Politics, Science / Tech

Beware the Emerging Field of ‘Trump Studies’

In a recent article at the usually excellent Washington Post “Monkey Cage” blog, Thomas Wood attempts to argue that, “Racism motivated Trump voters more than authoritarianism.”

To make his case, Dr. Wood leads with a chart showing that, contrary to popular narratives (like this one), Trump voters actually seem less supportive of authoritarianism than those who voted for Romney or McCain.

He then turns to racism—and in particular, data exploring partisan differences in endorsements of “symbolic racism” among whites. Here is the chart:

Now, for the sake of intellectual charity we’ll set aside questions about the extent to which his authoritarianism and symbolic racism scales are accurate measures of respondents’ mental states. We’ll just grant, for the sake of argument, that these are accurate measures of racist and authoritarian tendencies, and that the survey population is sufficiently representative that one could make sound generalizations about Trump and Clinton voters on the basis of his study. With these concessions in mind, what does the data show us?

One thing readers may notice about the second chart, which Dr. Wood makes the centerpiece of his argument, is that the gap between Democrats and Republicans is larger in 2016 than in previous cycles—suggesting race likely played a much bigger role in the 2016 election than in previous years. This is an interesting finding, to which we’ll return in a moment. In the meantime, however, readers should note that Dr. Wood’s data does not suggest that Trump voters were especially motivated by racism. In fact, it points in the opposite direction.

Among both Democrats and Republicans, whites were less likely to endorse symbolic racism in 2016 than they were in 2012. In other words, whites who voted for Trump are actually less likely to endorse symbolic racism than those who voted for Romney. The same holds for Democrats who voted for Clinton relative to those who voted for Obama (in either term).

So, not only are white Republican voters less authoritarian than they have been in previous cycles, it seems as though they’re less racist too. This is truly a newsworthy and provocative finding about Trump voters that would have certainly served as the springboard for rich discussion and further research.

Unfortunately, Dr. Wood never even addresses this phenomenon. He ignores Republican voters’ longitudinal trend and instead focuses on the gap between Democrats and Republicans on his chart—in the service of a conclusion his data does not support.

Misplaced Emphasis

Dr. Wood asserts, “Since 1988, we’ve never seen such a clear correspondence between vote choice and racial perceptions.” He goes on to concede that, “The biggest movement was among those who voted for the Democrat, who were far less likely to agree with attitudes coded as more racially biased.” This is an understatement.

In fact, the Republican symbolic racism trend was actually towards convergence with the other party. Change in Democrat sentiment drives literally the entire divergence effect he’s referring to…in an article that is supposed to explain Trump voters. This is not a good look.

To elaborate: the gap between Democrats and Republicans was caused, not because Trump voters were more likely to endorse symbolic racism than Republicans in previous cycles (again, the exact opposite is true), but instead because Democrats went out of their way to emphasize just how non-racist they were in 2016—(ostensibly) rejecting symbolic racism more powerfully than they ever have over the 28 years of data presented. They appeared to reject racism far more in this cycle than they did when an actual black candidate was on the ballot–from their own party, no less–the nation’s first black president, Barack Obama. Indeed, Dr. Wood’s data suggest that racist sentiment among Democrats actually increased in 2008, and more-or-less held at that higher level in 2012. This renders the unprecedented drop among Democrats in 2016 even more interesting.

Indeed, if Dr. Wood wanted to focus on the change in the partisan gap among symbolic racism indicators, given that Democrats are the ones driving that effect, the article should have been about them. Such a story would not suggest that Trump voters are especially motivated by racism—again, his own data confound that claim. Instead, perhaps Clinton voters were particularly motivated by…anti-racism? (This is disputed.) He could even still say something like “moving from the 50th to the 25th percentile in the symbolic racism scale rendered whites 20 percent more likely to vote for Clinton” if he wanted–only now he wouldn’t be making misleading insinuations about Trump voters in the process.

On the other hand, if Dr. Wood really wanted to make the article about Trump supporters instead, he should have been far more responsible in describing his results: the “unprecedented” partisan divide among symbolic racism indicators should not have been part of the story at all. Given that it is entirely a product of a severe aberration among Democrats, it isn’t clear what, if anything, can be inferred about Trump voters from the gap.

