Choosing a Power Analysis
Source:vignettes/choosing-a-power-analysis.Rmd
choosing-a-power-analysis.Rmdggpower organizes power analysis around the question being asked.
- Use a priori before data collection when sample size is unknown.
- Use post hoc when sample size is fixed and achieved power is needed.
- Use criterion when alpha should be selected to reach a target power.
- Use sensitivity when the smallest detectable effect size is the target.
- Use compromise when alpha and beta should be balanced with a chosen beta/alpha ratio.
Example: Planning a One-Sample Mean Test
Suppose a clinical scale has a baseline mean of 10, the expected mean is 15, and the standard deviation is 8. The effect size is:
d <- effect_size_d(mean_h1 = 15, mean_h0 = 10, sd = 8)
d
#> [1] 0.625For a one-tailed test with alpha = 0.05 and target power = 0.95:
power_compute("t_one_sample", "a_priori", d = d, alpha = 0.05,
power = 0.95, tails = "one")
#> ggpower result
#> Test: t test: Means - difference from constant (one sample case)
#> Analysis: a_priori
#>
#> Input parameters
#> tails: greater
#> effect_size_d: 0.625
#> alpha: 0.05
#> total_sample_size: 30
#> target_power: 0.95
#>
#>
#> Output parameters
#> noncentrality_parameter: 3.423266
#> critical_t: 1.699127
#> df: 29
#> actual_power: 0.9551444
#>
#>
#> Notes
#> - A priori sample sizes are rounded up to integer values and actual power is recomputed.Choosing tests
Use ggpower_tests() to inspect supported families,
domains, and modules. The package registers 48 tests
across workspace, biomarker, and clinical workflows.
ggpower_tests()[, c("id", "family", "domain", "module")]
#> id family
#> t_one_sample t_one_sample t tests
#> t_paired t_paired t tests
#> t_two_sample t_two_sample t tests
#> t_point_biserial t_point_biserial t tests
#> t_linear_regression t_linear_regression t tests
#> t_linear_regression_two_groups t_linear_regression_two_groups t tests
#> t_generic t_generic t tests
#> f_anova_one_way f_anova_one_way F tests
#> f_anova_special f_anova_special F tests
#> f_mreg_omnibus f_mreg_omnibus F tests
#> f_mreg_increase f_mreg_increase F tests
#> f_variance_two f_variance_two F tests
#> chisq_variance_one chisq_variance_one chi-square tests
#> chisq_gof chisq_gof chi-square tests
#> chisq_contingency chisq_contingency chi-square tests
#> exact_binomial exact_binomial Exact
#> exact_one_proportion exact_one_proportion Exact
#> exact_sign exact_sign Exact
#> exact_fisher exact_fisher Exact
#> exact_mcnemar exact_mcnemar Exact
#> exact_correlation exact_correlation Exact
#> exact_mreg_random exact_mreg_random Exact
#> z_corr_independent z_corr_independent z tests
#> z_corr_dependent_common z_corr_dependent_common z tests
#> z_corr_dependent_no_common z_corr_dependent_no_common z tests
#> z_logistic z_logistic z tests
#> z_poisson z_poisson z tests
#> z_tetrachoric z_tetrachoric z tests
#> wilcoxon_signed wilcoxon_signed nonparametric
#> wilcoxon_mann_whitney wilcoxon_mann_whitney nonparametric
#> roc_auc_one roc_auc_one biomarker
#> roc_auc_two roc_auc_two biomarker
#> diagnostic_acc diagnostic_acc biomarker
#> survival_logrank survival_logrank biomarker
#> cox_regression cox_regression biomarker
#> discovery_fdr discovery_fdr biomarker
#> ttest_biomarker ttest_biomarker biomarker
#> rct_superiority_continuous rct_superiority_continuous clinical
#> rct_superiority_binary rct_superiority_binary clinical
#> rct_noninferiority_continuous rct_noninferiority_continuous clinical
#> rct_noninferiority_binary rct_noninferiority_binary clinical
#> rct_equivalence_continuous rct_equivalence_continuous clinical
#> rct_equivalence_proportion rct_equivalence_proportion clinical
#> simon_two_stage simon_two_stage clinical
#> cluster_rct cluster_rct clinical
#> multi_arm_superiority multi_arm_superiority clinical
#> count_endpoint_poisson count_endpoint_poisson clinical
#> survival_pmu survival_pmu clinical
#> domain module
#> t_one_sample general workspace
#> t_paired general workspace
#> t_two_sample general workspace
#> t_point_biserial general workspace
#> t_linear_regression general workspace
#> t_linear_regression_two_groups general workspace
#> t_generic general workspace
#> f_anova_one_way general workspace
#> f_anova_special general workspace
#> f_mreg_omnibus general workspace
#> f_mreg_increase general workspace
#> f_variance_two general workspace
#> chisq_variance_one general workspace
#> chisq_gof general workspace
#> chisq_contingency general workspace
#> exact_binomial general workspace
#> exact_one_proportion general workspace
#> exact_sign general workspace
#> exact_fisher general workspace
#> exact_mcnemar general workspace
#> exact_correlation general workspace
#> exact_mreg_random general workspace
#> z_corr_independent general workspace
#> z_corr_dependent_common general workspace
#> z_corr_dependent_no_common general workspace
#> z_logistic general workspace
#> z_poisson general workspace
#> z_tetrachoric general workspace
#> wilcoxon_signed general workspace
#> wilcoxon_mann_whitney general workspace
#> roc_auc_one biomarker biomarker
#> roc_auc_two biomarker biomarker
#> diagnostic_acc biomarker biomarker
#> survival_logrank biomarker biomarker
#> cox_regression biomarker biomarker
#> discovery_fdr biomarker biomarker
#> ttest_biomarker biomarker biomarker
#> rct_superiority_continuous pharma clinical
#> rct_superiority_binary pharma clinical
#> rct_noninferiority_continuous pharma clinical
#> rct_noninferiority_binary pharma clinical
#> rct_equivalence_continuous pharma clinical
#> rct_equivalence_proportion pharma clinical
#> simon_two_stage pharma clinical
#> cluster_rct pharma clinical
#> multi_arm_superiority pharma clinical
#> count_endpoint_poisson pharma clinical
#> survival_pmu pharma clinicalSidebar modules
Not every research question belongs in the same module. Use this guide to pick the sidebar entry that matches your endpoint and study design.
