Sunday, October 19, 2008

Could The Polls Be Underestimating Obama's Support?

In our previous post, we cited evidence that the so-called "Bradley Effect"--the claim that 2-6% of white voters lie to pollsters about their intentions to vote for a black candidate--is nothing more than an urban myth.

That would certainly be a relief for Obama supporters, because it would mean his current lead in various polls is for real.

Now, what about another effect in the polls, one that could result in underestimating Obama's support. Let's call it the "landline effect."

If you go to, you can find all kinds of polling data on the presidential race, as well as many key Senate, House and gubernatorial contests. What you won't see is very much discussion of how those polls are conducted and whether the way they are conducted may introduce some distortions into the data.

All the major pollsters conduct their surveys with a very similar methodology: they generate a "random" list of telephone numbers, then call them, and then ask questions. One thing you need to know about those telephone numbers: often, they are only landline telephone numbers, not cell phone numbers.

Yet, there may be a difference between people who can be reached via landline and those who can only be reached via cell phone. One good analysis of this issue can be found in a recent National Journal article, "Dial A Cell, Reach A Dem."

According the NJ, nearly 15% of all households in the U.S. now are reachable only via cellphone. And those cellphone only households skew much younger than landline households. So, a survey that excludes cellphones is missing younger voters. Moreover, there is reason to believe that younger voters will vote at a higher proportion this year than in recent elections.

The NJ article also notes that some major pollsters, familiar with the problem, have investigated it, but including cellphones in some of their surveys. [They generally don't include cells because it is considerably more expensive to do so.] What did they find?

They found that including cell phones ups Obama's support by a couple of points and lowers McCain's by a couple points. Some pollsters seem to think that's no big deal, but we disagree. That's a four point swing.

Let's illustrate. Poll A surveys landlines and finds that Obama leads McCain by two points, 48%-46%. Close race, we say.

Now, Poll B surveys a weighted sample that includes roughly 15% cell phones. It finds that Obama leads McCain by six points, 50%-44%. Now it's not so close. In fact, that's a decent lead, and it's statistically significant, whereas the margin in Poll A is not.

When pollsters report their results to the public, they often include a lot of demographic breakdowns and other information so we can better understand the numbers. But we haven't seen anyone disclosing whether their sample includes cellphones, which, frankly, would be much more useful information.

Since MOST of the polls you are seeing exclude cell phones, you should anticipate that they are, to at least some degree, underestimating Obama's support. That's a good thing to keep in mind as the polls inevitably tighten in the coming days.

No comments: