Superpower (version 0.2.4)

emmeans_power: Compute power for emmeans contrasts

Description

Computes power based on t value and degrees of freedom for contrasts. Do not use to calculate "observed power" for empirical datasets (Hoenig & Heisey, 2001).

Usage

emmeans_power(x, ...)
# S3 method for emmGrid
emmeans_power(x, ...)
# S3 method for summary_em
emmeans_power(x, ...)
# S3 method for data.frame
emmeans_power(
  x,
  alpha_level = Superpower_options("alpha_level"),
  liberal_lambda = Superpower_options("liberal_lambda"),
  ...
)

Value

Returns dataframe with simulation data (power and effect sizes!), anova results and simple effect results, plot of exact data, and alpha_level. Note: Cohen's f = sqrt(pes/1-pes) and the noncentrality parameter is = f^2*df(error)

"dataframe"

A dataframe of the simulation result.

"aov_result"

aov object returned from aov_car.

"main_result"

The power analysis results for ANOVA level effects.

"pc_results"

The power analysis results for the pairwise (t-test) comparisons.

"emm_results"

The power analysis results of the pairwise comparison results.

"manova_results"

Default is "NULL". If a within-subjects factor is included, then the power of the multivariate (i.e. MANOVA) analyses will be provided.

"alpha_level"

The alpha level, significance cut-off, used for the power analysis.

"method"

Record of the function used to produce the simulation

"plot"

A plot of the dataframe from the simulation; should closely match the meansplot in ANOVA_design

Arguments

x

emmGrid-class. Grid of contrasts to estimate power from.

...

Other arguments passed to the function if object is not already a emmGrid object.

alpha_level

Alpha level used to determine statistical significance

liberal_lambda

Logical indicator of whether to use the liberal (cohen_f^2\*(num_df+den_df)) or conservative (cohen_f^2\*den_df) calculation of the noncentrality (lambda) parameter estimate. Default is FALSE.

Author

Frederik Aust

Details

Note that calculation of power is based on the F- and t-ratio assuming two-sided testing. Thus, the function does not honor adjustments of the testing procedure due to either one-sided testing (or two-one sided tests) or corrections for multiple comparisons via the p.adjust option in emmeans.

Power for one-sided tests can be calculated, if the means of the simulated dataset are consistent with the directional hypothesis, by doubling alpha_level. Similarly, power for Bonferroni-corrected contrasts can be calculated by adjusting alpha_level accordingly (see examples). ... Other arguments passed onto the function

References

Hoenig, J. M., & Heisey, D. M. (2001). The Abuse of Power. The American Statistician, 55(1), 19–24. https://doi.org/10.1198/000313001300339897

Examples

if (FALSE) {
# Set up a within design with 2 factors, each with 2 levels
design_result <- ANOVA_design(design = "2w*2w",
n = 40, mu = c(1, 0, 1, 0),
sd = 2, r = 0.8,
labelnames = c("condition", "cheerful",
 "sad", "voice", "human", "robot"))
 
exact_result <- ANOVA_exact(design_result, 
alpha_level = 0.05, verbose = FALSE, 
emm = TRUE, contrast_type = "pairwise")

# Power for pairwise contrasts
exact_result$emm_results

# Corresponding emmeans contrasts
exact_result$emmeans$contrasts

# Manually recalculate power
emmeans_power(exact_result$emmeans$contrasts, 
alpha_level = 0.05)

# Calculate power for Bonferroni-adjusted pairwise comparisons
n_contrasts <- nrow(as.data.frame(exact_result$emmeans$contrasts))
emmeans_power(exact_result$emmeans$contrasts, 
alpha_level = 0.05 / n_contrasts)

# Calculate power for one-sided custom contrasts
exact_result$emmeans$emmeans
custom_contrast <- contrast(exact_result$emmeans$emmeans, 
list(robot_vs_sad_human = c(0, 1, -0.5, -0.5)))
emmeans_power(custom_contrast,
 alpha_level = 0.05 * 2)

# Calculate power for follow-up ANOVA
follow_up <- joint_tests(exact_result$emmeans$emmeans, 
by = "condition")
emmeans_power(follow_up, 
alpha_level = 0.05 / 2)
emmeans_power(emmeans(exact_result$emmeans$emmeans, 
pairwise ~ voice | condition)$contrasts, 
alpha_level = 0.05 / 2)
}