I don’t say “control for” or “approaching significance”

Control for: These commenters on Andrew Gelman’s blog highlighted the fact that the technical meaning of this phrase almost directly contradicts it’s common-language meaning: In common English, having ‘control’ implies the ability to manipulate (or hold fixed) something. In contrast, statisticians and economists “control for” variables that they were not able to assign, and so must adjust for post hoc. I am following their reccomendation and using adjust for instead.

Approaching significance: Most know that you shouldn’t say that an estimate is “approaching significance” if it fails to hit your pre-specified threshold for significance. This said, most applied statisticians also know that the threshold of p=.05 is completely arbitrary, so it’s tempting to engage in this behavior when estimates are close to significant. I don’t have anything new to say on this front, but whenever I am tempted to say something like “approaching significance” it helps to imagine having to change all my marginally significant results to “approaching insignificance” for the sake of consistency, an idea I got from this pre-print.

Comments