Thursday 13 October 2011, 1.15PM to 14:45
Speaker(s): John Hey
This paper constitutes a call for econometrics to be done both before and after an experiment, and for the econometrics to be concerned both with predictive ability (rather than testability) and with the stochastic assumptions underlying the econometric analysis. The reason for the former comes from the fact that economists should be concerned with prediction (rather than testing); and the reason for the latter is that the power/ability of experiments to discriminate between theories depends crucially on the stochastic nature of the data. Problems with power can be remedied by carrying out simulations before an experiment is run in order to determine how many decision problems (to use a generic term) subjects should be presented with. The answer to this question depends crucially on the nature of noise, and its magnitude, in subjects behaviour. This is the final, and perhaps central, message of the paper the stochastics of behaviour are often more important than the deterministics: making wrong assumptions about the former can lead to incorrect inferences about the latter. Extensive econometrics is needed both before and after an experiment.
Location: ARRC Seminar Room (RC/010)