G*Power Analysis

G*Power analysis helps nursing researchers determine the sample size, statistical power, detectable effect size, or significance criterion required for a planned statistical test. For dissertation research, its most common use is an a priori power analysis, where you specify the expected effect size, alpha level, desired statistical power, and study-design parameters before recruitment begins.

G*Power is free statistical software developed for power analysis across many t, F, χ², z, and exact tests. It is widely used in nursing, health sciences, psychology, and other research disciplines because it allows researchers to translate a statistical analysis plan into a defensible sample-size calculation.

This guide is intentionally practical. It shows you how to use G*Power, which menu options to select, what values to enter, how to interpret the output, and how to report the calculation in Chapter 3 of a nursing dissertation.

If you first need to understand statistical power, Type I and Type II errors, alpha, beta, effect size, and the general logic of sample-size planning, read our Power Analysis for Nursing Research and Dissertations guide. This page focuses specifically on using the G*Power software correctly.

What Is G*Power Analysis?

GPower analysis is the process of using GPower software to calculate statistical power, required sample size, detectable effect size, or related quantities for a specified statistical test.

For example, a nursing researcher planning an independent-samples t test might tell G*Power that the study will use:

• a two-sided test;
• an expected effect size of d = 0.50;
• α = .05;
• statistical power of .80; and
• equal-sized intervention and control groups.

G*Power then calculates how many participants are needed to conduct the planned analysis with those assumptions.

The important point is that G*Power performs the calculation. It does not decide whether your statistical test, effect-size assumption, hypothesis direction, or study design is methodologically appropriate.

A perfectly executed G*Power calculation can therefore still produce a poorly justified sample size if the assumptions entered into the software are inappropriate.

Is G*Power Free?

Yes. G*Power is distributed free of charge by Heinrich-Heine-Universität Düsseldorf.

As of September 2026, the official site lists G*Power 3.1.9.7 for Windows and G*Power 3.1.9.6 for macOS. The developers also permit researchers to use screenshots of the software without requesting separate permission.

When downloading G*Power, use the official university source rather than an unofficial software-download website.

Official G*Power website and download page

What Can G*Power Calculate?

G*Power supports five broad types of power analysis.

Analysis Type	Information You Enter	Main Result	Typical Nursing Research Use
A priori	Effect size, alpha, desired power and design parameters	Required sample size	Planning recruitment before data collection
Sensitivity	Available sample size, alpha and desired power	Minimum detectable effect size	Existing datasets or fixed populations
Post hoc	Sample size, effect size and alpha	Calculated power	Available in the software but should be interpreted cautiously
Compromise	Sample size, effect size and relative weighting of Type I and II error	Alpha and beta/power	Specialized research situations
Criterion	Sample size, effect size and desired power	Required alpha	Uncommon in routine nursing dissertations

For most DNP, PhD, and master’s dissertation proposals, A priori: Compute required sample size is the relevant choice.

For a deeper methodological comparison of a priori, sensitivity, and post hoc approaches, see our main Power Analysis guide.

What You Need Before Opening G*Power

Do not start with G*Power and adjust settings until the sample size becomes convenient.

The software should be the last step in a chain of research decisions, not the first.

Before opening G*Power, determine the following:

1. Your research question

The research question should tell you whether you are comparing groups, examining an association, predicting an outcome, or evaluating change over time.

2. Your dependent variable

Identify whether the outcome is:

• continuous;
• binary;
• categorical;
• count-based;
• ordinal; or
• time-to-event.

This influences the statistical model you will eventually use.

3. Your independent variables

Know how many groups, predictors, conditions, or measurement occasions are involved.

4. Your study design

Examples include:

• independent groups;
• paired or matched observations;
• repeated measures;
• correlational designs;
• cross-sectional predictive designs; and
• experimental or quasi-experimental designs.

5. Your planned statistical test

The power analysis should correspond to the analysis actually specified in your data-analysis plan.

If your primary analysis is multiple linear regression, powering an independent t test simply because it produces a smaller sample does not justify the regression model.

6. Your effect-size assumption

The effect size should be defensible based on the scientific question and available evidence.

Possible sources include:

• meta-analyses;
• closely comparable studies;
• clinically meaningful differences;
• a prespecified smallest effect of scientific interest;
• pilot evidence; or
• conventional effect-size benchmarks when stronger evidence is unavailable.

7. Your significance level

α = .05 is common, but it should not be treated as an unchangeable rule.

8. Your desired statistical power

A power of .80 is commonly used. Some studies reasonably choose .90 or another value depending on the importance of avoiding a Type II error and applicable methodological guidance.

