Obsessing About Presidential Polls Before Labor Day Is For Suckers

-600Days -23Hours -20Minuts -36Seconds

That’s pretty much it. Carry on.

OK, maybe a little more. Couple of points:

1. Fundamentals. Trump is running a “campaign” only in the loosest sense of the term. No field, no money, no ground game, no ads, and he’s chasing around like an idiot claiming he’s going to compete in states he has no chance of winning. This will assuredly change, as Paul Manafort is not a moron. But Trump is months behind and tens of millions of dollars short of what he needs to organize a real campaign. It doesn’t happen overnight. These are enormous and arguably insurmountable problems.

2. There are so many polls this year that there are always going to be “OMG WTF DID YOU SEE THAT POLL?!?” moments. Restrain the urge to do this. The more polling, the more chance there will be for outlier results that are never replicated. Every poll must be viewed in the context of other recent polling. If one poll says Trump by 11 but there are 4-5 others that say Clinton by 5, the Trump poll has to be discounted unless and until someone else replicates it.

3. Know thy pollster. Polls by Rasmussen, a notorious hack of a polling outfit, has consistently had Trump leading for some time now. A new Rasmussen poll saying Trump is leading is not newsworthy. There are articles assessing the accuracy of many polling outfits all over the Internet; here’s one by Nate Silver from the immediate aftermath of the 2012 presidential election.

4. A “close” poll is not really close if it has been replicated many times. This is primarily for later in the race but it’s important. If one poll shows Clinton leading by 1.5 points in Florida, that’s close and within the margin of error. But if 20 polls all show Clinton leading by 1-2 points that is not really “close.” The confidence level on a large number of polls with similar results goes up quickly the higher the number of polls showing a similar result. This was the situation in 2012, when Obama led Romney by 1-3 points in a whole host of swing states, results that were consistent across multiple polls in each state over a period of weeks. The media persisted in saying the results were “within the margin of error,” but the high number of polls combined with the consistency of the outcomes meant that the confidence level approached 100%. Guess what? Obama won all the swing states.

5. Likely voter models four months out from an election are virtually useless. For many pollsters, a likely voter is one who says she’ll be voting. Well, who’s going to tell a pollster, “no, I’m not voting”? Not very many people. As we get closer, the models get more accurate and the results more meaningful.

6. As a result of all of the above, if Democratic leaders are freaking out about one set of dubious polls that are wildly different than other recent polling, they’re morons. Get a grip, folks. Act like you’ve been through a campaign before, would you please?

Now I’m done. Recommence normal breathing. 

2 thoughts on “Obsessing About Presidential Polls Before Labor Day Is For Suckers

  1. MadamX2016

    The congressional district polled should be included. Otherwise, this is just a pointless exercise. If I know the pollster polled in a majority republican district (Rasmussen ?) then we get a good read, right? The poll matters only in relation to the partisans polled.

    Bernie Sanders was supposedly ahead in California http://www.inquisitr.com/3050424/bernie-sanders-polls-desperately-seeking-momentum-sanders-may-now-have-it-new-analysis-predicts-a-61-percent-win-in-california/

    “One new analysis shows there could be hope at the end of the line for Sanders. An analysis of California polls from Political Analyzer found that Bernie Sanders is on pace to take more than 60 percent of the vote in the stateā€™s June 7 primary.”

    Humboldt County population134,493 (2013): Sanders won by a blowout 68% (13,245), Clinton 31% (6,032). http://graphics.latimes.com/election-2016-california-results/

    In other words the pollster seemed to find the one county where a landslide Sanders win would come and for whatever reason expanded that win outside of the Humboldt county/congressional district.

    Where the poll is taken matters therefore pollsters should provide the county/congressional district.

    June 5, 2015 Poll – http://www.newyorker.com/news/john-cassidy/bernie-sanders-raises-the-stakes-for-the-california-primary

    “These figures augur well for Sanders, as does some of the recent polling data. The Field Poll showed him running ahead of Clinton in the northern half of the state, including the Bay Area. Clinton was doing best in Los Angeles County and the Central Valley, both of which have large minority populations. However, the poll also showed that, among Latino voters, Clinton was leading Sanders by just four points, forty-six per cent to forty-two.”

    Primary Results, San Francisco Bay Area, California
    Population 17,755 (2013) Mariposa – Sanders 51.2%, votes 1,050, Clinton 47% votes 963
    Population 62,864 (2013) Santa Cruz – Sanders 55.3%, votes 23,171, Clinton 44.2% votes 20,105

    Los Angeles County population 10.02 million (2013):
    Los Angeles – Clinton 57% votes 590,502, Sanders 42% votes 434,656

    Final Results 2016
    Hillary Clinton delegates 269 votes 1,940,580 percentage 55.8%
    Bernie Sanders delegates 206 votes 1,502,043 percentage 43.2%

    Two issues (1) The total populations of Sanders blowouts were not comparable to county populations where Clinton won. (2) A pollster publishing such a non-comparative population poll should be totally discounted in the future unless providing the county/congressional district polled sample to compare population reality.

    Reply

Leave a Reply