In every election over the last 20 years, Appalachia shifted toward Republicans and the West revolted against the incumbent party’s candidate. These patterns continued in 2008 and 2012, but Seth Stephens-Davidowitz argues that these more recent manifestations are due to racism, since they correlate with the prevalence of racist Google searches. Although this explanation matches the data, it’s not persuasive. If a strong challenger in the West and a weak liberal in Appalachia is so common, there is no reason to assume that a similar phenomenon is attributable to racism once a black candidate is the party nominee, simply because it correlates with data related to racism. Otherwise, one could argue that racism cost Democrats in four of the last eight presidential elections, even though only one of those four elections involved a black candidate.
In his response to my initial criticism of his study, Stephens-Davidowitz asserts that the data doesn’t fit my hypothesis. But he doesn’t offer evidence contradicting my central point: Poor liberal performances in Appalachia and a strong challenger in the West are the norm, thereby raising serious doubts about whether racism is to blame. Since every recent presidential election is consistent with my hypothesis, Stephens-Davidowitz mainly resorts to hypotheticals, congressional elections, and other studies to resurrect his case. These arguments aren’t that persuasive, either.
First, Stephens-Davidowitz argues that polls showed Hillary Clinton and John Edwards doing better than Barack Obama in Appalachia. This hypothetical is relevant under the assumption that, if my argument about a long-term trend is correct, then “any” Democrat should have done worse in Appalachia. But my argument is not that “any” Democrat would have done worse in Appalachia: It’s that a northern, liberal Democrat would do worse—like Michael Dukakis, John Kerry, Walter Mondale, George McGovern, or Obama. This distinction is not lost on Stephens-Davidowitz, who concedes in the long version of his paper that whether a candidate comes from a Southern state is “important, as Southern states have been shown to prefer Southern Democrats.” Edwards is indisputably a Southern Democrat and Hillary Clinton took the name, legacy, and supporters of the most successful Southern Democrat since Andrew Jackson.
Perhaps Stephens-Davidowitz would take issue with characterizing Clinton, a senator from New York born in Illinois, as a Southern Democrat. That’s a reasonable objection, but beside the point: However we classify Clinton, her electoral appeal didn’t resemble a traditional, liberal, northern Democrat. The SurveyUSA polls cited by Stephens-Davidowitz found Clinton trailing in Kerry or Al Gore states like Washington, Oregon, New Hampshire, and Iowa while leading in Arkansas, Ohio, and West Virginia. Other polls showed Clinton losing in Minnesota and Wisconsin. I can’t think of any reason why Obama’s performance should be compared to such an unusual map, unless Stephens-Davidowitz would also like to argue that racism hampered Kerry, since he also outperformed in the West and underperformed in Appalachia with respect to Clinton.
To make his point, Stephens-Davidowitz needs to compare Obama to something like white Obama, not a Clinton or a southerner. Perhaps a better option might be Joe Biden. Maybe association with a black president has tarnished Biden’s reputation with white racists, but it’s worth observing that he’s performing no better than Obama among white voters in recent surveys—and far worse than Clinton.
Separately, Stephens-Davidowitz questions whether the long-term anti-Democratic trend in Appalachia explains Obama’s collapse, since “Obama should have performed worse in 2012 than in 2008 in such areas. He did not.” I’m not sure what Stephens-Davidowitz is arguing here: Obama did do worse last year in Appalachia than in 2008, both in absolute terms and with respect to the country. Obama suffered additional, even catastrophic losses in West Virginia, eastern Kentucky, western Pennsylvania, and western Virginia. In West Virginia, the most racist state according to Google search data, the trend against the president was more severe in 2012 than it was in 2008. Further south, in states like Arkansas, the downward trend continued but at a slower pace.1 In some places, the losses were staggering: Floyd County, Kentucky, trended 29 points to the right compared to the popular vote; Boone County, West Virginia, moved 39 points; 13 points in Jackson County, Tennessee; 19 points in McMullen County, Texas; 11 points in Clay County, Arkansas, and so on.
It’s hard to see why a slower trend complicates the argument that long-term trends explain Obama’s performance. A slower trend doesn’t prove that previous losses weren’t part of a trend, even if Republicans have finally reached the point of diminishing returns in Appalachia. In the past, the decline in Democratic support in Appalachia hasn’t been steady, so a faster trend in 2008 doesn’t prove that racism was at fault. In 2000, for instance, Gore suffered losses as large as Obama’s, even though the trend against Clinton in 1996 and Kerry in 2004 was somewhat smaller. And if a white candidate like Gore can suffer cataclysmic losses in Appalachia, then perhaps we shouldn’t automatically assume that racism is at fault once a black candidate suffers a similar fate.
The anti-Democratic trend in Appalachia should be slowing, anyway. Over the last twenty years, the highland South has shifted 30 points in the direction of the GOP. At some point, the trend has to slow—Democrats are down to about 25 percent of the white vote in some of these states, and it’s hard to argue there’s much more room to fall. One of the primary sources of the shift is coming to an end, too: the expiration of the heavily Democratic “Greatest” generation, the last legacy of traditional Democratic strength in this part of the country
Stephens-Davidowitz also argues that House Democrats should have underperformed in Appalachia, if there was a broad trend against Democrats. Not really. Although liberal, northern Democrats have struggled in Appalachian elections, most Democrats representing Appalachian districts were conservative. In part because parties can choose candidates well suited to their districts, trends in presidential politics don’t necessarily filter down to the House level. Many Southern Democrats hung around until 1994, even though no one could credibly argue that the region hadn’t trended away from the national party in presidential elections. Similarly, when Gore lost colossal ground in Greater Appalachia in 2000, Democrats and Republicans just traded two seats in the House election. And as with the belated Democratic fall in the South, 2010 brought an end to House Democrats in Appalachia—confirming the trend.
