In my previous post I argued that the indictments of former president Donald Trump did not in fact boost his standing with voters, despite the often repeated claim that they did.
Two folks I respect pushed back on the NY indictment, though agreeing the others had no effect. Here I respond, and agree in part. In any case, I think this shows a positive discussion is possible over such matters! My bottom line is Trump may have gained support in the GOP primary following the NY indictment. But I don’t find evidence of any gain in overall favorability with all registered voters. My new analysis (thanks to Philip and Sam) finds some evidence that the Florida indictment lowered his support and favorability, something I did not mention in my original post..
Since my original post, Philip Bump at The Washington Post published a nice story looking at my claim, and concluding there was in fact a boost from the first indictment in New York but agreeing there were no boosts from the subsequent indictments. See his piece here. (Link should not be paywalled.)
Sam Wang posted a similar point on Bluesky (sorry, don’t know how to copy a link to that)
Let’s start with the strong case, a rise in Trump GOP primary support following the NY indictment. Here is Trump’s percentage i.n GOP primary polls before and after each indictment. (I’ve updated the most recent polls since my original post, which affects only the post-GA data.)
We all agree the post-New York support is higher than pre-indictment. The only disagreement is that there was a trend of rising Trump primary support before the indictment and that rise continued for over a month after the indictment. My original post claimed the post-NY increase was primary due to the pre-existing trend, rather than an indictment effect. Bump and Wang disagree and see an indictment effect. Here is the chart with the full trend.
The red line shows the overall trend across all polls, the dots are individual polls. The high-frequency polls from YouGov (85 in all) make up the dense set of dots mostly above the red trend line. Those dense dots DO show a bump up immediately after the NY indictment. In my original post I discounted that, thinking the red line was increasing before and after and I gave more weight to that. But I discounted the dense set of polls that do rise. Wang raises the point but Bump dives into this and does more substantial analysis that I now think supports a NY indictment effect.
Here are my finding on this, in response to their points, and finding myself now agreeing there was an independent effect of the NY indictment even considering the trend before and after.
I fit several models, but here are the two that make me conclude NY had an effect. The dependent variable is Trump percent in primary polls. You do NOT want to use his margin over DeSantis because the latter has been declining steadily, so the margin confounds Trump support with DeSantis’ weakening. As you can see in the chart, Trump has been pretty flat since mid-May.
The models I fit are a polynomial in time, which allows for the curve of the support trend, which rises, then flattens. Obviously not a constant linear trend. I fit one quadratic and one cubic fit (the latter probably overfitting the trend but our focus in on the coefficients for the indictments.)
The first model is the quadratic trend.
The NY coefficient is a 4.12 percentage point increase in Trump support after the indictment, and is a statistically significant effect. So this agrees with Bump and Wang, and shows I dismissed this effect too quickly.
The model also suggests that the Florida indictment may have lowered Trump’s support by 2.8 points, a marginally statistically significant result. There is no evidence that DC or GA indictments changed his support. Both have negative but non-significant estimates.
An issue is whether the quadratic trend is sufficient to capture the trend over time, independent of the indictments. As a check I reestimate the model with a cubic in time. That model is shown next.
The NY effect here is a 4.8 point increase, which remains statistically significant. The Florida effect remains negative but falls short of statistical significance. So either way, the NY indictment seems to have boosted Trump with GOP primary voters over and above the trend leading up to the indictment.
On the other hand, Trump’s favorability ratings with all registered voters don’t seem to have gone up with NY, but may have declined slightly with the FL indictment.
Here the trend is nearly flat since January, so I estimate one model that is linear in time, and one that is quadratic as a robustness check. The linear time model is
The time trend is virtually flat, with no evidence of a NY indictment effect. However, the Florida indictment seems to have lowered favorability by 3.6 points.
Using a quadratic in time is similar:
The Florida estimate is a 3.8 point decrease in favorability, and marginally significant, with no evidence for a more complicated time trend. As a final check I ran a cubic time trend with the FL coefficient of -3.5, but p=.056 so not as convincing an effect.
Bottom line is the NY indictment didn’t show any evidence of boosting Trump’s favorability among all registered voters, but it does seem to have improved his standing in the GOP primary by 4.1 to 4.8 points depending on the model.
My thanks to Philip and Sam for pushing this point and adding to the analysis.