Abstract:
Null hypotheses in undergraduate econometrics courses are usually framed in terms of parameter values or distributions. But relatively simple techniques can also test for violations of good scientific practice. This is neatly illustrated for students by a reinterpretation of an influential paper by Sir Ronald Fisher, where a rejection region is formed on the left tail of a 2 distribution. This idea is extended to situations where dubious models fit 'too well'. In these cases, a high R2 may be taken as evidence that a non-random subset of regressions is being 'adversely selected' for publication.
