STUDYING elections is perhaps more challenging in India than elsewhere. Understanding the interplay of overlapping social cleavages in one of the world’s most demographically heterogeneous countries, high political fluidity in a multiparty systems, and complexities of a developing economy poses serious difficulties for anyone seeking to forecast election outcomes. Moreover, contrary to what many believe, opinion poll based studies of elections are not just about picking likely winners and computing margins of victory/defeat, but also to provide a nuanced understanding of the different factors which go into determining the verdict of an election – why people voted the way they did, the changes and continuities from the past, and what the future might look like.

The inability of pollsters in recent times to accurately predict the final outcome has only strengthened criticism of opinion polls as a method of decoding voter preferences. Many have even demanded a blanket ban on opinion polls during the elections. The allegation of political parties manipulating figures of some widely broadcast surveys for creating a bandwagon effect among the electorate has further added to this criticism. Earlier, political parties often commissioned surveys for selecting ‘winnable’ candidates and designing campaign strategy. It is alleged that they now use survey findings in advertisements and for propaganda in an effort to mislead and influence voters, thereby vitiating a ‘free and fair’ process.

The origins of election studies in India can be traced back to the 1950s when Eric D’Costa carried out an all-India election poll in 1957 just prior to the second Lok Sabha election. The election prediction based on that poll was quite accurate, thus paving the way for similar exercises in future. D’Costa’s study not only predicted the outcome but also provided a detailed explanation of voting behaviour. It clearly revealed that voting intentions of Indians varied a lot with their income, religion, and occupation. This study also measured the popularity ratings of political leaders based on voter assessment and feedback.

The team led by Rajni Kothari and others at the Centre for the Study of Developing Societies (CSDS), Delhi gave opinion polling and election studies an academic turn during the 1967 and 1971 Lok Sabha elections. In the 1980s, Prannoy Roy and his team provided a new momentum to the election polling exercise in the country. Their sophisticated analysis of 1984, 1989 and 1991 Lok Sabha elections influenced an entire generation of pollsters in the country. At the same time, the election studies programme at the CSDS, which had remained dormant for almost two decades, was once again revived under a new research programme – Lokniti, an all-India network of political scientists. Established in 1995, Lokniti paved the way for a series of national surveys beginning in 1996. This series was later rechristened as the National Election Study (NES).¹ Its most notable innovation was the use of dummy ballot papers and dummy ballot box to ask the vote choice question. Though the CSDS surveys were primarily intended to provide in-depth analysis of voting behaviour and attitudes of the electorate, but as the data on voting preference was collected prior to the counting day, it was also used for forecasting vote share and seats for different parties.

 

Sample surveys conducted at the time of elections fall into three broad categories. Pre-polls are generally done a few weeks or days before the start of the formal election process. These, at best, can provide a snapshot of voter preferences at that moment in time and inform us about trends and, if the trend holds, the likely outcome. It is no surprise that pre-poll estimates rarely match actual results given the volatility of political preferences and the fact that a significant proportion of electorate makes up its mind only close to the voting day. This, however, does not mean pre-polls are a futile exercise; they provide the baseline for us to figure out what might or might not change during the course of an election campaign up to the voting day. The pre-poll data becomes even more useful while carrying out the analysis after the results are known. Having these data points in time surveys helps in making credible interpretation of the trends that shaped the verdict.²

 

Post-polls are conducted after the last ballot has been cast and before the counting of votes. Post-polls have, in general, been useful in estimating the respective vote shares of individual political parties, assuming of course, that the exercise followed the rigorous methodological protocol associated with a sample survey: using a randomly selected sample of an appropriate size which is representative of population characteristics. Though in most democracies the time lag between voting day and counting day is usually too short to carry out a post-poll survey, a few days are always available in India, making post-polls a scholar’s delight. The third and most popular type of poll at election time is an exit poll, done on election day outside the polling station. However, over the years, it has been noticed that exit polls in India in general fail to collect a sample that is representative of the local population. Many of those who agree to be interviewed in an exit poll tend to be disproportionately male and from the upper strata of India society, thereby significantly skewing the representativeness of the sample.

Though the accuracy of election forecasting at the national and state elections in India has considerably improved in recent times, unfortunately it is the incorrect predictions which stick in public memory, creating a perception deficit about their reliability, competence and neutrality. This is largely due to various misconceptions about election related polling. Take, for instance, the common allegation that surveys are covertly used by agenda driven media houses or pollsters to influence voters, especially those who either vote strategically or are considered to be bandwagon voters. The former refers to voting for a relatively less preferred party if the voter does not expect his first preferred party to win. A bandwagon voter is one without strong policy preference and thus votes for a party most likely to win. Common to both types of voters is that in addition to policy preferences, expectations about the electoral verdict are likely to be a critical factor in decision making.

