known as linear mixed models. 0.05, alternative = c("two.sided", "one.sided"))

Cluster randomized trials (CRT) are a type of multilevel design • n= sample size (number of individuals per cluster)
• f= effect size (either main effect of treatment, or mean difference
where the entire cluster is randomly assigned to a control between treatment clusters and control clusters)
arm or one or more treatment arms. • J= number of clusters/sides. It tells how many clusters are considered
in the study design. At least two clusters are required
• icc= intra-class correlation (degree to which two randomly drawn
observations within a cluster are correlated)
• alpha= significance level
Example: Is there a difference in blood glucose levels between a treatment and
• power= statistical power
control?
• alternative= direction of the alternative hypothesis ("two.sided" or
• H0=0, H1≠0
• You don’t have background info, so you guess that there is a medium "less" or "greater")
effect size
• For f-tests: Effect size calculation NOTE: here we show a
0.1=small, 0.25=medium, and 0.4 large effect sizes 2 arm example
• Don’t know the icc, so will guess at 0.1 (0.5 is the default for repeated • 𝑓 = 𝜇𝐷 ൗ (𝜎𝐵2 +𝜎𝑊
2
) (treatment, control); to
measures, but we expect this to be lower, since the observations are from use a 3 arm design
different people) • 𝜇𝐷 =mean difference between (treatment1,
• Alternative is “two.sided” as we only care about difference treatment and control clusters treatment2, control),
• We can test for two sizes: number per cluster or cluster number
1. Try for 100 clusters • 𝜎𝐵2 =between-cluster variance use wp.crt3arm
2
2. Try for 15 individuals per cluster to get cluster number • 𝜎𝑊 =within-cluster variance
Multilevel Modeling: Cluster Randomized Trials
Results:

> #Multilevel Modeling

>
> #CRT sample size (number per cluster)
> wp.crt2arm(f=0.25, J=100, icc=0.1, alpha=0.05, power=0.80, alternative="two.sided")
Cluster randomized trials with 2 arms

J n f icc power alpha

100 9.456102 0.25 0.1 0.8 0.05
→ Round up to 10 individuals per cluster
NOTE: n is the number of subjects per cluster.
URL: http://psychstat.org/crt2arm
>
> #CRT sample size (cluster number)
> wp.crt2arm(f=0.25, n=15, icc=0.1, alpha=0.05, power=0.80, alternative="two.sided")
Cluster randomized trials with 2 arms

J n f icc power alpha

82.33782 15 0.25 0.1 0.8 0.05
→ Round up to 84 clusters (42 per arm)
NOTE: n is the number of subjects per cluster.
URL: http://psychstat.org/crt2arm
Multilevel Modeling: Multisite Randomized Trials
Description: Multilevel models are used when data are clustered R Code: WebPower -> wp.mrt2arm
within a hierarchical structure that will make them non-independent.
wp.mrt2arm(n = NULL, f = NULL, J = NULL, tau00 = NULL, tau11 = NULL, sg2 = NULL, power
Also known as linear mixed models. = NULL, alpha = 0.05, alternative = c("two.sided", "one.sided"), type = c("main", "site",
"variance"))
Multisite randomized trails (MRT) are a type of multilevel design • f= effect size (either main effect of treatment, or mean difference between treatment
where the entire cluster is randomly assigned to a control arm or one clusters and control clusters)
or more treatment arms, but then can be analyzed in a two-level • J= number of clusters/sides. It tells how many clusters are considered in the study
design. At least two clusters are required
hierarchical linear model. Can look at three types of tests: (1) The • tau00= variance of cluster/site means (must be positive); one of the residual variances in
"main" type tests treatment main effect; (2) The "site" type tests the the second level
• tau11= variance of treatment effects across sites (must be positive); one of the residual
variance of cluster/site means; and (3) The "variance" type tests variances in the second level
variance of treatment effects • sg2= level-one error variance; variance in the first level
• alpha= significance level
• power= statistical power
Example: Is there a difference in blood glucose levels between a treatment and • alternative= direction of the alternative hypothesis ("two.sided" or "less" or "greater")
control? • type= type of effect (“main”, “site”, or “variance”) with main as default. No tau00
• H0=0, H1≠0 needed for main effect; no tau11 needed for site effect; no tau or f needed for variance
effect
• You don’t have background info, so you guess that there is a medium effect
size Effect size calculation NOTE: here we show a 2
• For f-tests: arm example (treatment,
0.1=small, 0.25=medium, and 0.4 large effect sizes • 𝑓 = 𝜇𝐷 Τ 𝜎 2 control); to use a 3 arm
• Try a main effect, with a tau11 of 0.5 and a sg2 of 1.0 • 𝜇𝐷 =mean difference between design (treatment1,
• Alternative is “two.sided” as we only care about difference treatment2, control), use
• We can test for two sizes: number per cluster or cluster number treatment and control clusters wp.mrt3arm
1. Try for 100 clusters • 𝜎 2 =sample-specific variance
2. Try for 15 individuals per cluster to get cluster number
Multilevel Modeling: Multisite Randomized Trials
Results:

