Response to a Dylan Matthews article on Vox about bipartisanship
This is a disproportionately long takedown of an op-ed. It is relevant for judging Vox’s EA journalism and figuring out EA norms on politics and journalism. It also has a bit of information on American politics and culture if you are interested in that.
This article by Dylan Matthews (“The ‘why can’t we all just get along’ theory of politics”) was published two weeks ago under the “Future Perfect” column of EA-associated journalism, funded by the Rockefeller Foundation. Most of these articles have been good but this one is certainly the worst out of all that I have seen (n=25 or so, from multiple writers) and I believe it has negative expected value.
The article gives a basic line of argument that has been relatively common in Western political culture since 2016 or so—that we don’t need to try to get along with our political opponents. He nominally argues for this in cost-benefit terms. Indeed, the practical cost of making friends with extra people is perfectly real and relevant:
For average citizens and voters, [going out of your way to make friends with political opponents] is another burden to add to the list after work, schlepping the kids to and from school, taking care of elderly family members, and attending PTA and church/synagogue/mosque meetings, etc.
What the call for cross-partisan friendships asks people to do, essentially, is to make an altruistic sacrifice of time, perhaps money, and definitely emotional energy, in an attempt to heal our politics.
And that effort could be spent more productively. This makes perfect sense. Yet while this claim is defensible and reasonable, the article heavily pushes a much more poorly supported implicit point of view that Matthews’ side of politics is uncontroversially the correct side, that there is little or nothing to learn from others, and that the others are morally repugnant. In doing so, it goes well beyond the nominal argument for shrewdness and displays active antagonization instead.
First, Matthews begs the question by framing the main goal of bipartisan friendship as “attempt[ing] to persuade participants of certain specific, socially beneficial beliefs” and presenting his own views as the default good ones. This attitude is clearly an obstacle to having a mutually respectful friendship. Now we can certainly rank beliefs in terms of their expected social benefit. But to explicitly push this attitude and our own credences into our friendships with political opponents instantly destroys the necessary premise for any such friendship to achieve its primary teloses—making the political system work better for multiple points of view, or learning more from your opponent.
Matthews also commits a basic decision-theoretic error here in ignoring the robustness of our beliefs and the value of information. If our own political beliefs on any important issue are not very robust, then high value can be had by investigating unknown evidence to refine our expected value estimates. Matthews is not only assuming that his beliefs are beneficial, he is assuming that other people’s arguments lack enough potential merit to justify listening to them.
As for the beliefs in question, Matthews says the following:
Insofar as individual beliefs are deforming our politics, the beliefs that do so the worst involve bigotry — especially, in the American case, racist sentiment.
As we shall see, Matthews’ justification for this assumption is negligible. Of course it is by no means an uncommon or unsupportable assumption, but for Matthews to casually assert it with such high confidence and such poor justification reaches a criminally bad level of epistemic hygiene.
Matthews supports it only with a hyperlink to a story by another Vox journalist, German Lopez, who argues that “Trump won because of racial resentment.” Yet that article has major problems, both in its own right and in terms of the validity of its evidence for the purpose of Matthews’ citation.
Its first source, a Washington Post article by Fowler et al has data showing that Trump voting was predicted by white vulnerability, and white vulnerability is in turn predicted by “racial resentment”.
I do not see a paper showing Fowler et al’s actual analysis (!), but the survey data they used suggests a possible problem of soundness, albeit a relatively minor one that we may forgive Vox for overlooking. Their race attitude questions seem to be from Q20 to Q51, the majority of the survey. Their ‘white vulnerability’ questions are, according to the WaPo piece, Q29-Q31. But Q28A, Q28B, and Q32-Q35 can just as plausibly be construed as questions about white vulnerability, and nothing in the survey suggests preselection of Q29-Q31. It’s possible that the authors cherrypicked here, so that the idea of white vulnerability and the questions used to constitute it were chosen from other equally racial alternatives ex post, meaning that the correlation would probably be unrepresentatively high as well as overestimated (analogous to the Optimizer’s Curse). Similarly, it’s not apparent whether the authors used all of Q20-Q28 + Q32-Q51 to measure racial resentment or just some of them; again there is potential here for ex post fudging.
Looking up racial resentment scales online, I find this undergraduate’s article in a nonpartisan political journal summarizing and commenting on results from two studies which show that such scales can mistake general conservatism for racial resentment. Many of Fowler et al’s race questions seem to be relevantly similar, which suggests that they are not valid measures of racial resentment. More interestingly, Fowler et al presumably did not find that racial resentment predicted Trump voting; they merely found that it predicted the white vulnerability that predicted Trump voting. If they had found a significant direct connection, surely they would have mentioned that. The omission suggests that the connection is specious, e.g. resentment only predicted vulnerability among groups who were voting for Trump anyway or something of the sort. I think any good social scientist would agree that prediction is not transitive across social phenomena when you are controlling for confounding variables.
