Roughly a month ago, Atlantic senior editor Adam Serwer authored an article that MSNBC's Lawrence O'Donnell soon hailed as "mandatory reading."  Titled 'The Nationalist's Delusion," the piece challenges the popular narrative that Trump's electoral triumph was propelled by the economic estrangement of white working- and middle-class voters.  Rejecting this account, Serwer holds pervasive and deep-seated, if implicit, animosity toward non-white minorities as the ultimate fillip of the Trump phenomenon.  To borrow from MSNBC's Van Jones, the 2016 election outcome was, in Serwer's telling, just one big case of "whitelash."  Concerns over lax immigration policies, the flight of blue-collar jobs, Islamic terrorism (and obscurantism thereof), and an expressively stifling culture of political correctness were all a pretext for the maintenance of white supremacy and racial inequality.

A key data point Serwer draws on to advance this claim is Trump's "sweeping victory" across all income categories of white voters:

Trump defeated Clinton among white voters in every income category, winning by a margin of 57 to 34 among whites making less than $30,000; 56 to 37 among those making between $30,000 and $50,000; 61 to 33 for those making $50,000 to $100,000; 56 to 39 among those making $100,000 to $200,000; 50 to 45 among those making $200,000 to $250,000; and 48 to 43 among those making more than $250,000.  In other words, Trump won white voters at every level of class and income.  He won workers, he won managers, he won owners, he won robber barons.  This is not a working-class coalition; it is a nationalist one.

Incidentally, in a veritable "white male privilege coming out party," neo-conservative writer Max Boot recently credited both Serwer and these figures, more specifically, in helping him finally "see the light" of America's endemic racism and xenophobia.

There's just one problem: these data, by all indications, are spurious.  And, as a doctoral student who's been studying the reputable American National Election Studies (ANES) 2016 election survey for almost a year, the author found the data relatively easy to fact-check.

Below is a tabulated output of the ANES results.  For ease of interpretation, the winning vote margin is in boldface.

ANES 2016 Election Results: White Votes x Income Group

Note: ± margin of error listed in parentheses.  Each was calculated using 95% confidence intervals.

As shown, Serwer's claim begins to unravel as we surpass the middle-income bracket.  From $175,000 and onward, it's not even close: non-Hispanic whites voted for Clinton by sizeable margins (10.76% and 11.9%).  Including Hispanic whites (40.4% of whom voted Trump), the spreads are even wider (13.87%, 12.76%).  Such is consistent with the working- and middle-class "revenge against the elites" thesis but incompatible with that of an across-the-economic-board "whitelash."

Three objections might be made here.  First, the upper-income sample sizes are comparatively small (hence the wide margins of error).  Although ANES employs random sampling that, in theory, should ensure representativeness, a larger sample size is always preferable.  Second, one might argue that ANES's breakdown of the income groups does not exactly correspond with those mentioned in Serwer's article (for example, there is no $200K category).  While true, such hardly supports Serwer's claim that Trump bested Clinton among whites of every income category.  One need only point to the $250K+ bracket, which was included in Serwer's figures and which shows Clinton winning pretty handily.  Finally, how can I be so sure it's not the ANES data that are biased?

Addressing these concerns required that I first determine the source of Serwer's data.  For whatever reason, Serwer does not provide this information in the article.  I thus went ahead and emailed him.  To my delight, he responded fairly promptly and – after some pushback (because nobody likes being told they're wrong) – indicated that the data came from the Edison Research national exit poll, which he kindly attached.  Given their noted biases (e.g., self-selection, incomplete demographic data, exclusion of early voters) and the fact that roughly a year has passed since the 2016 election and more complete datasets are thus now available, it's puzzling that he opted for real-time election day polling.  Still, I couldn't assume a priori that the Edison data got it wrong.  Cross-replication, preferably on a larger sample, was needed to get at this.  I thus turned to Harvard's massive (N=64,600; roughly half of which voted) Cooperative Congressional Election Study (CCES) for a closer look.  In the end, my suspicions were confirmed:

CCES 2016 Election Results: White Votes x Income Group

Note: ± margin of error listed in parentheses.  Each was calculated using 95% confidence intervals.

As in the ANES, we once again find that white support for Trump trends markedly downward as we leave the lower- to middle-income brackets.  Beyond the $100,000 mark, white voters increasingly turn out for Clinton.  At $200K and onward, Clinton beats Trump in a landslide – by 15% and 17.84%, respectively. For some perspective, I've tabulated the exit poll figures cited by Serwer:

Edison Research Exit Poll Results: Whites x Income Group

As you can see, whereas the results of the ANES and CCES are substantively similar, the Edison exit poll Serwer cites hardly comes close.  Given how far they fall from the CCES's margins of error, the statistical probability that the Edison figures are accurate is exceedingly low (1).

So how do we explain this glaring discrepancy?  My best guess is that the Edison data – gathered in real time from those who agreed to be interviewed after exiting the voting precincts (hence the risk of self-selection bias) – oversampled white Trump voters.  Again, given the availability of alternative data sources, it's perplexing that Serwer relied on an exit poll.  But what's even more bemusing is that Serwer didn't bother to crosscheck it with other datasets.  While I won't go as far as to accuse him of cherry-picking what fits his narrative, such – at the very minimum – smacks of journalistic laziness with a hint of confirmation bias.

To be sure, there are many other issues with Serwer's assumptions that, for brevity's sake, I'll have to mostly leave for subsequent writing.  But very briefly, in foregrounding racial resentment, Serwer downplays the importance of many other variables that account for Trump's election.  For example, my own research – which I'll soon be submitting for publication – finds that even after controlling for various measures of prejudice (e.g. sexism, hatred of minorities, racial resentment), issue attitudes (economic discontent, immigration, refugees, etc.), and ideological orientations (authoritarianism, social dominance orientation), opposition to political correctness significantly positively predicts voting for Trump.

In other words, the hypothesis that such sentiment is all but a guise for white bigotry finds no support in the data.  Thus, while Serwer argues that Trump's presidency was made possible by racial resentment (2) and white ethno-nationalism, one could just as easily point the finger at "PC fatigue."  Instead, Serwer's explanatory model conveniently includes only those variables that absolve the left of culpability.

Why does any of this matter?  It matters insofar as Serwer is promulgating an inherently polarizing depiction of reality that rests on (in all likelihood) erroneous data and gross simplification.  Worse, this depiction is now being billed as "mandatory reading" to millions of Americans.

Genuine racial hostility undoubtedly motivated a minority subset of Trump voters.  But as a liberal alienated by the toxic identitarian political direction of our country, I worry that these broad-brush "whitelash" portrayals allows the left to demonize and dismiss the real concerns of millions of Americans.  Should this continue, the appeal of the Democratic Party will forever be confined to cosmopolitan bubble-land.

Zach Goldberg is a Ph.D. candidate in political science at Georgia State University.

(1) In fact, they even fall outside a wider margin of error, derived from a 99% confidence interval.  This means that the odds that the Edison figures accurately reflect the true distribution are less than 1 in 100.

(2) Incidentally, the degree of anti-minority sentiment among Trump voters also appears to be overstated.  According to the ANES out-group feeling thermometer data (scored along a 0-100 scale), Trump voters, on average, evaluated blacks (mean = 63.8) and Hispanics (mean = 64.4) in the "warm" direction.  By comparison, the mean ratings for both of these groups among Clinton voters were 75 and 73.9, respectively – warmer, but not as much as one might expect.