Thursday, October 6, 2016

10/6 Election Update: The NBC/SM National Poll

It's been quite the good week for Hillary Clinton. Not only did she crush the debate, but Trump has had several awful news cycles since (some self-inflicted, some not). The model has moved sharply in Clinton's favor this week, and now considers her the overwhelming favorite. If you're curious as to why, skip to the NBC/SM National Poll below, but the short version is:

  • A very large national poll came out that's favorable to Clinton
  • The state polls are also very favorable to Clinton
  • Time is running out
  • My model incorporates information from national polls into its poll aggregation in a way other models don't. This gives it more information to work with, and makes it more confident.


Nearly all of the key post debate polls have either been neutral or good for Clinton. Here's a sampling of them:

  • Good for Trump
    • Gravis National poll shows Clinton and Trump tied
    • Rasmussen national poll shows Trump +2
    • FL poll showing Trump +1
  • Good for Clinton
    • NBC/SM National poll showing Clinton +6
    • Reuters National poll showing Clinton +7
    • Economist/YouGov national poll showing Clinton +5
    • PA polls showing Clinton +5, +12
    • CO poll showing Clinton +11
    • FL Polls showing Clinton +7, Clinton +5
    • OH poll showing Clinton +2
  • Neutral
    • LAT/USC tracking poll holding at Trump +4
    • Lots of polls in states that won't impact the election

The NBC/SM National Poll

If you're not familiar with how state and national polls interact in my model, you can read about it here. The short version is that national polls are, at their core, individual polls of 50 states, and they can and should be treated as such.

NBC and SurveyMonkey release a large online-only national poll once per week. The sample size is huge (YUGE). The 12 NBC/SM polls released since July have had an average sample size of  17,000 voters, much larger than the typical national poll of 1,100 voters. Parsing that large sample out to 50 states ended up having a material impact many of those states. For example, the model is treating the most recent NBC/SM poll the same as if each of the following state polls were added to the model:
  • 660 voter poll in OH showing Clinton +5
  • 390 voter poll in AZ showing Trump +1
  • 180 voter poll in IA showing Clinton +7
  • 581 voter poll in NC showing Clinton +3
  • 163 voter poll in NV showing Clinton +8
  • 1142 voter poll in FL showing Clinton +4
For reference, the average sample size in all of my state polls is 900.
Adding each of those polls to the model, all at once, causes significant movement in the model, especially when they tell a different story than the one we're working with (in the case of the most recent poll, one more favorable to Clinton). Pair that with the influx of excellent state polls for Clinton, and the model feels very good about her chances.

No comments:

Post a Comment