Hidden Variable Sensitivity Analysis

Hidden Variable Sensitivity Analyzer

This tool explores how a hypothetical unobserved variable—such as one that might influence both smoking and cancer—could distort a predictive model’s conclusions. Grounded in the work of Cochran, Rosenbaum, and Rubin, it applies that logic to approval ratings for JFK in 1963. It also includes color-coded warning badges to highlight potential vulnerabilities.

How to use this tool:

Adjust the sliders to define the characteristics of a potential missing variable.
Click "Run Analysis" to simulate how it might influence model predictions.
Review changes in correlations, coefficients, margins, and predicted outcomes.
Click "Reset" to restore the default scenario.

Targeted Correlation with Outcome Variable How strongly should the synthetic variable correlate with the outcome?

-0.8 0.8 0

# Categories How many distinct categories should the synthetic variable have?

(2–5 categories allowed)

Max Correlation with Other Predictor Variables Controls maximum correlation between the synthetic variable and the other variables in the model

-0.8 0.8 0