Friday, October 28, 2022

Did Trump Sabotage the Midterms: A Statistical Question


I, like many other people, have interpreted Trump's involvement in the Republican primaries as increasing his power in the party at the cost of reducing the party's power in Congress, pushing candidates who were loyal to him but less electable than their rivals. We might be wrong. Most people badly underestimated Trump's ability to get votes in 2016. Perhaps we are underestimating the attraction of his political style this time as well. 

There is a way to find out.

Make a list of Republican candidates who won their primaries with Trump's backing. Estimate what the midterm vote should be for every seat in Congress based on data from previous elections, taking account of location, incumbent advantage, and whatever else you need to fit the data. After the election, see if the Trump candidates did better or worse relative to the estimate than the non-Trump candidates. 

Ideally, the first two parts should be done before the election results are known, but it's probably too late for that unless someone is already doing it.

5 comments:

back40 said...

Do you think that Trump took a deliberate action aimed at weakening an organization through subversion, obstruction, disruption, or destruction? That's what sabotage means.

Do you think that estimating what the midterm vote "should be" tells you anything beyond your own biases?

Is this not the methodology that has been so inaccurate for decades?

David Friedman said...

I don't think his purpose was to weaken the party in Congress. My guess is that he had that effect and didn't much care, but I could be wrong and was suggesting a way of finding out. I cannot tell from what you wrote whether you understood what I proposed or, if you did, why you were opposed to the methodology.

back40 said...

The methodology doesn't work. "Whatever you need to fit the data". That doesn't reveal anything. It's a just so story, another way to badly estimate elections. The use of the "electable" trope is a tell. It's a combination of mind reading and biases.

Geech said...

I feel like 538 (or perhaps some other outlet) does perform this style of analysis, where they look at base factors like demographics and try to estimate how a candidate is performing in a location according to what those base factors would suggest.

I don't know if I've seen them do this for senate races, though.

David Friedman said...

@Back 40:
"Electable" doesn't come into the fitting exercise — that's the point of what I am suggesting. We find out whether the Trump candidates are more or less electable than others by seeing how they do compared to how we would expect them to do if we knew nothing about the candidate, but had information on how that state or district voted in the past in midterms, how votes for incumbents compared to votes for non-incumbents, all of the available facts you would use to predict outcomes if you didn't have polling or information about the candidate, other than whether he was the incumbent.