Thursday, 16 April 2015

External validity in experimental economics


Lab experimenters worry a lot about external validity. OK, we say, it happened in the lab, but would we see it in the real world? Answering this question, by connecting the lab and the field, is a good way to get published.

Most of these papers look at individual-level external validity. For example, were the same people who were trustworthy in a "trust game" lab experiment,  also more likely to (e.g.) repay microfinance loans? Were people who cheated in an "honesty game" (throwing a die and reporting the result where you got paid more for some results) also more likely to cheat by riding without a fare on public transport? *

Policy-makers, though, are probably more interested in external validity of treatments. For example, in the lab, people may be more cooperative if others can observe their behaviour. Will the same thing happen if we (e.g.) gave people badges when they made a charitable donation?

Individual-level and treatment-level validity need not be related. Perhaps the same people are generous in the lab and in real life; but a particular treatment only works in the special atmosphere of the laboratory. Conversely, even if lab behaviour does not reflect a stable underlying trait of individuals, a policy intervention may still affect it. Both questions are interesting but it is important to distinguish them. The real test of external validity is probably: does your policy intervention work?


* Marie Claire Villeval's paper, not online yet.

Update: Ro'i Zultan points me at Vernon Smith's 1981 paper:
What parallelism hypothesizes in micro-economy is that if institutions make a difference, it is because the rules make a difference, and if the rules make a difference, it is because incentives make a difference.