Tuesday, January 23, 2007
Pet Peeve
I've been reading the text for my Epidemiologic Methods class this semester, and so far I really like it. First and foremost, it is so infinitely better than last semester's unedited monstrosity (which doesn't even dignify a link) that I was nearly elated to read it from the start. But then, almost 200 pages in, it begins to address one of my biggest research peeves EVER: the absolute preeminence of the P-value for absolutely everything.
For gushier stuff than T-cell counts (like cognitive function scores or pain indexes), where measures are necessarily subjective, I would argue that such dichotomization is totally inappropriate. Sadly, that's all that gets published. Unless you're already a super-established expert with bottomless funds who's probably sleeping with the editor.
...type I and type II errors arise because the investigator has attempted to dichotomize the results of a study into the categories "significant" and "not significant." Since this degradation of the study is unnecessary, an "error" that results from an incorrect classification of the study result is also unnecessary.As one who falls squarely on the quantitative side of things - scientifically, what you think is only as interesting to me as the numbers you can show - the reliance on P irks me for just this reason. If you see 15 separate studies of, say, Drug X on T-cell counts, showing a (non-significant) p of .1, to me that's nearly as interesting as two with p=.003.
For gushier stuff than T-cell counts (like cognitive function scores or pain indexes), where measures are necessarily subjective, I would argue that such dichotomization is totally inappropriate. Sadly, that's all that gets published. Unless you're already a super-established expert with bottomless funds who's probably sleeping with the editor.
Labels: epidemiology, rant, research methods, statistics, textbooks
