A myth that has been propped up by a great many people is that the polls were off in 2016. When one points out that Hillary Clinton lead in the average of national polls by 3.2 % and she won by 2.1% and so her popular vote margin was only off by 1.1 percent, then others say that the state polls of Michigan, Wisconsin, and Pennsylvania were really far off. The truth is that the polls were not off in those three states. Michigan, Wisconsin and Pennsylvania are trending red. So, a national lead of around 3 percentage points seems like it should have lead some to think that these states could be in jeopardy.
Now, Hillary Clinton had historically high unfavorables. The White House had been held by her party, the Democratic Party, for the last eight years. She had served in that Democratic White House as Secretary of State and had extensive experience in government. Trump had no experience in government. Therefore, she was a quasi-incumbent. Then, we have the fact that she was under investigation by James Comey. Put all of these facts together and one should conclude that undecideds were likely to break strongly against her. Some undecided voters won’t vote. In 2016, there were a high number of third party voters and it stuck even in late polling and was even realized at the ballot box. This also indicated her vulnerability. Now, some of the undecided voters might have chosen to vote third party, but that year third party candidates were better known and if the undecided had just decided to go third party, they would have said so to the pollster. Some of the undecided might have decided to simply skip the presidential part of the ballot or write in a candidate. However, most of the undecided who did vote were going to vote for one of the two candidates of the two major political parties. Due to all of the factors mentioned above, it should have been clear that the vast majority of the undecided voters who did vote were going to vote for Trump.
Big three states in polling and in votes
State |
poll for hrc |
vote pct for hrc |
difference |
michigan |
47.0% |
47.0% |
0.0% |
wisconsin |
46.8% |
46.5% |
.3% |
pennsylvania |
46.8% |
47.5% |
.7% |
Michigan, Wisconsin, and Pennsylvania were the three states that gave Trump the electoral college. So, in Michigan, the difference between what the polls on average predicted as her share of the vote and what her share of the actual vote was was exactly zero point zero percent. No difference. The average of polls predicted her share of the vote exactly correctly. In Wisconsin, the difference in the percentage of the vote that Secretary Clinton received and the share of the vote that the average of the polls predicted that she would receive was only three tenths of one percent. That’s very close. In Pennsylvania, the difference between the share of the vote that she received and the share of the vote that the average of the polls predicted that she would receive was only seven tenths of one percent. Again, that was extremely close to what the polls predicted for her !
And considering the factors I mentioned above, the rest of what happened shouldn’t be that surprising. It was not the case that almost all undecided voters would not vote for president or write in a vote for president or vote third party for president. Some undecided voters were going to vote for one of the two major party candidates. In fact, most of them did. And should it really be surprising given the unfavorables and the investigation and Comey’s sermon on morality and Clinton’s extensive time in Washington and Trump’s lack of a record that the undecided voters would end up mostly going to Trump ?
Is a poll inaccurate simply because it doesn’t allocate the undecided voters ? A great many of them are going to vote for one of the two major party candidates for president. The election outcome really makes sense when we see where Hillary was at in terms of percentage of the vote and her high unfavorables and Trump’s lack of record and the fact that the Democratic Party held the White House for the previous two terms. I don’t see how one can reasonably argue that the polls were inaccurate. All that happened was that many undecided voters went to Trump. And given the factors mentioned here throughout the diary, it shouldn’t be surprising that they ended up going to Trump. But a pollster can’t allocate the undecided themselves. If a person tells a pollster that they are undecided, then that’s what the pollster has. A forecaster like Harry Enten or Nate Silver has to predict how the undecideds will break. But that’s not the job of the pollster and if they do that, then they are not doing their job correctly. Therefore, it seems to me that the person who says that the polls were off in 2016 doesn’t under the job of the pollster very well.
.
Somebody might say, then how could a pollster be inaccurate ? Suppose Candidate A averaged 52% of the vote in the polls and Candidate B averaged 42% in the polls and the outcome was Candidate B won 52% of the vote and Candidate A won 42% of the vote. Here, the difference is not in the allocation of the undecided. A great percentage of the vote predicted to go to Candidate A went to Candidate B . They may have incorrectly modeled the election and weighted improperly because turnout was different. And we still have the same percentage of those who didn’t vote for the two main candidates. So, it can’t be the winning candidate boosting their percentage of the vote in the average of the polls. In general, perhaps a poll is inaccurate or the average of polls is inaccurate when one or more candidates are considerably below their predicted share of the vote. If one can’t easily or obviously explain the divergence in the outcome by the allocation of the undecided vote, then we may say that the average of the polls is off. But when it seems intuitively plausible that the allocation of the undecided voters is what accounts for the difference between the average of the polls and the results from the ballot box, then the polling should be regarded as likely to be pretty accurate.
