Last updated on January 9, 2011
Adeline Delavande, Xavier Giné, and David McKenzie have a new paper in the Journal of Applied Econometrics on the elicitation of subjective expectations during fieldwork:
“Eliciting subjective probability distributions in developing countries is often based on visual aids such as beans to represent probabilities and intervals on a sheet of paper to represent the support. We conduct an experiment in India which tests the sensitivity of elicited expectations to variations in three facets of the elicitation methodology: the number of beans, the design of the support (predetermined or self-anchored), and the ordering of questions. Our results show remarkable robustness to variations in elicitation design. Nevertheless, the added precision offered by using more beans and a larger number of intervals with a predetermined support improves accuracy.”
In other words, it does not seem to matter how you choose your visual aids to elicit subjective probability distributions, but increasing the number of tokens to distribute and the number of intervals to distribute them in increases the precision of your estimates.
I have collected a good amount of subjective expectation data the first time I went to the field in 2004. First, I asked landowners who are involved in land tenancy as landlords to distribute 20 tokens among two boxes — one labeled zero, and one labeled one — so as to elicit their subjective perception of the likelihood that they might lose their plot of land.
Second, I asked household heads to distributed 20 tokens among nine price intervals (1-100, …, 801-900) so as to elicit subjective expectation as to what the distribution of the price of rice might look like at harvest.
What Adeline, Xavi, and David are saying is that I would have reduced the variance of my estimates if I had given my respondents 50 or 100 tokens in the former case, but that I would also have reduced the variance of my estimates if I had given my respondents 18 intervals (1-50, …, 851-900) in the latter case. While the former seems obvious, the latter does not. Either way, I would have gotten roughly the same point estimate, which is good to know.
While I have used the former data to study the emergence of sharecropping, I have yet to work with the latter data.