Endogenous Sample Selection

Abstract:

This paper examines how incentives for data collectors shape the selection of
sampling units. We provide causal evidence that data collectors respond to
variation in effort cost across survey subjects by excluding high-cost subjects,
thereby breaching protocol. Exploiting the random assignment of eligibility for
individual interviews across 3.4 million households in 181 surveys worldwide, we
find that in 110 (39) surveys at least 5% (10%) of eligible subjects are missing
from the sample. Selection out of sample is systematic: missing subjects disproportionately
come from marginalised populations. Using three applications,
we illustrate how this selection undermines microeconomic and macroeconomic
analysis alike.

Torsten Figueiredo Walter, (NYUAD)