KonFound-It!: Sensitivity Analysis for Robust Inference

Contributor. Tools and methods for quantifying the robustness of causal inferences against potential biases.
R
Sensitivity Analysis
Causal Inference
Published

2024

IES-funded research · Award R305D220022 · Principal Investigator: Dr. Kenneth Frank

I contribute to Quantifying the Robustness of Causal Inferences, a cross-institutional project associated with Michigan State University. My work includes development and validation of the konfound R package, algorithm testing, reproducible code, documentation, and co-authorship of the 2024 JOSS article.

1. Project Background

Sensitivity analysis addresses when evidence is strong enough to justify action by testing the robustness of inferences against potential biases. For instance, sensitivity indices quantify how much of an observed effect would need to be bias-driven to alter conclusions. This project builds on these ideas by developing methods and resources, centered on the konfound R package, to:

  • Quantify the robustness of causal inferences (RIR, ITCV).
  • Provide tools for published studies (pkonfound()), R model objects (konfound()), and meta-analyses (mkonfound()).

2. Package Overview

Field Details
Package konfound
Current release See CRAN
Title Quantify the Robustness of Causal Inferences
Type Package
License MIT + file LICENSE
URL https://github.com/konfound-project/konfound/
https://konfound-it.org/konfound/
BugReports https://github.com/konfound-project/konfound/issues

Dependencies

  • Depends: R (>= 2.10)
  • Imports: broom, broom.mixed, crayon, dplyr, ggplot2, lavaan, purrr, rlang, tidyr, lme4, tibble, ggrepel, pbkrtest, ppcor
  • Suggests: covr, devtools, forcats, knitr, rmarkdown, mice, roxygen2, testthat, Matrix

3. Installation

Install from CRAN:

install.packages("konfound")

Install development version from GitHub:

install.packages("devtools")
devtools::install_github("konfound-project/konfound")

4. Usage Examples

4.1 Published Studies (pkonfound())

library(konfound)
# Compute robustness indices for published summary statistics
pkonfound(
  est_eff      = 2.0,
  std_err      = 0.4,
  n_obs        = 100,
  n_covariates = 3
)

4.2 R Model Objects (konfound())

# Fit a linear model
m1 <- lm(mpg ~ wt + disp, data = mtcars)
# Sensitivity analysis for coefficient 'wt'
konfound(m1, "wt")

4.3 Meta‑Analysis (mkonfound())

# Built-in example dataset
head(mkonfound_ex)
# Run sensitivity analysis across multiple studies
mkonfound(mkonfound_ex, t, df)

5. Authors & My Contributions

The konfound package authors include:

  • Joshua M. Rosenberg (aut, cre)
  • Ran Xu (ctb)
  • Qinyun Lin (ctb)
  • Spiro Maroulis (ctb)
  • Sarah Narvaiz (ctb)
  • Kenneth A. Frank (ctb)
  • Wei Wang (ctb)
    • Contributed to package documentation, vignettes, and unit tests for core functions.
  • Yunhe Cui (ctb)
  • Gaofei Zhang (ctb)
  • Xuesen Cheng (ctb)
  • JiHoon Choi (ctb)
  • Guan Saw (ctb)

6. Additional Resources

browseVignettes(package = "konfound")
  • File issues or feature requests: https://github.com/konfound-project/konfound/issues

7. Citation

Frank, K.A., Maroulis, S., Duong, M., & Kelcey, B. (2013). What would it take to change an inference? Using Rubin’s causal model to interpret the robustness of causal inferences. Education Evaluation and Policy Analysis, 35, 437–460.

Frank, K. (2000). Impact of a confounding variable on the inference of a regression coefficient. Sociological Methods & Research, 29(2), 147–194.