Analysis Modes

ggpower supports five analysis modes. Each mode solves for a different unknown given the others.

Mode	Solves for	When to use
`a_priori`	Sample size	Planning before data collection
`post_hoc`	Power	Fixed sample size, retrospective
`criterion`	Alpha	Choose significance level
`sensitivity`	Effect size	Minimum detectable effect
`compromise`	Alpha and beta	Balance \(\alpha\) and \(\beta\) via ratio \(q = \beta/\alpha\)

Restrictions: t_generic has no a_priori. simon_two_stage supports only post_hoc and sensitivity.

A priori — sample size

power_compute("t_two_sample", "a_priori", d = 0.5, alpha = 0.05,
              power = 0.8, tails = "two", allocation_ratio = 1)
#> ggpower result
#> Test: t test: Means - difference between two independent means (two groups)
#> Analysis: a_priori
#> 
#> Input parameters
#>   tails: two
#>   effect_size_d: 0.5
#>   alpha: 0.05
#>   sample_size_group_1: 64
#>   sample_size_group_2: 64
#>   target_power: 0.8
#> 
#> 
#> Output parameters
#>   noncentrality_parameter: 2.828427
#>   critical_t: -1.978971,  1.978971
#>   df: 126
#>   total_sample_size: 128
#>   actual_power: 0.8014596
#> 
#> 
#> Notes
#> - A priori sample sizes are rounded up to integer values and actual power is recomputed.

Post hoc — achieved power

power_compute("t_one_sample", "post_hoc", d = 0.625, n = 30,
              alpha = 0.05, tails = "one")
#> ggpower result
#> Test: t test: Means - difference from constant (one sample case)
#> Analysis: post_hoc
#> 
#> Input parameters
#>   tails: greater
#>   effect_size_d: 0.625
#>   alpha: 0.05
#>   total_sample_size: 30
#> 
#> 
#> Output parameters
#>   noncentrality_parameter: 3.423266
#>   critical_t: 1.699127
#>   df: 29
#>   power: 0.9551444

Criterion — alpha

power_compute("t_one_sample", "criterion", d = 0.5, n = 40,
              power = 0.8, tails = "two")
#> ggpower result
#> Test: t test: Means - difference from constant (one sample case)
#> Analysis: criterion
#> 
#> Input parameters
#>   tails: two
#>   effect_size_d: 0.5
#>   alpha: 0.02642633
#>   total_sample_size: 40
#>   target_power: 0.8
#> 
#> 
#> Output parameters
#>   noncentrality_parameter: 3.162278
#>   critical_t: -2.307422,  2.307422
#>   df: 39
#>   power: 0.8
#>   alpha: 0.02642633
#>   beta: 0.2

Sensitivity — effect size

power_compute("f_mreg_omnibus", "sensitivity", alpha = 0.05, power = 0.8,
              total_n = 100, predictors = 3)
#> ggpower result
#> Test: F test: Multiple Regression - omnibus (deviation of R2 from zero), fixed model
#> Analysis: sensitivity
#> 
#> Input parameters
#>   effect_size_f2: 0.1135624
#>   alpha: 0.05
#>   total_sample_size: 100
#>   predictors: 3
#>   target_power: 0.8
#> 
#> 
#> Output parameters
#>   noncentrality_parameter: 11.35624
#>   critical_f: 2.699393
#>   numerator_df: 3
#>   denominator_df: 96
#>   power: 0.8
#>   f2: 0.1135624

Compromise — alpha and beta ratio

power_compute("t_one_sample", "compromise", d = 0.5, n = 40, q = 1, tails = "two")
#> ggpower result
#> Test: t test: Means - difference from constant (one sample case)
#> Analysis: compromise
#> 
#> Input parameters
#>   tails: two
#>   effect_size_d: 0.5
#>   alpha: 0.0844535
#>   total_sample_size: 40
#> 
#> 
#> Output parameters
#>   noncentrality_parameter: 3.162278
#>   critical_t: -1.770542,  1.770542
#>   df: 39
#>   power: 0.9155465
#>   alpha: 0.0844535
#>   beta: 0.08445349
#>   beta_alpha_ratio: 1
#> 
#> 
#> Notes
#> - Compromise analysis solves alpha so beta / alpha matches the requested ratio as closely as possible.

Effect size conversions

Helper functions convert study parameters into effect sizes used by power_compute().

effect_size_d(mean_h1 = 15, mean_h0 = 10, sd = 8)
#> [1] 0.625

effect_size_f2(r2 = 0.1)
#> [1] 0.1111111

effect_size_w(p0 = c(0.25, 0.25, 0.25, 0.25), p1 = c(0.4, 0.3, 0.2, 0.1))
#> [1] 0.4472136

See the pkgdown site for the full effect size conversions article.

Calculator

The Calculator module evaluates distribution-function scripts via ggpower_calculator().

ggpower_calculator("zinv(0.975)")
#> [1] 1.959964

See the pkgdown site for the full calculator article.