Thursday, September 15, 2016

How do you tell the difference between a polling blip and a real shift?

New polls show Trump continuing to gain nationwide and in swing states.  Is that a real shift or just a temporary blip?  How do we tell?  Time.  That's it.  Sorry.  Everything else is bunk.  Let's debunk the bunk.

1)  Finding the best poll.

Yes, some polls are different.  We can't ask which ones are best.  Everyone has a different methodology, and they're all just guessing at how to figure out who is likely to vote (the "likely voter screen").  To make matters worse, the proper method of doing so can change from year to year.  How do we, as news consumers, handle it?  Take all of the polls together, in aggregate.  Ignore the fancy aggregators, like Nate Silver.  All you really need is a simple, arithmetic mean, like they do at RealClearPolitics.  It gets you the same thing that Silver gets you without the statistical "guitar face."  So, no, you can't just ask which polling organization has the best polls.  Just look at the simple arithmetic mean.  If that continues to shift over time, it's real.

2)  Demographic slicing

Every year, pundits and journalists like to find some specific, narrow slice of the population to call the critical demographic.  Sometimes, journalists like to put some stupid moniker on them.  "Soccer moms" (1996).  "NASCAR dads" (2004).  Bullshit.  Doesn't work that way.  Here's what they are doing:  we call it "the garden of forking paths."  Some data juggler starts slicing up the numbers in every imaginable way.  Statistically, there is bound to be some anomalous "result" somewhere in any data set.  That "result" gets reported even though there was no theory behind it.  It then gets turned into legend, and perpetuated.  Otherwise, some campaign just decides, on the basis of a focus group, to try to target some demographic slice because Person X liked some line in a speech, so obviously, everyone who looks like Person X will respond the same way, cuz LOGIC!

Either way, the faulty line of reasoning goes that if we can just track the supposedly critical demographic, we can distinguish between a blip and a trend.  But, since this whole thing is bullshit, there's no point.  It just takes time.

3)  Good, ole' post hoc ergo propter hoc

The lifeblood of journalism.  The logical fallacy that makes the profession work.  "After this, therefore because of this."  If you can point to an event before the numbers started to move, it must be the cause, therefore it's a shift, not a blip.  The problem, of course, is that you can always find something, if you look.  Yes, pneumonia sticks out, but between the Clinton Foundation and the emails...  there would always be something with Clinton.  That's sort of the point.  She has been under constant investigation for 25 years.  No matter when the numbers shift/blip, you can always find a supposed cause, giving you an excuse for post hoc ergo propter hoc, and you will always call it a shift, not a blip.

The polls have moved in Trump's favor, although Clinton maintains a narrow nationwide lead.  The betting odds have only moved a little, suggesting that the gamblers think it's probably a blip, but distinguishing a shift from a blip will take time.  Sorry.  (Not sorry).

Trump hasn't been very Trump-y lately...  I mean, going on Dr. Oz is pretty Trump-y, but not "rail at a gold-star family" Trump-y.  C'mon, man, I need my dose of Trumpiness.  Someone bait him by pointing out that Carlos Slim is richer than he is...

1 comment:

  1. Trump hasn't been Trumpy. Race returns to baseline.
    Voters are fucking stupid sheeple with goldfish memories.