I am interested in prediction markets, where traders have a monetary incentive to bet on who they think would win in an election. I think that the financial incentive in prediction markets makes them a better predictor of future election outcomes than polls.

I think that when people are asked how they would vote on the presidency if tomorrow were the election day, they tend to give you an answer which they would like to see happen rather than how they would actually vote. This discrepancy reduces the predictive value of polls. Alternatively, I think that if the incentives are set correctly the same people tell you more of the truth instead of their wishful thinking. I would call this phenomena "polls bias".


  • What is a scientific term for this problem?
  • Where can I read further about the phenomena?

2 Answers 2


It is the social desirability bias.

This is one of the reasons a behavioral study may give better results than a self-report study.

  • 2
    $\begingroup$ Could you also include a reference to a good survey on this if you know one off the top of your head? It would make the answer more complete and self-contained. Do you know any applications to prediction markets as described by the OP? $\endgroup$ Commented Jun 11, 2012 at 12:21
  • $\begingroup$ No, not off the top of my head. I will try to update the answer later. Please feel free to edit it if you wish. $\endgroup$ Commented Jun 11, 2012 at 12:28
  • $\begingroup$ The Iowa Electronic Markets constantly forecast US political outcomes. This market is in 74% of the time better than the polls. 100 days in advance of the predicted event 100% more accurate than polls. (Berg et al. 2008) sciencedirect.com/science/article/pii/S0169207008000320 $\endgroup$
    – Alex B
    Commented Jun 11, 2012 at 12:30

There are many reasons why stated intention can differ from behaviour (e.g., see Theory of Planned Behaviour). Social desirability is one possibility. However, there are many other factors. A person may not know who they are going to vote for or may not have decided. Factors may arise between the point in time when asked and when voting occurs that change the person's voting plan.

Furthermore, there are many potential limitations to polls that can lead their estimates of population voting intention to be inaccurate, and even biased. There is basic sampling error that arises whenever you estimate a population property from a finite sample. There are also a wide range of issues related to getting a sample representative of the voting population (see sampling bias).

Furthermore, polls form part of the information available in betting markets. Thus, in betting markets, people are able to use an ensemble of information including multiple polls, various historical trends regarding how polls change over time, various adjustment to poll predictions that might remove perceived biases and so on.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.