#> Question Module
#> 1 What sample size for a standard t test or ANOVA? Power Workspace
#> 2 Can my biomarker discriminate cases from controls (AUC)? Biomarker Discovery
#> 3 Is my classifier sensitive and specific enough? Biomarker Discovery
#> 4 Does a biomarker predict survival? Biomarker Discovery
#> 5 How many patients for a Phase III superiority trial? Clinical Trials
#> 6 Is treatment non-inferior to standard of care? Clinical Trials
#> 7 Is a new formulation equivalent (bioequivalence)? Clinical Trials
#> 8 Oncology single-arm Phase II with early stopping? Clinical Trials
#> Example_test
#> 1 t_two_sample
#> 2 roc_auc_one
#> 3 diagnostic_acc
#> 4 cox_regression
#> 5 rct_superiority_continuous
#> 6 rct_noninferiority_binary
#> 7 rct_equivalence_continuous
#> 8 simon_two_stage- Power Workspace — classical test families (t, F, chi-square, exact, z, nonparametric)
- Biomarker Discovery — ROC, diagnostic, survival, FDR
- Clinical Trials — superiority, NI, equivalence, Simon, cluster RCT
ggpower_tests(module = "biomarker")[, c("id", "label")]
#> id
#> roc_auc_one roc_auc_one
#> roc_auc_two roc_auc_two
#> diagnostic_acc diagnostic_acc
#> survival_logrank survival_logrank
#> cox_regression cox_regression
#> discovery_fdr discovery_fdr
#> ttest_biomarker ttest_biomarker
#> label
#> roc_auc_one ROC AUC: One sample vs null AUC
#> roc_auc_two ROC AUC: Compare two independent AUCs
#> diagnostic_acc Diagnostic accuracy: Sensitivity and specificity
#> survival_logrank Survival: Log-rank test
#> cox_regression Survival: Cox PH single covariate
#> discovery_fdr Discovery: Multiplicity-adjusted FDR screening
#> ttest_biomarker Differential expression: Two-group t test
ggpower_tests(module = "clinical")[, c("id", "label")]
#> id
#> rct_superiority_continuous rct_superiority_continuous
#> rct_superiority_binary rct_superiority_binary
#> rct_noninferiority_continuous rct_noninferiority_continuous
#> rct_noninferiority_binary rct_noninferiority_binary
#> rct_equivalence_continuous rct_equivalence_continuous
#> rct_equivalence_proportion rct_equivalence_proportion
#> simon_two_stage simon_two_stage
#> cluster_rct cluster_rct
#> multi_arm_superiority multi_arm_superiority
#> count_endpoint_poisson count_endpoint_poisson
#> survival_pmu survival_pmu
#> label
#> rct_superiority_continuous RCT superiority: Continuous endpoint
#> rct_superiority_binary RCT superiority: Binary endpoint
#> rct_noninferiority_continuous Non-inferiority: Continuous endpoint
#> rct_noninferiority_binary Non-inferiority: Binary endpoint
#> rct_equivalence_continuous Equivalence: Continuous endpoint (TOST)
#> rct_equivalence_proportion Equivalence: Binary endpoint (TOST)
#> simon_two_stage Simon two-stage Phase II design
#> cluster_rct Cluster-randomized trial
#> multi_arm_superiority Multi-arm superiority (ANOVA)
#> count_endpoint_poisson Count endpoint: Poisson regression
#> survival_pmu Survival endpoint: Log-rank primary analysisStay in Power Workspace when you need specialized classical tests (McNemar, tetrachoric correlation, Wilcoxon, etc.) even if the endpoint sounds clinical.