We are ten days away from the midterm election. At stake is control of Congress and a few states. Polls say that the Democrats are going to win the House and are very close to taking the Senate. But a lot has just happened to make polls look very questionable.
First, the trend in almost every midterm election is for the party not in power to do well. Voters tend to turn against a party that controls the White House and both houses of Congress unless things are going very well. Gas prices are up and people are generally not feeling great about the economy. So, every traditional sign is going against the Republicans.
Polls, too, confirm this. But polling is in a bad way.
Pollsters, as they are called, contact people by phone and ask them a series of questions. Apart from asking which candidate they plan to vote for (or have voted for now with early voting), they ask about age, race, religion, party affiliation, and if they have ever voted before. This last question is very important, because people who have never voted usually won’t vote in the next election.
Obviously, they aren’t going to call and speak with every person in the country, so what they have is called a “sample.” That sample probably won’t be exactly representative of the whole country. Older people are more likely to answer the phone and answer the questions. So the pollsters have to weigh this in their calculations. But older people are also more likely to vote, so they then need to swing things back in their direction. From the people they talk with, the pollsters have to move their answers into categories of things like likely voters and then break them into the proportions into the general population.
For instance, if three quarters of white women with children in grade school say they will vote for one particular party, the pollsters, regardless of how many such people they talk to, will assume three quarters of all women with children in grade school will vote that way, too. Not a bad bet, but iffy.
All of this has developed over the past 100 years into a highly sophisticated science. Then cell phones happened. Now a lot of people won’t answer unknown numbers, so this is making the sample very peculiar. In a recent New York Times poll, they called something like 230,000 people, but only 8,000 answered. How do you weigh that kind of sample? What inspired those 8,000 people to pick up the phone? They are already a very unrepresentative sample of the population.
One of the things that people have started to do is take a bunch of polls and average them out. This has been a pretty reliable way of increasing the likelihood of getting things right. It takes that 8,000 respondents and adds them to the respondents of a bunch of other polls.
The problem is that it also adds up the assumptions each pollster is making about how their sample corresponds to the actual electorate. If a lot of them are wrong in the same way, that mistake is amplified by adding them together.
One other big thing just happened. Four separate polling companies (as of today) have now admitted to making up their numbers. In some of these cases they didn’t even call anyone. While there might be many reasons for them to have done this, one prominent explanation is that they were trying to boost the numbers for one side and, therefore, discourage the other side from voting.
There is no reason to think that the polls are so far off that the Democrats won’t win at least the House. There is no way this will be a good election for Republicans. But some races might be closer, or the opposite, than they seem.
Polling can be a very useful exercise, but it can do only so much to tell us what is going to happen in an election. The underlying problem is our desire to predict the future, or to take rational control over the political world. (I’m reading Harvey C. Mansfield’s remarkable book, The Rise and Fall of Rational Control. Highly recommended.)
I once asked a leading expert on polls how accurate they are. He said about as accurate as an election. Elections show the opinions of voters on a particular day, even if voting is now stretched out over several weeks. It is a snapshot, or slow exposure shot. I leave it to my colleague to correct me.
But this is why we have a staggered system of elections. All members of the House are up for election next month, but only one third of the Senate and not the President. No single movement can sweep the government. There is always a check on the passions of the electorate. Polls simply give us the idea that we can control or predict or somehow know more than we really can know.
Enjoy following the race, but put not your faith in pollsters.