 

Critics of opinion polls argue that ‘paid surveys’ can be used by political parties to influence voter expectations about electoral prospects of different parties. While there might be some truth to this claim, the criticism, in our view, is grossly exaggerated. For a start, it is based on a limited understanding of the way bandwagon voters behave. While some bandwagon voters may opt for the party/candidate who in their perception is most likely to win in their constituency, others may prefer to vote for the party leading at the state level, and still others for a party leading at the national level. Findings from the 2014 NES data do identify a clear bandwagon effect in Indian elections – a little over four out of every 10 voters (43 per cent) seem to go with the hawa, while 45 per cent voted on their own, without any consideration for who was ahead in the race.

Data suggests that in the 2014 Lok Sabha elections, the BJP’s lead over the Congress was greater among those who preferred voting for the winning party as compared to those who were indifferent about this. The gap between the BJP and the Congress was 18 percentage points among the former as compared to just eight percentage points among the latter. This was also evident at the state level: the electorate voted for the party which they perceived to be leading in that state. In six states including Bihar, West Bengal and Karnataka, support for the leading party among those who preferred voting for the winning party was more than 10 percentage points greater than the vote share among indifferent voters. In another nine states, this gap was positive and there was a degree of bandwagon effect. There were only six states – Andhra Pradesh, Chhattisgarh, Jharkhand, Rajasthan, Tamil Nadu and Uttar Pradesh – where the leading party had no special advantage.

The 15 states where the leading party at the state level received higher support among those who voted for a party because it was seen as likely to win included Kerala, Telangana, Odisha and West Bengal – states where the BJP did not perform as well. It is interesting to note that in Odisha, Telangana, Tamil Nadu and West Bengal, the BJP had higher support among those who were likely to vote for the winning party. This points to an interesting dimension of the bandwagon effect. Most polls merely show national estimates for votes and seats aggregated from the states. Since constituency level forecasts are rarely shown by media houses, it is improbable that a national projection from a survey would influence the minds of voters.

 

Second, the outreach of opinion polls is limited than what is generally perceived by many. Data from the 2014 NES study shows that less than one fourth of the people actually knew about the result of any opinion poll. Essentially, those who knew about opinion polls were mainly those who regularly follow the news media. With the overload of other information – prime time newsroom discussions, field reports, op-ed columns and so on, it is difficult to isolate the independent effect of opinion polls. Moreover, most often the predictions of opinion polls are all over the place. For example, during the 2015 Bihar assembly elections, predictions ranged from an outright win by the BJP-led NDA, to a hung assembly, to a massive sweep by the Grand Alliance. Therefore, while it is likely that opinion polls may play a role in shaping the expectations of a small proportion of voters, it would nevertheless be an exaggeration to claim that they swing elections one way or the other.

 

Third, regulating the release of election forecasts has been a critical issue for the Election Commission of India (ECI). National elections, and many state elections, are often conducted over a month or more due to security concerns and logistical requirements. In national elections, different regions of a state often vote on different dates. Some have suggested that in phased elections, parties that perform well in the initial phase often see a momentum effect and tend to perform better in latter phases as well. While the empirical evidence for this claim remains thin, it is possible that the bandwagon voters may shift towards parties performing well in the initial phases, expecting them to emerge victorious. This possibility and the role of opinion polls in inducing and enhancing the momentum effect were advanced as a rationale for a ban on the release of opinion poll findings before the end of the last phase of an election.

While this specific restriction on the release of findings in between polling phases is justifiable, the idea of banning opinion polls is naive. Opinion polls are not the sole source of information that voters depend on for forming expectations about the election. Even if their telecast is banned, there would still be numerous political commentators sitting in television studios or writing newspaper columns, discussing their views on the upcoming elections. Are we going to ban these also?

Fourth, critics of opinion polls must look at their utility beyond seat and vote share projections, more specifically, the production of public knowledge. Data from polls help in providing crucial social science insights and have great academic value. Across the world, social scientists have used surveys to answer many important questions about the polity and society. Time-series survey data in particular has been useful in studying long-term trends of Indian politics: political participation, the ideological orientation of voters, trust in institutions, the efficacy of the vote, degree of belief in a democratic system, and leadership choices, to list a few.

 

Overall, election surveys over the last two decades have been a valuable and reliable source of information on India’s electoral politics. The unfortunate part is the shift in focus from psephology to prophecy. Election studies, including the business of forecasting and predictions, must not be reduced merely to election time entertainment. Forecasting based on scientific models is a common practice in many fields like pure sciences and economics. Survey evidence can also be used to prepare similar models to understand political and social events. Despite models on human behaviour being more prone to error, we must curb our intrinsic bias against forecasting since elections remain the most opportune moment to study politics and people.

 

Footnotes:

1. The NES is a scientific study of the political behaviour, opinion and attitudes of the Indian electorate. It incorporated innovative research techniques for understanding elections in India. The method note published by the Lokniti team in the special issues on the 2004 and 2009 Lok Sabha elections in the Economic and Political Weekly provides a detailed account of the NES tradition.

2. For example, the pre-poll survey conducted by Lokniti-CSDS during the Bihar 2015 assembly elections estimated a four percentage point lead for the NDA, but in the post-poll survey, the NDA trailed by a similar margin. See, Rahul Verma and Sanjay Kumar, ‘How the Grand Alliance Won’, Seminar 678, February 2016.