With regards to the relative significance of authoritarianism and racism, he could have more legitimately asserted that, “Trump voters seem to be more strongly influenced by racism than authoritarianism–although they were less motivated by either of these factors than Republican voters have been in previous cycles.” This latter part–which he omitted–is the significant story in his data on Trump voters (as compared to those who voted for other Republican candidates).

Dr. Wood should have also been explicit about how much of the total variance in voting behavior was explained by the factors he focused on, as reflected in the R2 of his regressions. I suspect a majority of the variance would remain unexplained by his authoritarianism and racism scales.

Following these standard best-practices would have had the added virtue of drawing readers’ minds to the following question: “if both racism and authoritarianism were less appealing to white Trump voters than to previous Republican constituencies—and if they don’t explain most of the variance in voting behavior anyway–is there some other omitted variable that may explain Trump voters even better?”

Omitted variables

My informed guess is that one factor that would likely predict support for Trump over Clinton far better than racism or authoritarianism would be something like, “opposition to the status quo” or “opposition to the political establishment” or “desire for major social change” or “distrust of major social institutions.” Indeed, Dr. Wood’s own data suggests that the antipathy between Trump voters and social elites was mutual. To quote: “the degree to which the wealthy disdained the 2016 Republican candidate was without recent historical precedent.”

But of course, social researchers overwhelmingly identify with the Left. They usually think of opposition to the political establishment or status quo as positive traits, and claim to champion those disdained by the rich and powerful. Therefore testing these omitted variables would require openness to the possibility that the very people progressives seem to despise most of all (Trump voters) might actually be motivated by–or outright embody–some trait they deeply admire.

In far too many instances, it is just assumed that Trump voters must be motivated primarily by some “bad” factor (ignorance, authoritarianism, racism, sexism, etc.); the task of social researchers is reduced to merely nailing down which of these morally-contemptible traits best explains the data. Meanwhile, by failing to discuss the extent to which Clinton voters were also motivated by negative impulses (the same ones or others), it is implied that their votes must have instead been guided primarily by benign traits such as love for their fellow man, patriotism, rational self-interest, etc. Dr. Wood’s essay was an exemplar of this genre.

Statistically, he was able to paint this picture because the decline in authoritarianism was greater than the decline in racism, while the decline in endorsing symbolic racism among Democrats was larger than the decline among Republicans. So even through there were actually declines across the board, symbolic racism could be made to seem relatively more significant–if that were the impression one wanted to convey. Of course, pointing to a decline in racism among Republicans in 2016 to demonstrate that Trump voters are particularly racist is an obviously bizarre and incoherent way of interpreting the data. Yet that is precisely what Dr. Wood did.

A Teachable Moment?

How is it possible that a scholar who presumably knows better could make such glaring errors–and subsequently have such obviously flawed work approved by a group of other smart people who also know better?

Although Dr. Wood is a Republican who worked on the Romney campaign, one of the big stories of 2016 was the disdain Romney’s faction of the party felt towards Trump and his supporters. These sentiments are also widely held on the Left, especially within the academy. A convergence of motivated reasoning on his part, and confirmation bias on the part of his reviewers, likely contributed to this weak piece of scholarship being published.

Of course, the essay in question was not just written by a professor of political science at a major state university and vetted by a team of prominent wonks—it was also published in the Washington Post, presumably with some editorial oversight from the paper in the process. Unfortunately, an additional layer of review would be unlikely to help in this instance given that the same ideological bubbles that plague institutions of higher learning also distort the work of many mainstream media organizations.

Given their ubiquity, combatting systematic biases in social research will not be easy.  However, getting a handle on this problem will likely prove necessary for ensuring the continued utility, credibility, and viability of the enterprise.

In the meantime, given how deeply passions about Trump run among social researchers (scholars are people too, after all), the burgeoning field of literature on Trump and his supporters should be approached with a healthy dose of skepticism and a critical eye (for instance, consider the obvious but largely unaddressed confounds of those “white supremacy” narratives about Trump’s election).

Indeed, this is especially important for Trump’s “resistance”: if they want a viable chance of beating him in 2020 (and, as things currently stand, the odds are not in their favor), then they’ll need a sober understanding of who voted for Trump and how best to reach them…even if this means being more charitable towards Trump voters than they feel naturally inclined to be.

