interflex performs estimation, diagnostics, and visualization of conditional marginal effects (CME) and group average treatment effects (GATE) of a treatment on an outcome across different values of a moderator. It supports linear, kernel, binning, lasso, and double/debiased machine-learning (DML) estimators, with optional fixed effects, multi-arm treatments, and bootstrap or simulation-based inference.

Maintainer: Yiqing Xu (yiqingxu@stanford.edu)

Authors: Jens Hainmueller, Jonathan Mummolo, Yiqing Xu, Jiehan Liu, Ziyi Liu, Licheng Liu, Tianzhu Qin

Repos: GitHub · CRAN

Documentation: User manual (Quarto book) covering installation, classic estimators, extensions, lasso, DML, discrete moderators, and a full plot-options cyclopedia.

Reference: Hainmueller, J., Mummolo, J., & Xu, Y. (2019). How Much Should We Trust Estimates from Multiplicative Interaction Models? Simple Tools to Improve Empirical Practice. Political Analysis, 27(2), 163–192.

Installation

Install the development version from GitHub:

# install.packages("devtools")  # if needed
devtools::install_github("xuyiqing/interflex")

Or install the released version from CRAN:

install.packages("interflex")

interflex depends on a number of CRAN packages that are installed automatically. If you need to install them manually:

pkgs <- c(
  "Rcpp", "mgcv", "sandwich", "pcse", "foreach", "doParallel", "doFuture",
  "doRNG", "future.apply", "lfe", "lmtest", "Lmoments", "ggplot2", "ggpubr",
  "ggplotify", "gridExtra", "RColorBrewer", "MASS", "mvtnorm", "pROC",
  "ModelMetrics", "patchwork", "xgboost"
)
install.packages(setdiff(pkgs, rownames(installed.packages())))

macOS toolchain notes

If you hit -lgfortran / -lquadmath errors during compilation, install the official R macOS toolchain from CRAN tools.

If you hit clang: error: unsupported option '-fopenmp', update R and the Xcode command-line tools, or install the r-macos-rtools bundle.

Quick start

library(interflex)
data(interflex)
# Linear estimator on simulated data with two covariates
fit <- interflex(
  estimator = "linear",
  data      = s5,
  Y         = "Y",
  D         = "D",
  X         = "X",
  Z         = c("Z1", "Z2"),
  vartype   = "bootstrap"
)
# Visualize the conditional marginal effect curve
plot(fit)

For an empirical example using app_hma2015:

fit <- interflex(
  estimator = "linear",
  data      = app_hma2015,
  Y         = "totangry",
  D         = "threat",
  X         = "pidentity",
  Z         = c("issuestr2", "knowledge", "educ", "male", "age10"),
  Ylabel    = "Anger",
  Dlabel    = "Threat",
  Xlabel    = "Partisanship",
  vartype   = "bootstrap"
)
plot(fit, xlim = c(0.25, 1), ylim = c(-0.2, 0.6))

The companion user manual walks through every estimator, the diagnostic plots, multi-arm treatments, fixed-effect models, lasso and DML, discrete moderators, and the full set of plot options.

Reporting bugs

Please open an issue on GitHub with a minimal reproducible example, or email yiqingxu [at] stanford.edu.