« More Divergent Than They Should Be? | Main | Weighting by Party »

September 22, 2004

So What Should a Junkie Do?

So what are we to make of polling, given that it allows for a wide variance of results at this point in the election? For average voters trying to make up their minds about who to support, AEI’s Karlyn Bowman has good advice: "I would urge most Americans not to pay attention to polls.”

But if you’re reading this, the odds are good you’re a political crack-head like me, long ago decided and badly addicted to following each twist and turn of the campaign. So given the inevitable and sometimes contradictory "divergence” in polls, what should a smart junkie do?

First, the classic advice, also from Carlyn Bowman:

Ms. Bowman advises going to a polling firm's website and looking at the methodology. How big is the sample? Over how many days was a poll taken? Do they weight for party identification, and if so, how? She also advises looking at the trend over time within one polling organization - and not to compare among different polls

Let me add a few more:

1) One easy shortcut is PollingReport.com. They show results for each pollster separately over time and obsessively report key methodological details like interview dates, number of interviews and reported sampling error. The also provide complete question wording and specify exactly who was interviewed (adults, registered voters, likely voters, etc).

2) Remember that no one poll or polling organization has a monopoly on wisdom or the art of defining likely voters. Look at the range of polls.

3) To reduce random variation of sampling error, try to look at very big samples.

Two specific suggestions. First, The Annenberg National Election Study is a massive ongoing daily tracking survey that conducts roughly 5000 interviews per month. It is an academic study, however, and its directors have chosen to withhold vote preference data until after the election. However, their occasional reports on the perceptions of the candidates are typically based on massive numbers of interviews and are worth watching.

Second, though a little unorthodox, it’s not crazy to effectively increase the sample size by averaging results for comparable polls over a comparable time period. Doing so also essentially pools methodologies, applying an "everyone’s right” approach to sampling likely voters. Realclearpolitics.com does a quick and dirty daily average of polls released over the most recent week. They also provide links to detailed reports released by the pollsters. However, their average includes only "likely voter” results when both are available. Also, it is also important to remember that a raw average may vary depending on which organizations report results in each week.

4) Finally, try to remember that the election will not be held today, that many voters are still considering their choices and that their answers to the horse-race question today may mislead us about what they are likely to do on November 2. I’ll have a lot more to say about this last point over the next week or so.

Related Entries - Divergent Polls, Interpreting Polls, Sampling Error

Posted by Mark Blumenthal on September 22, 2004 at 03:57 PM in Divergent Polls, Interpreting Polls, Sampling Error | Permalink


Any estimates on the percentage of voters who typically make up their minds during the last 48 hours of the campaign?

Posted by: lex | Sep 23, 2004 5:05:30 PM

The comments to this entry are closed.