Cynical statistician is cynical

[lederhosen]

Back when I was a postgrad student, my reaction to a high R² value* was "Yay, it works!" Now it's more like "It's a trick. Get an axe."

(Seriously, I am pleasantly surprised when admin data shows any sort of correlation to theory at all. Predicting admin data with R²=0.85 from a simple rule of thumb? That's just unsettling.)

*Translation: the data follows a nice straight line.

Comments

[nefaria.livejournal.com]

> the data follows a nice straight line

And we all know that a straight line usually precedes a bad joke.

[culfinriel.livejournal.com]

I get very suspicious of papers with high R2 values that don't let you have enough data to double check it yourself. Biology is not that accomodating.

[lederhosen.livejournal.com]

Yeah, I have not-so-happy memories of having to retract* my 'authorship' from a couple of presentations that did exactly that. I hope my experiences weren't typical, but these days I assume all data analysis is guilty until proven innocent, and I'm rarely disappointed.

*here meaning "point out that I never actually granted it in the first place, and wasn't about to change that position".