- The first of the strange one is the Fox News poll showing a 14-point swing over the last month in the generic ballot. There were some on the right today that were questioning an 11 point swing in a little over a month in the Rasmussen North Carolina Senate poll. Apparently, polling swings are only "proof" of bad polling when you do not like the results. What one would have to do is look at the cross-tabs objectively. The Fox News poll demonstrates that they think the electorate will be a net +2 Republican than in 2010. This puts them out on their own ledge. Most agree that the electoral make-up will be more similar to an average between 2012 and 2010.
-The Rasmussen poll is being criticized for using a D+8 sample. The electorate there is D +11 and in 2012, the turnout was D +6. Though, in that same year, the Female turnout was +4 of what this poll uses. This poll also uses a +4 more white electorate. It also uses an older electorate. The 18-39 age group is a net -6 of 2012. So, while the sample is more Democratic by 2 points, the sample is more male, white, and older.
-The D +8 sample is the same Demographic sample used in the final Rasmussen poll on the 2010 Senate race. In that poll, they over estimate the Republican candidate by 1.9%. So, in 2010, the same sample was used in a year where the environment favored Republicans (at the very least as much as it does now, unless you are the pollsters used by Fox News) and that sample still favored the GOP. An objective look at this sample shows that it, at least for Rasmussen, is a consistent mid-term sample that has produced good results in the past.
-I am aware that there was a Wendy Davis internal out of Texas today. While the top-lines were instep with last month's Rasmussen poll (Abbot +8); I did not include the poll as I couldn't even find the name of the polling firm that conducted this internal. As with any campaign, I won't just take their word that an unnamed pollster found a certain result. Even if that result is consistent with other polling.
-Finally, today we got the YouGov polls for the Gubernatorial races. While most of these interviews were conducted between 1 1/2 to 3 weeks ago, they are still the most recent polls we have for a handful of these races.
-The initial release of these polls today was a mess. The cross-tabs lead to Senate polls, the Kansas poll was redacted on accident, the wrong candidate in Ohio was highlighted for the lead (apparently 37 > 50?), etc. Most, if not all, of these mistakes seem to be fixed now. I was waiting to post today until these results got straightened out. They were such a mess that I figured I might as well not even touch them until they are sorted out.
-There are several of these polls (as has been the case this cycle) that are completely out of step with other polling done. Prime example would be Kansas where they have Brownback at +7 where all other pollsters have him down by 5-10 points. In fact, the only non-internal poll to find Brownback in the lead since April is the two YouGov polls. The only other firm to find him leading in the last year is Rasmussen back in April.
-Other strange results included PA being much closer, RI being a tie (also used the Republican that lost the primary so I won't include the poll), Maine showing the Republican a point ahead, WI showing Walker up by 4, and Carter being down by 8 (though closer to some polling done at the same time. Remember these numbers are a few weeks old).
-I will note that there were just as many races here that fit with recent polling within the MoE. Examples: FL, MI, IL, MN, IA, NY, CA, OH, SD, TN, NH, etc. So, not all of these polls were odd by any means. The MoE for these polls ranged from 2-8%. Anything over 5% is quite troubling as far as accuracy goes. Again, these are just one set of polls for these races. The point of an aggregate of polling is to average out results. This balances out the extremes on both ends. The more data, the better the aggregate.
-The ONLY reweighing of polls that I do deals with extreme mathematical outliers. I have yet to need to implement this to this point. In using all polling since August 1st, I find that the range for margins not to be at least a mild outlier to be D +0.5 - D +12.5. (1.5 * Q1-Q3 range). Finding extreme outliers gives us R +4 - D +17. (3 * Q1-Q3 range). Therefore, R +7 is an extreme outlier. What I do in this case, is "correct" the margin down to the closest non-extreme outlier. So this poll gets "corrected" to R +4. This will be calculated again when there is a new Kansas poll. The R +4 would become an R +7 again if the R +7 is no longer an extreme outlier in the new data set. To this point, I have not had to make this adjustment on any other poll this cycle.
-Even with the "correction", my margin for the Kansas Gubernatorial race moves from D +4.9% to D +3.5%. This is still a significant shift.
Here are the links to the polls used in these updates today:
Generic Ballot (Fox News)
New York Times/ CBS News/ YouGov
North Carolina Senate (Rasmussen)(Civitas)
Michigan Senate and Gubernatorial
New Hampshire Senate (D)
New Hampshire Governor
New Hampshire Senate
New Mexico Governor
New York Governor
North Carolina Senate
South Carolina Governor
South Dakota Governor