🔒All inputs are processed locally in your browser. Nothing is transmitted, stored, or logged. Inputs disappear when you close the tab.
SELECTED STUDIES & METHODS · #183

Response-Rate Sensitivity Lab

Enter an observed survey mean, drag the response-rate slider, and watch the worst-case identified interval sweep wider or narrower in real time - the core intuition behind nonresponse-bias sensitivity analysis, no dataset upload, no R. A second calculator answers the classroom follow-up: "what response rate would I need to trust this effect?"

📋 Pure closed-form interval arithmetic, no live data, no vintage
Honest framing: the underlying math is Manski-adjacent worst-case bounds. This is a classroom teaching demo for building intuition about nonresponse bias - not a substitute for an analyst-grade bounds estimator run on real survey microdata (e.g. bootstrap/MTR/MTS/MIV tooling). If you have a real dataset to bound, use dedicated statistical software.
Lab 1 · Drag the Response Rate
e.g. % who said yes, or a scale score
Lowest value the outcome scale allows
Highest value the outcome scale allows
60%
0100
Worst-Case Lower Bound
-
Worst-Case Upper Bound
-
Interval Width
-
Everything the true population value could be, given only the response rate and the worst-case scale bounds
Lab 2 · What Response Rate Would You Need to Sign This Effect?

Enter the same respondent mean and scale bounds, plus a threshold you want the true population value to clearly beat (e.g. a comparison group's rate, a pass/fail cutoff, "50%"). This finds the minimum response rate at which even the worst-case bound still clears that threshold - below it, nonresponse alone could explain away the result.

The value you want the true population mean to beat