Skip to contents

FEDAI

The tables, study datasets, and helper functions used in Field Experiments: Design, Analysis, and Interpretation, by Alan S. Gerber and Donald P. Green.

# install.packages("remotes")
remotes::install_github("acoppock/FEDAI")

What is in it

38 datasets. The hypothetical tables the book reasons over, the study datasets the chapters analyze, and the nine datasets the exercises ask the reader to download. Every dataset is documented with its dimensions, a description of each column, and the archive file it came from:

library(FEDAI)

table_2_1        # the seven villages, with both potential outcomes
arceneaux_2005   # precinct-level cluster-randomized canvassing
titiunik_2010    # senators assigned two- or four-year terms by lottery

?table_2_1       # dimensions, columns, and source
data(package = "FEDAI")

Potential outcomes are named for the assignment that would reveal them, so Y_Z_0 is the outcome if untreated and Y_Z_1 the outcome if treated. Where a second dimension of assignment matters the name grows: Y_A_1_Z_0 in Table 8.4 is the outcome for a hotspot whose neighbor is treated while it is not.

A few helper functions.

Function What it does
pop_var(), pop_cov(), pop_sd() Divide by N rather than N - 1, which is what a schedule of potential outcomes calls for
equation_3.4() The true standard error of a two-arm experiment under complete random assignment
difference_in_variances() Difference in outcome variances between arms, with the F test p-value
posterior_mean_fn(), simulate_posteriors() Bayesian updating from an experimental result, Chapter 11
make_studies() Simulate a research literature with a moderator, Chapter 11

Where the data came from

Every dataset except two is built from the ISPS Data Archive’s study D081, Field Experiments: Design, Analysis, and Interpretation, on Yale Dataverse. data-raw/ holds one build script per group of datasets; each names the archive file it reads by its numeric id, so any dataset can be traced back to a citable file:

source("data-raw/build_all.R")   # refetch and rebuild every .rda
source("data-raw/verify.R")      # check each rebuild against what is shipped

The two exceptions are Tables 12.2 and 12.5, which are new in this edition and authored in the manuscript rather than drawn from the archive; their values are written out in data-raw/book_tables.R. The hotspot coordinates behind Table 8.4 are a third case: they are not in the archive as data, but the archive’s own Chapter 8 code draws them from set.seed(1337), so the build regenerates them exactly.

Regression adjustment helpers that used to live here now live in estimatrTools; see NOTES_moved_to_estimatrTools.md.