Back when Barack Obama was still fighting to become the Democratic nominee for president, there was worry—from supporters and opponents—that the “Bradley effect” would take hold once he moved to the general election. Were white voters voicing support for Obama out of a sense of obligation to egalitarian norms? Would that change when they actually had to cast a vote? In other words, could Obama poll well in the lead up to the election, but then lose as a result of bias on part of voters?
Of course, those fears were unfounded. Obama handily won the 2008 election with a solid majority of the popular vote. What’s more, when it came to white voters, Obama improved on the performance of John Kerry in 2004 and Al Gore in 2000.
But we shouldn’t think that this rules out a racial bias effect in the final election results. It’s possible that Obama underperformed relative to where he could have been absent racial animus. In a newly published paper, Seth Stephens-Davidowitz—a political scientist from Harvard University—attempts to measure the extent to which racial animus had an effect in the 2008 election. His answer? Up to 5 percent of the vote was lost to Obama because of his race.
Crazy, right? Stephens-Davidowitz explains his method:
This paper develops a new proxy for a local area’s racial animus: normalized Google search volume for racially charged language. To test the effects of racial animus on voting for a black candidate, I compare the Google-based measure to an area’s change in Democratic vote shares from the 2004 all-white presidential election to the 2008 biracial presidential election. […]
Google-based measures of racial animus may be less subject to social censoring than those from surveys: Google searchers are online and likely alone, both of which make it easier to express socially taboo. Furthermore, individuals say they are forthcoming with Google. Relative to measures from the General Social Survey, Google-based measures are also meaningfully available at a finer geographic level, use more recent data, and aggregate information from much larger samples.
To ensure that he hasn’t captured an “Obama effect” in Google searches – reverse causation stemming from Obama’s prominence on the national stage in 2007 – Stephens-Davidowitz limits his data to an area’s total searches from 2004 to 2007 that include the word “nigger” or “niggers” (terms used to constrain the amount of data). His analysis includes data from nearly 200 media markets, including over 99 percent of voters.
The results are dramatic. “A one standard deviation increase in an area’s racially charged search is associated with a 1.5 percentage point decrease in Barack Obama’s vote share, controlling for John Kerry’s vote share,” writes Stephens-Davidowitz. The results suggest that, relative to the area witht he lowest level of racial animosity, racial animus cost Obama 3 to 5 points in the popular vote.
Put another way, Obama would have won between 56.7 and 58.7 percent of the vote if the country were as tolerant as its most tolerant area. The results check out even when you control or the population of African Americans in a given area (whose use of racial slurs differs from whites and other groups), prior vote share in presidential elections, changes in House voting over the same period, Census-related effects, and other demographic controls.
It’s hard to say what this means for 2012, if anything. If racial animus is costing Obama support among the public, then it’s likely that it already shows up in polls. What’s more, at this point, Obama’s performance as president probably matters more than his race. Either way, this study sheds important light on the dynamics of voting in the 2008 election, and provides a useful reminder that we are not—and never have been—a “post-racial” society.