9. One-sided or two-sided testing

A one-sided test should not be chosen simply because it reduces the required sample size. It should follow from a genuinely directional hypothesis specified before examining the results.

10. Design-specific parameters

Depending on the analysis, G*Power may also require:

• number of groups;
• number of predictors;
• allocation ratio;
• number of measurements;
• correlation among repeated measurements;
• nonsphericity correction;
• degrees of freedom;
• null correlation;
• proportion parameters; or
• assumptions about additional predictors.

A defensible workflow is therefore:

Research question → study design → statistical test → effect-size rationale → alpha and power → G*Power

How to Do a G*Power Analysis Step by Step

The exact fields change depending on the statistical test, but the general workflow remains similar.

Step 1: Select the Test Family

The first major dropdown is Test family.

Common G*Power families include:

t tests

Used for analyses such as:

• independent-samples t tests;
• paired-samples t tests;
• one-sample t tests;
• point-biserial correlations; and
• selected regression-slope tests.

F tests

Used for many:

• ANOVA procedures;
• ANCOVA procedures;
• repeated-measures ANOVA procedures;
• multiple linear regression tests.

χ² tests

Used for procedures involving chi-square distributions, including contingency-table analyses.

Exact tests

Examples include:

• bivariate-normal correlation;
• tests of proportions;
• McNemar-type procedures;
• Fisher’s exact test;
• sign tests; and
• some other exact procedures.

z tests

G*Power places several procedures here, including:

• multiple logistic regression;
• Poisson regression;
• comparisons of Pearson correlations; and
• selected correlation models.

The official G*Power manual lists these procedures by test family.

Step 2: Select the Statistical Test

The second dropdown determines the exact analysis being powered.

For example, if your nursing dissertation uses multiple linear regression and the primary hypothesis concerns whether the complete set of predictors explains outcome variance, select:

Linear multiple regression: Fixed model, R² deviation from zero

If your question instead concerns whether a particular group of predictors explains additional variance beyond variables already entered, the appropriate option may be:

Linear multiple regression: Fixed model, R² increase

These are different hypotheses and can produce different required sample sizes.

Step 3: Choose the Type of Power Analysis

For prospective sample-size planning, select:

A priori: Compute required sample size – given α, power, and effect size

Use Sensitivity when your sample is already fixed and you want to determine the smallest effect the available sample can detect.

Post hoc power is available in G*Power, but it should not be used casually to explain why a completed study produced a nonsignificant result. We discuss this later in the guide.

Step 4: Enter the Effect Size

Effect-size fields vary by test.

Examples include:

• Cohen’s d for independent-group mean differences;
• dz for paired means;
• Cohen’s f for ANOVA;
• f² for multiple regression;
• r or ρ for correlation;
• Cohen’s w for chi-square tests.

These metrics are not interchangeable.

If G*Power asks for f² and your source reports R², you must convert the value rather than entering R² directly.

For detailed effect-size guidance, see our Cohen’s d and Effect Size guide.

Step 5: Enter Alpha

For many dissertation analyses:

α err prob = .05

is used.

However, your alpha may differ if your design includes multiple primary comparisons or another prespecified error-control strategy.

Step 6: Enter Desired Power

Typical planning values include:

Power (1 − β err prob) = .80

or

Power (1 − β err prob) = .90

Increasing the desired power normally increases the required sample size when other assumptions remain constant.

Step 7: Enter Design-Specific Values

This is where many incorrect G*Power calculations occur.

For example:

• an ANOVA needs the number of groups;
• regression needs the number of predictors;
• repeated-measures ANOVA needs the number of measurements and correlation among repeated measures;
• an independent t test may require an allocation ratio;
• correlation requires the alternative and null correlations.

Do not assume the default value is appropriate simply because G*Power automatically displays it.

Step 8: Select Calculate

G*Power produces the required sample size and additional statistical output.

Depending on the test, the output may include:

• noncentrality parameter;
• critical test statistic;
• numerator and denominator degrees of freedom;
• total sample size; and
• actual power.

Step 9: Convert Required Analyzable N Into a Recruitment Target

The G*Power result usually represents the number of analyzable observations required, not necessarily the number you must initially recruit.

For longitudinal or intervention research, anticipated dropout may require a larger enrollment target.

