Can asking about people’s social circles inform election predictions?

Relying on the wisdom of crowds can be useful for predicting future events. Election polls could rely on the wisdom of crowds by asking people which candidate they think will win. However, most election polls do not do this. Traditionally, most election polls have only asked people for which candidate they plan to vote. The 2016 USC Dornsife / LA Times Presidential Election Poll, in addition to asking people about their own voting preferences, harvests the wisdom of crowds to complement its forecast of the election.

While the full analysis of how wise the crowds really were will have to wait until after the election, we can look at some of the initial results.

Two ways of harvesting the wisdom of crowds

In our Presidential Election Poll, we rely on the wisdom of crowds by asking for people’s expectations about election results in two ways.

First, we harvest the wisdom of crowds by asking people about their expectations of how all Americans will vote. Here, we ask people about the percent chance that each of the candidates will win the election. Previous experience with this approach has shown it to yield accurate results; for example, in the 2012 election, people’s answers to this question have consistently pointed to the actual election winner.

Why might the question about the percent chance that each of the candidates will win work so well? It is believed that it taps into the crowd’s wisdom about their social contacts. Indeed, people can give quite accurate reports about various properties of their social circles.

This leads us to the second way we harvest the wisdom of crowds: using a question that asks people directly about voting behavior of their adult friends and family. People are asked to report what percent of those social contacts who are likely to vote might choose each of the candidates.

How might asking people about their social circles be helpful?

We think that people’s reports about the percent of their social contacts voting for each of the candidates can improve our predictions of voting behavior in at least three ways. First, these reports can give us a glimpse about what’s going on with the parts of the population that are not well represented in the polls. Second, reporting about social circles might be more truthful than reporting about own voting intentions. And third, the reports can help us understand whether people supporting different candidates live in ‘echo chambers’ in which they mostly socialize with like-minded people.

Social circle reports might be less affected by sample peculiarities and by media reports on national polls

Election predictions based on reports about own voting intentions will often be affected by peculiarities of a particular survey sample. If a sample by chance includes a bit too many Democrats, the resulting survey estimates might be skewed towards Democrats’ preferences; and vice versa.

Election predictions based on expectations about how all Americans will vote (the first way of harvesting wisdom of crowds mentioned above) could also be skewed, this time towards the national polls people hear about from the media. To the extent that the polls are erroneous, people’s expectations will also be inaccurate.

In contrast, election predictions based on social circle reports might provide information about voting intentions of a larger and more diverse sample than the original survey sample, and are therefore less likely to be affected by the sample specifics. Because they are about people’s close social contacts, they are also less likely to be affected by the national polls.

Predictions based on social circles might therefore fall somewhere in-between more extreme polling predictions. Is that indeed the case? Figure 1 shows preliminary results from the three waves of the Presidential Election Polls in which the social circle questions were asked so far. Predicted lead (or lag) of Trump compared to Clinton based on social circle reports typically falls in-between own vote intentions on the one hand, and population-level win expectations and national polls on the other hand.

Of course, only time will tell which of these predictions were more accurate about the elections.

Figure 1. Different election predictions (and confidence intervals) from the Presidential Election Poll, compared with average results of national polls published in the same time periods. * “What is the percent chance that you will vote for Clinton, Trump, or someone else?” ** “Of all your social contacts who are likely to vote, what percentage do you think will vote for Clinton, Trump, or someone else?” *** “What is the percent chance that Clinton, Trump or someone else will win?”

Reducing social desirability of answers about likelihood of voting

The Presidential Election Poll asked about the likelihood of voting in two questions. One question asked people about the percent chance that they will vote in the upcoming elections. The other asked them about the percent of their adult social contacts that they think will vote.

Voting is typically seen as a socially desirable behavior, and people might be reluctant to admit that they won’t vote. But when reporting about their friends, people don’t have to embellish their answers. Therefore, the social circle reports might be more accurate indicators of actual election turnout than reports about own likelihood of voting.

Our initial results suggest that social circle reports might indeed be more truthful than questions about own voting intentions, because they produce a lower percentage of likely voters (approximately 73% of social circles) than the responses about own likelihood to vote (approximately 79% chance of voting). The social circle reports are closer to the turnout in past elections, which has rarely been larger than 60%.

Echo chambers of people voting for different candidates

The social circle question also allows for investigating how homogenous are the perceived social circles of Trump and Clinton voters. Is one or the other group more likely to live in an ‘echo chamber’, surrounded by people with similar views? Our results suggest that both groups live in quite homogeneous social circles, with a large proportion of their social contacts sharing similar political views.

Comparing people who would vote for Clinton with 100% certainty with those who would vote for Trump with 100% certainty, the results suggest that staunch Clinton supporters might live, or perceive to be living, in slightly more homogenous social environments. Among the social contacts of these strong Clinton supporters, on average 17% are Trump supporters. Among the social contacts of the staunch Trump supporters, on average 22% are Clinton supporters. The difference persist when we adjust for the baseline percentage of Trump and Clinton voters in average social circles.

These results can be interpreted in two ways: Clinton supporters may indeed be surrounded by more people sharing their political opinions. Or, Trump voters surrounded by staunch Clinton supporters might be reluctant to admit that they support Trump, creating an illusion for Clinton supporters that more people think like them than it is really true.

Looking ahead

In further analyses after the elections, we will be able to determine how useful the additional information form social circle reports really is for election predictions. Meanwhile, these reports might already provide valuable information about the social desirability of voting and about echo chambers of different groups of voters.