Weather-Shock Selection Correction
RCTClinical Trial
In labor market hiring experiments and field trials, non-random participant attrition—where individuals fail to return with referrals—biases estimates of how financial incentives affect candidate quality.
Picture this
Think of evaluating who attends an optional weekend training session. If bad weather makes traveling difficult, only the most dedicated employees show up. Using rainfall as a decision variable allows analysts to mathematically correct for travel reluctance without distorting actual indoor test performance.
What the evidence says
One additional day of rainfall during the 3-day window reduced participant return probability by 20.7 percentage points (p < 0.01, chi-squared > 12.6), providing a valid instrument that eliminated non-response selection bias without affecting indoor cognitive task scores.
- Who was studied
- N = 561 original labor market participants tracked across 3-day recruitment windows in Kolkata, India.
- How
- Heckman two-step selection model utilizing the number of rainy days (0 to 3) during the recruitment window as an exclusion restriction instrument.
What to do
Utilize exogenous environmental friction variables (such as local precipitation) as exclusion restrictions in selection models to adjust for non-random participant attrition.
From the source
"One extra day of rainfall within the three-day referral cycle makes an OP 21 percentage points less likely to return with a referral to the laboratory."
468 Job Referrals in India AER Dec12.pdf