Power analysis two proportions

power.R

library(pwr)
library(dplyr)
library(ggplot2)

compute_sample_size <- function(
    p1, p2, power = 0.8, sig.level = 0.05, alternative = 'two.sided'
  ) {
  h <- ES.h(p1, p2) # Cohen's h
  result <- pwr.2p.test(
    h = h, power = power,
    sig.level = sig.level, alternative = alternative
  )
  ceiling(result$n)
}

compute_sample_size <- Vectorize(compute_sample_size)

get_df <- function(baselines, effect_sizes, alternative = 'two.sided') {
  df <- expand.grid('baseline' = baselines, 'effect_size' = effect_sizes)
  # floating point comparison issue below otherwise
  df$baseline <- round(df$baseline, 3)
  df$effect_size <- round(df$effect_size, 3)
  
  df$n <- compute_sample_size(
    df$baseline, df$baseline + df$effect_size, alternative = alternative
  )
  df$type <- ifelse(alternative == 'two.sided', alternative, 'one.sided')
  df
}

baselines <- seq(0.10, 0.45, 0.05)
effect_sizes <- seq(0.05, 0.15, 0.01)

df <- rbind(
  get_df(baselines, effect_sizes),
  get_df(baselines, effect_sizes, alternative = 'less')
)


p <- ggplot(df, aes(x = effect_size * 100, y = n, color = factor(baseline))) +
  geom_line(size = 1.2) +
  scale_color_brewer(name = 'Baseline (unclustered)', palette = 'Set2') +
  scale_x_continuous(breaks = seq(5, 15, 1)) +
  scale_y_continuous(breaks = seq(0, 1600, 200), limits = c(0, 1600)) +
  facet_wrap(~ type) +
  labs(
    title = 'Required Sample Size per Group vs. Effect Size',
    x = 'Effect Size (Percentage Point Increase)',
    y = 'Sample Size per Group'
  ) +
  theme_minimal() +
  theme(
    legend.position = 'top',
    plot.title = element_text(size = 14, hjust = 0.50)
  )

p

tab <- df %>% 
  filter(
    baseline %in% c(0.15, 0.30, 0.45),
    effect_size %in% c(0.05, 0.10, 0.15)
  )

tab

tab2 <- df %>% 
  filter(
    baseline %in% c(0.15, 0.30, 0.45),
    effect_size %in% c(0.05, 0.06, 0.07, 0.08, 0.09, 0.10)
  )

tab2