Growing expertise

THE Assembly elections of November represent another significant landmark for the burgeoning industry of opinion surveys and exit polls. Although intuitive predictions of a qualitative nature often had a fair degree of success, the diversity of poll exercises that were conducted and the range of forecasts they provided offer interesting insights into the process and the growing expertise available in India.

The most significant pitfalls that opinion surveys and exit polls have to steer clear of are those connected with sample size and selection. The question of size is addressed through a statistical formula applied to the electorate as a whole. Yet, even with the most careful system of selection, as by using various forms of stratification by region, locality and class, poll samples often fail to reflect the bewildering complexities of electoral behaviour in India.

At another level, opinion surveys and exit polls are known to be fallible on account of inherent questionnaire biases. Even the manner in which a question is posed can often influence its answer. Evidently, the exit polls that were conducted in the recent round of Assembly elections were influenced by one of the overwhelming realities of recent Indian politics - the incumbency disadvantage.

Amidst signs of increasing instability, forecasters have begun falling back on this one element of stability in electoral politics. With rare exceptions, it has been the rule over the last 15 years, since the Congress(I) fell from its position as the stable pole around which all politics revolved, that the incumbent governments both at the Centre and in the States have rarely won a second consecutive term in office. The Left Front in West Bengal has repeatedly bucked this trend, as it has on occasion in Tripura.

Inherent questionnaire biases exert a particularly strong influence when the margin between the principal contestants in terms of popular vote share is very thin. Although the picture is considerably simplified in a context of bipolar contests - as the recent Assembly elections by and large were - the conversion of popular vote share into seats won is always a tricky exercise.

The picture that emerges from the exit polls conducted in the recent round of elections is that they underestimated the magnitude of the Congress(I) margin in both Rajasthan and Delhi. In Madhya Pradesh, they called the results wrong - by a rather large margin in the exit poll conducted by the Centre for the Study of Developing Societies (CSDS) and very narrowly in the exercise carried out by the Development Research Services (DRS) for Doordarshan.

The CSDS poll gave the Congress(I) 39 per cent of the popular vote in Madhya Pradesh, while the DRS gave it 40 per cent - rather modest figures in a bipolar contest. These were close enough to the actual figure of 41 per cent that the electorate returned. But the CSDS erred rather grossly in giving the BJP no less than 47 per cent of the vote - one point higher than the share it garnered in the 1998 Lok Sabha elections. But then, rather inexplicably, it gave the BJP only 178 seats in the Assembly, whereas in the Lok Sabha polls, it had established leads in no fewer than 222 Assembly segments.

This points to the fallibility of all formulae that seek to convert popular vote share into seats won, even in a bipolar contest. The DRS, which called the popular vote share with a greater measure of accuracy, gave the BJP 169 seats against 133 for the Congress(I). Between them, the two main contestants were expected to wrap up 302 of the 320 seats at stake. In the event, they got 291 between them, indicating again that the clout of third contestants is often under-reported in exit poll predictions.

The DRS gave the Congress(I) 46 per cent of the popular vote in Rajasthan and 111 seats. The BJP was, in its estimation, likely to get 37 per cent of the vote and 50 seats. Both the estimates of popular vote share were inflated - in the case of the Congress(I) by 3 percentage points and in the case of the BJP by 5 percentage points. But this did not seem to matter much to the Congress(I), which won 39 seats more than forecast. For the BJP, however, the difference was crucial, since it obtained 17 seats less than forecast. This State offers a reversal of the syndrome that was witnessed in Madhya Pradesh - third contestants in a bipolar contest were given greater influence than they actually happened to possess.

The CSDS exit poll in Rajasthan forecast a 14 percentage point gulf between the Congress(I) and the BJP in terms of popular vote share - higher even than the actual figure of 11 points returned by the electorate. But the difference in terms of seats won was modest in comparison to the final outcome - 116 against 61. This again is a paradoxical situation which calls attention to the inability of poll forecasts to translate vote shares into seats accurately.

In the case of Delhi, the DRS translated a 7 percentage point advantage for the Congress(I) into a virtual landslide victory - 51 seats to the BJP's 14. The seat forecast was remarkably accurate, falling only one seat short in the case of the BJP. But the estimate of popular vote share was way off the mark - the actual gap, it turned out, was in the range of 15 per cent.

The CSDS too returned the same 7 point difference in popular vote shares, but gave the Congress(I) five seats fewer and the BJP three seats more than they eventually won. Although both exit polls concurred on the magnitude of the swing in aggregate voter preference, they obviously adopted widely divergent criteria in converting this figure into seats.

Since in electoral forecasts it is only the final figure that sticks in memory rather than the minute details, the pre-poll opinion survey carried out by ORG-MARG for the weekly newsmagazine India Today will perhaps endure as the big winner of these elections. Alone among all surveys, ORG-MARG predicted a conclusive three-nil shutout of the BJP in the three northern States. It accurately forecast a two-point advantage for the Congress(I) in the popular vote in Madhya Pradesh, and converted this, though less accurately, into a seat margin of between 17 and 22.

The popular vote distribution in Delhi too was called with a fair measure of accuracy - ORG-MARG reported a 17 per cent gap translating into a virtual landslide. It was on less sure ground in Rajasthan, converting a 10-point difference in popular vote shares into a 38-seat advantage for the Congress(I). The actual result, it turned out, was that an 11 percentage point margin worked itself out into a stunning advantage of 117 seats - 150 seats for the Congress(I) against a rather dismal 33 for the BJP.