Data over dogma: A reply to Matt Yglesias

Further thoughts and data on the declining electoral benefit to ideological moderation. Yglesias wants Democrats to fire LeBron!

On Monday, the political blogger Matthew Yglesias published a “response” to my article last week on the small and shrinking electoral reward to ideological moderation, “Moderation is not a silver bullet.” Unfortunately, in his piece Yglesias (a) does not respond to any of my actual arguments (it seems he only loosely read the intro; Yglesias told me over Twitter DMs that he does not have a paid subscription to this newsletter and “is not responding in any detail to [my] analysis” — even though the subhead of his article is “G Elliott Morris is wrong on the internet”) while (b) raising one new methodological objection that is easily dealt with (in fact I’ve already addressed this with him on social media. he seems not to have cared).

We do agree on a few things: Moderation does increase vote margin by about one percentage point, heterogeneity in moderation exists, and candidates’ brand/“vibes” can be decisive. But do those conclusions present actionable steps, stronger than other levers strategists can pull? Do our conclusions change when we acknowledge uncertainty?

I do not want this publication to devolve into an endless debate about moderation (see footnote one of my first article, as well as this one), but I did feel the need to address Matt’s “response.” Yglesias is a big figure in political media, so when he gets my work wrong I feel pressure to correct the record. Many of his characterizations of my argument are simply wrong, though it’s clear that some areas of my first post bear repeating or repackaging for clarity and impact (or, again, maybe he just didn’t read the original analysis).

So here goes:

Editor’s note: Thank you to the individuals who provided feedback. You know who you are.

---

The big picture: Small effects, large uncertainty

As a refresher, my analysis finds that more moderate Democratic candidates (think Maine’s Jared Golden) outperform the median Democratic candidate on average by between 1 and 1.5 percentage points in vote margin, but with a huge confidence interval that ranges from the moderate gaining 10 points to losing 8. As far as I can tell, this is similar to the political science consensus on the value of moderation, and the first effort to quantify the uncertainty about the change in vote margin a candidate might get by moderating their views (or being replaced by someone else who is more moderate), adjusting for other factors.

In a nutshell, Yglesias argues that this 1-1.5-point effect size is large enough to craft political strategy around: He ends his article with “The most significant thing, though, is that the sign of the effect is clear: moderation works!”

I disagree. I think the uncertainty here matters a lot, and I believe that the refusal of Yglesias to acknowledge this uncertainty reflects underlying incentives or an ideological prior¹ more than it reflects the value of moderation. You do not get to ignore uncertainty just because it’s convenient to do so. By focusing just on the effect size and not uncertainty, Yglesias biases the debate in his favor. The uncertainty has consequences for how these statistics get used, so it’s fairly important that he’s ignoring it.

In my article, I make the point that the uncertainty around the estimated effect size is large enough to make recommendations for campaigns based on an analysis of ideological scores practically useless. This is particularly true when you consider the other factors determining House outcomes — above all, whether the national environment favors them. The uncertainty in the analysis is so large that it changes the political calculus being made, relative to the advocacy of the larger Yglesias-adjacent crowd that ignores uncertainty.

There is also a lot of uncertainty about what a “moderate” even is, making the argument that candidates should move to the center borderline tautological. As you’ll see, Yglesias’ argument essentially boils down to “good candidates perform better.” I assume we can agree on that, too.

In addition to some other interpretative errors,² I think the biggest problem here is a classic case of people focusing on the point estimate and ignoring the uncertainty interval. This is the biggest mistake.

---

Stop! Read this before you Tweet!

My first article got taken out of context by people who did not read the full analysis. Because I don’t want people to take this article out of context too, here is a short box of what I’m *not* saying:

- I’m not anti-moderate.

- I am not denying heterogeneity in outcomes.

- I am not lazily relying on my own formulation; I have tested all findings with other data sources for robustness, and I get the same results.

- I _am_ simply pointing out that parties have higher-impact levers to pull (fundraising, experience, research, district fit) when uncertainty about the value of moderation is high.

---

The rest of this article is paywalled for paying members of Strength In Numbers. I use the paywall for two reasons: (a) to reward people who have paid with premium content that others cannot access, incentivizing them to stick around, and (b) to incentivize readers who aren’t currently paying to do so.

If a paid membership is out of the question for you, reach out to me via email, and I can give you a free trial. Otherwise, please support the site. We are planning to expand our data and editorial offerings soon, and membership growth is key to our plans.

Alternatively, because I recognize this is in the public’s interest to read, I will unlock this article if 20 new people sign up for a paid membership.

---

Moderation is not the only variable we care about