Filed under: Politics, Science / Tech


Musa al-Gharbi is a Paul F. Lazarsfeld Fellow in Sociology at Columbia University and a research associate with Heterodox Academy. Readers can connect to his research and social media via his website: https://fiatsophia.org/


  1. Tony Sinclair says

    This article makes the mistake of conflating self-reported survey answers with actions; just because Democrats were less likely to give racist answers in surveys in 2016 does not mean that they “went out of their way to emphasize just how non-racist they were in 2016”.

    Even if we concede this, though, there’s another problem with the argument being made: if Democrats really did go out of their way to emphasize non-racism in 2016, why did their non-racism this alienate people? Why did people vote for Trump instead for the non-racist side? The obvious answer is because they’re motivated by racism. Thus focusing on the way the Democrats changed rather than the Republicans leads you back to the original conclusion that Dr. Wood was arguing: it’s the gap that matters.

    I’m not sure you can actually make specific claims about e.g. whether Democrats in 2008 were actually more racist than Democrats in 2004 without having more detailed data, knowing the variance in each poll so as to make a test for statistical significance, and so on. However, a linear regression on the data in the graphs shows that the general trend over the last 28 years is that Republicans are getting more racist, while Democrats are getting less racist, and this holds true across every survey question.

    • 11411181 says

      “if Democrats really did go out of their way to emphasize non-racism in 2016, why did their non-racism this alienate people?”

      “Why did people vote for Trump instead for the non-racist side? The obvious answer is because they’re motivated by racism.”

      The classic case of circular reasoning masquerading as a priori knowledge, writ large.

      • Tony Sinclair says

        It wasn’t a circular argument, I was just reiterating the problem the article’s reasoning. If you want to talk about a posteriori knowledge, let’s talk about what the surveys show:

        1. Republicans are generally more racist than Democrats, and this has remained true in every election since 1988

        2. In 2016. racism decreased more among Democrats than Republicans.

        The increase in the gap between the racism of Democrats and Republicans means that the correlation between racism and voting is stronger than it was before. Granted, correlation does not prove causation, but there’s a lot of evidence that this correlation is not just a coincidence.

        • 1. It is bad social science to take an effect driven 100% by one group and use it to impute things about some other group. That’s my first point, and it is not really disputable. If this same kind of move were used to feature minorities in an unfavorable light, for instance, it would have been flagged in a journal for the kind of error that it is.

          2. By his own data, Trump voters were less racist than Romney voters. And by some of his measures, they were less racist than McCain voters as well. Similarly, by his own data, Trump voters were less authoritarian than Romney voters, and by some of his measures, less authoritarian than McCain voters as well. This is just a fact. By his own measures, they were less racist. In absolute terms.

          3. Now let’s talk about relative terms: given that there were declines in both racism and authoritarianism, it is bizarre and unnatural to describe Trump voters as being more racist than they have been. Again, you can say racism was “relatively more significant than authoritarianism” as long as you concede that “both racism and authoritarianism mattered less to Trump voters than previous Republican electorates.” This last part is really **the** thing that separates 2016 Republicans from 2012 or 2008 Republicans.

          4. Now, more broadly, at the outset we said that for the sake of argument we can assume that his SRS scale measures what he claims it measures. This was a very generous concession. For the record, there are huge and well-established problems with symbolic racism measures like these–especially when applied to political questions like these. For a great elaboration of this point see, Sniderman & Tetlock (1986), “Symbolic Racism: Problems of Motive Attribution in Political Analysis.”


          So frankly, not only were his inferences really bad, but his method was terrible to begin with.

    • “Why did people vote for Trump instead for the non-racist side? The obvious answer is because they’re motivated by racism.”

      Your statement here is painful in its flawed logic — it presumes a correlation when innumerable variables exist. You are doing the exact flaw of generalities that the author is bringing to light here.

  2. Brett Hawthorne says

    So what if Romney voters were slightly more racist than Trump voters? Is that really the argument you want to make if you’re trying to prove that Trump supporters aren’t racist? You’re making it look like Republicans have always been racist, and that Romney lost because he made the mistake of not capitalizing on it the way Trump did.

    • Only one of the symbolic racism questions even looks like a likely proxy for actual racism, the article is merely taking it at face value.

