We had a couple of new polls come out this week in the competitive states. I want to start off and explain how the margin of error works and what typical errors are. The margin of error is calculated based on a variety of assumptions (independence of observations, identically distributed probabilities, and no sampling error) that I don’t believe always hold. So the margin of error doesn’t explain all of the error inside of a poll. FiveThirtyEight analyzed polling error and found an (weighted) average error of 5.4 for Senate elections for polls 21 days before the election. An average error works like a standard deviation in interpreting error. This has an interpretation of a poll being about plus or minus 10.8% the expected mean about 95%. But since the election is far away and it is a midterm year, we should expect a little higher error in the polls right now, so anything within 5-10 points is still somewhat realistic for the other candidate. But with multiple polls combined the errors get exponentially smaller.
In West Virginia, a new poll shows Manchin with a 10 point lead. This combined with multiple other polls suggests this race is not competitive and the one poll with a Morrisey lead was an outlier. I am no longer considering the West Virginia as a competitive race.
In Arizona, a set of polls on 7/9, focused on the Republican Primary candidates. McSally was in the lead, and she had the best poll numbers against the presumed Democratic nominee. This seat is currently held by Senator Flake who is retiring.
A Missouri poll shows a Republican primary candidate, Josh Hawley, ahead of Senator McCaskill. I am not sure if Hawley will win the primary, but this poll shows that the race is competitive.
In Montana, the race is still looking close. A new poll had Tester with a 3 point lead. This is 5 points closer than the last poll, but all or most of this difference could easily be explained by sampling error. Rosendale (R) will have to appeal more to moderate and independent voters to win.
In Tennessee, a poll shows the presumed Democratic Senate nominee and former Governor Bredesen with a 6 point lead over the presumed Republican nominee Blackburn, a US Representative.
In Florida, a new poll showed Senator Nelson with a lead of 4 points against Rick Scott (current Governor of Florida), technically outside the margin of error. The polls have been going back and forth. The polls are over different periods of times, use different polling methods, and are by different pollsters, which means it’s not unusual that they are different. My guess is that right now the result would be between +4 for Nelson and +4 for Scott. The primary is coming up soon, and after that more people might decide even though Nelson (D) and Scott (R) are the only viable candidates.
These changes along with the other changes from last week mean the competitive races are now: Arizona, Nevada, Tennessee, Florida, Missouri, Montana, North Dakota, Ohio, Michigan (temporarily until primary), Wisconsin (temporarily until primary). I know that one poll has a large error, but when you see multiple large leads without external evidence for a close race the odds of the polling being wrong to get smaller. In my model competitiveness matters because it determines the prior which can affect the final call. I want my groups to be as homogenous as possible.