There is common perception among pundits and the public the polls were hugely wrong in 2016. To be fair, I have been extremely harsh on pollsters recently for past failures in 2018 and their samples for this year. I can see them being off again like 2016. But, this is conjecture on my part. However, I recently viewed a fun video on Youtube positing what would happen this year if the polls were off by as much as 2020? In the video, Biden pulls over the 270 mark for a win but it has caveats. First, the author pulls the 2016 polling results from RCP but the 2020 average from 538. Secondly, the 2016 polling average to final results is incomplete because RCP aggregates its polling average periodically depending on time frame and number of surveys.
So, for my analysis there are some ground-rules to make it a more even playing field. First, both the 2016 and 2020 polling aggregates will come from RCP and not 538. I have nothing against 538 but the site often pulls in many partisan surveyers and those of dubious quality. Secondly, the average of 2016 polling results will be capped at one month from the election irregardless of the number of surveys done in this time-frame. Obviously, this means some states will have seen more polls than others but due to being unable to control the number of surveys we can at least control the time-frame. Finally, the states consider “swing” in this analysis were considered competitive in both or either 2016 and 2020.
Results Trump =/- (rounded up/down by 10ths)
|State||2016 Polling Average||2016 Result||2016 Trump +/- Swing||2020 Polling Average||2020 Result w/ 2016 Swing||2020 Winner|
* Due to lack of surveys, polls were averaged that were taken longer than a month ago.
** No recent polls taken in the last three months
A number of things stand out from the table above. A number of states – Minnesota, Missouri, Nevada, New Hampshire and Ohio have been scarcely polled this cycle. As a result, a single poll significantly skews the average. Secondly, compared to the abbreviated RCP average of polls on the eve of the 2016 election, the average of polls 30 days before the election shows Clinton often had larger advantages. For example, in Florida, RCP had Trump ahead .2 percent on the eve of the election but a 30 day average showed Clinton up .9 percent. In North Carolina, a 30 day average of polls had Clinton up 1.4 percent. But RCP’s abbreviated average had Trump up a point. Tied to these differences is when the polls are looked at in total, it is clear Trump was closing the gap in the run-up to November. Using Michigan as an example, Clinton had a double-digit lead in some polls in early/mid October but on the eve of the election most polls saw her with a five to six point edge. Another example is Pennsylvania where Clinton started October with high single digit to low double digit leads. But on the eve of the election, polls saw her with one to four point leads (note: Trafalgar had Trump up by one and they were ultimately proven right). This partly reflects the tumultuous nature of the 2016 election and how the media’s see-sawing negative coverage of Trump and Clinton drove the polls.
But for fun, assuming the polls and averages held until November, Trump would fall short in in every state he lost in 2016 – including Minnesota – while also seeding Florida and Pennsylvania to Biden. Arizona is tied. Biden’s two flips would mean he would have 282 EV, irregardless of how Nebraska’s 1st CD or Maine’s 2 CD voted and win him the White House. However, it is unlikely if the polls are off so much in the Midwest it would not filter down to other red leaning states such as Florida and AZ.
All in all, it was a fun exercise and a longer term analysis of polling can provide hints about 2020 as well as show a pattern we missed in the past election. For the media, pollsters, and analysts, the data paints a picture they missed. Trump was closing the gap in the run-up to the election and it allowed him to be competitive enough to win with late deciders. Further, and this is a warning to pollsters yet again, it points to how even though pollsters miss non-traditional issues gaining salience in elections – globalization, trade, etc. – underlying trends in the data point to these issues coming to the fore.
Recently, history may be repeating itself again. Trump is gaining in the polls, arguably due to the impact of BLM, the ineffectiveness of the Biden campaign, and widespread looting rocking the nation. Even voters who don’t like Trump personally may be, just like 2016, deciding to throw their lot in with a man they know may be vain and vulgar, but is serious about protecting American cities. We will see.