There are so, so many public opinion polls out there these days, particularly for the presidential race and the most hotly-contested Senate races.
Putting aside the many technical aspects of public opinion polls, here’s some rules of thumb for reading them, plus some resources:
1. Don’t cite one poll if there are many polls on the same race. Instead, use poll aggregators and models. Why? Because it’s too easy to simply cherry-pick the poll that makes your case.
According to the statistical model underlying the poll, one in twenty polls that were conducted perfectly will be wrong outside of the margin of error. So why would you base your view of the race on what could be an outlier?
2. The trend is your friend. As you look at the collections of polls, look at how polling is going overall. If one candidate is surging, that will show up. If the polls, taken altogether, show movement toward one candidate, take that seriously. Again, do not pick out one poll to make your case!
3. Never forget that polls are snapshots. Campaigns are not static. Stuff happens. Candidates give great speeches. Candidates make gaffes. Between external events, advertising, and what candidates say and do, opinion can change and often does.
4. Not everyone votes, but predicting who will do so is not so easy. At some point in high-profile races, pollsters start presenting results for “likely voters.” Each pollster has a different way of classifying people as likely voters and different methods can be more accurate than others. These likely voter screens can matter a lot for who is included and can be problematic. Your solution: Watch the trend and, when many polls are available, don’t focus on just one poll.
5. Field work affects who votes and won’t be fully reflected in polls. All those things that campaigns do — like call voters and knock on their doors — matter. Field activities, the get-out-the-vote work, can provide the edge for close campaigns, but not show up in the polls. Knowing this should lower your certainty when you read the polls.
6. In the presidential race, keep an eye on the state polls and the electoral college. In our system, it’s possible to win the popular vote and lose the election. It’s happened three times, including as recently as 2000. (The other two times were in 1876 and 1888). Focusing on the national polls can lead you astray.
Resources: What I read
1. HuffPost Pollster. This started out as an independent site and then was acquired by the Huffington Post. They’re very transparent about how they operate. They use a statistical model that plots and projects trends, but you can see each and every presidential and Senate poll that’s out there now and you can adjust the trend’s sensitivity.
The articles are well-done and, starting halfway down on the right, you’ll find links to all the polls you could possibly want, plus an color-coded electoral college map and Senate map. If you want to fool around with the polls, you can see what changes if you leave out certain type of polls or you can narrow the time period you’re examining.
2. RealClearPolitics. Besides gathering all sorts of articles from around the country, the site gathers lots of presidential and Senate polls and shows poll averages, an electoral college map, and a Senate map. One nice feature is that you can see the electoral map with toss-ups and no toss-ups. For the latter, if a candidate is ahead by any amount in the most recent poll average, he wins that state.
While I read the site every day and recommend it, do note that they don’t include every poll that comes out and it’s not clear why some are left out. Moreover, the time period for the poll average seems to vary. Sometimes you’ll have polls that were conducted weeks ago and other times only more recent ones are included.
3. Talking Points Memo Poll Tracker. This is the new kid on the block, having just started this year. They provide lots and lots of poll data and, like HuffPost Pollster, use a statistical model to show trends. That model looks to be pretty sensitive at times, perhaps too so.
But it’s very easy to see what’s going on in the states for the presidential and Senate races. And not only are they fully transparent about their approach, but have an app for various mobile devices.
And two more, which use poll data to model the presidential race:
1. The Princeton Election Consortium. This site is run by Princeton Professor Sam Wang and uses state polls to provide a snapshot of the race. Wang has done this in several past presidential races and proved to be extremely accurate. He provides a snapshot of the race and shows a prediction zone. Methods are fully disclosed. The site attracts wonky types who know statistics but the main elements are not hard to understand.
The key items here are two charts. One gives the median electoral votes estimated by the model. The other, called the meta-margin, shows how much polls would have to shift for there to be a tie between the presidential candidates.
2. Nate Silver’s FiveThirtyEight. Silver started out independent and then moved to the New York Times. Like the Princeton site, he was very accurate in 2008. However, his method, which is fully disclosed, is different. He uses polls and economic indicators and indicators about each state’s tendencies to show where the presidential race is now (the Nowcast) and to predict where it will be on election day.
Silver’s model has gotten more complicated over the years and we don’t yet know if it’s better. Readers should note he’s giving probabilities for who will win certain states and races, not saying what definitely will happen.
Both the Nowcast and forecast are shown in electoral votes and the percentage of the votes. Silver also provides a lot of information about each state and its status, along with many polls.