“The NDTV-Indian Express-AC Nielsen poll survey all but puts the NDA back in the saddle” – The Hindu, March 2004.
This report on the eve of the 2004 Lok Sabha election, mirrored the overall sentiment of exit polls and surveys in the lead up to that election. Put in other words – the last time an NDA government was seeking re-election, this is what the pundits predicted. We know how poorly these predictions panned out, as the UPA government rode to a decisive victory and formed the government from 2004 to 2014.
As we read the tea leaves for the 2019 Lok Sabha polls, this election bears a striking resemblance to the 2004 election. There are two main factors behind this assessment – one, that opinion polls are a poor science that are further exacerbated by the silent voter effect; two, the “Modi wave” as a phenomenon is overstated.
Opinion polls are a poor science
Though opinion polls rule media coverage in the run up to an election, the verdict on the quality of these predictions is clear – they lack clear credibility and accuracy. For the 2004 Lok Sabha Elections, organisations ranging from Aaj Tak to NDTV to Zee News had predicted a NDA victory. However, the NDA got 68 fewer seats than the polls predicted and the UPA got 36 more seats than polls predicted.
The increasing inaccuracy in electoral predictions is due to multiple factors. Firstly, most forecasters rely on the old style of forecasting using a vote share to seat share model, which has lost its relevance today. Secondly, polling agencies lack the scale to conduct accurate polls. The ideal way to forecast is to sample every single constituency with a decent sized stratified sample. This needs heavy resources in terms of people, money and time, which most agencies don’t have access to and hence have to resort to extrapolated models. Thirdly, the rise in the number of political parties has reduced the fidelity of polling. In 2014 alone, 464 political parties contested, around twice the number that competed in 1998.
Finally, there is also some form of moral hazard in polling as there is no accountability for wrong forecasts. Our internal analysis has shown that exit polls have been wrong 85% of the time in state elections post-2014.
Further, there is no clear pattern in the errors. For example, in 2018 Rajasthan Vidhan Sabha election, BJP bagged 73 seats when Cvoter predicted they would win 57 and CSDS had predicted they would secure 84. i.e. Cvoter was on the lower side, and CSDS on the higher side.
The Indian electorate is largely unaware of the inaccuracy of these polls – in the short run, we must take all polls with a grain of salt; and in the long run, we must hold the pollsters to a higher standard of performance and accountability.
In the upcoming 2019 election, one must also be cognizant of the silent voter, or the Bradley effect. The Bradley effect is a theory concerning observed discrepancies between stated voting preferences and actual voting behavior – for example, a voter stating their intention to vote for the BJP to a pollster but actually casting their ballot for the Congress.
This phenomenon is exacerbated in autocratic regimes, and the most recent case study of this was in Malaysia’s 2018 general election. In an unprecedented victory, the PH coalition defeated the incumbent BN coalition – the first regime-change in Malaysia’s history, as the BN had enjoyed an uninterrupted reign over the country since 1957.
Coming into the election, independent pollster Merdeka Center said the BN was set to win 100 seats to the PH’s 83. These figures failed to capture the mood of the nation, as the actual tally was 79 for BN while the PH secured an astounding 121 seats.
The “Modi Wave” Misnomer
To start off with, many consider 2014 to have been a “Modi wave” election. The data belies this fact, as the BJP garnered just 31% of the votes - which is the lowest a party has polled to form a majority government in India since independence. Though Modi had a clear impact on BJP’s rise, its effect has been overstated and categorizing it as a “wave” does not line up with the statistics.
If we now look at 2019, there are clear fault lines that exist with Modi’s image. Though opinion polls have proved inaccurate in predicting election outcomes, survey results are useful in gauging the overall pulse of the electorate.
The most recent “Mood of the Nation” survey in 2018 by CSDS-Lokniti found that nearly 50% of respondents believed that the Modi government “does not merit another opportunity” to govern the country.
Moreover, the popularity of a leader is not enough to power a party to victory. In 2004, Atal Bihari Vajpayee had extremely high favourable ratings – higher than even those of Modi in 2014, at the peak of his popularity. However, only 23% of the populace stated a preference for the BJP, and the BJP ended up losing that election.
The most recent state elections in the Hindi heartland, where the Congress unseated 3 BJP governments to come to power in Rajasthan, Madhya Pradesh and Chhattisgarh, provide a precursor into the possible outcome that awaits us in the 2019 Lok Sabha election.
(The authors work for the Data Analytics Department of the Indian National Congress of which Praveen Chakravarthy is the chairman)