The Senate voted on cloture today. Manchin crossed over to vote with the GOP, and Murkowski voted with the Democrats. I have made some remarks about gender and representation, about which I will elaborate again soon, but being a statistician, I couldn't resist trying to crunch some numbers.
Here's what I did. I put all of the Senators into a file. Three variables: Vote for cloture, coded as 1 for yes, 0 for no, as is tradition in roll call analysis. Party, coded as 0 for Democrats and 1 for Republicans, so that the expectation is a positive coefficient in regression analysis. Again, as is tradition. I treated King and Sanders as Democrats. Gender, flipping the script, coding men as 0, women as 1, because why not? The Doctor is a woman now anyway, so get over it.
I started with a bivariate "probit" model, which is the standard model to use when your "dependent variable" is dichotomous. Does party predict vote on Kavanaugh? Duh. The only defectors were Manchin and Murkowski. The question was what gender adds, given that the defectors were male in the pro-Kavanaugh direction, and female in the good/correct direction. (What, like you couldn't tell what I thought of this?)
My statistics program, Stata, told me to go fuck myself, and get back to work on my next book. It was slightly more polite than that, but not much. What happened was, in econometrics-speak, the model wouldn't converge. I tried it with logit too, and same thing. Logit and probit models are estimated iteratively, and the computer went through so many iterations without reaching a conclusion that it gave up. That's the long and short of it. Why? Party does all the work, and two odd cases, even if they match gender expectations, don't mean shit in the face of all of the evidence that this is basically party.
Party, not gender.
OK, technically, the computer didn't give me a zero coefficient for gender. It's just that with two deviations from party that both fall along gender lines, trying to make the gender argument is going out on an econometric limb.
