Ref's Ruling: Recent NBC Polls Seem to Show Flawed Voter Turnout Models
The NBC/WSJ poll released Wednesday found Hillary Clinton leading Donald Trump by six points among likely voters. The NBC/Survey Monkey poll released Tuesday found a five-point lead for Hillary among likely voters. I would not care about this if these were registered voter polls, but because they are likely voter polls and a good filter was allegedly applied, these need more scrutiny.
NBC reported that Hillary had stopped the bleeding from her bad weekend and reversed the tide, now firmly back in control of the race. NBC also found the same level of voter intensity for Hillary Clinton as for Donald Trump. Huh, seems out of sync with everything else we are hearing, right?
All other national polls taken at the same time show a closer race, some with Trump leading, some with Clinton leading. The Economist shows a +2 for Clinton. Reuters shows a +2 for Trump. LA Times shows a +4 for Trump. UPI daily tracking shows a +1 for Clinton. Averaging those polls, you find a 0.8 point lead for Trump. So comparing to other national numbers shows us that the NBC polls probably are applying a flawed turnout model.
Another way to assess the NBC polls is to compare them to state polls from states that are bellwethers. Ohio's popular vote has nearly matched the national popular vote each year since 1996, as the chart demonstrates to the right. Florida also matched the popular vote very closely, although it tends to slightly favor Republicans.
Generally, the national polls should roughly match Ohio's poll results. This is true so long as a candidate is not from Ohio or is appealing to a particular voting tendency in Ohio. Trump may be doing just that by running a blue collar themed campaign. But Hillary is also mobilizing her supporters in the state and it is likely that Ohio will closely match the national result this year.
Florida favors the GOP a bit, by about a point and half above the national result. It is consistently off the national number by about a point and a half. One thing is true for both states, they have each picked the winner in the last five elections.
Both of these states are heavily polled. Also, the population is much smaller than the nation at large, so it is easier to get an accurate measure of voter opinion.
To determine if a poll is an outlier due to a flawed turnout model, a three-prong test will be applied. First, the poll must be within five points of the average of all other national polls. Second, it must be within five points of the current Ohio average. Last, it must be within five points of the current Florida poll averages. Both NBC polls failed each prong of the test.
Once a poll fails this test, a weight will be applied to that poll in the national averages. The poll will only count half as much as the other polls that do not indicate a flawed voter turnout model.
Follow my new account on Twitter for daily updates on the election!
Update: Nate Silver doesn't seem to think the NBC polls mean much either
FAILED: Other national polls +0.8 Trump, NBC polls +5 and +6 for Clinton
FAILED: OH+2.6 Trump, NBC polls are 7.6 and 8.6 points off
FAILED: FL+0.33 Trump, NBC polls are 5.36 and 6.36 points off