Use:

Recruitment target = Required analyzable sample ÷ (1 − anticipated attrition proportion)

For example, suppose G*Power indicates that 128 participants must complete the analysis and you reasonably anticipate 15% attrition:

128 ÷ (1 − .15)

= 128 ÷ .85

= 150.59

Round upward:

Recruitment target = 151 participants

If equal group allocation must be preserved, you may need to round further. For two equally sized groups, recruiting 152 participants would provide 76 participants per group.

For a one-time online survey, longitudinal attrition may not apply, but you may still need to account for nonresponse, incomplete questionnaires, exclusions, or unusable responses.

G*Power Worked Examples for Nursing Research

The following examples are intentionally presented as tutorial scenarios. Unless a study is specifically cited, effect sizes are illustrative assumptions rather than claims that the value has been established for that nursing intervention.

In an actual dissertation, replace an illustrative effect size with a defensible value or rationale appropriate to your study.

Example 1: Independent-Samples t Test

Nursing research question

Among adult surgical patients, does a nurse-led preoperative education intervention result in different postoperative pain scores compared with usual care?

Planned analysis

Independent-samples t test.

G*Power settings

• Test family: t tests
• Statistical test: Means: Difference between two independent means (two groups)
• Type of power analysis: A priori
• Tail(s): Two
• Effect size d: 0.50
• α err prob: 0.05
• Power: 0.80
• Allocation ratio N2/N1: 1

G*Power result

The required sample is:

128 participants total

or:

64 participants per group

Adjusting for 15% attrition

128 ÷ .85 = 150.59.

Because equal allocation is required, round to:

152 participants total

or:

76 per group

Example Chapter 3 wording

An a priori power analysis was conducted in G*Power 3.1 for an independent-samples t test. Assuming a two-sided test, an effect size of d = 0.50, α = .05, power = .80, and equal allocation between the intervention and comparison groups, the required analyzable sample was 128 participants, or 64 participants per group. After allowing for 15% anticipated attrition, the recruitment target was increased to 152 participants, or 76 participants per group. The effect-size assumption should be supported with the study, meta-analysis, clinical rationale, or other planning evidence used for the dissertation.

Example 2: Paired-Samples t Test

Research question

Do burnout scores among staff nurses change from before to after an eight-week mindfulness intervention?

Planned analysis

Paired-samples t test.

G*Power settings

• Test family: t tests
• Statistical test: Means: Difference between two dependent means (matched pairs)
• Type: A priori
• Tail(s): Two
• Effect size dz: 0.40
• α err prob: 0.05
• Power: 0.80

Required sample

G*Power requires approximately:

52 participants with complete paired measurements

This means 52 usable pairs, not 52 observations at each measurement considered independently.

Recruitment adjustment

Assuming 10% loss of complete paired data:

52 ÷ .90 = 57.78

Recruit at least:

58 participants

Important effect-size warning

The paired-test effect size dz differs from Cohen’s d for two independent groups.

Do not take an independent-groups d from a published article and enter it directly into the paired-samples dz field without confirming that the effect-size definition is compatible with the planned analysis.

Example 3: One-Way ANOVA

Research question

Do patient outcome scores differ among three nursing units using standard rounding, hourly rounding, and technology-assisted rounding?

G*Power settings

• Test family: F tests
• Statistical test: ANOVA: Fixed effects, omnibus, one-way
• Type: A priori
• Effect size f: 0.25
• α: 0.05
• Power: 0.80
• Number of groups: 3

Required sample

For three equally sized groups, use:

159 participants total

or:

53 participants per group

Why the effect-size field matters

One-way ANOVA uses Cohen’s f, not Cohen’s d.

If a previous study reports η², it can be converted using:

f = √[η² ÷ (1 − η²)]

For example, if η² = .0588:

f = √(.0588 ÷ .9412)

≈ .25

Do not enter η² directly into a field asking for Cohen’s f.

Example 4: Pearson Correlation

Research question

Is moral distress associated with intention-to-leave scores among registered nurses?

G*Power settings

For a Pearson product-moment correlation under the bivariate-normal model:

• Test family: Exact
• Statistical test: Correlation: Bivariate normal model
• Type: A priori
• Tail(s): Two
• Correlation ρ H1: .30
• Correlation ρ H0: 0
• α: .05
• Power: .80

The bivariate-normal correlation procedure belongs to G*Power’s exact-test family; point-biserial correlation is a different procedure found in the t-test family.

Required sample

For these settings:

N = 84

Chapter 3 example

An a priori power analysis for a two-sided Pearson correlation was conducted in G*Power 3.1 using the bivariate-normal correlation procedure. With an anticipated correlation of ρ = .30, a null correlation of 0, α = .05, and power = .80, the required analyzable sample was 84 participants.