      • Aapje says


        Indeed. It’s more a test of how much the respondent agrees with one particular narrative. Namely the narrative where the problems of black Americans are primarily caused by continued racism by white people. There is another narrative where black Americans have developed a culture that holds them back.

        Neither narrative is really more racist than the other, as neither claim that any race is inherently inferior to the other or less deserving. They both claim that people with a certain skin color have a problematic culture because of the past. The one narrative claims that white people primarily have the power to change this and the other narrative that black people primarily have this power.

        IMO, it’s much better to discuss these narratives in terms of (the extent to which they are) right and wrong (and by looking at the evidence with an open mind), than to classify them as racist and anti-racist, where certain factual claims are considered morally wrong and others morally right, regardless of whether they are true.

        I would argue that a black supremacist who favors murdering all white people would score very low on the symbolic racism questions, but I would consider this person very racist for his unwillingness to the most basic human right to white people.

  3. Thomas Wood says

    Dear Mr Nusra

    Thanks for your kind summary of my post.

    I’m not overly motivated to address your argument, since your text wildly extrapolates beyond my piece’s very limited inference.

    Instead, would it be possible for you to take down the very poor reproductions of my plots? I’d be thrilled (enthused even!) to provide you with better copies. Your readers might benefit from some exposure to these data.

    Tom Wood

    • I agree, I would like to see better cuts of the tables. They are not labeled on all axis, leaving the reader to make assumptions. Example: red circles and blue circles are…? I guess the reader is to infer based on political colors.

      A link to Dr Wood’s article would be greatly appreciated as well so we might research further with ease.

  4. Claiming something is so flawed you won’t respond to it is a response. A very arrogant response.

  5. Plenty of Democrats voted for Obama simply because he was black. There is plenty of blame to go around on both sides. Many Trump supporters hate inner-city blacks as much as inner-city blacks hate rural whites.

    The issue with Trump and the GOP is economic. GOP policy is directed toward the economic elite, and any GOP candidate will get the wealthy vote everywhere, no matter how despicable. Policy that favors the oligarchs does plenty of damage to every race that is not in that economic class.

  6. Interesting that Wood called the author “Mr. Nusra”, as Nusra is a well known terrorist group. A Freudian slip that could be interpreted as latent racism?

  7. sanjiv says

    This article misses the point. Whether Trump’s voters are racist is, to use the title of one of your sections, misplaced emphasis. Without being intellectually charitable, let us assume that these symbolic scale tell us nothing about the mental states of Trump voters. You know whose mental state we know everything we need to know about? The president’s. It was on full display during his campaign and continues to be so every single day since he has assumed the oath of office. The fact that he was explicitly racist did nothing to deter his supporters’ opinion of them. One has to ask: why? Trump’s voters might or not be particularly racist but Trump most certainly is. The fact that they continue to back him is a reflection of their mental states. I want to hear from Trump supporters in sizeable numbers who lack a college degree, unequivocally condemn Trump’s racism/xenophobia but feel they had to vote for Trump anyway due to other reasons (anti-establishment, economic, Hillary is worse, etc.). Until that happens , I will desist from entertaining ridiculously naive notions like Trump voters might be motivated by traits progressive deeply admire.

    • Ok, how about I refuse to teach my daughters to believe that they can lie, flout the rules, and cheat on tests (debates) and still expect to be given a reward simply because they lack an Y chromosome? How about I refuse to teach my daughters that you don’t vote for President of the US like you do the student counsel where they rah rah about free pizza and 1/2 days for Seniors that they don’t have ANY ability to actually deliver.

      You claim Trump is racist, well so is Hillary by that definition. Ms. “Super Predators” herself! Wanting to secure the borders against an influx of immigrants from the South or of unvetted refugees from the Middle East = “racist.” Here’s a tip, want to know why many voted for Trump? Because FINALLY there’s a candidate who is immune to these defamatory attacks that MANY of us have endured for 8 years. I didn’t DARE to discuss why I disagreed with President Obama’s policies because any disagreement was CLEARLY because i’m Hitler reincarnate. Gee, surprise, the Left’s entire campaign was … “i’m not Trump” because “i’m not Bush” worked in 2008.

