Karnataka Assembly Election: Pre-poll surveys often miss the mark in predicting tri-party contests; here's why

The difference between opinion and numbers is that one is subjective while the other is not. The pre-poll surveys have always been viewed with a hint of doubt, especially when they predict vote share and seats. Such surveys, at best, can predict a trend or perception of the voting population. But pollsters, in their eagerness to predict outcome and justify their existence, go where even statistician are scared to set foot.

Take the most exhaustive poll published on Thursday: this has more than 22,357 respondents, let's call it 'Poll A'. This poll predicts a vote share of 46 percent for the ruling Congress party, an upswing of more than nine percent from the last state elections. Likewise, the total number of seats assigned to Congress is 126, which amounts to 56 percent of the total 224 seats in the Karnataka legislative Assembly.

Typically, in a closely fought tri-party contest, where no clear wave favouring one party or another can be perceived, a swing of 2 to 3 percent is usually enough to assure victory for a party. It should be noted here that in the last Lok Sabha elections BJP got 43.37 percent share of votes and won 17 out of the 19 (89.5 percent) of the seats.

The pollsters could give the argument that it is a legislative election with smaller constituencies, hence the overall vote share is not the operative for number of seats won. But if vote share is not the operative, simple extrapolation of sample data to predict number of seats won may also not be correct.

Another survey released on 23 April is even more interesting, let’s call it 'Poll B'. Here the number of people polled are in excess of 4,000 through stratified sampling method. This could mean that on an average 20 people in each of the 224 constituencies were covered. Doubtlessly, this one has a much smaller sample size than Poll A but since it was released first, it has inadvertently led the discussions on these elections. This survey assigns a vote share of 38.6 percent to Congress (an upswing of just 2 percent) and gives it 91 seats, almost 31 seats less than what it got in the last election in 2013. This means that a 2 percent swing in votes can result in a loss of 31 seats, and can also mark the difference between losing and winning, which seems plausible looking at the past elections.

Another interesting bit is that the survey assigns BJP a vote share of 35.03 percent, an upswing of more than 15.14 percent from the 2013 Assembly elections but substantially less than what the party managed in 2014 Lok Sabha elections. The survey gives BJP 89 seats, just three less than Congress, but not enough to form a government on its own. Poll A predicted that 31 percent votes will be polled in favour of the BJP, fetching it 70 seats. This meant an upswing of 11 percent in vote share for the saffron party.

The challenge with pre-polls in a keenly fought tri-party election is that the actual vote share cannot assumed to be a simple extrapolation of the sample data. Winning margins are much smaller. In a two-party contest, the math is much more simpler and even data from small sample size could show the trend. Here once the raw data comes in, pollsters would have to take a call on swing constituencies one by one. And this is precisely what Poll B does. It separates out the data from south, central and other parts of the state to predict its results. But even then it would need somebody to look at each constituency with a much larger sample size to even begin predicting the seats. Another survey tries to avoid the vote share conundrum but still predicts seats.

Another factor these pre-poll surveys ignore is that BJP's star campaigner Narendra Modi is yet to step into the state. He is stated to address multiple rallies across the state and that could alter trends in way or another, especially in swing constituencies. The campaign trail is yet to start in full swing. Hence, these pre-poll surveys should be taken with a pinch truckload of salt.


Updated Date: Apr 26, 2018 20:26 PM

Also See