As we approach another general election - OK, not just another election, but a mighty important one - the role, beneficial or deleterious, of pre-election opinion polls is being much talked about. Much of the debate centres on the philosophical question of whether one ought, or ought not, to ban or regulate such polls.
For some, it would be illiberal to do so, whatever the possible impact of polls on voters' behaviour, and the matter ends there. For others, if polls could be shown to influence the way that voters might cast their ballots, then there might be a legitimate case to intervene, in the interests of ensuring a fair and unbiased election process.
Yet, much of the debate stops there, without investigating the empirical question, the answer to which would inform a reasoned policy debate: do polls, in fact, influence voters' behaviour? Some respected columnists (not writing in this newspaper!) have asserted, without basis, that there is no evidence of an impact. They are, evidently, unaware of new research which suggests that the opposite might be true.
An important new paper by economist Manasa Patnam ("Learning from exit polls in sequential elections: Evidence from a policy experiment in India", CREST, France, February 2013), studies the impact, not of pre-election polls, but of exit polls conducted in the middle of multi-stage election contests.
Mid-election exit polls pose unique problems not shared by pre-election polls, in that they create the possibility that the order of voting, which is arbitrary, may have an impact on the election outcome. If the choices made by early voters - at any rate, as approximated by exit polls - influence later voters, then earlier voters will have a disproportionate impact on the final outcome.
For instance, if early voters heavily favour party "A" over "B" - merely because early-voting regions contain a disproportionate share of "A" supporters - there might be a momentum swing in favour of "A", which would not have occurred if a different region had voted first. On the other hand, heavy early support for "A" could galvanise supporters of "B" voting in later phases of the election to come and support their party in greater numbers, and so there might be the opposite impact.
Analysing data from the 2004 general and state elections, Patnam finds that voters in later phases were, at least in a limited way, affected by exit poll results from earlier phases. In particular, she finds that voters attach importance to exit poll results from within their own states, but largely discard poll information from other states. Further, the impact occurs not due to the exit poll results themselves, but whether the results are surprising vis-à-vis beliefs that voters held prior to the polls being released.
For instance, if pre-election polls showed a likely poor result for the United Progressive Alliance (UPA) and a good result for the National Democratic Alliance or NDA (which they did), but early exit polls showed the UPA doing better than expected and the NDA doing worse, such news, according to Patnam's research, would indeed have an impact on voting behaviour in later stages of the election. In other words, voters, seeing the UPA doing better than they had anticipated and the NDA doing worse, would take a more positive view of the UPA and a more negative view of the NDA, and this - again, according to Patnam's analysis - would affect actual voter choices in later phases in favour of the UPA and against the NDA.
Patnam concludes her analysis with a fascinating counterfactual exercise: in the absence of exit polls, by how much would the vote shares for the UPA, NDA, and others have shifted? The results are striking. In the simulated model without exit polls, her prediction is that the NDA would have gained a five per cent greater vote share and the UPA would have lost a whopping 15 per cent, with the UPA's loss coming largely from ceding vote share to other parties (which would have gained 11 per cent).
In other words, within the framework of Patnam's model, the NDA suffered a bit, but the UPA gained enormously, from the fact that the latter polled well early on, with the UPA's gains coming largely at the expense of other parties but also to some extent from the NDA.
Since the final difference in vote shares between the NDA and the UPA was a matter of a few percentage points only, this research raises the possibility that the NDA's poor polling in early phases of the election might have contributed to their surprise defeat and to the UPA's unexpected victory.
While this research does not speak directly to opinion polls, it is, at least, suggestive that opinion polls could well influence voters' behaviour. Further, research from other countries, such as the US and the UK, does document the impact of opinion polls on voting behaviour, largely supporting the idea that good poll performance magnifies the lead of the party polling well.
Of course, such research does not, of itself, determine the case in favour or against banning or regulating polls. But proponents on both sides would be well served by knowing the facts.
The writer is an economics professor at Carleton University in Ottawa, Canada, and is co-author of Indianomix: Making Sense of Modern India (Random House India, 2012) Marginalia makes research from the academic world accessible to all our readers
Disclaimer: These are personal views of the writer. They do not necessarily reflect the opinion of www.business-standard.com or the Business Standard newspaper
