As an aside to my discussion of ART and the fragility of high dimensionality models, you can find a related discussion in this post on Gelman's excellent statistics and political science blog.
His post relates to the fragility of even Chi-squared tests (among the most robust tests out there) in the face of data with odd cluster structures. Public management survey research faces these sorts of odd clusters all of the time. If Chi-squared can be fragile in these cases, multinomial probit does not stand a chance.
The moral of the story is know your data and your models.
No comments:
Post a Comment