Get ready for one dry blog post!
I've been curious about how accurate the polls I'm following are. I rely on www.electoral-vote.com for my polling data. He does a great job of posting the polling results only from non-partisan pollsters. He also does this great thing where he averages the most recent polls for each state, so you can see in aggregate what numerous polls (if there are more than one) are saying.
So, I thought I would compare their average polls to the actual results. Some interesting findings here. Sorry, we are just looking at the Democrats.
There were 16 states we had at least one poll for in the final week. In 10 states, the averaged polls were pretty accurate, within a point or two of what I'm guessing is an average margin of error of 4-5% (unfortunately, the poll margin of errors are not posted to the site).
That leaves 6 other states, where the polling results were off by quite a bit more. Here are the amounts the polls were off:
- AL: 14 points
- CA: 8
- CO: 33
- DE: 13
- GA: 19
- MN: 42
The two "most wrongest" projections were in states that had caucuses (CO and MN). Lesson: don't try to poll a caucus state. It's futile.
So it's probably more accurate to say that the polling was fairly good in 10 of the 14 states with primaries. Let's look at what might explain what happened in the other 4 states.
In DE, there was only one poll to rely on. Generally, the more polls there were to rely on, the more likely you'd have accurate results (the glaring exception being CA). Some of the states that were 6 pts off (the high end of my acceptable margin of error) also had only one poll. So it may be true that you'd want to make sure there were two, if not 3 or 4 polls to average together before you started believing anything.
CA wasn't too far off, but with 10 polls in the final week to rely on, you'd think they might have gotten closer. Maybe CA is just so huge and diverse that it's like trying to survey 3 or 4 states, and so statewide polling is just going to be less accurate. It could be that some of the factors were in play there that I will mention below for AL and GA, though.
The real stumpers here are AL and GA. There were 4 polls in AL and 7 polls in GA during the week leading up to Super Tuesday. I looked to see if they had large numbers of undecided voters in the polling, but they did not when compared to the other states voting that day (overall, for all the primaries looked at here, there was no correlation between number of undecideds and poll accuracy). in fact, these polls were so wrong that even if every single undecided voter had gone for Obama, the original polls still would have been wrong. So what caused this? Well, I hate to say it, but once again I'm going to posit that race is a factor, since the problem with these polls is that they underestimated Obama support. GA and AL were the only states with sizable minority populations (perhaps especially rural minorities). Are the polling techniques being used not capturing minority voters? This might also explain the discrepancy in the CA polls, if this is true for Hispanic voters.
Overall, though, the lesson of Super Tuesday seems to be that if you average multiple polls from the week before a primary election, you're very likely to come within the margin of error (that happened in 11 of 14 states). If you are a state with a large minority population, though, this might not work out so well for you.
Comments