The estimates imply that racial animus in the United States cost Obama three to five percentage points in the national popular vote in the 2008 election.
The Google methodology is a viable way to grapple with people’s unwillingness to reveal racial prejudice in polls and surveys. Of course, one can criticize it—as Rebecca Greenfield does here—but an even better strategy is simply to see if Stephens-Davidowitz’s results are confirmed by recently published research using other kinds of measures. Here’s an example, from a recent paper in Political Psychology by political scientist Brian Schaffner (ungated; see also the rest of the issue, also ungated thanks to Wiley-Blackwell publishers):
In this paper, I introduce a relatively unobtrusive measure of racial salience to examine whether these initial interpretations are correct. I find that when race was a more salient factor for White voters, they were substantially less likely to vote for Obama and were more likely to think that Obama was focusing attention on African Americans during the campaign. I estimate that the salience of race for some Whites may have cost Obama as much as 3% of the White vote. Thus, this paper indicates that even in Obama’s historic 2008 campaign, African American candidates continue to face barriers to winning White support.
Emphasis mine. The effect is similar in size to what Stephens-Davidowitz find. Schaffner’s measure began with this prompt:
If you had to vote in an election but did not know any of the candidates competing, which pieces of information would be most useful for helping you decide who to vote for?
Then respondents simply asked people to rank several factors, including the candidates’ race, in terms of their usefulness. Schaffner finds that the higher people who ranked race as more useful were less likely to report voting for Obama—a relationship that was most evident among racial conservatives (in this study, opponents of affirmative action).
In fact, the effects in these studies are comparable to the effects in other studies that do rely on explicit measures of racial prejudice. Although it is true that some people may be reluctant to reveal those sentiments in a survey interview, many people will. So Lewis-Beck, Tien, and Nadeau estimate that racial prejudice cost Obama 5 percentage points. Josh Pasek and co-authors arrive at the same estimate: 5 points.
However, a lot hinges on the relevant counterfactual. Stephens-Davidowitz imagines that every media market has a level of racial prejudice equal to that of the least prejudiced market. Pasek et al. study simulate the absence of both pro- and anti-black sentiments.
But another counterfactual is: what if we replaced Barack Obama with a white candidate—say, Hillary Clinton? In this study by Simon Jackman and Lynn Vavreck, they find:
The Democrats would have won the 2008 election regardless of who they nominated, but the average Democratic party nominee from the last 16 years, and either of Edwards or Clinton, would have done better against McCain than Obama, although in Clinton’s case, not by much.
What is “by much”? Half a point. But Edwards would have done about 3 points better—an estimate commensurate with those above.
So contra my very preliminary back-of-the-envelope analysis a few months after the election, race may have cost Obama in 2008. But the hypothetical Clinton vs. McCain race is important. It’s not enough just to imagine a world without racism—whether measured with explicit questions, implicit questions, or Google searches. It may be equally relevant to imagine a world without a black candidate.
UPDATE: See also the study by Benjamin Highton in the comments.