So what do Fowler et al really tell us? That Trump voting is predicted by Q29-Q31, namely the beliefs that 1) “through no fault of their own, whites are economically losing ground today compared to other racial and ethnic groups,” that 2) “discrimination against whites has become as big a problem as discrimination against blacks and other minorities”, and that 3) minorities overtaking whites by 2050 would weaken the country.
The second study cited by Lopez is Schaffner et al, which shows that Trumpism is strongly predicted by the combined metric of 4) rejection of the claim that “white people in the U.S. have certain advantages because of the color of their skin,” 5) endorsement of the claim that “racial problems in the U.S. are rare, isolated situations,” and 6) not being “angry that racism exists.” Other evidence cited by Lopez shows that Trump voters were vulnerable to subconscious political bias by racial triggers.
Still more arguments are given by Lopez but they are going to be ignored here because even if we take them at face value they do not give clear support to Matthews’ assertion that racist sentiment among voters elected Trump. Now, assuming that the basic results of these studies are correct and assuming that Lopez has not cherrypicked evidence to fit his point of view, let’s stop and review Matthews’ position.
He has not included any rationale for considering Trump to be a particularly bad president in the first place, although many brief arguments can be found scattered across the writings of himself and others on Vox so we can perhaps leave that aside.
He completely ignores the harms that can be caused by all sorts of other beliefs. Even with the assumption of Trump being a particularly bad president, it is non-obvious that the beliefs that elected him are more harmful (especially on a per-person basis) than beliefs which create other political problems. For instance, consider issues of farm legislation, obligations to future people, foreign aid, and support for Effective Altruism.
And to summarize the above review of Lopez’s article, all that we can say with rigor about the sentiment at play among Trump voters is that it involves the six aforementioned beliefs (or maybe some subset) as well as implicit racial fears and biases.
Matthews believes that it is painful and costly for a minority person to have to listen to someone who has the above attitudes. But there are three problems with this. First, it is not clear how they will make someone unusually painful or difficult to befriend. After all, they are not sufficient for the orthodox rigorous definition of racism, Jorge Garcia’s volitional account, which is especially the most relevant when we are talking about interpersonal relations. (Note: even if you believe that the other racial questions given by Fowler et al are predictors of Trump voting, most of them are similarly very weak in this regard, and none of them are particularly strong.) Going by personal experience, I am Armenian, yet if I meet a Turkish person I will not feel unduly emotionally troubled merely because they are not sufficiently angry about the Armenian Genocide, and so on for similar analogies.
Second, the correlations still do not show that Trump voters have a sufficiently severe case of these traits so as to make relationships sufficiently costly. What is the threshold?
Third, Matthews uses the example of a Muslim woman hearing about a travel ban as an intuition pump to show how costly it is to befriend a conservative. Yet the conservatives may only have racial sentiments with respect to some minority groups and not others, and those of us who are making friends with Trump voters may not be minorities at all. In that case, the force of Matthews’ argument dissipates further. In line with his generous interpretation of the phrase “racist sentiment”, Matthews characterizes support for the travel ban as “hate” for a Muslim woman and her family.
Matthews then argues that conflict, such as that on Twitter, can be beneficial. Matthews looks to Bruneau and Saxe as supposedly showing that conflict in the form of “airing grievances” works in making people more moderate, but reading their study reveals that, far from being reminiscent of Twitter mudslinging, it actually appears to be constituted by the sharing of carefully-written political views followed by a rudimentary version of the Ideological Turing Test (!). That sounds exactly like a bipartisan friendship to me. His second point is his personal recollection of how white journalists like himself learned from blacks on Twitter. But it should be apparent by now that Dylan Matthews is probably not someone who has racist sentiments against blacks, making the anecdote invalid.
Matthews suggests that the value of conflict provides evidence against the value of bipartisan friendships, yet it is a weak inference. It could plausibly be the case that both friendships and conflicts are useful, maybe in different contexts. Indeed, the other study that he cites here clearly acknowledges that “cooperative behaviors are socially and economically beneficial across a large variety of contexts.” Matthews also gives a metanalysis which says that friendly contact has a real but weak effect in reducing racial prejudice (the latter being measured in many ways across a variety of studies); he leaves out the more optimistic metanalysis by Pettigrew and Tropp but that one is older so we can let it pass.
