Do biased polls skew elections? Experimental evidence says yes
As an election looms in the UK and a presidential vote approaches in the US next year, opinion polls are everywhere. Citizens will have to navigate a lot of noise as they decide which party or candidate to choose.
As our research has found, the way citizens receive information about polls through the media can affect the outcome of elections. We conducted several laboratory experiments to test whether voting behavior is affected by biased reporting of poll results. The results suggest it is.
In our experiments, we asked 375 student volunteers to vote in a series of elections, choosing between two parties—party K and party J. They did so under two conditions: an unbiased control condition, where they saw all polls, and a biased treatment condition, where they saw only the polls where party K's popularity was the highest. The biased scenario conferred a considerable benefit to party K. In it, party K won 80% of the time. In the control scenario, the party won only 60% of the time (this relatively high proportion was due to pure chance).
In another experiment we explicitly informed participants beforehand that they would be receiving biased information about the polls. Remarkably, party K still gained considerably from the existence of the bias, winning 64% of the time compared to 57% in an unbiased setting. Even when our voters knew they were receiving biased information, they didn't seem to discount it enough when casting their vote.
Does media slant matter?
The traditional media selects which polls to emphasize from a large pool of results. Sometimes outlets do this with an eye to make interesting news or pander to the expectations of the public. The latter means that journalists may decide not to publish a poll showing an unexpected result, for example, even if they believe it to be true, out of concern that readers might see them as less credible.
In our research we also found that social media propagates poll results in a biased way. On Twitter, in particular, "good news" for left-liberal parties and politicians spreads more widely than good news for conservatives.
And, as our experimental results show, even when we know this is happening, we citizens struggle to take it into account in the voting booth. Our research tells us this, because we asked participants how they thought candidates of party K and J would perform in the imminent election, getting their estimates on the vote share of each candidate. These expected vote shares were very similar to average poll results, even when voters knew that these results were biased.
Knowing that they've only seen the good results for party K, a fully rational voter would have expected a much lower election vote share for party K than average poll results, but this is not what we discovered in our experiments.
Our participants were provided with considerable information and developed considerable experience—voting in 15 elections. The fact that even in our rich information environment people do not discount biased polls suggests that political experience and sophistication is, unfortunately, not enough to undo the effect of biased feedback.
There is concern about the role of voting intention polls in modern democracies. Our results indicate that this concern may be justified. This was an experimental piece of research and we still need to establish whether the results change in noisier election environments in the real world, but they nevertheless provide important food for thought for voters, media outlets who report polling, and even the governments of countries seeking to hold free and fair democratic elections.