The Thin Blue Line That Stays Strangely Horizontal

Last night I was pondering how to test the Reactionary claim that society was drifting inexorably to the Left, and I wondered whether maybe someone had created a mathematically rigorous model of people’s positions along the left-right spectrum and then applied it to every legislator in every session of Congress from the founding of the United States to the present day.

This being the Information Age, of course the answer is yes. NOMINATE is a statistical system that takes every legislator’s vote on every issue that comes before Congress, then number-crunches it to see which legislators are closest or furthest away from one another. The classic left-right dimension naturally falls out of such a model, allowing a quasi-objective way to determine which legislators are more liberal or conservative than others.

It would appear that this could only compare two legislators serving at the same time and voting on the same issues. But an enhanced version of the system, DW-NOMINATE, gets around this limitation with a clever trick. Political scientists know that legislators’ views generally change little over time. Therefore, long-term legislators can be used to “bridge” different sections of Congress. If Senator Bob is the most liberal Senator in 1950, but by 1970 he’s somewhere in the middle, then we know the Senate became more liberal from 1950 to 1970. If Senator Mary is the most liberal Senator in 1970, but by 1990 she’s somewhere in the middle, then we know the Senate also became more liberal from 1970 to 1990. Therefore, we can conclude that the Senate in 1990 is more liberal than in 1950, even if no single Senator served in both bodies.

DW-NOMINATE has been called “one of the great achievements of modern political science”, has won a bunch of awards, and most tellingly is routinely used by Nate Silver. It is recognized as broadly accurate by people from both sides of the political spectrum. So let’s see what it finds:

A few things here stand out as correct. Congress is getting more polarized recently – true. The Republicans took a big turn to the right starting in 1980 – true.

But from this it looks like there has been zero liberal drift of either party over the past 120 years. And don’t get me wrong, if it’s true I will relish rubbing it in the Reactionaries’ faces – but it seems kind of implausible. That whole “expansion of the welfare state” was a thing that happened, right? And Martin Luther King Jr speeches aside, I’m pretty sure the civil rights movement wasn’t literally a dream. Gay marriage? Also, either we got an income tax early in the 20th century, or else some guy with the initials I.R.S. has been pulling a very successful con on me.

Normally I’d dismiss this. I’d say it probably wasn’t kosher to use DW-NOMINATE data to make time comparisons. But a political analyst tried making that argument last year and was shot down by the developers of the DW-NOMINATE system. And Nate Silver seems pretty happy to draw DW-NOMINATE comparisons across presidents. And from what I understand of the long but gratifyingly non-mathematical guide to using DW-NOMINATE, intertemporal comparisons are entirely acceptable as long as they don’t go back further than 1879. And the system’s ability to pick up other secular trends, even smallish ones, seems quite good.

I’ve got to admit it. I’m stumped on this one.

Edit: There’s a similar measure called Martin-Quinn scores for the Supreme Court that finds a similar result