      The problem is that the Left ran Hillary — the polling even showed that Bernie would beat Trump — that’s how bad Hillary was! It is no different than the sour grapes from the Right in 2008 saying “Obama only won because he’s black … they voted for him because he’s black” — no, he won because he had a positive, morale building message and the GOP ran Dumpy McCain. I say Dumpy because McCain in 2008 was not the McCain of 2004. My in-laws are all hard-left and THEY figured McCain was a lock no matter who the DNC ran in 2008. I honestly believe that Obama was “given” the DNC nomination in 2008 over Hillary because they expected McCain to be unbeatable and didn’t want to tarnish the Hillary brand. That explains the chaos immediately after he won trying to organize and get things started legislating and it isn’t out of the question considering the behind the curtain look at how the DNC nomination process occurs thanks to Wikileaks.

      Trump ran as Obama, with a positive, morale building message. Take your blinders off and read about the NYU play where they put a male actor in for Hillary and a female actor in for Trump and reenacted a debate both in response and mannerisms. Female Trump still SOUNDLY beat male Hillary, and the very left-leaning audience was in shock because “they got it” once they took off their “but…but…TRUMP!” blinders.

  8. Anonymous says

    I hesitate to even validate this more with a comment but…
    I forgot the part of stats course where you can poke holes in a constructed statistical *argument* with hijacked pixelated pictures of graphs and leading assumptions.

    Believe the original thesis was “Racism motivated Trump voters more than authoritarianism” among a discreet set of ‘popular narratives’, so this is to be superseded with …. maybe the stats are more confusing than this one social scientist was able to elaborate on in one Washington Post article?

    I came here for a good old fashioned character assassination, not a flaccid finger wagging and overly elaborate linkedin meme of “there’s lies, damned lies, and statistics”.

    • Ethan says

      Exactly. His argument boils down to “you should have made a different emphasis?” lawl, idiot, it was a 500 word blog post.

      Also–didn’t the original Woods article come out in April? Why this tripe now?

  9. The questions that are supposed to serve as indicators for racism are absurd. One would have to have such a broad conception of racism for an affirmative answer to any of the questions to render one a racist.

    The author was righr to expose the laziness of Wood’s post.

    Also Imputing large swathes of the country with bad motives probably isnt an effective electoral strategy, which is why the dems will continue to lose

    • “One would have to have such a broad conception of racism for an affirmative answer to any of the questions to render one a racist.”

      The point is to emphasize the difference in answers over time though, isn’t it? The power of ANES is that is large and stable (there is historical data for responses) – not that they really cracked the code for spotting racists. And the racial perception questions had a stronger correspondence to voter outcome than authoritarianism (the limited set of narratives in question). I suppose the criticism could be that those are not the most popularly spread narratives? I personally definitely heard both of those narratives being thrown about though.

      But yes, how lazy of Wood to not offer us the full report of every survey variable tracked by ANES! I will bet the lazy Wood didn’t even do any analyses himself and just grabbed some lo-fi pictures of graphs from the Washington Post! I will bet he is not even a “sociologist at an Ivy League school” as Musa al-Gharbi reminds us he is, how can I trust his public institution drivel

  10. Eric Oliver says

    This piece has so many problems not sure where to begin, but to be brief:

    Data is a plural term, Datum is singular. If you don’t know the difference between the two, you probably shouldn’t even begin to talk about someone else’s statistical analysis.

    The data come from the America National Election Studies. If you don’t know what that is you shouldn’t comment on the measures. Either way, Wood did not invent the survey questions.

    Symbolic Racism did not really drop by any statistically significant margin among Republicans between 2012 and 2016. It did drop, however, among Democrats. What Wood showed was how the gap between the two greatly increased.

    Wood is not a Republican. He is Australian. And if he is a Republican than how can he also be one of these researchers who “assumed that Trump voters must be motivated primarily by some “bad” factor”?

    What you characterize as “Trump Studies” is actually what political scientists do after every election — they compare sets of explanatory variables to see which are most predictive of voter behavior and attitudes. There is a long history in political science of looking at constructs like symbolic racism and authoritarianism the far predate the 2016 election (hence the ability to draw on data from previous elections).

    The biggest predictor of whether or not someone voted for Trump or Clinton was not dissatisfaction with the status quo but whether they were a Democrat or Republican. Your “informed” guess is not very informed.

