Turnout

October 30, 2008

Obama won by a larger margin than polls just before the election had predicted, though compared against long-term averages his lead was roughly as expected.

Pollsters must not only determine what people think, but how likely they are to act on their views by casting a vote.  The former aspect of the pollster’s profession involves fairly simple statistical modelling; the latter is mostly based on educated guesses. As a consequence, it is hard to guess Democratic Senator Barack Obama's lead over Senator John McCain.

People lie

As former Democratic pollster Mark Blumenthal has explained, the crux of the problem is that people lie: not so much about whom they intend to vote for, but on whether they will turn off the television, trudge down to the polls (often in unpleasantly cold conditions), and stand a the queue to cast a ballot. A survey in 1980 found that people continue to lie about whether they voted, even after the election.

Divining intentions

This is why even polls that draw their random samples from registered voter lists still need to distinguish between the views of voters in general and ‘likely voters’.  This is usually done via a series of screening questions that are statistically correlated with an increased proclivity to vote.  For example, pollsters often ask respondents whether they voted in the last election, and how closely they are following the current contest. 

Obama-induced complications

Yet this is a highly abnormal year.  Obama has drawn an unusual degree of support from young voters and blacks -- two groups that usually cast ballots at a significantly lower rate than their share of the registered electorate.  However, in the Democratic party primaries, voters from these groups participated at much higher rates than expected in some states. 

This phenomenon has produced a schism in the pollster conclave.  Approximately half of pollsters have continued to apply the likely voter screen questions that worked well in 2004, based on long-term turnout patterns.  But a large dissenting faction has adjusted the weighting of their screening questions, anticipating higher relative turnout among Obama-supporting groups.  For its part, the venerable Gallup poll has refused to choose, and publishes separate polls based on ‘traditional’ and ‘expanded’ likely voter models.  The former, as expected, gives Obama only a two-point lead; the latter has him ahead by seven.

Who is right?

Both the traditional and expanded models have arguments on their side, but early voting patterns hint -- just hint -- that the expanded model may be sound.  (Early voting in-person or by mail is legally allowed in many states weeks before election day.)  For example, in Georgia, blacks cast 35.2% of the early votes counted through October 29 -- over six points higher than their share of registered voters (29.0%).  And, as Michael McDonald of George Mason University has demonstrated, early voters in 2004 were disproportionately more Republican in every state except Iowa.  This year, according to McDonald’s compilation of early voting statistics, the Democrats’ share of pre-election day balloting is much higher.

Of course, statistics like early voting in Georgia might also simply be an artefact of improved Democratic party ‘get out the vote’ operations designed to encourage early voting.  But if the black share of the national electorate is even a point higher than the norm, it will substantially advantage Obama -- and confirm the expanded model of likely voters.