While Stephens-Davidowiz raises many questions about Appalachia, he only contends that the anti-incumbency explanation for Obama's performance in the West is illogical, since Kerry was also a challenger and Obama didn’t face an incumbent president. This argument either misreads the empirical record or my claim, so I’ll reiterate both: Since 1980, nine of the ten least-racist states have tended to shift against the party in control of the White House compared to the previous election. This tendency holds in elections when one party has held the presidency for eight years, like 2008, 2000, and 1988. In other words, it is not unusual for the West to shift against the incumbent party in consecutive elections, even after an incumbent president leaves the ballot. This is illustrated in the following chart, where orange shades indicate the few instances when states have moved toward the incumbent party.
Even if one is not persuaded by the compelling empirical evidence supporting the Western anti-incumbent theory, there’s still the fact that Democrats were generally expected to do particularly well in the West in 2008. Recall, for instance, that Denver was already selected as the site of the Democratic National Convention. There’s no way to know whether another liberal Democrat would have done as well with respect to the rest of the country, but it certainly seems possible.
Finally, Stephens-Davidowitz argues that he tested for my hypotheses by controlling for long-term trends. These controls are buried in Table 8 and aren’t explained in great depth, but it appears that Stephens-Davidowitz controlled for the change in Democratic support between 2000 and 2004, as well as 1996 and 2000. This not sufficient. By controlling for the national trend, the model effectively turns a blind eye to Gore’s epic collapse in Appalachia, since he also fared poorly in the anti-incumbent West. Although controlling for the 2004 election helps a bit, since the change in Kerry’s support was roughly similar to Obama’s, the relative stability of the 2004 election means that the control underestimates the long-term trend against Democrats in Appalachia. Perhaps a better approach would have been to independently control for the long-term Democratic decline in Appalachia and anti-incumbency in the West.
Would a correlation remain if Stephens-Davidowitz had controlled for anti-incumbency or Appalachia? I’m not sure. But a different way of getting at this question—excluding the Appalachian and anti-incumbent states—raises real questions.
If racism is hurting Obama, rather than regional trends, then shouldn’t there be a correlation after excluding the Appalachian and anti-incumbent states? According to Google search data, racism isn’t restricted to Appalachia, yet there’s not much evidence of a correlation between racist searches and Obama’s performance outside of the region. This suggests that Stephens-Davidowitz is observing a phenomenon particular to Appalachia and the West—like anti-incumbency and an anti-Democratic trend—not just a simple relationship with racially charged searches. Perhaps Stephens-Davidowitz would argue that there is something unique about Appalachian racism, but that’s not supported by the rate of racist searches.
So how many votes did Obama lose to racism? Even before considering Stephens-Davidowitz’s study, five million votes seems too high. The idea that the president ought to have won reelection in a landslide is a little tough to swallow, and it’s wildly inconsistent with the expectations of the economic-based, fundamental models. President Bill Clinton, with a far healthier economy, won by about the same amount as Stephen-Davidowitz expected for Obama. President George W. Bush, with a modestly healthier economy, actually did worse than Obama. If anything, there’s a case that Obama outperformed the fundamentals in 2012.
Stephens-Davidowitz calculates the cost of racism—4.3 percentage points—based off of Obama’s strong performance in the West, under the assumption that Obama did so well due to low-levels of racism. If anti-incumbency is a better explanation, then the West sets an inflated baseline. Correcting for an inflated baseline probably cuts quite a bit into his total. In fact, the correlation between racially charged Google searches and the change in Democratic support between 2004 and 2008 hinges more on Obama’s strong performance in the West than you might think—notice how the slope of the best-fit line in the chart on the left appears to be driven by the clear trend among the least racist media markets (try putting your hand over the media markets with a racist search rate of less than 40).
Even after correcting for an inflated Western baseline, some portion of the remaining votes lost due to racism must be attributed to long-term trends in Appalachia. Exactly how many votes Obama might have lost due to racism in Appalachia is harder to say. To the extent that Obama’s decline went above and beyond the longer term Democratic in Appalachia, racism could be part of the cause—although that’s tough to pinpoint, since Gore suffered just as much as Obama.
In my original response, I conceded that Obama’s struggles in Appalachia seem excessive. Gore’s collapse coincided with a change in GOP messaging aimed at traditionally Democratic, but conservative Appalachian voters concerned with coal and guns. The severity of Obama's decline doesn’t correspond with a similarly persuasive explanation. From this perspective, racism almost certainly hurt the president, at least to some extent. But by conflating long-term trends with racism, Stephens-Davidowitz’s study exaggerates the losses that Obama suffered due to racism.
The difference between West Virginia and, say, Arkansas might be attributable to Obama’s struggles in “coal country” and therefore could be fairly distinguished from a continuation of the broader Democratic collapse. But it doesn’t really matter: Obama did worse than 2008, both with respect to the country and in absolute terms, in every state of the upland South except Oklahoma.