Mahalanobis points us to a couple of interesting articles by Peter Kennedy on the nitty-gritty of applied econometrics. 'Oh no! I got the wrong sign! What should I do?' is a very nice summary of the data and modelling problems that can generate misleading estimates, and anyone who has had to deal with puzzling estimates - that is to say, anyone who has ever done applied work - would benefit from reading it.
But there's a niggling question that has to be addressed. Suppose you get the right sign for a given parameter estimate. Would you still go through Kennedy's checklist of problems that might generate the wrong sign? No? Why not?
Comments