The dissertation should additionally explain why an effect of approximately .30 was considered an appropriate planning assumption.

Example 5: Multiple Linear Regression

Research question

To what extent do age, comorbidity burden, and health-literacy scores predict a continuous 30-day readmission-risk score among adults with heart failure?

G*Power settings

• Test family: F tests
• Statistical test: Linear multiple regression: Fixed model, R² deviation from zero
• Type: A priori
• Effect size f²: .15
• α: .05
• Power: .80
• Number of predictors: 3

Required sample

N = 77

Converting R² to f²

If a comparable model reports:

R² = .13

then:

f² = R² ÷ (1 − R²)

f² = .13 ÷ .87

= .149

≈ .15

Do not enter .13 directly into a field requiring f².

Overall regression versus incremental prediction

The option:

R² deviation from zero

asks whether the complete predictor set explains outcome variance beyond zero.

The option:

R² increase

is appropriate when the research hypothesis concerns whether one or more tested predictors explain additional variance after another predictor set has already been included.

For example, a dissertation may ask whether health literacy adds predictive value after age and comorbidity burden have already been controlled.

That requires careful specification of:

• total number of predictors;
• number of predictors being tested; and
• expected incremental R².

Power the hypothesis you actually intend to test.

Chapter 3 example

An a priori power analysis for fixed-model multiple linear regression was conducted in G*Power 3.1. The analysis evaluated R² deviation from zero using three predictors, an assumed effect size of f² = .15, α = .05, and power = .80. The required analyzable sample was 77 participants. The selected effect size should be supported by prior evidence or another prespecified substantive rationale.

Example 6: Repeated-Measures ANOVA

Research question

Do anxiety scores change across baseline, immediately preoperative, and 24-hour postoperative measurements among patients receiving a guided-imagery nursing intervention?

Planned model

A single-group repeated-measures ANOVA examining change across three measurement occasions.

G*Power settings

• Test family: F tests
• Statistical test: ANOVA: Repeated measures, within factors
• Type: A priori
• Effect size f: .25
• α: .05
• Power: .80
• Number of groups: 1
• Number of measurements: 3
• Correlation among repeated measures: .50
• Nonsphericity correction ε: 1.00

Required sample

With these assumptions, the required sample is:

N = 28 participants

This corrects a common error in worked examples that report 24 participants for the same configuration.

Why repeated-measures calculations deserve extra care

The result depends not only on effect size but also on assumptions about:

• number of measurement occasions;
• within-person correlations; and
• sphericity.

Do not enter r = .50 simply because it appears in a tutorial.

If a prior study using the same or a closely related outcome instrument suggests a different correlation among repeated observations, use the better-supported assumption.

Similarly, ε = 1 assumes sphericity. If a more conservative value is justified, the power analysis should reflect that.

What if you have an intervention and control group?

If the primary hypothesis concerns whether change over time differs between groups, your target effect is usually the group × time interaction, not simply the within-subject time effect.

In that situation, select the G*Power procedure that corresponds to the within-between interaction rather than powering only the within-subject effect.

This distinction is crucial because the required sample can differ substantially.

Example 7: Chi-Square Test

Research question

Is receipt of nurse-led discharge coaching associated with 30-day readmission status?

Both variables are categorical:

• coaching: yes/no;
• readmission: yes/no.

G*Power settings

• Test family: χ² tests
• Statistical test: Goodness-of-fit tests: Contingency tables
• Type: A priori
• Effect size w: .30
• α: .05
• Power: .80
• Degrees of freedom: 1

For a 2 × 2 table:

df = (2 − 1)(2 − 1)

= 1

Required sample

For w = .30:

N = 88

Effect size

If previous research reports expected cell probabilities, use those values to derive Cohen’s w rather than automatically assuming a medium effect.

G*Power Test Selection Cheat Sheet

Research Goal	Planned Test	G*Power Family	G*Power Procedure	Main Effect-Size Input
Compare two independent means	Independent t test	t tests	Difference between two independent means	d
Compare paired/pre-post means	Paired t test	t tests	Difference between two dependent means	dz
Compare 3+ independent means	One-way ANOVA	F tests	Fixed effects, omnibus, one-way	f
Test within-person change over time	Repeated-measures ANOVA	F tests	Repeated measures, within factors	f
Test group differences in change over time	Repeated-measures interaction	F tests	Repeated measures, within-between interaction	f
Pearson correlation	Pearson r	Exact	Correlation: Bivariate normal model	ρ
Overall multiple regression	Multiple linear regression	F tests	R² deviation from zero	f²
Test added predictors	Hierarchical regression	F tests	R² increase	f²
Test categorical association	Chi-square	χ² tests	Contingency tables	w
Compare adjusted group means	ANCOVA	F tests	Fixed effects, special, main effects and interactions	f
Predict a binary outcome	Logistic regression	z tests	Logistic regression	Model-specific parameters/odds ratio

How to Choose an Effect Size for G*Power

Effect-size selection is one of the most important judgment calls in a power analysis.

