Over the last three days, we at Lokniti-CSDS have reported the findings of our post-poll survey, which have been published in these pages. We published our national analysis on the first two days and State-wise patterns the following days.
On June 3, we reported the estimated vote share for different parties based on the post-poll survey data. It is important for the readers to know how the survey is conducted, how vote shares for different parties are estimated, and how seat shares are estimated.
Voting choice
During the survey, we asked respondents which party they had voted for. This question was not asked in the open; we used a method which allowed the respondent to maintain the secrecy of the response about their vote. The question was asked using an app which had the names and symbols of all the political parties. All the responses were collated. Based on the responses to the question on voting, we estimated from the survey what we call the raw vote share.
The next step was to look at the demographic profile of voters of each State to estimate the variance in the demographic profile of voters and the sample. The variance in the demographic profile of voters (if that happened) was adjusted (weighted) by the actual demographic profile of all the voters in the State. This resulted in an automatic adjustment in vote share estimates though it was very marginal in case of our survey in view of the approximation with actual demography reported on the first day.
Blank responses
Often, not all the people we interviewed responded to the voting question. This means that we got blank responses to the ‘vote choice’ question. We referred to this as ‘don’t know, no opinion, no response’ in the survey research method. The ‘don’t know’ responses for voting questions were distributed to different parties looking at their responses to other questions related to political opinion. It was a subjective exercise, but needed to be done carefully. That was another stage of adjustment of vote share where the researcher did not arbitrarily decide the respondent’s answer but through a uniform procedure applicable to all similar cases. In the latest Lokniti-CSDS post-poll survey for 2024, we had 11% ‘don’t know’ responses. In some States, it was 3-4%, but there were States where the share was much higher.
In the survey, we often got underestimations of vote for independents and smaller regional parties. In the last stage of vote share adjustment, we tried to fix the vote share of smaller regional parties and independents based on previous election results and other available information. Once that was done, we arrived at the vote share estimates for different parties at the State level as well as the national level.
Estimating seat share
Finally, as per the standard method of estimating seats (which Lokniti did not do in this survey), seats are estimated based on vote share estimates using projection models. Since the vote-seat relationship is very tricky, Lokniti-CSDS tries to keep away from seat forecast. The vote share of the Bharatiya Janata Party declined only by 1% in 2024, but the party lost 63 seats. There could have been a possibility of the party not losing such a large number of seats with the vote share declining by just 1%.
We hope that this information helps clarify the process as well as the doubts about the survey data not being reliable. Lokniti believes in complete transparency and non-interference in its survey exercise by any outside agency and assures its readers and the academic community about the complete reliability of the findings as is admissible within the scientific parameters of the survey method.
Lokniti believes in complete transparency and non-interference in its survey exercise by any outside agency and assures its readers about the complete reliability of the findings.
