I took public (not internal) pollsters who polled at least four states in the presidential race the last two weeks of the campaign to determine who was good, and who sucked as bad as Rasmussen. Without further ado, the results from Public Policy Polling (IVR, or robo), Ipsos (internet), SurveyUSA (IVR), Opinion Research for CNN (live callers including cellphones), YouGov (internet), Marist (live callers including cellphones), Gravis (IVR), Zogby (live callers, I think), and Rasmussen (IVR):
Remember, this isn't all pollsters, just those who polled a multitude of races.
The winner should come as no surprise—our very own Public Policy Polling. The loser should also come as no surprise—Rasmussen.
Rasmussen has always been bad, but it's hard to understate just how awful they were this year. Of the nine battlegrounds and national numbers, they called the winner in just three races, or 30 percent. They were flat out wrong in five of the ten (they had a tie race in the other two, so not right or wrong). In contrast, PPP called the winner in nine of the ten races, and got none wrong (one was a tie).
The biggest miss of the lot was Rasmussen's—their 8.4-point flub of Colorado. Second place was Rasmussen, their 6.9-point flub of Wisconsin. Third place was Rasmussen, their 6.8-point flub of Iowa. If you want more awful, look at Zogby. Not only did their numbers suck, but they somehow thought Romney would win the popular vote by two points while losing Florida by five, Ohio by six, and Wisconsin by eight. Nonsensical. Gravis, one of the baby right-wing Rasmussens that infected the polling aggregators this cycle, tweaked their numbers late and still had some of the worst results.
PPP's accuracy was particularly noteworthy given the frequency of their polling. While many pollsters were selective about the states they hit, PPP was everywhere that mattered. And despite the kvetching about their methodology (robopolling) and lack of cellphone calling (much of that on this site), they nailed yet another cycle.
The internet pollsters—Ipsos and YouGov—had fantastic results for an unproven technology. They are the future of polling. Traditionalists who hate robopolling are going to now have to hate internet polling, but it's clear that it can be done, and it's certainly here to stay.
One final note: There was one pollster who polled several of these states that did better than even PPP—the Mellman Group (D, of course). The average error of their final-week results in Florida, Iowa, Nevada, Ohio and Virginia was a miserly 1.7 points. However, I excluded them from the chart above because as an internal campaign pollster they were selective in what was released and what wasn't. For all we know, they polled New Hampshire and withheld results for partisan reasons. However, the numbers they did release were stellar.
Update: Added the conservative robo-pollsters We Ask America to the chart.