The Shy Trump Effect

I’m taking a bit of a mental health day, today, so I’ll cheat early by copying an interesting twitter thread by Philip Pilkington (h/t commenter FreeFrench). If the name Pilkington seems familiar, it may be because I’ve cited him several times in the past, mostly regarding geo-politically related matters (he’s a macro-economist), and most recently here: Assessing the Economic Value of Military Materiel.

I’m sure most readers have heard of the Shy Trump Effect, and may even have anecdotal evidence to support the idea. It’s a fairly well recognized phenomenon, but Pilkington, as an economist and therefore a stats savvy guy, puts some meat on them bones. In particular, he’s able to show a clear pattern of polling misses that always work against Trump and and quantifies that effect in swing states. Based on Nate Silver’s analysis, there should not be patterns like this—it should be more random, just based on the nature of most public (as opposed to internal) polls. So …

Philip Pilkington @philippilk

1/ The US election is here. If you analyze the polls properly you'll see that @realDonaldTrump is ahead of @KamalaHarris by anywhere between 2.4% and 6.9% in the swing states. The reason the media won't tell you this is because they don't understand the Shy Trump Effect

3:15 AM · Nov 4, 2024

2/ The polls show consistent bias against @realDonaldTrump. This is not due to cheating but due to quirks in the polling. More later. For now, let us just look at polls versus outcomes. Here are the November polls in 2020 versus the actual results. They were way off.

[Mark Mitchell from Rasmussen has noted something like this phenomenon as well.]

3/ Despite these not being the swing states in 2016, we see a similar pattern of polls missing their mark by a long shot in 2016. Again, the polls always showed @realDonaldTrump way behind what his actual results would be.

4/ The difference between polls and election outcomes can show us the Shy Trump Effect in the polling. We have used the 2020 Shy Trump Effects to adjust the 2024 polling and this is what gives us the 2.4%-6.9% lead in swing states.

5/ Why the Shy Effect? There are a few hypotheses. One is that Trump voters are less inclined to take surveys that they associate with the government. Basically, Trump-voting Boomers hang up the phone while Harris-voting Boomers love having their opinions heard. [Check]

6/ Another hypothesis is that there is a significant pool of Trump-voting men whose wives would kill them if they found out that they were voting for Trump. When they get a call from a pollster they either lie about their voting intention or hang up. [Nope]

7/ Surely though, the pollsters would have corrected for this effect? Not so. The effect got worse between 2016 and 2020. The reality is that there is no way for a pollster to change methodology to correct for this effect. If people lie in polls, they lie. [As Shipwreckedcrew explained in our post yesterday.]

8/ Pollsters know this. They tell the newspapers. What can they do? They can only report what people are telling them. If statisticians want to adjust the polls for the Shy Trump Effect that is for statisticians to do, not pollsters. But few do - because they lean Democrat.

9/ The smart money knows about the Shy Trump Effect. The Polymarket Whale - a French trader who bet $30m on the election and moved the odds - told the @WSJ that he was betting based on the Shy Trump Effect. It is simply in the liberal media's self-interest to lie about it.

10/ Could the Shy Trump Effect disappear this cycle? Yes. But there's no reason to think it will. When doing analysis the standard scientific practice is assume that what happened in the past will happen in the future - unless we have strong reason to think otherwise.

[I.e., if it does disappear, think: hanky panky, shenanigans.]

11/ And so we're back to adjusting the polls. We take the polling error from 2020 and apply it to the 2024 polls. Standard practice and uncontroversial. But here's the thing: it shows a Trump landslide. Let the experiment begin!

[Very similar to what Rasmussen noted—cf. link above.]