Another day, another poorly described study advertised on Yahoo!
What do we learn? We learn that there is a correlation between coffee drinking in middle age and dementia in later life in some unknown Finnish panel data set. We also learn that the researchers have no model of why caffeine should affect Alzheimers, they are just fishing around in the data set.
When I was an undergraduate, I worked for a few years at a "remote" site of the campus computer center located in the Warren G. Magnuson (a former porkmeister senator from Washington State) Health Sciences Center. There was one researcher who would literally regress "all on all" - apparently you can do this in SPSS, which is another reason to prefer Stata - print out the results on fan-fold paper (this was back in the day), and then stand by the printer circling all the regressions with statistically significant coefficients with a red pen. Pre-test bias, you say? Never let that stand in the way of publication.
What do we not learn? We do not learn anything about initial response rates or subsequent attrition in the Finnish panel data. We do not learn how many other treatment variables besides coffee the researchers looked at. The write-up suggests indirectly that the identification strategy here is linear selection on observed variables but we learn nothing about what else was conditioned on that might serve to make such an identification strategy plausible.
And, of course, the article provides no information on how to actually find the study so that you can read it yourself and fill in the important bits omitted by the reporter.
Who was my favorite student this term?
4 years ago