A month ago, Election Metrics had remarked that the Karnataka elections were hard to call, and that they “may turn out to be anybody’s game". Historically, exit polls have had greater success in predicting election results than opinion polls. This time round, though, even exit polls are largely inconclusive, though we might interpret them in aggregate to mean that the Bharatiya Janata Party (BJP) will be the single largest party.
Following the previous assembly elections in Karnataka in 2013, various polling agencies had been divided in their forecasts of the number of seats each party would get, but were rather unanimous in their vote share forecast. As it happened, the “consensus forecast" was rather accurate in its prediction of the vote share of the Congress, which romped home with a comfortable majority in the wake of a three-way split in the BJP.
This time round, opinion polls are widely divided in both their vote share and seat share forecasts. While some pollsters such as CVoter, JanKiBaat and Today’s Chanakya have given a seat share range that includes a majority for the BJP, others such as Axis have given a seat share range that includes the possibility of a Congress majority.
Interestingly, polls that broadly agree on the seat distribution don’t agree on the vote share predictions (for example, CVoter and CNX), and polls that agree on the vote share don’t agree on the seat share distribution (for example, Axis and VMR). At one level, this indicates the difficulty of calling these elections. However, this also calls into question the sampling processes of these opinion polls, and the models used to convert vote shares to seat shares.
Back in 2014, Election Metrics had commented on the difficulty of converting votes to seats. This is because in a first-past-the-post system such as the one we use in India, not only does the overall vote share matter, but also the distribution of this vote share among constituencies. The latter is a hard thing to predict, especially when there might be some fundamental changes on the ground that might result in the respective stronghold of parties.
Yet, in all the confusion among the different exit polls, there are some straws we can clutch at to try and extract some signal. For starters, we can divide the exit polls into two. VMR and Axis predict that the Congress will be the single largest party, with the latter giving the Congress a possibility of getting an absolute majority. The rest of the major exit polls predict that the BJP will be the single largest party, with some of them giving a range that includes an absolute majority for the BJP (interestingly, no exit poll predicts a certain majority for any party).
Next, we can look for consistency between a poll’s vote share and seat share predictions based on a rudimentary vote swing model.
As Election Metrics had remarked in the last edition, the Karnataka polity is asymmetric— unless there is a major difference in the way a party’s votes are distributed, the Congress needs a much higher overall vote share to get a majority compared to the BJP. This is a consequence of the former being a pan-Karnataka party while the BJP has its regions of strength.
This is where the Axis poll falls short. While it is similar to VMR in predicting a 4 percentage point lead for the Congress over the BJP in terms of overall vote share, the seat share it predicts for the Congress is disproportionately large (VMR’s vote and seat share predictions, on the other hand, are internally consistent). This either means that Axis is predicting a massive realignment of party strongholds in the state, or is simply using an incorrect model to convert votes to seats.
Coming to the Karnataka exit polls that predict that the BJP will be the single largest party (Today’s Chanakya, CNX, CVoter and JanKiBaat), the interesting thing is that they all predict that the BJP will have a higher overall vote share than the Congress. If this prediction turns out right, it will be a really significant development since it would be the first time this would have happened in a Karnataka election (even in 2008, when the BJP formed the government, the Congress had a higher vote share).
That said, the vote and seat share predictions by these four pollsters are all internally consistent to a high degree of confidence (going by Election Metrics’ rudimentary swing model). So, given that the prediction range of all these polls include a hung assembly, it might just be possible that the BJP might have the highest vote share and still fall short of a majority.
So where does this leave us? We have five polls whose vote share and seat predictions are internally consistent, one of which (VMR) predicts that the Congress will be the single largest party, while the others predict the BJP will be ahead. In the face of this evidence, a prudent analyst would be inclined to go with the majority opinion, and predict that the BJP will be the single largest party.
However, given the margins of error inherent in exit polls and the variation in prediction across the polls, it is impossible to say if the BJP will get enough seats to form a government on its own, or even if it will form a government at all. To know that we will have just to wait for the results.