Multiple-output agency incentives in data production: experimental evidence

Abstract

This paper interprets the production of sample surveys as a multiple output agency problem in which interviewers act as agents for the principal investigator of the survey. The central aspect of data production that gives rise to the agency problem is that sampling error cannot be separated out from interviewer performance. We argue that contract effects on data production are important because the quality-quantity incentives facing interviewers affect the survey, in terms of its bias and its sample size. Direct evidence pertaining to these quality-quantity tradeoffs in the production of attitudinal surveys is examined through consideration of an experiment in which piece-rate and fixed wage contracts were randomly assigned across interviewers. Using these randomly assigned contracts, we find that observable quality dimensions do not respond to quantity incentives but unobservable quality dimensions do respond weakly in a negative way.