Package {ri2}

ri2: Randomization Inference

Description

Tools for conducting randomization inference for experiments.

Author(s)

Maintainer: Alexander Coppock acoppock@gmail.com

Authors:

• Alexander Coppock acoppock@gmail.com

See Also

Useful links:

• https://alexandercoppock.com/ri2/

• https://github.com/acoppock/ri2

• Report bugs at https://github.com/acoppock/ri2/issues

Conduct Randomization Inference

Description

This function makes it easy to conduct three kinds of randomization inference.

Usage

conduct_ri(
  formula = NULL,
  model_1 = NULL,
  model_2 = NULL,
  test_function = NULL,
  assignment = "Z",
  outcome = NULL,
  declaration = NULL,
  sharp_hypothesis = 0,
  potential_outcomes = NULL,
  studentize = FALSE,
  IPW = TRUE,
  sampling_weights = NULL,
  clusters = NULL,
  permutation_matrix = NULL,
  data,
  sims = 1000,
  progress_bar = FALSE,
  p = "two-tailed"
)

Arguments

Details

1. Conduct hypothesis tests under the sharp null when the test statistic is the difference-in-means or covariate-adjusted average treatment effect estimate. 2. Conduct "ANOVA" style hypothesis tests, where the f-statistic from two nested models is the test statistic. This procedure is especially helpful when testing interaction terms under null of constant effects. 3. Arbitrary (scalar) test statistics

Examples

# Data from Gerber and Green Table 2.2

table_2.2 <-
    data.frame(d = c(1, 0, 0, 0, 0, 0, 1),
               y = c(15, 15, 20, 20, 10, 15, 30))

## Declare randomization procedure
declaration <- randomizr::declare_ra(N = 7, m = 2)

## Conduct Randomization Inference
out <- conduct_ri(y ~ d,
                  declaration = declaration,
                  assignment = "d",
                  sharp_hypothesis = 0,
                  data = table_2.2)

summary(out)
plot(out)
tidy(out)

# Using a custom permutation matrix

permutation_matrix <-
 matrix(c(0, 0, 0, 0, 0, 0, 1,
          0, 0, 0, 0, 0, 1, 0,
          0, 0, 0, 0, 1, 0, 0,
          0, 0, 0, 1, 0, 0, 0,
          0, 0, 1, 0, 0, 0, 0,
          0, 1, 0, 0, 0, 0, 0,
          1, 0, 0, 0, 0, 0, 0),
        ncol = 7)

conduct_ri(y ~ d, assignment = "d", data = table_2.2,
           permutation_matrix = permutation_matrix)


# Randomization Inference for an Interaction

N <- 100
declaration <- randomizr::declare_ra(N = N, m = 50)

Z <- randomizr::conduct_ra(declaration)
X <- rnorm(N)
Y <- .9 * X + .2 * Z + 1 * X * Z + rnorm(N)
dat <- data.frame(Y, X, Z)

ate_obs <- coef(lm(Y ~ Z, data = dat))[[2]]

out <-
  conduct_ri(
    model_1 = Y ~ Z + X,
    model_2 = Y ~ Z + X + Z * X,
    declaration = declaration,
    assignment = "Z",
    sharp_hypothesis = ate_obs,
    data = dat, sims = 100
  )

plot(out)
summary(out)

# Randomization Inference for arbitrary test statistics

N <- 100
declaration <- randomizr::declare_ra(N = N, m = 50)

Z <- randomizr::conduct_ra(declaration)
X <- rnorm(N)
Y <- .9 * X + .2 * Z + rnorm(N)
dat <- data.frame(Y, X, Z)

balance_fun <- function(data) {
    f_stat <- summary(lm(Z ~ X, data = data))$f[1]
    names(f_stat) <- NULL
    return(f_stat)
}

out <-
  conduct_ri(
    test_function = balance_fun,
    declaration = declaration,
    assignment = "Z",
    sharp_hypothesis = 0,
    data = dat, sims = 100
  )

plot(out)
summary(out)
tidy(out)

Objects exported from other packages

Description

These objects are imported from other packages. Follow the links below to see their documentation.

generics

tidy()

randomizr

block_and_cluster_ra(), block_ra(), cluster_ra(), complete_ra(), conduct_ra(), declare_ra(), obtain_condition_probabilities(), obtain_num_permutations(), obtain_permutation_matrix(), simple_ra()

Compute a randomization inference confidence interval

Description

Inverts the RI test over a grid of sharp null hypotheses to find the set of hypotheses that cannot be rejected at level alpha. The bounds of that set form the confidence interval.

Usage

ri_ci(..., alpha = 0.05, n_grid = 40)

Arguments

Details

The permutation matrix is generated once and reused across all grid points, so the cost is roughly n_grid times the cost of a single conduct_ri call (without the permutation matrix generation step).

Currently only supported for two-arm trials (single-term formulas). For multi-arm designs, call ri_ci separately for each pairwise comparison by setting condition1 and condition2 in the formula.

Value

A data frame with columns term, ci_lower, ci_upper, and alpha.

Examples

declaration <- randomizr::declare_ra(N = 40, m = 20)
Z <- randomizr::conduct_ra(declaration)
Y <- 0.5 * Z + rnorm(40)
dat <- data.frame(Y, Z)
# sims and n_grid are kept small here so the example runs quickly;
# use larger values in practice for a finer, less noisy interval.
ri_ci(Y ~ Z, declaration = declaration, assignment = "Z", data = dat,
      sims = 100, n_grid = 20)