> #MRT sample size (number per cluster)

> wp.mrt2arm(f=0.25, J=100, tau11=0.5, sg2=1.0, alpha=0.05, power=0.80, alternative="two.sided")
Multisite randomized trials with 2 arms

J n f tau11 sg2 power alpha

100 14.24177 0.25 0.5 1 0.8 0.05
→ Round up to 15 individuals per cluster
NOTE: n is the number of subjects per cluster
URL: http://psychstat.org/mrt2arm
>
> #MRT sample size (cluster number)
> wp.mrt2arm(f=0.25, n=15, tau11=0.5, sg2=1.0, alpha=0.05, power=0.80, alternative="two.sided")
Multisite randomized trials with 2 arms

J n f tau11 sg2 power alpha

98.2174 15 0.25 0.5 1 0.8 0.05
→ Round up to 100 clusters (50 per arm)
NOTE: n is the number of subjects per cluster
URL: http://psychstat.org/mrt2arm
Multilevel Modeling: Cluster/Site Size
• While the webpower documentation says it can
be used for clusters or sites for 2+, it cannot be
used for small cluster number unless effect size (f)
is large enough and the inter correlation
coefficient (icc) is low enough
Multilevel Modeling: Practice
Calculate the sample size for the following scenarios (with α=0.05, and power=0.80):
1. You are interested in determining if a drug A could lower blood pressure for patients with
hypertension using 50 hospitals across the county, separated by cluster. From trial data, you
found blood pressure to be lowered by 6.90, with a between-cluster variance of 58 and a
within-cluster variance of 243.
2. You are interested in determining if a drug B changes blood pressure for patients with
hypertension using 6 hospitals the state, randomizing at each site. From trial data, you
found blood pressure to be different by 2.5, with a variance of treatment effect across site
of 2 and a person-specific variance of 1.
Multilevel Modeling: Answers
Calculate the sample size for the following scenarios (with α=0.05, and power=0.80):
1. You are interested in determining if a drug A could lower blood pressure for patients with hypertension using 50 hospitals
across the county, separated by cluster. From trial data, you found blood pressure to be lowered by 6.90, with a between-
cluster variance of 58 and a within-cluster variance of 243.
• Number of clusters (J) is 50
• One-tailed test -> “less”
• Effect size= 𝑓 = 𝜇𝐷 ൗ (𝜎𝐵2 +𝜎𝑊
2
) =6.90 / (58 + 243) = 0.40
• Inter-class correlation = 𝜌 = 𝜎𝐵2 Τ(𝜎𝐵2 +𝜎𝑊
2
) = 58/(58+243)=0.19
• wp.crt2arm(f=0.40, J=50, icc=0.19, alpha=0.05, power=0.80, alternative="less")
• n =16.45-> 17 samples per cluster
2. You are interested in determining if a drug B changes blood pressure for patients with hypertension using 6 hospitals the
state, randomizing at each site. From trial data, you found blood pressure to be different by 2.5, with a variance of
treatment effect across site of 2 and a person-specific variance of 1.
• Number of sites (J) is 6
• Two-tailed test -> “two.sided”
• effect size= 𝑓 = 𝜇𝐷 Τ 𝜎 2 =2.5/ 1 =2.5
• tau11= 2
• sg2= 1
• wp.mrt2arm(f=2.5, J=6, tau11=2, sg2=1, alpha=0.05, power=0.80, alternative="two.sided")
• n =3.86-> 4 samples per site
Generalized Linear Mixed Models
Description:
Combination of a Generalized Linear Model (GLM) and Mixed Model
• GLM: can be used with non-normal data
• Mixed Model: include both fixed and random effects
These models can be made very sophisticated and cover a very large range of
models
• Need to understand how to create model and define variables
• Therefore, it requires a Module of their own
• Look for the second sample size module in R: Sample Size Calculation
with R: GLMMs
 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).
 References
• In this module, many of the
functions that I show, I’ve General References:
refrained from including all the • https://www.statmethods.net/stats/power.html
options for simplicity
• https://www.graphpad.com/guides/prism/7/statistics
• More detailed descriptions (and /index.htm?stat_sample_size_for_nonparametric_.ht
m
sometimes examples) can be
found in the package manuals Packages:
• https://cran.r-project.org/web/packages/pwr/pwr.pdf
• https://cran.r-
project.org/web/packages/WebPower/WebPower.pdf
• https://webpower.psychstat.org/wiki/_media/grant/
webpower_manual_book.pdf