Today I heard about a female biostatistician who went on a date with a man who advertised on craigslist that he was looking to go out with a woman who knew what logistic regression was.
As a public service to my nonquanticognoscenti female readers out there who wouldn't want to let a catch like this slip away, it's an analogue of linear regression for binary outcome variables; it can be derived by positing the outcome to be the binary manifestation of a latent continuous variable whose distribution conditional on the explanatory variables is logistic. If asked about it in a speed-dating context, you could probably just wink and say "linear in the logit, baby" and be fine, at least so long as you know that it's pronounced LOW-jit.
If you would like to be sassy, you can also retort that you won't go out with him until he proves he knows how to interpret an interaction effect from a logistic regression model, as all kinds of people mess that up, including hordes of smartypantsed economists and, well, me. If you show him interpretations and tests based on derivatives, he will almost certainly swoon; and if you combine these with a revealing graph of predicted probabilities, you may well stop his heart.
12 comments:
So, when Larry Laffer the Logit Lover says he needs to see at least a D-cup, he's really more interested in inspecting a woman's Double Derivatives?
You might just as well go out with Skilling or Lay.
a revealing graph of predicted probabilities
Hey, my book has one of those. My heart goes "BLUP, BLUP, BLUP" whenever I see it.
nonquanticognoscenti readers should note that Jeremy is making things needlessly complicated; you don't need a hypothetical latent variable to motivate the logit model- you can explain it just as a transformation that conveniently constrains your probability outcomes to between 0 and 1. Down with unnecessary hypothetical latent variables!!!
-Corrie
I feel a tasteless "probit" joke is imminent.
But if JF did not make things so hopelessly complicated life would be so dull, dull, dull.
Boy, this post makes me feel dumb.
Corrie: Perhaps you will notice that the "transformation that conveniently constrains your probability outcomes to between 0 and 1" is included in my post--namely, inside the heart. Note that there are various other transformations that "conveniently constrain" to the same end, leading to the inevitable, but still un-uttered, probit joke.
inevitable, but still un-uttered
I think the word you're looking for here is "latent."
Hmmm. JF seems at long last to have hogtied himself in his quest for a date.
what is the first derivative of the logit so you can derive the marginal effect of variable with interactive terms? And how do you form confidence intervals around it?
The gent has higher standards than most male quantiods. So many of these guys are just looking for simple "dummy codes" to mess around with.
Post a Comment