The US election: gone completely wrong

Another election in the United States has ended and criticism of opinion polls is back on the agenda. The latest average polls from RealClearPolitics and FiveThirtyEight showed Biden’s lead of +7.2 and +8.4 percentage points, respectively, nationally. So why was the race as smooth as it got? How could the polls fail so much?

The short answer is that It was never as simple as one had the impression and that opinion polls did not lose the result after all. The long answer includes votes-by-mail, extremely even statuses, and a continuing mystery about why opinion polls consistently underestimate Trump’s support.

The election result ultimately showed a relatively convincing victory for Biden. It was always about twelve whiplash states (Texas, Iowa, Ohio, Georgia, North Carolina, Florida, Arizona, Pennsylvania, Nevada, Michigan, Wisconsin, Minnesota) that would distribute a total of 193 delegates. Of these, Trump had to win over 142 delegates, and thus Biden only needed 53 of these to walk away with the winner of the election. This was why my colleagues at Ipsos USA thought that the most likely outcome was a button for a moderate victory for Biden, and why most others also thought that he would win.

Trump foresaw this

Nationwide look As of this writing, Joe Biden is approaching 80 million votes, while Trump is likely to end up with around 75 million. This, with all the votes from other candidates, will likely give Joe Biden a margin of victory of between 3.5 and 5.5 percentage points of the votes nationally.

We would have woken up This result after election night, no one had considered equally exciting. The reason it still looked so even was, first of all, because it took a long time to count all the mail-in votes, and these were posted largely by those who voted for Biden.

It is also a point that several states were very even, especially Georgia, Arizona, Pennsylvania and Wisconsin, but it must also be remembered that Trump had to win three of these and victory had to include Pennsylvania since that state has the most delegates . Therefore, Trump had the odds against him all along.

The main reason for that we are left with the impression of a fair election is due, however, to opinion polls. Although the vast majority of opinion polls, and all of the prediction models that were based on them, generally predicted the correct outcome in 48 out of 50 states, one was under the impression that the victory would be significantly greater and clearer. As mentioned in the introduction, Biden was aiming for a national victory of around +8 percentage points. When all the votes have been counted, it can become about half of this.

In the 2016 elections, we saw the largest blisters in Pennsylvania, Wisconsin and Michigan, where the average poll had Hillary Clinton leading by +3.6 percentage points, but where she lost by -0.7. In Wisconsin the advantage was +5.4, but ended with -0.8, and in Michigan with +4 and a score of -0.2.

Of course, Trump could lose

Few measurements were made here in 2016, leading to many more measurements being carried out in these states this year. Therefore, it is worrying that the result looks even worse.

In this year’s election, Trump was early with wins in Florida and North Carolina, while Biden passed him at the finish line in Georgia and Pennsylvania. Aside from the latter, the measurements here were fairly uniform. For example, Biden led Florida with +2.5, which is by no means safe for those of us who work with opinion polls.

However, it was worse in the states of Wisconsin, Michigan and Ohio. It was here that Trump surprisingly won in 2016, which became the subject of all explanations of why he might be elected.

This is where we must redirect our attention; here it went completely wrong again. Biden’s lead at Wisconsin was +8.4, but he appears to win by less than a percentage point. In Michigan, the lead was +7.9, ending at around +3. In Ohio, Trump had the advantage with +0.8, but finished with more than +8.

These are many Numbers and details, but the big picture is pretty clear: Trump was underrated in polls in almost all so-called tip states and many safe states as well.

A common explanation is that respondents are hesitant to answer that they will vote for Trump when pollsters call them. This is quite doubtful. In that case, the differences should be smaller in clearly Republican states where one is not as afraid to vote for Trump, but where the discrepancy is often as great or greater. At the same time, we see that the Congressional elections also appear to have underestimated Republicans. What stands out as the best explanation so far, and which may be similar to the mistake made in 2016, is who answers when opinion polls call: there are far more frequent respondents with higher education.

In Norway we see in Ipsos the same. If we call 1000 respondents completely at random, we end up with more than 50% of the respondents with higher education. The actual share in Norway is around 33%. We solve this problem by weighing the results. So a respondent with a lower education will receive more weight than one with a higher education. They also did this in these states of the United States before the elections, but they still didn’t get it right.

However, the trend is unequivocal, and it is that respondents with higher education want to answer surveys to a greater extent. When this is why opinion polls continue to underestimate Republicans, it testifies first and foremost that America is becoming increasingly polarized. This also shows the greatest weakness of opinion polls, and in the industry we must continue working to improve, if we want to have the confidence of the population.

America is a special case. In Norway, the industry repeatedly manages to measure correctly before elections, because opinion polls are, in general, a very good and reliable tool. But one must be aware of the weaknesses and limitations that it entails.