Why polls go wrong
Psephology as a discipline was invented and evolved in the United Kingdom in the late 1940s and was first used in the early 1950s to read and predict elections. As a tool to read elections, it is essentially a western concept. Given the fact that in most of the western democracies the nature of politics and society is very different from India, it was always a huge challenge for psephologists to adopt this tool in the Indian context. Even in countries like Britain and the US, where polls often predict election outcomes very correctly, they have gone off the mark several times. For instance, exit polls projected John Kerry’s win in 2004 but he lost.
In India, opinion polls start pouring down a couple of months prior to the elections, projections keep taking different shapes and colours till the exit polls. Mostly, the actual results are very different from any one of these predictions.
No wonder, a TV channel recently called an opinion- poll based programme Siyasat Ka Sensex. Different agencies come up with entirely opposite projections at the same point of time.
Thus, if psephology is a science and uses scientifically designed statistical tools then at given point of time all agencies should have the same projections and exit polls should never go wrong. As a famous psephologist from the US once said, “Exit polls should make news when they go wrong, not when they are right.”
The big question is, why do exit polls go wrong in India. There are many factors which limit the use of psephology in the Indian context. Politics in Britain and the US are dominated by two parties while India has a multi- party system besides dominance of individuals at the cluster, Assembly and Lok Sabha levels. This makes calculation of the swing factor a complex exercise due the shift of votes between more than two parties.
Swing analysis is further complicated by the plurality of the electorate across socio- economic segments. Then, variations due to the education level, occupation, level of urbanisation etc. create multilayers of voters and their motivations and influencers.
Western societies have a relatively homogenous electorate, though there are variations due to ethnicity, region and pressure groups. Thus, due to relative homogeneity, only a 2,000- respondent sample can be representative to make predictions with very nominal “margin of error”.
In India, due to the plurality of the electorate, a representative and statistically significant sample may run into hundreds of thousands and may still have a large margin of error. Thus when opinion polls in India are done with just a few thousand samples; how can we expect them to be correct?
In the West, every election is closely fought as there very few instances when a huge shift in vote takes place. This is due to the fact that a majority of the electorate is strongly aligned with their parties. If a poll manages to capture the mood of the “swing electorate”, which isn’t large, it can make an accurate projection. On the other hand, due to the lack of strong political affiliations and dominance of leadership, caste and several contemporary issues, we often see a huge shift in every election in India.
For example, the BJP increased its votes from 18.8 per cent in 2009 to 31.3 per cent in 2014 and the Congress tumbled from 28.6 per cent to 19.5 per cent. This makes the use of swing analysis very difficult because of the huge shift in the base vote in every election.
The BJP base shrank considerably between 2009 and 2014, which made the task of pollsters very difficult in predicting the post- 2014 Assembly elections. Particularly in states ruled by the Opposition parties, most of the opinion and exit polls underestimated the BJP’s performance. The main reasons for this seem to be methodological compulsions in estimation of the swing and its application in the base vote ( votes polled by a party in the previous election).
In most of the Opposition- ruled states before 2014, the BJP was either non- existent or was a considerably weak force. The BJP had a base of just 15 per cent votes in UP, 9 per cent in Haryana, 16 per cent in Assam and 20 per cent in Jharkhand. The BJP needed a more than 20 per cent positive swing for victory in all these states.
Practically speaking, no pre- poll survey can estimate such a huge swing after taking into account sampling errors. Probably this was the reason that most of the pollsters couldn’t predict the BJP’s victories.
In all these states, the BJP had a significantly higher vote share in the Lok Sabha elections. If the Lok Sabha votes were used as a base, the predictions could have been more accurate. But making the Lok Sabha election vote as a base for the forecast would have been like comparing apples with oranges and overestimating the Modi factor.
The Karnataka elections are the most recent example where the polls went haywire. Due to the split of votes by the KJP and the BSR Congress, the BJP vote was reduced to just 20 per cent in the 2013 state polls. With this as the base vote, the BJP needed a 16- 17 per cent swing to get a simple majority. If the 43.4 per cent votes polled in 2014 was made the base, the BJP could have reached close to majority even with a negative swing of 7- 8 per cent, but then it would have been the apples vs oranges story.
These are some of the technical issues which limit the ability of psephologists to read the Indian elections. Due to the potential to the opinion polls to influence the electorate’s mind, the objectivity of psephologists has come under the scanner. Given the history of election predictions in India, inherent limitations of psephology as a discipline and the objectivity of psephologists, opinion polls and exit polls should be read carefully, including those for the five states for which results will be declared on December 11.
Though psephology is a science, no two polls throw up similar results. Also, given the complexities of politics in India, this essentially Western concept has been found to flounder more than once. Nevertheless, it has emerged as a necessary evil. Also, the use of swing analysis is very difficult in India because of the huge shift in the base vote in every election
IN INDIA, DUE TO THE PLURALITY OF THE ELECTORATE, A STATISTICALLY SIGNIFICANT SAMPLE MAY STILL HAVE A LARGE MARGIN OF ERROR