README.md

In gwb/RGroupFormation: Randomization tests for group-formation experiments

RGroupFormation

This R package provides utilities for running randomization tests of peer-effects in group-formation experiments. It also includes functions for designing such experiments.

Installation

Install the package from GitHub with:

# install.packages("devtools")
devtools::install_github("gwb/RGroupFormation")

Example

The following code describes a simulation computing the power of a conditional randomization test for different alternatives of the form H1: Y(1) = Y(0) + tau. It uses the functions find.good.L0, summarize.exposures, test.FRT.strat, and T.dim from the package.

library(RGroupFormation)

set.seed(123) # for reproducibility

# initialization
N <- 90 # total number of units in the experiment
M <- 3 # numher of units in each (equal-sized) group
A <- rep(c(0, 1), each=N/2) # each unit has a binary attribute
L0 <- find.good.L0(N, M, 1000) # draws 1000 seed room assignments.

# Displays the exposure for each seed assignment.
summarize.exposures(W.star.fn(L0, A))

# Initializes the potential outcomes
# Each entry in tau.10.ls is a value for the effect of having a single neighbor 
# with attribute `A=1`. These values are used to define the alternative hypotheses. 
# tau.20 is the effect of having 2 neighbors with attribute `A=1`.
Y0 <- rnorm(N)
tau.10.ls <- c(0, 0.1, 0.2, 0.4, 0.6, 0.8, 1) 
tau.20 <- 1


# Initializes result vector
res.ls <- vector('list', length=length(tau.10.ls)) 

# Iterates over values of tau.10.ls and computes a p-value for each
for(i in seq(length(tau.10.ls))){
    print(paste0("tau spill = ", tau.10.ls[i], collapse=""))

    # constructs a science table
    science.i <- cbind(Y0, Y0 + tau.10.ls[i], Y0 + tau.20) 
    colnames(science.i) <- NULL

    # Computes the p-value
    res.ls[[i]] <- test.FRT.strat(L0, science.i, T.dim, A, 1, 0, n.rand=300, n.draws=1000)
    print(paste0("power = ", mean(res.ls[[i]] <= 0.05), collapse=""))
}

# Computes the power!
power.ls <- sapply(res.ls, function(res.i) mean(res.i <= 0.05))

Recipes

The following snippets show how to use the functions provided in this package by isolating a number of use-cases.

Room assignments, group assignments and exposures

There are 3 `assignment' quantities of interest in Basse et al (2020). The room assignment vector L, the group assignment Z = Z(L), and the exposure vector W = W(Z) = W'(L). The package provided functions for easily generating these.

library(RGroupFormation)

# setup parameters
N <- 90 # total number of units
M <- 3 # number of units per room
A <- rep(c(0,1), each = N/2) # attribute vector

# Room assignment vectors 
L <- rL(N, M, 1)
Lmat <- rL(N, M, 100) # draws 100 random room assignment vectors.

# Group assignments
Z <- Z.fn(L)
Zmat <- apply(Lmat, 1, Z.fn)

# Exposures
W <- W.fn(Z, A)
Wmat <- apply(Zmat, 1, function(z) W.fn(z, A))

# Exposure (alternative approach)
W.alt <- W.star.fn(L)
Wmat.alt <- W.star.fn(Lmat)

all(W == W.alt)
all(Wmat == Wmat.alt)

gwb/RGroupFormation documentation built on Sept. 13, 2020, 4:41 p.m.