Rating the Pollsters

People on the left spent the year yelling that Rasmussen was really a Republican pollster that was influencing the election. I doubt the polls had any effect. People don't make up their mind based on a poll. There were races where all the pollsters were wrong. Did people not listen to those polls?

I took all the gubernatorial and senate polls published within three weeks of the election. That's close enough to be a pollster's best effort, considering early voting had started. There were 10 pollsters that polled at least 7 races, with Rasmussen leading the pack with 45 polls. They only failed to poll the Minnesota gubernatorial contest.

When it came to accuracy, the poll being within 4 points of the final result, the pollsters didn't do very well. This should be within the margin of error. I'd expect the pollsters to be within 4 points one way or another. Yet all of the pollsters except Quinnipiac and Mason Dixon were within the range of 47%-61%. Quinnipiac was better. Mason Dixon was worse. I wouldn't trust any of these pollsters for the next election.

When we set aside the spread and only look at the pollster getting the correct winner, the pollsters did better. Here 8 of the 10 pollsters were above 80%, with 5 of them being between 86%-90%.

When it comes to bias, the critics were correct. Rasmussen had a Republican bias, with their polls being 3.3 points more Republican than the final results. They were no more Republican biased than McClatchy/Marist, Reuters/Ipsos, or Fox News. Fox used a subsidiary of Rasmussen for their polls, however. Suffolk, on the other hand, had a 3.0 bias in favor of the Democrats. I'd normally expect the pollsters to have similar numbers on each side of the bias curve, but it seems they were the ones who were influenced by the narrative and adjusted their polls too far to the Republicans.

Mason Dixon and CNN/Time had the least bias. It's worth noting, however, that Mason Dixon was the worst pollster. Their polls were inaccurate, but they were as likely to be biased in favor of the Republicans as they were the Democrats. Three of their polls were biased an average of 6 points to the Republicans, while four were biased an average of 5.3 points in favor of the Democrats. I'll take a biased pollster who is more accurate over one that could be way off in either direction. At least I know how to adjust Rasmussen's polls.

Who was the best? Quinnipiac was accurate most often, did fairly well picking the correct winner, and didn't have a strong bias.