The methodology

THE Frontline-CMS Pre-election Public Opinion Survey, commissioned by Frontline, was conducted by the Centre for Media Studies (CMS). The opinion poll seeks to forecast the outcome of the elections to the twelfth Lok Sabha. The survey was conducted between January 27 and 30, 1998. CMS had earlier conducted two rounds of surveys - the first, between December 3 and 5, 1997 and the second, between January 7 and 10, 1998.

The Frontline-CMS Survey followed a unique methodological framework taking into account the complexities of the current Lok Sabha election scene. CMS researchers interviewed 8,900 registered voters spread over 61 Lok Sabha constituencies in 15 States and Delhi in four regions. In all, 543 villages and 244 urban localities in some 181 Assembly segments were covered. The field work was done by nearly 200 researchers after a day-long briefing on State-specific peculiarities in the election scene. The margin of error of the predictions of the survey is expected to be 3 per cent in either direction.

Each of the 61 parliamentary constituencies chosen was representative of a cluster of constituencies reflecting similar electoral trends or voting behaviour in the recent past.

Systematic random sampling procedures were used to select the requisite number of villages and urban localities within each Assembly segment of each Lok Sabha constituency and of the requisite number of households within each village/urban locality. At each level, care was taken to ensure that the sample represented the population.

The constituencies were chosen in such a way that different combinations of contesting parties were represented in the sample. Such a design was necessary to assess the impact of gains or losses among partners in alliances since the last election.

In each household, one registered voter was interviewed with the help of a structured questionnaire, but with a couple of open-ended questions. The proportion of voters belonging to different castes/communities/occupations in the population was represented in the sample. Similarly, while drawing the sample, urban/rural and gender ratio compositions were maintained.

An interactive interview-based approach with sensitive questions on voting intentions was adopted for the survey. The respondents' replies to questions on their past voting behaviour were analysed to estimate the likely swing in vote shares of different parties in the 1998 elections. The forecast of voting intentions in the 1998 elections was obtained by applying the swing factors to the past election data for constituencies. This resulted in a party-wise and region-wise forecast. The projections were then put to a sensitivity analysis with cross-checking, taking into account the ground realities.

MARGIN OF ERROR

All sample surveys, however scientific, are subject to sampling errors as well as non-sampling errors. The margin of error of the voting percentages estimated by the survey at the national level is 3 per cent in either direction. At the regional levels, the margin of error of the forecast is likely to be higher as the projection is based on a smaller sample.

CMS has attempted to minimise the sampling errors through innovations in sampling procedure. Non-sampling errors have been reduced by carefully designing and pre-testing the questionnaire. Quality control was ensured through intensive training and supervision of field researchers. To ensure the accuracy of data on voting intentions, CMS has done a validation of data by analysing the voter's response to an array of related questions, including an open-ended probe into shifts in voting preferences.

According to CMS, the size of the sample, the spread and the selection procedures adopted are the strengths of the methodology. According to CMS chairman Dr. N. Bhaskara Rao, the size of the sample in the Frontline-CMS Survey is more than adequate to forecast the outcome of the national election with equal reliability.

The CMS-Frontline Survey was done by the psephologist Arun K. Behuria. He was assisted by K. Dharmaraju, Dharmendra Singh, Vinod K. Das, Pradosh Sharma, Dinesh Singh, Mumtaz Ahmad and Mohamed Ahsan. The team worked under the guidance of Dr. Bhaskara Rao.