Volume 6 Number 300
We all know how the pollsters screwed up royally in 2016, after calling the prior 16 elections correctly since Truman and Dewey in 1948. They almost all got it wrong this last time, by predicting Hillary would beat Trump. There was a lot of egg on their pusses. Finger pointing identified “shy” voters, by hidden Trump supporters who were embarrassed to say they were voting for him, as a primary reason for the widespread miscalculation. It is thought that this moved Trump into the ‘win” column in close, ‘battleground” states, and there were a handful of those to move him over the top.
Others pointed to some inaccuracies in the way samples were drawn: over-sampling the college educated, while under-sampling non-educated white voters and undecided voters who decided at the last minute to go for Trump.
Some blamed it on factors the polls couldn’t take into account: Democratic voters being so sure Hillary was going to win that they didn’t turn out, voter suppression preventing primarily lower income people, who usually vote Democratic, from getting to the polls, and Republican operatives doing an unexpectedly good job of getting out the vote.
Still others blamed the pollsters for not adequately considering the vast number of non-responses from people not wanting to participate in the polls. All of these can lead to distortions, so that the polls don’t represent an accurate sub-set of the voting preferences of all the American people.
To give the devil his due, in 2016 most of the polls in one sense did have it right. Hillary got more votes nationally than Trump, just as Gore had more votes than Bush in 2000. But we all know that the electoral vote is the one that counts, and the Republicans won in both those presidential elections. The 2016 mistake occurred because certain individual states have too few votes to be adequately represented in a national polling sample. Overall, while this balances out to the national parameters, each state may not be accurately represented in smaller sub-samples. This can be disastrous when the electoral vote must consider each state and not the popular vote. For financial reasons, the pollsters can’t do a humongous sample with, perhaps, a thousand people in each state, and then weigh the results.
So, one thousand people are used to represent the whole country, which is adequate in total, but doesn’t reveal the nuances of each state. That is why the polls got it wrong in 2016. They weren’t adequately representing “battleground” states like Pennsylvania, Michigan and Wisconsin, that went for Trump by small amounts. Separate large subsamples are needed in these states, to truly represent what is happening “on the ground.” These should then be weighed in with the remaining states to give a more accurate picture of what is going to happen.
On second thought, think of how much valuable polls would be if they did an adequate state-by-state poll, of let’s say, one thousand per state. Then the polls could be reporting what the electoral vote is instead of the popular vote. Imagine if the campaign manager had an accurate reading of how they were doing in each state. The costs sound prohibitive: having to poll 50,000 people by telephone rather than 1,000 or so in a current national poll. The cost would probably be driven down if this was all done electronically. But, at present, most polls rely on telephone interviews because electronic samples are self-selecting and are considered more biased than telephone polls where each respondent has an equal chance of being interviewed.
Another factor that should be addressed is the questioning procedure generally being used. According to the Pew Research Center the following questioning sequence is used in most polls:
- Both the presidential and vice-presidential candidates are included in the questions.
- The party affiliation of each ticket is mentioned explicitly.
- In states where independent candidates are on the ballot, the leading one or ones and their running mates are included in the choices read to respondents.
- The order of presentation of the Democratic and Republican tickets is randomized so that some respondents hear the Democratic ticket first and others hear the Republican ticket first. The independent candidates always follow the major party candidates.
These features are an effort to make the presentation of the options as similar as possible to what voters would experience when casting their ballots. That is all fine and good historically. But it’s a different story in 2020, where starting in September states are started to have absentee or mail-in ballots. It is anticipated, largely because of the coronavirus pandemic, that two-fifths of voters will prefer not to go to the polls and will have voted by election day. The total among Democrats is even higher. It makes no sense to ask a hypothetical question if the person had already voted. What should be sought in polls, where appropriate, is what they did.
So, the polls should take this into account for those voters in states where early voting has already occurred. At the beginning of the preference sequence, the following should be also added:
- Whether or not the respondent has already voted.
- If yes: The candidates voted for presidential and vice-president are read.
- If no: Then the above-mentioned procedure is followed.
If they haven’t done it already, it is urgent that pollsters take into account whether the individual has voted already, and the questioning sequence should also take this into account. Having the actual voting information available would be of great interest to both parties, and to the public. One can conjecture that this information might keep people away from the polls. This shouldn’t be the case–usually less than half will have voted by the time polls open, and that is plenty to swing votes in person.
The “shy voters” effect also shouldn’t be underestimated. It has been found that in telephone polls, when a person is asked to press a button to record preferences, rather than talk to a live person, Trump scored two three points higher than if an interviewer asked them out and out for their preference. Some of the few polltakers who got 2016 right were using impersonal electronic polling. This was somewhat verified by post-election polls, when 35 percent of Trump voters said they had been unwilling to talk about their vote. Therefore, if they haven’t done it already, pollsters should be switching to some kind of press-the-button technique to record preferences.
What can make poll watchers feel good, going into the 2020 elections, is that polls are usually quite reliable. Over the years they have come within an average of two percent of accurately doing the head counting. And hopefully they will have corrected some of the systematic errors that were in the 2016 polls. But, and it’s a big but, there is nothing to guarantee they won’t underestimate Trump again.
“I haven’t trusted polls since I read that 62% of women had affairs during their lunch hour. I’ve never met a woman in my life who would give up lunch for sex.”