    Before you make gross and inaccurate generalizations of someone based on a months old, 500 word Monkey Cage piece, you might want to familiarize yourself a little more with statistical analyses and studies of voting behavior.

  11. *yawn*

    When did I suggest Dr. Wood invented the survey questions? He developed the scale. To summarize you, “if you don’t know the difference there…”

    Ditto with the ANES data. I knew the origin, never said I didn’t. Fun fact: it’s on the side of the pictures.

    As far as “how could he be against Trump or Trump voters if he’s a Republican?” as I explained in the article, there are a bunch of those.

    As far as him working on the Romney campaign previously, that bit of information came from the Monkey Cage editor in conversations leading up to this piece.

    As far as Dr. Wood showing the gap increased…uh…yeah. I kind of made that a big point of the article, explaining what could be soundly inferred from that.

    And of course, I know that “symbolic racism scales” predate Dr. Wood and this election. As I pointed out in a previous comment–for the sake of intellectual generosity, I ignored the well-established problems with symbolic racism scales to explain political positions. But I flagged in the comments an article from 1986 explaining why this approach is problematic…so you could say I’m familiar with the fact that these things have been around…

    Fun fact about my informed guess: I was one of the few people who actually did predict Trump would win. On the record, consistently, throughout 2016. Before the primaries were even over. My model keyed in on this dissatisfaction as being the most predictive drivers.

    And by the way, I didn’t say my proposed scale was the biggest conceivable explainer. I said it would explain more than racism.

    Also, with regards to my familiarity with statistical analyses…you could say I know a thing or two. You know…my work in this area was cited in State Department dossiers, academic textbooks, etc. And I’m a sociologist at an Ivy League school…what was it you said…before you make “gross and inaccurate generalizations about someone” ?

    • Thank you for the Lols says

      “Also, with regards to my familiarity with statistical analyses…”

      oh my god I thought this post could not get funnier, but my god. I haven’t heard “I go to an Ivy League” in a while.

      • You’re right. That last part was pretty lame. I wanted to delete it almost immediately, but they don’t let you. I was trying to underscore that you can’t really operate in this department at this level without having a pretty good grounding in stats, etc. Which is true. But yeah…

        I came from University of Arizona previously. And if someone had said something like that to me back then, “I’m at an Ivy” I’d have had the same reaction. Ditto with throwing around things like “I’m a PhD” or “I’m a tenured professor” etc. Stand on your arguments and your work, right?

        I was perplexed/ disturbed that the previous responder was making all these bizarre inferences and assumptions…but definitely a better way to push back on that point.

  12. Also, there is a certain irony in this comment:

    “The biggest predictor of whether or not someone voted for Trump or Clinton was not dissatisfaction with the status quo but whether they were a Democrat or Republican. Your “informed” guess is not very informed.”

    Uh…you basically made the same argument yourself previously. That populism was likely the best predictor of support for Trump. And you defined populism as basically comprising the variables I flagged there as predictive.


    That’s kind of…awkward. Was your informed guess also not very informed? Given that, of course, Democrats will tend to vote Democrat and Republicans for Republicans?

  13. This piece has so many problems not sure where to begin, but to be brief:
    Data is a plural term, Datum is singular.

    You don’t know where to begin – so you begin with that?

    You didn’t fault him on his choice of font or something really important before getting around to his actual argument?

  14. Carl Sageman says

    Very enjoyable article to read, Musa.

    I’m torn over a significant theme our society struggles with today. Identity politics. I noticed you used the term “symbolic racism” (ie. specifically narrowing this as racism toward blacks). That’s identity politics 101. In spite of this, your analysis was unusually unbiased. Putting it another way, identity politics usually throws fact and impartiality out the window immediately.

    I’m not keen on identity politics. It’s being used to cause significant social harm.

    I’m not sure whether to congratulate you or not. Your integrity seems high, your use of identity politics served a specific purpose. Your article wasn’t divisive from my interpretation.

    Definitely a worthy read. Thank you!

  15. John Quiggin says

    The distinction between Trump voters and Romney voters is silly. There is about 90 per cent overlap between these groups, and even the people who switched from Obama to Trump had mostly voted Republican in the past,

Comments are closed.