It should answer:

What effect should this study be adequately capable of detecting?

There is no universal hierarchy that makes one source automatically correct in every study.

Useful approaches include the following.

Meta-analytic evidence

A well-matched meta-analysis may provide a useful estimate because it combines evidence across studies.

However, confirm that the included:

• populations;
• interventions;
• outcomes;
• follow-up periods; and
• statistical models

are sufficiently similar to your planned study.

Comparable empirical studies

A previous nursing study can provide an effect-size estimate when the research population, exposure or intervention, outcome, and design are reasonably similar.

Do not choose a paper solely because it reports a convenient effect.

Clinically meaningful differences

For clinical outcomes, the effect that matters may be the smallest difference considered important to patients or clinicians, rather than the average effect reported in one previous study.

Smallest effect of scientific interest

Researchers may prospectively specify the smallest effect that would meaningfully support the hypothesis.

A study can then be designed to reliably detect that effect.

Pilot data

Pilot estimates may be useful but are often imprecise because pilot samples are small.

Use them with appropriate caution.

Cohen’s conventional benchmarks

Conventional benchmarks such as small, medium, and large effects can be useful when stronger information is unavailable.

They should be presented honestly as planning conventions, not falsely described as effects established by previous research.

For more detail, read our Cohen’s d and Effect Size guide.

G*Power Sample Size and Recruitment Size Are Not the Same

Suppose G*Power indicates that your analysis requires 100 complete participants.

Your recruitment target depends on anticipated data loss.

Expected Data Loss	Required Complete N	Calculation	Initial Target
5%	100	100 ÷ .95	106
10%	100	100 ÷ .90	112
15%	100	100 ÷ .85	118
20%	100	100 ÷ .80	125
25%	100	100 ÷ .75	134

Always round upward.

The percentage itself should be justified when possible.

For longitudinal studies, use evidence about participant dropout.

For cross-sectional surveys, separately consider:

• expected response rate;
• incomplete responses;
• eligibility failures;
• unusable records; and
• missing data.

Do not label every type of data loss “attrition.”

A Priori Analysis in G*Power

For a dissertation being planned before data collection, select:

A priori: Compute required sample size

You normally enter:

• effect size;
• α;
• desired power; and
• design-specific parameters.

G*Power returns the required sample.

A priori analysis is frequently used to support sample-size justification in research proposals, protocols, funding applications, dissertation methodology chapters, and ethics submissions where a sample-size rationale is requested.

Do not state that a particular IRB or committee universally “requires G*Power.” Requirements differ by institution and design.

The methodological question is whether the proposed sample has been appropriately justified.

Sensitivity Analysis in G*Power

Sensitivity analysis is useful when N cannot be changed.

Examples include:

• an existing electronic-health-record dataset;
• a completed survey dataset;
• a fixed cohort;
• a finite eligible clinical population.

Suppose 180 usable records are available.

Rather than pretending an a priori analysis determined N = 180, use:

Sensitivity: Compute required effect size

and ask:

At N = 180, α = .05, and the desired power, what is the smallest effect this study can reliably detect?

This provides a transparent description of the statistical sensitivity of the fixed dataset.

Post Hoc Power: Use With Caution

G*Power can calculate achieved or post hoc power after data have been collected.

However, calculating post hoc power from the observed effect size is generally not a useful explanation for a nonsignificant result.

Hoenig and Heisey (2001) showed why observed post hoc power is closely tied to the observed test result and can be misleading when used to interpret completed studies.

For completed research, more informative quantities usually include:

• the estimated effect size;
• its confidence interval;
• the direction of the effect; and
• the precision of the estimate.

Read the broader discussion in our Power Analysis guide.

How to Interpret G*Power Output

After selecting Calculate, G*Power may display several values.

Total sample size

This is usually the most important result during an a priori analysis.

It represents the sample required under the assumptions you entered.

Actual power

Because participants cannot be fractional, the final integer sample size may produce power that is slightly higher than your requested value.

