Diary

Are polls showing a Trump lead reliable?

Dan Spencer has posted that Trump has gotten some good polls of late.

Leon Wolf has tweeted that the electorate in the poll sample does not match what we saw on Election Day 4 years ago.

Let’s take a closer look, and also compare to the last round of Quinnipiac polls from 3 weeks ago.

In Pennsylvania, Hillary was up 1 and is now down 2.

Ohio was a tie then and is a tie now.

Florida, Hillary has gone from up 8 to down 3.

 

In Florida, the distribution in the latest polls is

31% Republican, 29% Democratic, 32% Independent, and 7% other.

Three weeks ago it was  FL 31-33-30

Ohio: 30-30-34-6      Then OH 27-34-33

Penn: 34-35-25-6      Then PA 33-37-26

 

So in Ohio Trump has gone from tied in a heavy Democratic sample to tied in a balanced one, suggesting he is doing worse there now.

Now how does this compare to 2012?

In Florida, it was 33-35-33, a little bit better for Democrats.

In Ohio, it was 30-38-31, much better for Democrats, and similar to June.

In Penn, it was 35-45-20, much better for Democrats, even more than June.

So is the polling wrong, or is it possible that this is what the new electorate is?  Keep in mind that in 2012, Republicans felt that the polls were wrong and terribly off in Party ID, when they were actually accurately identifying the Obama turnout operation.

In Pennsylvania, to get an electorate of 34-35 split, instead of 35-45 on Election Day, would have required reducing Democratic turnout by 500,000.

In Ohio, it would have required 400,000 ‘missing’ votes.

There is some evidence that the Obama turnout operation achieved results of this magnitude.

Particularly, in 2008, Obama netted 480,000 votes in Philadelphia county, which was an improvement over Kerry’s record 410,000 margin, itself so much better than Gore’s 350,000.  This was while the county population was flat.  The gap was a smaller 467,000 in 2012.

Combined with suburban counties and Pittsburgh and elsewhere, it is reasonable to think Obama’s turnout operation boosted votes by 200,000.

Then you have the people who switched parties ahead of the primaries, reported as more than 60,000.  These people would count double, since they are being subtracted from the Democratic column, and added to the Republican one.  I am assuming they continue to identify with the new party to the pollster and didn’t just switch to wreck the GOP primary(though they are not exclusive).

Total Republican registration has increased by 150,000 since last year, so there is another 90,000 Republicans.

So that is estimated 200,000 + 60,000 * 2 + 90,000 = 410,000 balanced by people who switched from Republican to Democrat and new Dem registrations, and extra voters produced by Romney’s turnout operation.

There is also the possibility of people identifying with the new party to a pollster without actually switching,

 

It can’t be ruled out that this party ID is accurate, though more likely it is something similar to the June numbers, where Democrats have a slightly larger share of the electorate.