Everyone wants to be right.
As Election Day draws nearer, political campaigns and commentators begin talking more and more about recent poll data, attempting to interpret the results to indicate their candidate is winning. As a result, favorable poll numbers are often exaggerated and unfavorable results are “explained away”.
It’s a perfect example of the “filter” theory of politics, arguing that we all have a political “filter” based on past experiences and existing policy preferences. When information is presented that contradicts the already-held ideas, it simply “bounces off” the filter and is quickly dismissed.
Both sides attempt to blame polling organizations for flawed methods or deliberate bias, depending on the results.
One such conservative website, unskewedpolls.com, blames all polls for being biased toward Democrats, releasing their heavily criticized versions of “unskewed” polls that show distinct Republican advantages.
On the other side, now that the race has become closer, Democrats blame the polling organization Rasmussen for a conservative bias. After Gallup came out with the “Romney 51%, Obama 45%” poll after the debates, accusations again started flying about Gallup’s alleged bias.
Although polls play a crucial role in American politics, there are a number of statistical flaws that cannot be neutralized.
One of the most obvious problems is the problem of registered voters vs. likely voters. Polls of registered voters tend to lean Democratic, because there are more Democrats registered than Republicans. However, many of those registered voters do not turn out on Election Day. Among “likely voters,” a screen developed by polling agencies to attempt to solve this problem, Republicans have an advantage. This explains why Democrats are generally more active in “get-out-the-vote” efforts.
However, there is still no perfect way to measure who is a “likely voter.” People may feel pressure to lie to pollsters, or they even may expect to vote but find themselves too busy on Election Day.
There are other problems as well. Are people who answer their phones to pollsters truly representative of the voting population? Do the results skew one direction depending on what time of day the call comes?
As a result of these factors and more, the polls normally carry a 2-4% margin of error. If a poll comes out with Obama at 49%, Romney at 48%, and a 3% margin of error, it is still very likely that Obama in fact has 48% and Romney has 49%.
One way to eliminate much of the statistical uncertainty is to average the polls together, as RealClearPolitics has done. Their average is consistently used by media outlets. Still, it is not perfect. If many polls in the average are flawed, the average won’t magically fix the problem. As long as the election outcome depends on what people actually do, not what they say on the phone, we will never be perfect at prediction.
Polling, though it can offer us general predictions of outcomes, will not reach the peak of accuracy until right before Election Day (when many have already voted). Until then, take the polling criticism from either side with a grain of salt. According to Erick Erickson on RedState,
“I’ve been in politics long enough to know that the louder one side gets complaining about the polls, the more likely it is that this is the side that, in reality, actually is losing.”
A variation of this article was originally published in the Baylor Lariat on 10/30/12.