For example, you might request:

Power = .80

and G*Power may return:

Actual power = .802

or another nearby value.

This does not mean you performed a post hoc analysis.

“Actual power” is simply G*Power’s label for the power corresponding to the integer sample it selected.

You may report it for reproducibility, but the essential methodological information remains:

• target power;
• effect size;
• alpha;
• statistical test;
• important design parameters; and
• required sample.

Critical value

The critical t, F, χ², or other statistic indicates the rejection threshold used in the calculation.

It rarely needs to be reported in a routine dissertation sample-size paragraph.

Noncentrality parameter

G*Power uses noncentral distributions internally for many power calculations.

The noncentrality parameter helps define the relevant alternative distribution but is normally not required in Chapter 3.

Degrees of freedom

Degrees of freedom can be useful as a quality-control check.

For example, if a regression is meant to contain three predictors and the output does not reflect the intended model structure, revisit the input settings.

How to Report G*Power Analysis in Chapter 3

A good power-analysis paragraph allows another researcher or committee member to understand and reproduce your calculation.

At minimum, report:

1. analysis type;
2. statistical test;
3. software;
4. effect-size assumption;
5. rationale/source for that assumption;
6. alpha;
7. target power;
8. important design parameters;
9. required analyzable sample;
10. recruitment adjustment, if relevant.

Weak example

G*Power showed that 128 participants were required.

This does not explain what was calculated.

Improved t-test example

An a priori power analysis was conducted in G*Power 3.1 for a two-sided independent-samples t test. The calculation assumed an effect size of d = .50, α = .05, power = .80, and equal allocation between two groups. The required analyzable sample was 128 participants, or 64 participants per group. The effect-size assumption was based on [insert actual source or substantive rationale]. Allowing for 15% anticipated attrition increased the planned recruitment target to 152 participants, or 76 participants per group.

Improved regression example

An a priori power analysis was conducted in G*Power 3.1 for fixed-model multiple linear regression testing R² deviation from zero. The model included three predictors. Assuming f² = .15, α = .05, and power = .80, the required analyzable sample was 77 participants. The selected effect size was justified using [insert the actual evidence or substantive rationale].

Improved repeated-measures example

An a priori analysis was conducted in G*Power 3.1 for a repeated-measures ANOVA examining the within-subject effect across three measurement occasions. The calculation assumed f = .25, α = .05, power = .80, a correlation of .50 among repeated measurements, and ε = 1.00. The required analyzable sample was 28 participants. The assumed repeated-measures correlation and effect size were based on [insert supporting evidence].

Notice how the reporting template makes the assumptions explicit instead of presenting G*Power as an authority that independently decided the correct sample size.

Common G*Power Mistakes

1. Powering the wrong statistical test

If your dissertation uses regression, power the regression hypothesis rather than a simpler comparison involving the same variables.

2. Confusing d, dz, f, f², and w

Different G*Power procedures require different effect-size metrics.

Always check the label next to the input box.

3. Entering R² directly as f²

For regression:

f² = R² ÷ (1 − R²)

unless the procedure specifically requires a different form.

4. Choosing a large effect because it produces a manageable sample

This reverses the logic of power analysis.

Choose the effect the study should meaningfully be able to detect, then evaluate whether the resulting study is feasible.

5. Using “medium effect” without explanation

A conventional medium effect can be acceptable as a planning assumption when stronger evidence is unavailable, but say that explicitly.

6. Selecting a one-sided test to lower N

One-sided testing requires a substantive and prespecified directional hypothesis.

7. Confusing overall and incremental regression tests

R² deviation from zero and R² increase test different hypotheses.

8. Entering the wrong number of predictors

Use the variables actually represented by the tested model.

9. Confusing repeated-measures groups and measurement occasions

A design with:

• two treatment groups; and
• three measurement times

does not have six groups.

10. Ignoring repeated-measures correlation

The expected correlation among repeated observations can materially affect the required sample size.

11. Assuming ε = 1 without considering sphericity

A less-than-perfect nonsphericity correction may be appropriate for some designs.

12. Powering the within-subject effect when the hypothesis concerns group × time interaction

These are different effects.

13. Forgetting attrition or incomplete data

G*Power may calculate the required complete sample, not the number that must initially be approached or recruited.

14. Applying attrition language to a one-time survey

A cross-sectional study may face nonresponse or incomplete records rather than longitudinal dropout.

15. Using the study’s observed effect as though it were a prospective assumption

An a priori effect should be specified before the results of the study being planned are known.

16. Reporting only the final N

A sample-size number without its assumptions cannot be independently evaluated.

