How to read the 2019 exit polls
Given their methodological opacity, they are more like black box forecasts than a subtle science
An overwhelming majority of exit polls has predicted a return of the Narendra Modi-led National Democratic Alliance (NDA) government. We will have to wait till May 23 to know whether they are right or not. In 2014, most exit polls got the trend right, but underestimated the extent of the Bharatiya Janata Party (BJP) and the NDA victory. In 2009, when the Congress-led United Progressive Alliance (UPA) came back to power, most exit polls grossly underestimated its performance. 2004 was an election when most pollsters got even the trend wrong. The UPA unseated Atal Bihari Vajpayee’s NDA-I, contrary to what most exit polls predicted. Given their track record, political parties are likely to cherry-pick from history to accept or reject the latest exit polls. Should this be the correct way to read exit polls in India?
Exit polls are not the only tool used in election forecasting in India. There are pre-poll surveys, even tracker surveys which are conducted much before elections. The reason why exit polls are taken more seriously is that they are supposedly based on a survey of voters returning from the polling booth. This is more likely to capture voter preferences which can change between a pre-poll survey and the day of polling. The protection against a shift in political mood during the campaign notwithstanding, exit polls face other significant challenges. At least two among these are worth highlighting.
First is the question of converting vote shares into seat shares. Then there is the challenge of ensuring that the sample is indeed representative.
A couple of examples can be given to explain these points. In the 2017 Gujarat assembly elections, the BJP had a vote share of 50%, two percentage points more than what it had in 2012. Yet, the BJP’s number of seats came down from 115 to 99 between 2012 and 2017. What the exit polls are expected to have are vote share numbers. Even if they are accurate, there is no guarantee that seat share projections will be accurate. What makes matters even worse is that most exit polls do not reveal their methodology to convert vote shares into seat shares. So, there is hardly any academic/scientific evaluation of this crucial step in arriving at final numbers. Even the vote share numbers are not easy to arrive at.
The first challenge is getting the caste math right. Some agencies such as Today’s Chanakya have been releasing caste-wise support estimates for different parties/coalitions and then extrapolating these using the approximated population share of these groups. Such a method is bound to dilute the principles of random sampling, which, simply speaking, tries to eliminate human bias in the selection of a sample. Attempts at generating a truly random sample, even at the constituency-level, might give a misleading sample. For example, we do not know whether the roughly 1,400 sample per constituency which India Today-Axis claims, has a balanced proportion of different social groups. If methodological details such as these were shared transparently, the discourse around predicting election outcomes would have been much more informed in India. Because polling agencies, except a few such as CSDS-Lokniti refuse to be transparent, exit polls lean more towards gambling based on what could be termed as black box forecasts rather than a subtle science.
With these caveats in place, there is no harm is summarising three key trends captured by the exit polls for the 2019 general election. The first big takeaway is that the BJP seems to be completely demolishing the Congress in direct contest states, in contrast to what we had seen in the assembly elections in Gujarat, Rajasthan, Madhya Pradesh and Chhattisgarh. This, if it were to happen, has to be inferred as the success of Narendra Modi to convert the 2019 race into a presidential-style contest between him and Rahul Gandhi. The second big trend is the BJP emerging as a big challenger in the eastern states of West Bengal and Odisha at the expense of the Left and the Congress respectively. This will be a first in the history of the party. The third key takeaway is a probable revival, and the polls seem to be split on this, of the SP-BSP from the near rout they suffered at the hands of the BJP in both the 2014 and 2017 elections in Uttar Pradesh. This bit is the most puzzling. If the BJP seems to be riding a wave in almost all states except Kerala and Tamil Nadu, Uttar Pradesh should not have been an exception, even in the face of an SP-BSP alliance. Is there an element of preference falsification – voters not being truthful to the surveyors – in favour of the BJP, something Praveen Chakravarty, the chairperson of the Congress’s data analytics department pointed out in an interview with Hindustan Times last month? We will know on May 23.