So, for the question of the value of bipartisan dialogue, Matthews:
Ignored one of the main goals of bipartisan dialogue, systemic changes to political culture
Ignored the other main goal, the possibility of learning from one’s political opponents
Did not include justifications for his view that Trump’s election had particularly bad consequences
Provided no comparative argument to show that Trump’s election and the relevant beliefs are the most harmful issues to consider
Provided very weak support for his claim that Trump voters tend to hold sentiments that will often offend well-meaning minority friends, with the relevant studies turning out to be investigations of attitudes which do not seem likely to cause this effect
Ignored non-minority readers and readers who are of minorities that are not targeted by Trump voters’ attitudes
Provides mediocre support for his claim that cooperation and contact lack sufficient efficacy to prefer them over conflict
Then, for the EA community itself, this article exhibits a destructive attitude as it suggests that internal friendship and cooperation with conservative EAs is a waste of time, even though these relationships are vastly more productive than an ordinary friendship between a Republican voter and a Democrat voter. Remember that this is not part of the regular Vox writing—it is labeled as part of the “Future Perfect” series of EA journalism.
Had Matthews argued carefully on a broader point that bipartisan friendship is ineffective at improving the political system, this problem would not obtain. Yet his immediate leap to narrow partisan point of view, the poor quality scholarship (even by op-ed standards) and his epistemic unfairness on the very topic of friendship with political rivals combines uncharitability with irony to make it strongly exclusionary. The article heavily contradicts Ozymandias’ recommendations for welcoming conservatives in EA, although that was posted only yesterday. And we should recall that Matthews once wrote an article that indicated negative attitudes towards the prevalence of whites in EA—it is reminiscent of some of the attitudes sampled by Fowler et al and Schaffner et al. This looks like a double standard. Of course, some people believe that racial issues should be held to different standards based on the history of a particular race. But even if you take that view, it should only apply to institutional questions of whole racial communities which are intrinsically connected with racial history. When we are working with small teleological communities and individual people who have far more relevant characteristics beyond their racial identity, we need to remember to treat them as such.
And lastly, what is Matthews trying to achieve with this article? Are people spending lots of time and money trying to make friends with political enemies just for the sake of fixing America? The nominal rationale for making this argument—that it will save young altruists from expending lots of time and effort on these civic duties—seems specious given how rare of a phenomenon that actually is.
The article got minimal attention on social media, and I’m probably making conservatives feel worse about EA overall by bringing it to their attention. And it probably seems weird that I’ve spent so much time and effort analyzing a relatively unnoticed op-ed from a journalist. But I think it’s important to set a clear early example for what sorts of behaviors are considered good or bad, and to build trust that some of us are monitoring this stuff and holding it to serious epistemic standards.
When Vox launched I was very excited, as I thought it would be a good source of high-quality journalism, even if they did censor authors for having the wrong conclusions. However, it seems like virtually every article, even when otherwise high quality, contains some unrelated and unnecessary jibe at conservatives—an unusually direct example of Politics is the Mindkiller. Perhaps this lead to their being in something of an echo chamber, where conservatives stopped reading?
Here’s a recent example, to help make the above more concrete:
1) Trump signed a good law this week. Yes, really. - why does this need the snark in the title? The meaning would have been clearer, and less insulting, if they had just written “Trump signed a good law about HIV this week.”
I worry about this in general with Future Perfect. This behaviour is not something the EA movement wants, but if Future Perfect ends up producing a very large volume of ‘EA’ articles, we risk getting tarnished by association.
“Trump signed a good law this week. Yes, really.” presents conflict: here’s a person who you usually expect to be doing harmful things, and here they are doing something good. It can’t make that hook without assuming something about their readers, and the hook draws people’s interest. It’s not an “unnecessary jibe”; it’s the sort of thing that draws far more interest than a headline like “Trump signed a good law about HIV this week.”
It’s not a tradeoff I would make in my writing, but Vox is a left-leaning outlet and it seems pretty reasonable to me for them to write for a left-leaning crowd.
That’s a good point, but I think Larks is annoyed that they do make that tradeoff—they’re okay putting in jibes at conservatives because it potentially helps them with leftists.
My impression is that article titles are often not chosen by journalists, so politicised titles are a risk inherent to writing a column for Vox—the actual article on PEPFAR didn’t seem problematic to me.
Sure, that’s why I criticized Vox, not the individual author. I suspect the author did not complain about the title though.
Note: I have a feeling that ‘policing tone’ is an annoying meme for a forum and something more appropriate for moderators than for readers so I’ll post this one and default to refraining from doing it again.
Quick few thoughts on the tone of this, feel free to ignore if it doesn’t change your mind:
This part, right at the top and at a few other points, I felt a little uncomfortable. If I were the author and I read this, I feel like I would feel more ‘attacked by my allies’ than ‘constructively critiqued’.
