I thought it would be interesting to compare how the polls are right now compared to where they ended up in 2016. Now I realize many people think the polls are wrong, or even faked, but I personally I do not buy that. There is a lot of misconceptions on what polls actually are, and how to use them. Many people seem to forget they have a margin of error, and indeed, some sources are more accurate than others. Margin of error on polls is easily 3% and can be even higher. Some use Adults, some use Registered voters, some use Likely voters. Some have small sample sizes of less than a thousand, others are bigger. And no poll will always 100% reflect who actually decided to vote on election day. Still, one can look at trends. Of course, polling and trends can change too, so what will be most interesting is the final averages for 2020. But we don't have those yet.
First, look at the final RCP average for 2016 for a bunch of the swing states and compare to the actual vote. Real Clear has the standard RCP averages for today as well as for 2016. Later for current data I will do better and go to Five thirty Eight, who does a better job - they actually rate and eventually will weigh polls based on past performance and biases. But for now, lets just start with the RCP averages. I used the final RCP average for 2016 compared to the actual election results. And if I flubbed a number or a calculation somewhere, apologies. Lots of numbers.
2016 Final RCP versus Actual
Overall: 46.8 Clinton 43.6 Trump (Clinton +3.2) | Clinton 48.18 Trump 46.09 (+2.09)
WI: 46.8 Clinton 40.3 Trump (Clinton +6.5) | Trump 47.22 Clinton 46.45 (+0.77)
FL: 47.0 Trump 46.6 Clinton (Trump +0.4) | Trump 49.02 Clinton 47.82 (+1.2)
PA: 46.8 Clinton 44.7 Trump (Clinton +2.1) | Trump 48.18 Clinton 47.46 (+0.72)
NC: 46.5 Trump 45.7 Clinton (Trump +0.8) | Trump 49.83 Clinton 46.17 (+3.66)
MI: 47.0 Clinton 43.4 Trump (Clinton +3.6) | Trump 47.50 Clinton 47.27 (+0.23)
AZ: 47.0 Trump 43.0 Clinton (Trump +4.0) | Trump 48.67 Clinton 45.13 (+3.54)
OH: 46.2 Trump 44.0 Clinton (Trump +2.2) | Trump 51.69 Clinton 43.56 (+8.13)
IA: 44.3 Trump 42.3 Clinton (Trump +3.0) | Trump 51.15 Clinton 41.74 (+9.41)
I could do more, but it takes time to look the data up. So what do we see here when looking at ALL the polls averaged? A lot of close states, many within the margin of error. Trump won all of the close states, more or less sweeping the battleground states for the Electoral College system. Wisconsin, Ohio and Iowa seems to be the big outliers. But remember, RCP average does not look at poll rating, bias, past performance or sample type. What one could do is create bias or offsets given this data to then apply to current RCP averages, or one could go to FivethirtyEight and use only the best polls. Regardless, we have our RCP baseline.
So here is the final 2016 offsets (Delta in Poll vs Final)
Overall: +1.11 to Trump
WI: +7.27 Trump
FL: + 0.8 Trump
PA: + 2.82 Trump
NC: + 2.86 Trump
MI: + 3.83 Trump
AZ: + 0.46 Clinton
OH: + 5.93 Trump
IA: + 6.41 Trump
For comparison here are the CURRENT (not final obviously) RCP averages, then adjusted with the 2016 offset:
Overall: +9.1 to Biden | +7.99 Biden
WI: +6.4 Biden | +0.84 Trump
FL: +7.8 Biden | +7.0 Biden
PA: +7.4 Biden | +4.58 Biden
NC: +3.0 Biden | +0.14 Biden
MI: +8.4 Biden | +4.57 Biden
AZ: +4.0 Biden | +4.46 Biden
OH: +1.5 Biden | +4.43 Trump
IA: + 1.5 Trump | +7.91 Trump
Note that for some reason both Ohio and Iowa have very little polling right now, so take those current numbers with a big asterisk.
However, we can already see that Biden is currently well ahead of where Clinton was, albeit we are comparing NOW to final RCP 2016 average. Now note that it would only take 3 states flipping from 2016 to change to results for 2020, and currently we see 5 among just this group. But as I said earlier, we can do better. Next time I will look at the polling from FiveThirtyEight and use only the Grade A and Grade B polls.
Nate silver does a great job (my opinion only) in running data. If you go to his page and look at latest polls, he gives you the source, sample size, type of sample and a grade letter rating.For his overall analysis and predictions he adds and subtracts biases in the polls to create his own average.But for now, I will look into some averaging of his top rated polls.I think my next step will be averaging out only his recent polls that have a B rating and above. Unfortunately I don't have his final 2016 data at the moment to create offsets, but that would be even more interesting step, if possible.
So that will all be Part 2 when I get around to it.
Can we draw any conclusions? That's a tough call, but even with the large 2016 offsets, it does appear Biden is doing well. And one could certainly argue those offsets are actually too large. But why did all the polls end up offset in 2016, even if many are within the error margin? Enthusiasm issue with Clinton voters? Hidden Trump voters? Late swing not seen in the polling averages? Good questions. But even with the rather large offsets from 2016, Biden still is ahead for 2020. And if those offsets turn out to be smaller - or none - right now you would be looking at a landslide. And I think one must remember a poll is just a tiny snapshot at a moment in time with a certain chosen, or random, set of people.
2018 Leaf SV Pearl White with Tech and All Weather, Purchased New on 5/3/19