Extreme Value Top-Coding Truncation Protocol
RCTClinical Trial
A small number of exceptionally large recruiting firms can generate massive hiring spikes that severely distort linear treatment effect estimates in randomized trials.
Picture this
Think of calculating the average height of students in a classroom. If a ten-foot giant walks into the room, the average jumps drastically and no longer represents a typical student, so statisticians cap the maximum recorded height to keep the calculated average realistic.
What the evidence says
Capping administrative distributions at the 99th percentile eliminated extreme hiring outliers, ensuring linear treatment effect estimates (+0.047 vacancies, +0.046 hires) reflected representative firm behavior across the sample.
- Who was studied
- N = 7,438 establishments in France tracked through administrative database records.
- How
- Stratified Randomized Controlled Trial applying 99th percentile top-coding truncation to administrative vacancy and hiring metrics.
What to do
Apply 99th percentile top-coding truncation to administrative labor market outcome variables prior to running linear treatment regressions in randomized trials.
From the source
"Finally, we top-code all administrative data at the 99th percentile of their distribution in order to make sure our linear estimates are not driven by a few, very large recruiting firms."
Are_Active_Labor_Market_Policies_Directed_at_Firms_Effective_Evidence.pdf