17. Treating G*Power defaults as research evidence

A value appearing automatically in the interface is not automatically appropriate for your study.

18. Forcing a complex design into a simpler G*Power procedure

When the final analysis is substantially more complex than G*Power can represent, use a method capable of modeling the real design.

When G*Power May Not Be Enough

G*Power is extremely useful, but it is not a universal sample-size solution.

More specialized methods may be required for:

Multilevel or hierarchical models

Examples include:

• patients nested within nurses;
• nurses nested within units;
• units nested within hospitals.

Clustering changes the effective information available in the sample.

Cluster-randomized trials

If entire clinics, wards, schools, or hospitals are randomized, sample-size planning must account for intracluster correlation and the number of clusters.

Complex longitudinal models

Mixed-effects growth models, irregular measurement schedules, and complicated covariance structures may require simulation or specialized formulas.

Survival analysis

Time-to-event outcomes often require methods based on:

• number of expected events;
• hazard ratio;
• follow-up time;
• censoring; and
• accrual pattern.

Structural equation modeling

SEM sample-size requirements depend on more than the number of observed variables.

Mediation and moderation

Simple approximations may be inadequate, particularly for indirect effects.

Simulation or methods developed specifically for mediation may be preferable.

Complex logistic regression

G*Power does provide a logistic regression procedure under the z-test family, but its setup is more complicated than simply entering an odds ratio.

Depending on the model, assumptions may concern:

• baseline event probability;
• predictor distribution;
• odds ratio;
• relationships between the predictor of interest and additional covariates.

The official G*Power materials describe logistic and Poisson regression as z-test procedures and show the additional model assumptions involved.

Complex survey designs

Sampling weights, stratification, clusters, and finite-population design effects may require dedicated survey-design methods.

Prediction-model development

A study developing a clinical prediction model should not automatically use a generic hypothesis-test calculation based solely on a conventional number of participants per predictor.

Modern prediction-model sample-size methods consider factors such as:

• number of candidate predictor parameters;
• outcome prevalence;
• anticipated model performance; and
• overfitting.

Precision-based sample-size planning

Sometimes the objective is not to achieve power for a null-hypothesis test but to estimate a mean, proportion, effect, or confidence interval with a specified level of precision.

That requires a different sample-size rationale.

G*Power vs IBM SPSS Statistics Power Analysis

G*Power is not the only software that can perform prospective power calculations.

Current IBM SPSS Statistics releases include Power Analysis procedures within Statistics Base for supported analyses such as one-sample, paired-sample and independent-samples t tests and one-way ANOVA.

Feature	G*Power	IBM SPSS Statistics
Cost	Free	Commercial software
Primary purpose	Dedicated statistical-power software	Full statistical-analysis package
t tests	Yes	Yes for supported power procedures
ANOVA	Yes	Supported power procedures available
Multiple regression power	Yes	Capabilities depend on procedure/version
Correlation procedures	Extensive G*Power options	Depends on SPSS power procedure
Exact-test power	Several supported procedures	Different implementation/capabilities
Installation	Stand-alone application	Installed as part of SPSS
Reproducibility	G*Power protocol and screenshots/settings	Syntax and statistical output
Best choice	Dedicated power-analysis workflow	Convenient when already using SPSS and the required procedure is supported

Neither program is automatically more valid.

The important questions are:

• Does the software support the hypothesis you are testing?
• Are your assumptions defensible?
• Can another researcher reproduce the calculation?

For help after data collection, see our SPSS Data Analysis Help service.

G*Power Analysis Checklist for Nursing Dissertation Students

Before submitting your methodology chapter, confirm that you can answer yes to each item:

• My power analysis matches my primary statistical hypothesis.
• I selected the correct G*Power test family.
• I selected the correct statistical procedure.
• I used the correct type of power analysis.
• My effect-size metric matches the selected test.
• I can explain where the effect-size assumption came from.
• I did not increase the effect size simply to reduce my required sample.
• My alpha level is appropriate for my analysis plan.
• My selected power is justified.
• My one-sided or two-sided choice matches my hypothesis.
• My number of groups is correct.
• My number of predictors is correct.
• My repeated-measures assumptions are justified where applicable.
• My G*Power result refers to the analyzable sample.
• I adjusted recruitment appropriately for expected data loss.
• My Chapter 3 explains all important G*Power inputs.
• Another researcher could reproduce my calculation.
• I have not forced a complex design into a G*Power procedure that does not represent the actual analysis.

Need Help With Your G*Power Analysis?

