August 28, 2020

Here's why THEY are wrong

Previously, I explained why I didn't believe my own filtering of the REalClearPolitics (RCP) average of polls and why I am inclined to think president Trump will win, despite the polling evidence.  It's a stretch for me, I know.  I'm an evidence guy.  I'm a data guy.  Yet I don't agree with the evidence I've collected. So in the spirit of self-checking and to dispel any sort of confirmation bias on my part, I thought I'd run an experiment of sorts.  Or rather just do some checking.  I took a look at the RCP polls for 2016 Trump vs. Clinton and compare it to 2020 Trump vs. Biden.  I specifically drilled down on Pennsylvania as a battleground state.  More specifically, I looked at the polls and compared August vs August.  Here's what I found.

The results are remarkably similar in both election years.  Consider - Trump did not lead in either state in either year, in any poll.  But in 2016 Trump defeated  Hillary Clinton in Pennsylvania.  Famously so.  If you compare the 6 polls from 2016 to the 6 from 2018, Clinton was ahead of Trump by a combined +42.  Biden is ahead by +33. If you remove the meaningless Registered Voter (RV) polls, Clinton was ahead by +31, Biden only by +26.  In other words, Biden is under-performing Hillary Clinton at the same time period (August) in the election cycle.

Some might argue that pollsters have gotten smarter and are polling better than in 2016.  Show me proof after the election.  Right now there is no evidence to support that claim.  

In fact, when you look at individual pollsters who polled both 2016 and 2018, CBS News/YouGov has Clinton +8, Biden +6.  He's under-performed. Franklin and Marshall had both Clinton and Biden +7.  They've even got a lower margin of error in 2020, so Biden is doing better right?  No, because in 2020 they polled Registered Voters instead of Likely Voters like they did in 2016. Why is that?  Registered Voters always favor Democrats versus Likely Voter polls, and they are always less accurate.  The only pollster in 2020 that has good news for Biden is Emerson.  Emerson has Biden +9 and it only had Clinton +3.  Interestingly in this case, Emerson showed the lowest spread of Clinton over Trump and shows the largest spread of Biden over Trump.  Is it a perennial outlier or the one pollster we can say is an apples-to-apples comparison?  There's no evidence it's the latter without investigating the cross tabs (which I have not yet done).

Evidence, though only circumstantial, would indicate that the pollsters showing a Clinton lead and a Biden lead could be suffering from the same innate bias.  Trump beat Clinton despite the polls (in August anyway), and the same is quite potentially true in 2020.  And Biden is under-performing Clinton at the same point in the race.  The directional indication thus is Trump beat Clinton in Pennsylvania, despite trailing in August.  He trails Biden by less than he trailed Clinton in August.  Therefore it's possible he will defeat Biden by more than he defeated Clinton in Pennsylvania.  That's not an impossible trail of breadcrumbs to follow. 

Now time for the glass half empty view.  In the polls the gap between Trump and Clinton appear to be due to her own ceiling.  Remember, she was not a likable candidate.  Biden, while oafish, is more likable than Hillary was.  Trump's numbers have come up vs. 2016 but Biden has a higher polling average than Hillary did.  It seems he has a higher top end.  The Emerson poll is an interesting example.  Trump scored a 43% in both 2016 and 2020 but whereas Clinton got 46% in the pol, Biden got 52% support.

I eye that even with suspicion.  It is possible that a pollster could (NOT would) hold an option static and then smooth out the results based on that.  In other words, they could have held Trump at 43% and then took the remainder of voters and got fewer not sure/uncommitted and more Biden support.  I don't think they did that.  It's more plausible that Biden is more likable than Hillary was.  That does not translate to more votes in November though.

I'll leave it with one last thought.  The final RCP average in 2016 had Clinton +2.1%, Trump won by 0.7%.  The only two polls that had Trump competitive in the final days of early November were Trafalgar Group (Trump +1) and Harper (Tie).  Most of the pollsters (excluding the RV polls) had Clinton +2.  Morning Call had Clinton +6.  Monmouth had Clinton +4.  Not included in the RCP average but still final week polls, were CNN (Clinton +5), Quinnipiac (Clinton +6) and Franklin and Marshall (Clinton +11).  Remember: The only poll that matters is election day (plus all the monitored mail in voting but that's a story for another day).

No comments:

Post a Comment

Disagreement is always welcome. Please remain civil. Vulgar or disrespectful comments towards anyone will be removed.

Related Posts Plugin for WordPress, Blogger...

Share This