Friday, September 28, 2012

The Problem with Polls

According to the latest polls, the election is all over. President Obama leads in all battleground states. In some of them he has an unassailable margin of nine points.

There are a couple of polls that show a different race. Rasmussen and Pew show the race much tighter. Rasmussen has the race essentially tied over all with Obama having a slight lead in some swing states.

It is easy to dismiss these polls as outliers, especially since Rasmussen himself is a noted conservative. Electoral-Vote.com has a Rasmussen-free version. But before we do that let's remember the history of polls in tight races.

In 1980 pollsters declared the race too close to call. Reagan won by a landslide. He won so decisively that the election had been called before the California polls had closed.

That was a long time ago. For a more modern example, look at 2004. Democrats went to the polls sure that John Kerry would win easily. Instead Bush won by a decisive margin. Even in Ohio where the vote was close, Bush's margin of victory was large enough that there was no automatic recount.

Rasmussen called that one. He and Pew were also the closest to the actual results in 2008. So, they must be taken seriously.

Some conservatives are calling "foul" on the other polls, alleging that they are being manipulated to make Romney supporters loose faith. Liberals are calling this unfounded conspiracy theories.

So, what is going on? Part of the problem is that polling had gotten harder.

In a year with high-turnout, less than 2/3s of the population votes. If there is a low turn-out then barely half of the population votes. Democrats lead among apathetic voters so the first thing that a poll must do is try to winnow out the likely voters from the registered ones (and eliminate unregistered voters completely).

The next problem is in getting an even sample. During the 2004 race it was noticed that Kerry did better in polls taken during the week and Bush did better in polls taken over the weekend. A close look at the number showed that the percentage of Democrats in the polls was higher during the week.

There are ways of compensating for sampling errors. If you know that Democrats make up 1/3 of the population but are 1/2 of the group polled then you need to adjust the result.

This is where the conspiracy theories come in. Every poll that shows Obama with a significant lead has a disproportionately high number of Democrats in the sample. Currently the country is split nearly evenly between Democrats and Republicans with the Republicans having a slight advantage for the first time ever.

But recent elections have complicated this. For a long time voter affiliation was a fairly constant percentage at the polls. That changed in 2006 and 2008 when record numbers of Democrats voted. The result was that the Democrats took over both houses of Congress (2006) and the White House (2008). In 2010, fewer Democrats turned out but more Republicans voted. The Republicans took the House and made significant gains in the Senate.

The polls that favor Obama are based on the assumption that the mix of Democrats and Republicans in 2012 will be similar to 2008. This assumes that Obama's name on the top of the ballot was the driving force for a Democrat turn-out.

Republicans point out that in 2008, having a black man on the ballot was a first and many people turned out just for the privilege of voting for him on that basis. In addition, Obama was the candidate of Hope and Change in 2008. Now he is just another political hack trying to hold onto his job. The excitement that brought people to the polls in 2008 is gone and people registered as Republicans is at an all-time high.

Democrats insist that the voter turn-out in 2010 was much less than in a presidential election and that the only reason the Republicans did so well was that Democrats were dispirited and stayed home. A lot of energy has gone into increasing voter turn-out so the mix will be closer to 2008 than 2010.

Both arguments are valid and neither one requires ill-will on the part of the pollster.

Another factor is that more people have dropped land-lines in favor of cell phones (technically "mobile" phones). Pollsters are prohibited by law from calling these phones. The people who only use cell phones are younger than the general population and skew Democrat (the old rule of thumb is that people start out idealistic and liberal and become more realistic and conservative over their lifespan). Different pollsters have ways of adjusting for this group but none of them are satisfactory.

A final factor is that polls present three choices - Obama/Romney/Undecided. There is no "undecided" box on the ballot. People who say that they have not decided often stay home or have already decided to vote for the challenger but do not want to admit it to a pollster. Any state where Obama has less than 50% should be considered in play no matter what his lead is.

So, the election is still too close to call. Michelle should wait a few weeks before buying her inaugural dress.

No comments: