Wednesday, March 9, 2016

Yesterday's normality and weirdness

Trump keeps winning.  Betting odds still put Trump's nomination chances at around 70%, or, roughly 10 points lower than before the brokered convention "strategy" (I use the term loosely) developed.  No surprises there.  Is 70% too low?  Maybe, but perhaps it just factors in what is now my standard 2016 disclaimer:  everything could go out the window because in 2016, down is up, left is right, and pop music doesn't suck.  Wait.  Scratch that last one.  Pop music still sucks.  We may be in bizarro world, but some things are immutable, universal constants that even Q can't change.

But let's talk briefly about Michigan.  It wasn't just one poll that got it wrong.  It was every poll.  Clinton had a massive lead in every poll, and Sanders pulled out a narrow victory.  What happened?  Either public opinion changed dramatically and instantly, or the polling methodology got it wrong.  There are two main ways that could happen-- sampling error, or the "likely voter screen."  Sampling is hard.  We can't just put every registered voter's name and phone number into a big hat, pull them out at random, and legally compel people to respond, and do so honestly.

Why not?  Because I don't rule the universe.  Yet.  Someday.  And when I do, pop music will stop sucking.  Oh yes, it will happen...

I'm back.  Anyway, sampling errors are not constant across polling organizations, so in aggregate, those usually cancel out.  That is why we look at polling averages rather than individual polls.  If organization A oversamples a group, and organization B undersamples the same group, the polling average gets it right.

Ah, but the likely voter screen.  Which respondents are likely to vote?  That one is just plain hard.  We know how to predict turnout in a general election.  Primaries are weird.  Turnout is lower, and not randomly so.  And we don't always know the direction of the bias.  Most likely, turnout among Sanders supporters was just higher than anyone expected.  Or maybe Hillary supporters were overconfident, and hence lazy.  Who knows?

There's no such thing as "statistically impossible."  We are always uncertain.  Or, at least, we always should be.

What does that mean for Trump and the Republican nomination?  It means defeating Trump is getting to the point that it requires something as weird as what happened with the Democrats in Michigan-- not impossible, but we should be surprised at this point if the nominee is anyone other than Trump.  And the Democrats?  Be realistic.  Here is the current delegate count.  Bernie is an unusually strong protest candidate, but not a plausible victor.  Hillary is just too far ahead.

No comments:

Post a Comment