Just how many experiments are getting published in economics?
And just how many are rooted in theory?
David Card, Stefano DellaVigna, and Ulrike Malmendier plead for a little more theory in this forthcoming JEP piece.
Presentation to the Joint Chiefs Operations Directorate
November 11, 2022
4 Responses
More and less theory, please. Design? Yes. Analysis? No.
We didn’t have enough theory the last 200 years already?
I’ll have a positive view : During the last 6 years we have started to see a number of papers using competings models to try to explains thing, and that’s a *first* ever.
Also it’d be intersting to understand better why during 10 years the numbers of any kind of non-descriptive experiments has been zero ? Is that really a correct measurement ?
Lets turn the perspective: The initial goal of an experiment is (under the condition of internal validity) establish the possibility to draw causal inferences from a sample on a population. Afterwards, one (or others) replicates the experiment on another sample, and another sample, and…, and hopes that the results will be replicated. At least that’s what natural scientist do (economists truly want to be counted to this group). Only then your are approaching the main goal – establishing external validity (or the even more ambitious goal ecological validity).
The (mainstream) economists remain interested in establishing external validity regarding their structural parameters of their models, to derive predictions about future outcomes when the parameters are variied (as claimed by Acemoglu 2010 in his JEP piece). The problem is that confirming a model does not tell you anything about the population. You always can find ways to establish model fit via calibration or addition of variables in your empirical tests – it won’t tell you anything about the external validity.
I guess most will not agree with that, the point is (and has been made by Binmore criticism of Fehr et al. calibration strategies) it may be just for the best replicating a finding with a single model or remain descriptive (in the sense of the article above) on other data, at least different data than the one that was used to test the initial model. Though it won’t be as ground breaking, it will simply lead to more information concerning external validity. Experimental Economics certainly does not suffer from a lack of theory, rather a lack of replication.