If you're an obsessive poll-watcher like me, you've probably noticed something interesting in the past three weeks. For the first time since Gallup began its daily tracking poll on the presidential race back in April, President Obama has held a lead over Mitt Romney for more than just a few days.
It's not much of a lead, to be sure: only two points on average over the three-week period from June 18 through July 8. But in every other non-overlapping three week period before the most recent one, the race had been tied at 46 percent each. This has made the Gallup Poll somewhat of an outlier among national polls using a similar methodology to Gallup's -- live interviewers with calls to cell phones as well as landline phones. On average, these other polls have shown Obama with a two- to three-point lead over Romney. And Gallup's three week samples are so large, that a two-point swing in the margin is almost certainly statistically significant.
So what explains this shift? Most if not all of it appears to be due to one thing -- an increase in the percentage of nonwhites in Gallup's sample of registered voters. While Gallup does not provide the percentages of whites and nonwhites in its weekly compilation of tracking poll results, these can easily be calculated based on the racial breakdown of the vote.
There is a little bit of uncertainty about the exact results because a very small percentage of Gallup's respondents, perhaps two to three percent, are not classified by race but are included in the overall results. But this should not affect the estimates of the racial composition of Gallup's registered voter sample very much.
According to my calculations, the percentage of nonwhites in Gallup's registered voter samples for each of four non-overlapping three-week periods was as follows:
- April 11-May 6: 21.8
- May 7-May 27: 21.1
- May 28-June 17: 21.8
- June 18-July 8: 23.6
For the first three of these three week time periods, the nonwhite share of registered voters averaged 21.6 percent. For the final one, it was a full two percentage points higher. And for the first three of these time periods, Obama and Romney were tied at 46 percent each. For the final one, Obama led Romney by two points.
Is this just a coincidence? No -- the increase in the nonwhite share of registered voters actually accounts for about two-thirds of the two-point swing in margin. The main reason President Obama has been doing better in the Gallup tracking poll recently is that the percentage of nonwhites among registered voters has increased.
I have no idea why the racial composition of Gallup's registered voter sample has changed since mid-June or whether this greater representation of nonwhites will continue in the future. What I can say with some confidence is that the results from the most recent period appear to be more realistic in light of evidence from the U.S. Census Bureau which put the nonwhite share of registered voters at 24 percent in 2008. If anything, one might expect the nonwhite proportion of registered voters to increase between 2008 and 2012 as it has in every inter-election period since 1996 due to the steady increase in the nonwhite share of the voting age population.
Based on what we know about the likely racial composition of registered voters in the U.S., the results from the most recent three-week period are probably a more accurate estimate of the current state of the 2012 presidential race than the earlier results. They certainly bring Gallup's results more closely in line with the results of other national polls using similar methods.