View my bio

Now the Details

Media, ethics, and journalism. What works. What doesn't.

Jeffrey Dvorkin

Monday, October 20, 2008

Why The Polls Could Be Wrong

In the last two weeks before the election, the polls are showing that Senator Barack Obama increasing the gap with Senator John McCain.

Most recently, a Wall Street Journal - NBC News poll now shows Obama with a ten point lead - the biggest since polling started in this election cycle.

The poll also shows that Governor Sarah Palin is now a drag on the McCain campaign and a majority of Americans polled now say she is not presidential material.

That has the usually nervous and cautious Democratic activists breathing a little easier. As one long time party worker told me, "it looks like we can't mess up even this one." Tradition has it that Democrats lose the race in the last few weeks of a presidential campaign due to overconfidence or a last minute upset that polls and politicians could not anticipate.

But polling is not an exact science, to say the least. There are good reasons why both the McCain and the Obama camps should keep pushing right up to November 4th.

While there are some dependable polls that are done with integrity, many are not.

Here are some of the variables in world of polling:

1. Sample size. Most social scientists assume that a national poll cannot be considered accurate with fewer than 1000 respondents. But many polls are done with 700 or less. Some even claim to be accurate with less than 400.

2. Return to sample. This means that when a polling company calls and no one answers, a return call at a different time of day must be done to ensure accuracy. But many polling companies operate on a 9 to 5 basis. This guarantees that the people who are reached between those hours are mostly homemakers or retired people, thus skewing the sample and the demographic.

3. Cell phones vs land lines. Cell phone users were rarely contacted by polling companies. This means again, a skewed demographic. Recently that practice has changed but the sample remains too small.

4. National polling. A true national poll needs to reflect the population distribution through different regions. In the past, polls were conducted by concentrating in New York and Los Angeles with a few samples in Houston or Chicago. Hardly a national poll.

5. Push polls. These are polls in which the question is phrased in order to obtain a specific answer. Example: "Would you vote for an African-American if you knew we went to a church where the pastor was anti-American?"

6. Omnibus polls. These are polls in which political intention is tacked on to a commercial poll. If Proctor & Gamble is doing a survey about a new brand of soap, some political operative add their own question. It's an inexpensive way of polling, but it runs into the same limitations about the return to sample (see #2.)

7. Truth in advertising. Does the poll inform us about who commissioned the poll? What was the sample size? When was the poll conducted? How were the questions phrased? If we don't know the answers, we shouldn't trust the poll.

In the last week of a campaign, newspapers and broadcasters will be flooded with polling results that are offered for free. Usually they attempt to sway the few undecided voters who might still be making up their minds.

In this election, the relatively small number of undecideds may mean that lousy polls may not have the same impact as in other years.

But one thing is certain, this election may still be too close to call.

No comments:

Post a Comment