Recently, the Election Commission (EC) solicited the views of political parties regarding the imposition of a ban on opinion polls before elections (a plethora of which have been and will likely be conducted in advance of state elections and next year’s general election), purportedly for their undue influence on the electorate. This has predictably spawned a discussion on the merits of the action alongside the understandably cynical observation noting a suspicious correlation between the action of the Congress and other pro-ban parties and the predictions made by the polls. There have been a few articles that have striven to understand and explain the motivations behind such a proposal. In what follows, I attempt to analyze the arguments propounded through the prism of rational (voter) choice theories.
Opinion polls are based on a survey of a suitably large and representative random sample, although the methodology of several of these has been called into question. Most media agencies try to outdo one another in the frequency with which they commission such polls, and sometimes even present conflicting findings. Whether they provide accurate forecasts of elections is a matter of study, but it is their informative role that concerns us here. The proposal by the ECreveals its belief that opinion polls affect the decision of undecided voters, i.e. voters who have not decided whom to cast a ballot for (or indeed whether to cast it at all). This is linked to theories of voter turnout, and the ‘Paradox of voting’. This term, introduced by Anthony Downs in 1957, refers to the puzzle that if voting is costly (e.g. cost of registering, opportunity cost of time spent participating in the electoral process etc.) and a vote matters for its impact on the election outcome, citizens will not vote if they are not likely to be ‘pivotal’, i.e. decisive to the outcome. This is a particularly acute problem in large elections, where intuitively a voter is less likely to be pivotal, and theories suggest that uncertainty regarding pivotal probabilities accentuates the problem. But if nobody votes, the probability that a voter’s ballot is decisive is one…. hence the paradox.This challenge to the paradigm of rational choice has produced a considerable literature on turnout that attempts to resolve this problem with varying success. A prominent approach has been to postulate that voters derive some intrinsic benefit from voting that outweighs the cost, for example a sense of civic duty.
The arguments put forward in defense of a ban hinge on the impact on turnout. They contest that impressionable voters will latch on to the results of such polls when making their voting decisions. There are two forecasts at work here: one where voters back a possible winner (as suggested by the polls), resulting in a clear victory for the frontrunner as determined by the opinion poll (the ‘Bandwagon effect’); and the other where voters support the trailing party, which in turn can attempt to salvage the situation by intensifying its efforts to gain power(the ‘Underdog effect’). Note the role that the opinion poll is playing here, namely serving as a guide to a voter of how pivotal they could be, with the pivotal probability being captured by the difference in mandates for the parties involved. A close call would suggest a larger pivotal probability, inducing more voters to incur the costs and participate. This idea has some backing from laboratory experiments as well as several theoretical studies.
More interesting is why we observe ‘Bandwagon’ and ‘Underdog’ effects in the first place. The main reason suggested by researchers is the informative nature of opinion polls, particularly when information acquisition is costly in terms of either time or money or both. For example, uncertainty on the part of some voters regarding the frontrunner of the election or the voting behavior of other voters could be alleviated by their forming expectations of the outcome based on the results of the poll. As they encounter more information (either via a sequence of opinion polls or other sources), the expectations formed by the uninformed voters are increasingly likely to correspond to those formed by the informed voters. So, if the informed voters tend to back a particular party and this is reflected in the opinion poll, this feeds back into the voting decisions of the uninformed voters, leading to a cascade. Another channel is when voters have preferences over different parties and use opinion poll results as guides to the possible outcome of the election. If the results suggest a close call between, say, a left-of-center party and right-of-center party in a multiparty contest,‘leftist’ voters who might back some other party will line up to support the least harmful option according to them. Recent research has also suggested that a preference for ‘conformity’, i.e. to have the same views and make the same choices as others in a group, is vital to understand bandwagons, as is intuitive. One would expect these effects to be attenuated in the event that voting is costly, as knowledge of a decisive victory for one party would lead its supporters to free ride and not incur the cost of voting.
Which of these forces are at play in the Indian system is a matter of study that would benefit from conducting opinion polls that are open about their methodology and also reveal micro-level information about the factors affecting individual voter decisions. Besides freedom of expression issues that are certainly important, opinion polls also serve as low-cost informative signals to voters and this role should be given deeper consideration during the evaluation of arguments that precedes a decision.
References for the curious reader:
On turnout, the interested reader can consult either a non-technical article like Feddersen (2004) or a more demanding survey by Dhillon and Peralta (2001). Bandwagon and underdog effects were introduced by Herbert Simon in 1954, and have been studied in technical articles by Myerson and Weber (1993), McKelvey and Ordeshook (1985) and recently Callander (2007). Palfrey (2009) surveys laboratory results.