Thursday, April 21, 2011

Don't Transform

One of my pet peeves about my ecological colleagues is their tendency to transform binomial data using arcsine of the squareroot of the proportion in order to use a linear model. OK, once upon a time, it might have made sense to do this. But we have better tools now, honestly! Travis Hinkelman brought a great paper by David Warton and Francis Hui to my attention this morning. I'm just going to quote one line, which sort of says it all:
The most striking result in power simulations was that logistic regression and GLMM always had higher power than untransformed and arcsine transformed linear models ...
So, don't transform your binomial data. And please, if you are collecting proportion data, write down both the numerator and denominator! This will be required reading in my Ecological statistics class next fall.

No comments:

Post a Comment