Why do we need opinion polls?

Why do we need opinion polls? It is a good question. We need air to breathe, but we do not need opinion polls, right? While I believe it is a fair question, I also believe it is often misguided or misleading to say that opinion polls are unnecessary. The reason being that we tend to overestimate what we already know about public opinion and underestimate the value of opinion polls.

If people had correct ideas and full information about what the public believes, we would not need opinion polls. The reason we need opinion polls in the first place is that we do not know what those opinions are. If we care about public opinion, opinion polls are useful. There is no reason to believe that we can make valid inferences about public opinion without such polls.

The issue is that people and politicians alike are bad at estimating public opinion when asked. In a recent study, Walgrave et al. (2022) demonstrated how politicians in four countries misperceive what a majority of the voters wants. Accordingly, without the help of opinion polls, we should have no confidence that politicians will have correct perceptions about public opinion.

One explanation could be that politicians simply do not care about public opinion. However, there is no evidence that this is the case. In a recent working paper, Senninger and Seeberg (2022) show in a field experiment that political candidates do care about public opinion. In other words, politicians do not get it wrong because they do not care about public opinion. Consequently, to make sure politicians actually know what the public believe, we need opinion polls.

Next, citizens are not good either at estimating what public opinion is on different topics. In a recent study, Sparkman et al. (2022) showed that while 66–80% of Americans support climate mitigating policies, Americans estimate the prevalence to around only 40%. This is a significant gap between actual public opinion and perceived public opinion. And the only reason we know that? An opinion poll. In another study, Mastroianni and Dana (2022) found that Americans do not know how attitudes towards contentious issues have changed over time, as people, on average, only estimated attitude changes over time correctly on two attitudes (4% of the attitudes in the study). Specifically, people have a tendency to overestimate how much attitudes have changed over time.

Again, if politicians and ordinary citizens had correct believes about what public opinion looked like, we would not need opinion polls. The problem is that once we have opinion polls on specific topics, the numbers all make sense and we would not need the poll at all. To quote Duncan Watts, everything is obvious once you know the answer. And if an opinion poll shows something we disagree with, we are more likely to engage with the methodological details and find out why it does not align with our attitudes. This is not to say that we should not be critical towards opinion polls (we should and we are), but that we need opinion polls if we want to be informed on public opinion.

The other day, I saw a tweet saying: “Polls are fine, but this analysis of cars visible on Google Street View finds voting patterns can be predicted based on sedans vs. pickups”. I have a problem with all these “Polls are fine, but …” statements. If you look at the study in question, there is only a correlation of .75 between cars visible on Google Street View and voting patterns. Trust me, if results from opinion polls only correlated .75 with actual election results, we would throw them out without hesitation.

In our book, we do discuss alternatives to opinion polls to capture public opinion, such as vox polls and social media. My reading of the literature is that we are a long way from having any sensible alternative to opinion polls to capture public opinion, and it is not even close.

We do need opinion polls. Not because they are perfect (they are not), but because we overestimate our own ability to estimate public opinion and underestimate how good opinion polls, on average, actually are.