I feel like some quick changes to the tone, particularly early on (e.g., ‘I found this one distasteful’ rather than ‘this is the worst I’ve seen’) feels less aggressive. Perhaps adding an extra paragraph at the beginning saying a few positive things about their column in general (if you have those views) and saying that you only mean your comments in a constructive way. Personally, that would be enough for me to take the feedback well. Maybe no one on their team reads it, maybe one person reads it and forwards it to the whole team. Seems worth assuming the latter is the case and it’s that scenario that prompted me to make this comment.
Given in particular that Future Perfect is not funded by donors that explicitly identify with EA ideas and that run by Vox, my quick guess is that careful constructive criticism is far more valuable / low risk than more assertive / slightly aggressive criticism (apologies if I’m already preaching to the choir here). I’m currently still really glad that Vox, Ezra etc., have chosen to do this column taking lots of EA ideas into account.
Funnily enough I had a similar opinion about one of the mobile thumbnails for their anti-mars piece. The thumbnail read “Elon wants to go to Mars, here’s why that’s a bad idea” which didn’t seem worth it.
For whatever it’s worth, I read “I found this one distasteful” as more aggressive than “This is the worst of the 25 articles, most of which were good”. One article has to be the worst after all, and so that sentence alone might not even be a real criticism. I see the actual criticism in the next half-sentence “and believe it has negative expected value”, which feels to me pretty non-aggressive.
I understand disagreement about how harsh or gentle of a tone is appropriate here, but we must at least accept the clear expression of extreme rankings lest we lose the ability to share meaningful credences. We should not make it impossible or overly difficult to say that something is the worst, or that we are certain about that fact, because sometimes a thing really is the worst (SOMETHING must be! It’s trivially true, assuming ordering!) and losing that information will bias us. If you ever find yourself writing from a similar position to mine, I’d urge you to find a graceful way to express your belief that something has negative expected value and is inferior to all the other things, and make sure that you aren’t moderating yourself to expressing positivity or uncertainty which you do not really believe.
I want to make it clear that I would express these beliefs with more grace and cushioning to an author who I could give the benefit of the doubt (e.g., Kelsey Piper, or a source I had never seen before). My approach here is partially informed by Matthews and Vox’s track record, they have been criticized before.
I hope my early, explicit statement that most of their articles are good makes it clear that I don’t wish for Matthews or Vox to be run out of this whole thing. And hopefully it’s implicit that I don’t really worry about similar articles that they write in non-EA contexts, I just file that into the bin of permissible propaganda. So all I have to do is lead them to coming up with the very easy compromise of keeping this sort of thing on the other parts of their site, which means that (a) a stiff response won’t back them into a corner, and (b) I don’t have to complete the Sisyphean task of truly changing their minds about conservatives in America.
Good point, but Vox feels this risk as well, which is why a response like this will encourage them to take the easy compromise rather than face the risk.
(Also, since Matthews’s own opinion is that across-the-aisle friendship is overrated, and that “airing grievances” can be a good learning experience, surely I get a bit of extra leeway here.)
I share you irriation with this article. This struck me as a normal opinion Vox opinion piece, which never should have been postet on Future Perfect.
I think some of your points of criticism might be explained by the fact that we had to/wanted to keep the article below a certain length. But i also believe that when Dylan writes for Future Perfect about such political topics, he should make sure to argue every point carefully as well as be especially rigorous and careful in his arguments.
Even if your policy views are correct, having friends on the other side of the aisle will do wonders for your predictive abilities, which should influence how you vote in party primaries where electability is at issue. I’m a staunch Democrat currently living in a similarly liberal area but born and raised in a much more conservative area. I was always more bullish on Trump’s odds than my friends here, and every time I hear them say they can’t understand how he got elected or how he’s still popular with the base, I wonder what other easily avoidable mistakes they’re currently making. And there’s no special magic in my improved predictive ability; I just talk to my grandma regularly.
Another problem with this article: implicitly making inferences about populations based on what you read on Twitter. There’s no particular reason to believe that Twitter users are a representative sample of the population (or that your Twitter feed is a representative sample of Twitter users, even).
How did Dylan Matthews become associated with EA? This is a serious question—based on the articles of his I’ve read, he doesn’t seem to particularly care about some core EA values, such as epistemic rationality and respect for “odd-sounding” opinions.
He wrote one of the early articles about earning to give, as far as I know “became associated with EA” in the same way that other EAs do—by getting interested in the ideas and starting to act on them. For example, he donated a kidney.
You make many good points here! One note: I’d suggest changing the title of the piece, which is quite ambiguous at the moment. Maybe something which refers to the topic of bipartisanism, or to journalism that isn’t careful enough with logic or statistics?