If there’s one thing that makes a pollster very, very nervous, it’s seeing a survey come back with results that dramatically challenge the conventional wisdom. Releasing survey results that match what everyone else is saying is easy. Being the outlier is not.
Normally, upon seeing a survey like today’s Bloomberg poll that shows Obama with a significant advantage over Romney, I’d dismiss it as an unusual outlier, likely the product of a skewed sample or odd methodology. However, Bloomberg’s pollster, Ann Selzer, has a strong reputation in the polling industry. The poll’s partisan makeup also appears reasonable, lining up slightly more Republican than the 2008 election. So what happened?
There are a few things that caught my eye as potential reasons the poll is so favorable to Obama. The survey is of adults, which results in a more Democratic-leaning sample than registered or likely voters, though the ballot numbers that are grabbing the headlines are among likely voters. However, the results only show the demographic breakdown of the full adults sample, leaving us to guess about the makeup of the likely voter subgroup. Looking at the adults sample, there are a few groups that are significantly over-represented compared to who actually turns out in elections.
For instance, the Bloomberg survey has 22 percent of respondents under the age of thirty. Much to my chagrin, this group is a more Democratic-leaning age group. Even in the 2008 election, only 18 percent of voters were under thirty. The likelihood that the proportion would increase in this election is slim. Furthermore, even with energized turnout among black and Latino voters, some 74 percent of voters were white in 2008, only down three points from 77 percent in 2004. In the Bloomberg poll, white voters fall to only 67 percent of the sample. Even with America’s changing racial makeup, I am skeptical that white voters will fall so significantly as a proportion of all voters.
The truth is that there’s no sure way to know why this survey is so much more favorable to Obama without seeing the makeup of the underlying likely-voter sub-sample.