A G*Power calculation can take only a few minutes to run, but selecting the correct assumptions often requires much more methodological judgment.

If you need help with:

• choosing the appropriate statistical test;
• selecting the correct G*Power procedure;
• identifying a defensible effect size;
• converting reported effects into the metric required by G*Power;
• calculating your minimum sample size;
• adjusting for attrition or incomplete data;
• conducting sensitivity analysis;
• checking a calculation requested by your supervisor;
• reporting the analysis in Chapter 3; or
• aligning the power analysis with your dissertation hypotheses,

you can contact our nursing dissertation statistics team or order nursing dissertation statistical support.

Frequently Asked Questions About G*Power Analysis

What is G*Power analysis?

GPower analysis uses the GPower statistical program to calculate required sample size, statistical power, detectable effect size, or other power-related quantities for a specified research design and statistical test.

How do I do a power analysis in G*Power?

For a typical a priori analysis:

1. identify your statistical test;
2. select the corresponding Test family;
3. choose the exact Statistical test;
4. select A priori;
5. enter the effect size;
6. enter alpha;
7. enter desired power;
8. enter design-specific parameters;
9. select Calculate; and
10. adjust the resulting analyzable N for expected data loss when appropriate.

Does G*Power calculate sample size?

Yes. Select A priori: Compute required sample size, enter the relevant effect-size, alpha, power, and study-design inputs, and G*Power calculates the required analyzable sample for the selected test.

What effect size should I use in G*Power?

Use an effect size that represents the effect your study should reasonably be capable of detecting. The rationale may come from meta-analytic evidence, comparable studies, clinically meaningful differences, a smallest effect of scientific interest, pilot evidence, or conventional benchmarks when stronger evidence is unavailable.

Should I use .80 or .90 power?

A power of .80 is commonly used, while .90 may be appropriate when stronger protection against Type II error is warranted. The appropriate target depends on the research question, consequences of failing to detect an important effect, disciplinary expectations, feasibility, and design.

How do I use G*Power for multiple regression?

For an overall fixed-model regression hypothesis, select:

F tests → Linear multiple regression: Fixed model, R² deviation from zero

Enter f², alpha, power, and the total number of predictors.

If your hypothesis concerns the additional variance explained by particular predictors after controlling for others, use the R² increase procedure instead.

How do I calculate sample size for repeated-measures ANOVA?

Select the appropriate repeated-measures procedure and enter:

• effect size f;
• alpha;
• power;
• number of groups;
• number of measurement occasions;
• correlation among repeated measurements; and
• nonsphericity correction.

If the hypothesis concerns whether treatment groups change differently over time, make sure you power the within-between interaction, not only the within-subject time effect.

Is G*Power appropriate for logistic regression?

G*Power includes a multiple logistic regression procedure under the z-test family, but the analysis can require several model assumptions beyond an odds ratio. Complex models may be better powered using specialized methods or simulation.

What is an a priori G*Power analysis?

An a priori analysis estimates the sample size required before data collection based on a prespecified effect size, alpha, desired power, and study-design parameters.

How should I report G*Power in Chapter 3?

Report the:

• analysis type;
• statistical test;
• G*Power version;
• effect size;
• rationale for the effect size;
• alpha;
• target power;
• relevant design inputs;
• required analyzable sample; and
• recruitment adjustment where relevant.

Can G*Power be used for every nursing research study?

No. Multilevel models, cluster-randomized trials, complex longitudinal models, survival analyses, SEM, prediction-model development, and some mediation or logistic-regression designs may require specialized methods.

References

Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Routledge. https://doi.org/10.4324/9780203771587

Faul, F., Erdfelder, E., Lang, A.-G., & Buchner, A. (2007). G*Power 3: A flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behavior Research Methods, 39(2), 175–191. https://doi.org/10.3758/BF03193146

Faul, F., Erdfelder, E., Buchner, A., & Lang, A.-G. (2009). Statistical power analyses using G*Power 3.1: Tests for correlation and regression analyses. Behavior Research Methods, 41(4), 1149–1160. https://doi.org/10.3758/BRM.41.4.1149

Hoenig, J. M., & Heisey, D. M. (2001). The abuse of power: The pervasive fallacy of power calculations for data analysis. The American Statistician, 55(1), 19–24. https://doi.org/10.1198/000313001300339897

Kang, H. (2021). Sample size determination and power analysis using the G*Power software. Journal of Educational Evaluation for Health Professions, 18, 17. https://pmc.ncbi.nlm.nih.gov/articles/PMC8441096/