Neil Gorsuch, who definitely understands regression

For context for others - regression analysis is an estimation method in statistics that allows you do describe a relationship between two variables, and importantly in the case of multiple regression to try to describe that relationship after “controlling” for other factors...

For example you might be interested in the relationship between race and educational attainment, but be worried that because race and income are associated, the relationship between race and education you observe also includes the influence of income.

If you can measure income, you can put it in a regression with the measurements you have on race and education and get closer to isolating the true relationship between race and education. There are at least a few things wrong with what Gorsuch is saying:

First, as you may be thinking to yourself, there are many many things that could influence any relationship you’re interested in, and it is probably hard to capture all of them. That’s why regression alone is generally not believe to provide good causal evidence.

Two, Gorsuch’s diction here is just way off. Which is perhaps a bit petty but suggests he doesn’t understand the underlying theory. And he needlessly throws in the 95th percentile comment in what seems like an attempt to demonstrate his knowledge but does the opposite

First - nothings really getting disaggregated. That’s more when you’re looking at the same relationship for different groups or something. Two, the 95th percentile thing is referring to a 95 percent confidence interval, which statisticians use to estimate the margin of error

The problem is that 1) percent is different from percentile and 2) the percent in 95% confidence interval comes not from your level of confidence but the fact that in 95% of cases you would expect the true estimate to fall within your margin of error.

To say that something is identified “to a 95th percentile of confidence” parses literally as there are a range of confidences one can have in their estimates and this one is greater than 95 of the rest. Which is not at all what is going on. One might even call it gobbledygook.

tfw when you tweet fast to explain how someone misspoke and make lots of grammatical errors but also you’re not a Supreme Court Justice.

A few reactions from people I respect saying: look, it's hard to speak extemporaneously about a topic out of your field of expertise. And fair enough, but for comparison here's Elena Kagan discussing the simulation methods used by @gerrymandr to evaluate partisan gerrymandering