So, Will 272+ Lotuses Bloom?

There has been a rash of exit polls as soon as polling ended on 12 May 2014 in the Great Indian Elections. Barring minor differences, all exit polls have predicted a victory for the NDA, the alliance led by the BJP.

2014 Exit Polls

Agency	BJP+	Congress+	Others
TimesNow	249	148	146
News 24	340	70	133
India Today	261-283	110-120	150-162
India TV	289	101	148
CNN IBN	270-282	92-102	72-82

On the basis of these exit polls, it seems reasonably certain that the lotus, the symbol of the BJP, will sway proudly in the wind for the next five years. With such unanimity, one would expect the final results (to be declared on 16 May) to replicate the results of the exit polls. However, in the absence of any information on the methodology adopted by the exit polls, it is difficult to evaluate them. Right now, the exit polls are like a black box: some data has been fed in and some results have been obtained but we don’t know anything else.

 Can there possibly be a slip at this stage between the proverbial the cup and the lip? It turns out that unanimity in exit polls and being spectacularly wrong has been the norm for the last two general elections in India. See the tables below for the exit polls of 2004 and 2009.

  

2004 exit polls

Agency	BJP+	Congress+	Others
NDTV	230-250	190-205	100-120
Aaj Tak	248	190	105
Zee	249	176	117
Star	263-275	174-186	86-98
Sahara	263-278	171-181	92-102
Outlook	280-29	159-169	89-99
Actual result	189	222	132

2009 exit polls

Agency	BJP+	Congress+	Others
STAR News	197	199	136
CNN	165-185	185-205	165-195
NDTV	177	216	150
Headlines Today	180	191	172
News X	199	191	152
Times Now	183	198	162
Actual result	159	262	79

In 2004, all pollsters got the victor wrong. Almost all were taken in by the exuberance of the India Shining campaign. The India Shining campaign was a grand urban narrative that did not play out well in rural India.

In 2009, all pollsters got the margin of victory wrong. Did the full roll-out of the Mahatma Gandhi National Rural Employment Guarantee Act (MGNREGA) in 2008 have a role in giving the UPA victory? Once again, was the inflated support for the BJP and its allies more an urban phenomenon?

It is worth reading Praveen Rai (see here) for the problems with Indian opinion and exit polls. Various statistical issues, such as sampling method, sample size, response/non-response errors, measurement errors come into play. There is also a strong urban bias in the coverage of respondents while conducting the exit polls. Equally, it is not clear how the constituencies that make it to the sample are chosen.

Apart from the issues noted by Rai, the problem is compounded in India where most constituencies witness contests between numerous parties. In a two-party situation, the connection between votes polled and seats won is pretty straight forward: the party that gets more than 50% of the vote-share wins the seat. But with more than two parties, vote-shares often do not cross the 50% mark and the seat is won by the party with the highest vote-share. At exactly what vote-share percentage does a party win the seat is not easily predictable. Given that exit polls only collect data on votes cast by the voter, all that they can possibly get is an estimate of the vote-share for a party. From this vote-share to estimate to seats won is a complicated process: the functional relation between the two is not clearly understood. Errors are likely to be substantial. This is where the danger of exit polls lies.

In the last two elections, exit polls have, both, underestimated the performance of the challenger (UPA in 2004) and overestimated it (NDA in 2009). Will it be any different this time? All will be revealed on 16 May. Till then, perhaps, it might be a good idea to keep the champagne in the fridge. In the meantime, NDA and its supporters should make do with the brew – tea – that their prime ministerial nominee reportedly sold in his childhood.