Let’s look at one example of the distribution of the undecided voters. Trump went up in Pennsylvania from 44.7% in polling to 48.2% at the ballot box, a growth of 3.5% whereas Clinton only went up by .7%. Thus, it is possible among undecided voters who voted for either Clinton or Trump around 5/6 of them decided to vote for Trump [it is conceivable despite the mental gymnastics involved that a demographic or a person who supported one candidate in a poll switched to support the other candidate.] The two main candidates in polling attracted on average only 91.5% of the vote and at the ballot box they won 95.7% of the vote. Remember that we are now weighting by education as well.
.
Biden in Michigan. He looks very hard to beat in that he not only has a seven point lead in the average of the polls, but his share of the percentage of the vote in the average of polls is 49.7% and he almost always has traversed around 50%. Trump is at 42.8% but the rounding makes it exactly seven points in the margin of the average of the polls. Biden is so close to 50% that it is hard to see how he would not get there and Trump is so low and Trump would have to get almost all of the vote not already assigned to him or Biden or else he would have to get some of Biden’s voter to defect. .3% from 50% . Biden’s favorables are positive nationally and in these three states and Trump is underwater in all three states. It is, therefore, difficult to see how Trump gets almost all of the undecideds. Biden’s main job is to not lose his voters.
Biden in Wisconsin looks really hard to beat. It’s not just the better than six point margin which Biden has held for months in the average of the polls, but rather it is his tendency to be at or above fifty percent for months. That must just terrify Trump pollsters. Biden is at 50.3% in the average of the polls with a lead of 6.5%. Biden did have some occasional drops below 50%, but he gradually gained ground and has been at or above 50% for a while now. Obviously, if Biden keeps the share of the vote he has in polls in the election, then he can’t be beaten. Trump has to win over Biden voters according to the polls. That’s an indisputable observation of the polling data. Clinton had what looked like an unbeatable lead, but she only had 46.8% of the vote in polling. That was her weakness.
Biden in Pennsylvania has an average percentage in polling of 49.6 to 44.9% for Trump giving us a sum of 94.5%. That sum will likely increase at the ballot box. Biden simply needs to get a little over 4/10ths of one point to get past 50% and lock Trump out. If Biden retains his vote percentage in the polls which in Pennsylvania has mostly (not always) remained consistent around 49.5% and then gets a little more, then he wins. Another way of saying this is unless if Biden gets x% of the other vote (ie. the vote not for either of the two main candidates) and Trump gets 4.7% +x% +y % where 4.7%+2x% +y% <5.5% where x is an element of the Reals such that 0<=x<.4 and where y is an element of the Reals where 0<=y<.8 and 0<=2x+y<.8, then Biden wins. Unless Biden hemorrhages votes (ie if Biden retains his pct that he has in the polls) , then the condition listed above has to happen and it does not seem to be likely to me. Remember, the polls were pretty accurate about what they predicted for Hillary. In Pennsylvania, she got a greater percentage of the vote than her polls had assigned her. We have about 40 days left until November 3. This looks pretty good.
Bottom line: the polls were pretty damn accurate both nationally and in each of Michigan, Wisconsin, and Pennsylvania.
I really want to emphasize the importance of turnout. If we turnout our voters in these three states, then we win the electoral college. There are just not enough republican voters in these three states if we turn out all of our voters. These states were lost in 2016 because we did not turn out our voters We can see that because the simple fact is that President Obama won more votes in both in 2008 and in 2012 in each of Michigan, Wisconsin, and Pennsylvania than Donald Trump did in 2016 despite population growth. In Wisconsin, it was so bad that Romney won who lost by seven percent to President Obama won more votes in 2012 than Donald Trump did in 2016. So, this was lost by our failure to turn out our voters in 2016. If we turn out our voters in 2020, then we will win all three states and we will shut off all paths for Trump to 270 electoral college votes; Biden will win more than 270 electoral college votes. This is still our most obvious path to the White House. Thankfully, we have numerous paths. 538 says we are just barely more likely to win Ohio than Trump, just barely more likely to win North Carolina than Trump, slightly more likely to win Florida than Trump, and more likely to win Arizona than Trump. We are competitive in Iowa, Georgia, and Texas. We are leading in the polls in NE 02 and Maine’s rural electoral college vote.
Let’s GOTV !