Why Delhi 2020 verdict is equally important for both poll pundits and political parties

By Atanu Biswas

The number of opinion poll results for the 2015 Delhi assembly elections listed in Wikipedia is 27, while that for 2020 is only eight, most of which were published within the week before elections. Were pollsters extra cautious and hesitant to forecast Delhi results this time around? Most opinion polls performed disastrously both in 2013 and 2015, in the important but small state.

The fact of the matter is that opinion polls fail to predict the winner in many elections worldwide. Sometimes, despite correctly predicting the winner, pollsters grossly miscalculate the quantum of the victory margin.


In the case of Delhi, caste and other social structural features are notoriously complicated. This may have caused gross miscalculations in the last two Delhi assembly elections. However, it is customary that the sample size be determined a priori, to achieve a pre-fixed level of accuracy (usually within ±3% limit), keeping the heterogeneity of the concerned population into consideration. Then the samples are supposed to be selected at random, maintaining the proportion of people in each group similar to that in the population. However, it is very unclear how many polling agencies conduct their polls maintaining such statistical requirements and standards.

It is often argued that predicting seats is tricky in a first-past-the-post system like India’s. However, vote shares of different parties are expected to be predicted quite accurately. But that, too, didn’t happen in Delhi in 2015.

Explanations, such as voter intent having changed prominently by voting day, have been trotted out, and this is true especially in ‘wave’ elections due to a ‘late swing’. But shouldn’t at least exit polls portray aclear picture? Exit polls are indeed, on an average, more accurate than other opinion polls, since they measure actual ‘voter behaviour’, not ‘voter intent’.

However, in the case of the 2015 Delhi assembly elections, exit polls were drastically off the mark. The average of seven exit polls mentioned in Wikipedia gave 43 seats to Aam Aadmi Party (AAP), 25 seats to BJP, and two seats to Congress.

The closest one gave 53 seats to AAP, while the least accurate one had put down 31 seats for AAP. AAP would go on to win 67 of the 70 seats (with BJP claiming three and Congress drawing a blank).

Did many AAP supporters shy away from responding to the poll surveyors in the previous two Delhi assembly elections? Or did many of them lie? In Britain, Conservative Party supporters are usually reluctant to take part in opinion polls, resulting in a traditional underestimation of Tory votes. This is known as the ‘Shy Tory factor’, a term that became popular during the election that brought John Major to power in 1992.

Was there any such ‘Shy AAP factor’ in the last two assembly elections in Delhi? For example, say, in a sample of 1,000 respondents, there are 400 supporters of Party A, 300 of Party B, 200 of Party ‘C’, and 100 support ‘Others’. If 120 supporters of Party A do not respond, any prediction based on the remaining 880 samples will exhibit 34.1% support for Party Bagainst a 31.8% of Party A — a completely misleading picture.

Did anybody examine the possibility of such a case for Delhi in 2020 for Party A, B or C? Psephologists should use suitable statistical adjustments for such missing information, if any.

Again, there is a belief that opinion polls tend to generally provide a conservative prediction for ‘wave’ elections. How comfortable would pollsters feel to report, say, 67 seats favouring a particular party — whichever party that might be — in a 70-member assembly, even if their calculations yield such a result? Since voting day on Saturday, pollsters have been trying to estimate ‘voter behaviour’ in ‘Delhi 2020’ with full rigour, even though very few polls were conducted to gauge ‘voter intent’.

Exit polls don’t usually consider the ‘late-hour polling’ — which may be about 15% of the total votes — and there can be other kinds of biases too. Interestingly, the average result of exit polls announced so far — 53 seats for AAP, 16 for BJP — are very similar to the average of the opinion polls — 55 seats for AAP, 13 for BJP. Confusion — and suspicion — however, continues, as far as predicting today’s results is concerned.

We know that psephology is an inexact science. Still, the Delhi 2020 verdict will provide a sort of agni pariksha for serious pollsters. Along with winners and losers in the political fray, there will be victory and defeat among pollsters too today.

(The writer is professor of statistics, Indian Statistical Institute (ISI), Kolkata)