The Big Picture   

Methodology of National Election Study 2004

NES 2004 was the largest and most comprehensive social-scientific study conducted on Indian elections. The CSDS team explains methods used to find out what the electorate thinks and why it behaved in the way it did.

The analysis of the national and State-level verdict presented here draws upon a post-poll survey conducted by the Centre for the Study of Developing Societies (CSDS) as part of its National Election Study 2004 (NES 2004). The NES 2004 is the largest and the most comprehensive social scientific study of Indian elections. It adopts the best international practices to try and get the most realistic understanding of how the electorate behaves and feels about the key political issues of the moment. The findings of the NES 2004 are being reported here for the first time.

The CSDS has been using surveys to study Indian electoral politics since the 1960s, and carried out its first National Election Study in 1967. The series of National Election Studies have provided a comprehensive source for academic studies of social and political change in India, and have been used as a basis for international comparative studies. As such, the NES 2004 is situated within a distinguished academic tradition, and conforms to rigorous international standards of methodology.

The post-poll survey, whose findings are presented here, is not an exit poll, in which voters are approached outside the polling booth where they have just pressed the button. It is not used for predicting the number of seats that the Congress-led alliance or the BJP-led coalition is likely to get. Rather, it is a survey that takes the results as given, and then tries to understand the reasons why voters chose the parties they did. Rather than picking on whoever is walking out of a polling station and is willing to answer a few questions when the interviewer happens to be there, we select a sample randomly from the whole electorate. We do not use our data to predict the outcome in terms of seats. This is fraught with difficulty, since a small change in vote share between parties can lead to large swings in the number of seats returned. Instead, we seek to explain the underlying motivations of the voters across the country.

If we were only concerned with the national outcome, the process would involve putting the names of the whole Indian electorate, numbering 671,524,575, into a [very large] hat and drawing names out randomly. We would then try to interview each of these people. However, because we feel Indian politics must first be understood at the State-level, and then the national, for this survey we have selected samples of voters at the State-level. For this State-level analysis, we have made a random selection of Lok Sabha constituencies (adjusted for the size of electorate), and within these constituencies selected at random Assembly constituencies and polling booths. This procedure, known to statisticians as stratified random sampling, allows us to gather opinions from as wide a cross-section of the electorate as is practically possible.

For this survey, we selected samples from 32 States and Union Territories: we chose 420 of the 543 Lok Sabha constituencies. This is perhaps the first-ever survey of political opinion and behaviour, academic or commercial, carried out in all 28 States of the Indian Union, besides some Union Territories. Within the parliamentary constituencies, 932 Assembly segments were selected, and then a further 2,380 individual polling station areas, typically villages or urban wards, were selected for conducting interviews. Selection at each of these stages was by a random procedure, adjusting the probability of choosing a particular constituency or polling station according to the size of its electorate. From each polling station, 15 persons were selected from the electoral rolls.

Once we identified our sample of the electorate, trained investigators were sent to meet them. They were asked to interview only those whose names were given to them. In special circumstances, they were permitted to substitute the person concerned with another person of the same gender in the same household. The interviews were conducted after the votes had been cast in each constituency, but before the results were known. As such, we hoped to find out what motivated each person whether to vote or not to vote, and what their political opinions were, without knowledge of the outcome influencing their answers to the questions. Our investigators sat down in the homes of people whose names were selected from the electoral register, and asked them a detailed set of questions, which could take up to one hour. We did not just ask them how they voted, which could alter the way people reply, but gave them a ballot paper [in this election it was made to look as close to the actual EVM they would face as possible] on which they could mark their choice. This was then placed in a ballot box. The process was designed to ensure that the people interviewed knew they would remain anonymous. This enabled us to collect detailed information about the respondents' work and background, and allowed us to place voting decisions and political opinions within the context of broader social and economic factors.

The questionnaire we presented to our sample of voters was carefully designed. The questions were phrased so that they did not lead the person being interviewed to any particular answer. The questionnaire was in the language mainly spoken in the region in which it was presented — not just in the official State language. The translation process was carefully monitored, so that a question in one State did not have a different meaning in another.

Our aim was to provide a sample of the electorate across the States of India which was as representative of the whole Indian electorate as was practicably possible. In order to minimise the risk of sampling error, we interviewed a very large number of people: 25,433. We are confident that we used the best possible methodology to achieve this task. The sample was broadly reflective of the Indian population, in terms of the country's general demographic profile.

In order to get an accurate and detailed picture of the Indian electorate, we took very large samples from all States, even the very smallest. Therefore, when grouping all the States and Union Territories together, for our all-India analysis, we adjusted the figures using a statistical technique known as weighting, which means that each State was proportionately represented in the analysis. This means that we were able to produce an accurate assessment of regional and State level situations, as well as having a balanced and authoritative over-view at the national level.

(We would be happy to answer any queries relating to the National Election Study 2004 on lokniti@csdsdelhi.org)

Printer friendly page  
Send this article to Friends by E-Mail

The Big Picture

News: The Big Picture | State by State | Opinion |
Advts: Classifieds | Employment | Updates: Breaking News |

Stories in this Section

• Radical shift in the social basis of political power
• Methodology of National Election Study 2004
• A vote for secular politics
• The turnout was low but not alarmingly so
• Reform: the elites want it, the masses don't
• Caste matters, but so do a whole lot of other things
• Vajpayee's popularity alone was not enough
• Alliances did the trick for the Congress

Copyright © 2004, The Hindu. Republication or redissemination of the contents of this screen are expressly prohibited without the written consent of The Hindu