If you beat the egg long enough,
it will confess.

A very wise statistics professor of mine at Iowa State University made this statement while discussing the pros and cons of excessive model fitting, bizarre transformations, and unnecessary tweaking of the data just to get the desired results. I recalled those words today as I sat at my desk generating scores upon scores of pages of statistical output. I'm in the midst of analyzing data from a cluster of related experiments. And, while the experiments were relatively easy to execute and the data rather simple in scope, their analysis has proven to be quite the bugger!

I always strive for simplicity, for parsimony. And yet, that is tough to do when I request a simple answer of the software, and out pops reams upon reams of bizarre tests, each with their own unique probabilities. Akaike Information Criterion here, Huynh-Feldt Epsilon there. I roll my eyes as I attempt to wade through the gobbledygook. "Oh but such things are now required by the reviewers and editors of the journals!" cry out colleagues. "We are well beyond simple Student's T-Tests."

And yet, when I peruse Science and Nature, the two journals regarded as finest in all of the lands, I frequently see complex results analyzed in the most simple of ways. I see means followed by their standard errors, and a simple t-test used to determine if those means are any different from each other.

Often when I get frustrated over such matters, I consult with that wizened statistician. His observation in an e-mail this morning: "I am very tired of the Statistics profession burdening biologists with complexities I doubt are any more than a new way to get some young stat professor promotion and tenure."

Amen.

I think I'll drink a Guinness and then conduct some 'simple' Student's T-Tests (invented almost a century ago by a brewer of that fantastic stout). If its good enough for Science...


Romneya coulteri