tests/testthat/test-empirical-probabilities.R

In randomizr: Easy-to-Use Tools for Common Forms of Random Assignment and Sampling

library(randomizr)

# Test against empiricial probabilities -----------------------------------

blocks <- c("A", "A", "A", "B", "B", "C")
prob_each <- c(.1, .2, .7)

analytic_probs <-
  rbind(
    c(0 / 2 * 2 / 3 + 1 / 3 * 1 / 3,
      0 / 2 * 2 / 3 + 1 / 3 * 1 / 3,
      2 / 2 * 2 / 3 + 1 / 3 * 1 / 3),
    c(0 / 1 * 1 / 2 + 1 / 3 * 1 / 2,
      0 / 1 * 1 / 2 + 1 / 3 * 1 / 2,
      1 / 1 * 1 / 2 + 1 / 3 * 1 / 2),
    c(0 / 1 * 0 / 1 + 1 / 3 * 1 / 1,
      0 / 1 * 0 / 1 + 1 / 3 * 1 / 1,
      0 / 1 * 0 / 1 + 1 / 3 * 1 / 1)
  )


# construct analytic probs from randomizr arguments
num_arms <- length(prob_each)
N_per_block <- table(blocks)

for_sure_allocations <- floor(N_per_block %*% t(prob_each))
possible_allocations <-
  matrix(1 / num_arms, nrow = length(N_per_block), ncol = num_arms)

for_sure_probs <- sweep(for_sure_allocations, 1, N_per_block, `/`)
possible_probs <-
  sweep(possible_allocations,
        1,
        (N_per_block - rowSums(for_sure_allocations)) / N_per_block,
        `*`)
final_probs <- for_sure_probs + possible_probs

#
# # Conduct Simulation
# perm_mat <- replicate(1000, block_ra(blocks = blocks, prob_each = prob_each))
#
# prop.table(table(perm_mat))
# prob_T1 <- apply(perm_mat, 1, function(x)mean(x=="T1"))
# prob_T2 <- apply(perm_mat, 1, function(x)mean(x=="T2"))
# prob_T3 <- apply(perm_mat, 1, function(x)mean(x=="T3"))
#
# simulated_probs <-
# cbind(
#   tapply(prob_T1, blocks, mean),
#   tapply(prob_T2, blocks, mean),
#   tapply(prob_T3, blocks, mean)
# )
#
#
# simulated_probs


dec <- declare_ra(blocks = blocks, prob_each = prob_each)
block_ra_probabilities(blocks = blocks, prob_each = prob_each)

dec$probabilities_matrix

analytic_probs
final_probs



# figure out prob_each ----------------------------------------------------



prob <- 0.1
N <- 11
m_floor <- floor(N * prob)
m_ceiling <- ceiling(N * prob)

m_floor * (1 - prob) + m_ceiling * (prob)
(prob * N)

prob_each <- c(.1, .2, .7)
m_each <- floor(N * prob_each)

m_each / (sum(m_each))



# complete_ra -------------------------------------------------------------

complete_ra(5, prob = .1)
complete_ra_probabilities(5, prob = .1)

Z_mat <- replicate(n = 100, complete_ra(5, prob = .1))
mean(Z_mat)

ceiling(5 * .1)
floor(5 * .1)