A Few Tips On Reading Political Polls

I’m not an expert, but I’ve been following elections closely since the 2000 Presidential Race. That’s really the first race that could be followed closely on the Internet. As a result, it’s relatively easy to follow the campaigns of various candidates and get an idea in advance of who is going to win. Of all my friends who discuss this stuff, I was the only one who predicted that George W. Bush was going to be re-elected in 2004. All of the others misread the poll results. As you probably know it was a very close election all the way until election day. So, why did I go to bed on election night confident that GWB would defeat John F. Kerry? It’s not that I was so smart, but I had picked up on one thing that was later confirmed by someone who reads polls for a living. A single poll is meaningless. I’ll say that again. A single poll is meaningless. You have to average the polls out, or better yet figure the mean of all of the polls, to get a real picture of what is going on. In 2004 a few polls showed Kerry ahead by a small margin, but most had George Bush ahead by a still small, but larger margin. Looking at the trending over the last two or three weeks of the campaign, there was no lead for John Kerry and consistent, but small lead for George Bush. Edited To Add: It’s important to not that even thought George Bush’s lead was within the margin in every poll, there was no case where John Kerry had a lead. One would expect that a lead that some polls would have shown Kerry leading, while some would have shown Bush leading, but such was not the case. In the current election look at the current numbers for the CA Governor’s race. Whitman has a small, within the margin of error, lead in every poll. Which doesn’t mean that Whitman will win since there are still several weeks to go, but if that lead holds, Whitman wins.

There are a couple of important things to know about polling samples. There are three major ways that sampling is done. The first, which is not very accurate is “adults of voting age”. That is a random sampling of, as the name suggests adults 18 and older. The poll does not try to determine if the sample population is registered to vote or is likely to vote. These polls are often used by agenda driven media outlets since they tend to favor Democrats. The second method is “registered voters”. These are people who are registered, but no attempt is made to determine if the sample population actually does vote. This method is more accurate that “adults of voting age”, but not as accurate as the next method. “Likely voters” is the most accurate of the three methods. The sample is vetted using election results (anonymized) to determine if the sample population generally votes in elections. Because the sample uses people with a history of voting, it’s predictive value is better than the first to methods.

It’s also important to look at the margin of error. Generally in political polls it’s five percent, although a few use four percent. Not tremendously accurate if you’re testing a new medication, but for politics, given the whims of the voters it’s pretty good. If the difference between two candidates is under the margin of error, the election is a statistical tie. Someone ahead by 1 or 2 percent is not a significant lead in and of it’s self. Again, the trending data is what’s important in close races. If candidate A is consistently 2 percentage points ahead of candidate B, it’s a good indication that even though within the margin of error, A is going to be the likely winner.

You should also know that some polls will oversample members of one party or the other. Generally it seems that Democrats are oversampled more than Republicans, but that’s not always the case. Keep that in mind, especially if the poll is sponsored by a major media outlet like ABC or the New York Times.

Also, look at the track record of the polling organizations. The two that seem to me to be the most accurate are Rassmusen and Quinnipiac University. Rassmussen does all sorts of polls, not just political ones. Rassmussen makes some of the polls available for no charge.

The best sight I’ve found for getting a good look at polls is Real Clear Politics. The site offers a number of different polls and also political commentary. As the election gets closer, I’ll be checking it daily or even several times a day to see what’s going on.

So, there you have all my political polling secrets. Well, such as they are. As I wrote at the beginning of this post, I’m not an expert, just a political geek.