After two clear victories and “Bradley Effect” no-shows, it would seem that the fear that racism could cost Barack Obama the presidency was overstated. But Seth Stephens-Davidowitz, a Ph.D. candidate in Economics at Harvard, begs to differ; he thinks that racism significantly hindered Obama. With a novel approach that uses racially charged Google searches as a proxy for a region’s racial animus, Stephens-Davidowitz argues that racism cost the president more than five million votes in 2008 and 2012.
I’ll take the under.
At risk of oversimplifying, Stephens-Davidowitz’s argument is that these two maps look similar:
The first map shows the rate of "racially charged searches." Unfortunately, these searches are disturbingly common. Google searches containing the n-word are as frequent as those for “Lakers,” and far more common than “Republicans.” The deepest reds are difficult to differentiate, but the areas with the highest rates of racist searches are upstate New York, West Virginia, eastern Ohio, rural Illinois, western Pennsylvania, southern Oklahoma, and southern Mississippi.
There’s reason to assume that the frequency of racist searches would be related to the presence of racism in a given area, since search volume can be a strong indicator of the presence of an underlying variable. For instance, there is a very strong correlation between the size of an area’s Jewish population and the frequency of searches including the word “Jewish.” Unfortunately, there’s no way to prove a relationship between racist searches and racism, since there isn’t data on the prevalence of racism in the United States, like there is on the Jewish population. But that’s part of why the approach is so novel and potentially useful. There isn’t solid data on racism because people aren’t willing to divulge socially unacceptable beliefs in surveys. Google searches, conducted in privacy and anonymity, might finally provide a window into prejudice.
The second map shows the difference between Obama’s performance in 2008 and John Kerry’s four years earlier. As you can see, the extent that Obama improved over Kerry’s performance seems to correlate with the prevalence of racist searches. Many of Obama’s best states, like Utah and Hawaii, rank at the bottom racist searches; his worst states rank near the top, including the five states where Obama actually did worse than Kerry: West Virginia, Louisiana, Kentucky, Tennessee, and Arkansas. The relationship is far from perfect; it has a correlation coefficient of .24, suggesting that racial prejudice explains about 24 percent of Obama’s performance relative to Kerry. But based on Obama’s tendency to perform best in the areas with the least racial prejudice, Stephens-Davidowitz estimates that racism cost Obama 4.2 percentage points of the national popular vote—more than enough to sway a close presidential election.
The problem? Much of the correlation between racist searches and Obama’s performance can be explained by two important trends in political geography that pre-date the first black presidential candidate. Many liberal challengers would have produced a similar map.
The first trend is the long term decline in Democratic fortunes in the South and Appalachia—the country’s most racist region, according to the Google search data. Take West Virginia’s Ohio County, home to Wheeling, which Stephens-Davidowitz highlighted in a 2012 New York Times op-ed. Between 1992 and 2004, Ohio County swung 19.7 points toward Republicans compared to the country as a whole—an average of 4.9 points per election. Once Obama was at the top of the ticket, Ohio County lurched an additional but utterly average 5.02 points toward John McCain. If Obama’s poor performance reflects the steady continuation of a longer term trend, it’s difficult to confidentially attribute the most recent decline to racism.
In fairness to Stephens-Davidowitz, Wheeling is an exception: The shift against Obama was larger in Appalachia was generally larger than those faced by prior Democrats. But even though the shift was larger than average, it wasn’t necessarily unprecedented. West Virginia and Kentucky swung more decisively against Al Gore than Obama. Presumably, racism was not to blame. Although the movement away from Obama in Tennessee, Oklahoma, Arkansas, and Louisiana was, indeed, larger than any other recent election, the swings against Kerry and Gore weren’t that much smaller.
Stephens-Davidowitz’s argument doesn’t just rest on Obama’s poor performance in racist media markets; it also depends on Obama’s strong performance in media markets with lower rates of racist searches. Most of these less racist markets are west of the Mississippi, including 11 of the 12 least racist states. Should we assume that Obama’s superior performance in the west was because of lower levels of racism? I don’t think so. Many believed that the west was “going in the Democrats’ direction,” as the Washington Post declared, long before Obama won the nomination. For that reason, the party had chosen Denver as the site of its 2008 convention. Western states also possess an anti-incumbent streak: In the 9 elections since 1980, the 10 states with the lowest levels of racism (minus Hawaii) shifted toward the challenging party with respect to the country 78 percent of the time—83 percent if you exclude 1992, when Ross Perot’s third-party candidacy played well in the West. In 2008, seven of the ten states with the lowest levels of racism moved toward Obama—an utterly typical result for a challenging party candidate.
Between the well-established anti-liberal trend in the South and the West’s tendency to favor challengers, any liberal Democratic challenger would have been expected to perform best in the states with low racism and struggle in the states with high racism. The performances of the prior three liberal, Democratic challengers is consistent with this hypothesis. Racism probably didn’t cost Kerry the presidency or save George H.W. Bush from an extremely close race with Michael Dukakis, yet that’s basically what the Stephens-Davidowitz’s model would suggest:
It’s important to emphasize that the slope of the best-fit line is steeper in 2008 than either 2004 or 1988, since the decline in support for Obama in Appalachia was more severe. Even so, the relationship between racially charged searches and the performance of a liberal, Democratic challenger is far too consistent to attribute to racism. As a result, Stephens-Davidowitz’s argument that racism cost Obama five million votes seems far too high: The typical strength of challengers in the West creates an unfair baseline, while the longer-term decline of Democrats in Appalachia explains much of Obama’s problem there.
But it doesn't explain everything. Obama’s collapse in Appalachia, though not too much worse than the fate suffered by Gore in 2000, was indeed worse than the historical trend. Its severity begs for a better explanation than a simple continuation of a longer term trend. In 2000, Gore’s collapse coincided with a strategic decision by the GOP to appeal to Appalachia. As a result, Bush, a southern candidate, elevated coal and gun control as wedge issues. There wasn’t a similar change in messaging in 2008, so Obama’s Appalachian struggles in the Democratic primaries indicate that his challenges went beyond coal or the Democratic Party’s platform. I’m willing, then, to entertain the possibility that racism could be responsible for Obama’s decline in the region. But by a margin of 4 million votes? Not even close.
Correction: An earlier version of this story mischaracterized Stephens-Davidowitz's study as saying Obama lost four million votes due to racism. The study claims racism cost Obama 4.2 percentage points of the national popular vote—around 5 million votes.