Why the Bihar polls are too close to call
With the announcement of polling dates for the Bihar elections, the battle lines have been drawn for an intriguing contest that will pit a coalition led by chief minister Nitish Kumar’s Janata Dal (United), or JD(U), against the National Democratic Alliance led by Prime Minister Narendra Modi’s Bharatiya Janata Party (BJP).
Two opinion polls have been released in the past week, indicating that pollsters are also getting into business. Yet, as this piece will argue, these elections are going to be incredibly hard to call, perhaps second only to the two Delhi assembly elections in the last two years.
The first-past-the-post system used in Indian elections means that an opinion poll prediction consists of two steps—forecasting vote share for each party and forecasting seat share. Historically, pollsters in India haven’t done too badly on the former. For example, in the 2013 assembly elections in Karnataka (around the time Election Metrics made its debut), most exit polls broadly concurred on the vote shares of different parties, and this consensus wasn’t far from the actual results (mintne.ws/1abZelB ). Where the exit polls erred back then was in forecasting seats, and that is going to be a massive problem in the Bihar elections.
Given the number of constituencies (Bihar has 243), it is virtually impossible to do an opinion poll of any significance in each and every constituency. The way most pollsters work around this is to select a small set of constituencies (partly at random) where the opinion poll will be conducted. From this opinion poll, overall state-wise vote share for each party is forecast, and then a model is used to extrapolate the seat forecast from a vote forecast.
All this means is that there are a large number of assumptions to be made, and the way pollsters usually get around it is by building a model based on the previous elections. A popular model for forecasting seats based on votes is the “swing model”, where the swing in favour of or against a particular party or alliance is calculated by comparing the opinion poll to the previous elections (this swing may be calculated on a regional basis). This swing is then applied uniformly to results of previous elections to get the seat-wise forecasts.
The problem with the Bihar elections is that with a massive redrawing of battle lines, previous elections are all but unusable. When there is a split in a party, or a newly formed alliance, we can use some transformations with data from earlier elections to make the predictions. The problem with Bihar, though, is the degree to which alliances have changed.
In 2010, it was a battle with the BJP and the JD(U) on one side, fighting in a three-cornered contest against the Rashtriya Janata Dal (RJD) and the Congress. In the Lok Sabha elections in 2014, the JD(U) went its separate way, with Ram Vilas Paswan’s Lok Janshakti Party joining the BJP in alliance. That alliance still stands (with the addition of the party of former chief minister Jitan Ram Manjhi), but the JD(U), the RJD and the Congress have joined hands.
And there is the Samajwadi Party, from neighbouring Uttar Pradesh, which first aligned with the JD(U)-RJD-Congress alliance and then withdrew.
This massive realignment means that results from 2010 are pretty much useless in forecasting these elections. For example, the BJP then only contested 102 seats, of which it won 91, but there is little that can be done with that information since that was in partnership with the JD(U), which is now a rival.
And this is not all. The way the formations have lined up in these elections makes it a straight fight between the BJP-led and the JD(U)-RJD-led alliances, or, in other words, a two-cornered contest. In an earlier edition of Election Metrics (mintne.ws/1e2EML9 ), we had seen that two-cornered fights tend to have a winner-takes-all result. In such contests, a small difference in vote share can result in a massive difference in the number of seats won. There have been several instances, for example, where a party has got over 40% of the popular vote, but less than 10% of the seats.
The opinion poll by Swarajya predicts that the RJD-JD(U)-Congress alliance will get 42% of the votes, while the BJP-led alliance will get 39% of votes (mintne.ws/1Kf2fHO ). The poll conducted by CVoter and Huffington Post gives the BJP-led alliance 40% of the votes and the RJD-JD(U)-Congress alliance 43% of votes (mintne.ws/1M1rEEg ).
Both polls indicate a straight and close fight.
Without a proper model (on account of changed alliances) and with a close bilateral contest, the Bihar elections are going to be extremely hard to call. Good luck to the pollsters.