1E: Degrees of freedom: WTF?!

Jeff Witmer (Oberlin College)


Degrees of Freedom are the Wrong Topic of Focus

Getting the degrees of freedom right means getting the right p-value; but if you use the wrong df the p-value barely changes -- and you didn't need an exact p-value anyway. Other things are more important, so just use +/- 2 and don't worry about being more precise. Don't let mathematical pseudo-precision supplant statistical thinking.

Rather than teach degrees of freedom and the like, what should we be teaching?

Here is a starter list:

  • Effect size
  • The regression effect
  • The sign test
  • McNemar’s test (easier to understand than a paired t-test, and easier to do)
  • Simpson's Paradox -- and controlling for confounders
  • Bayesian inference -- at least an introduction

There are many other things that could be listed here. We'll brainstorm, compare experiences, and maybe reach some level of agreement (or not).