Assembly Polls 2023: Exit polls a comedy of errors? Seven reasons why they can go wrong

Assembly Polls 2023: Exit polling can go wrong for a number of reasons.

The exit poll predictions are out and instead of providing clarity they have added to the confusion. Agencies have given conflicting projections. The heavyweight duo, Axis My India and CVoter, gave exactly opposite estimates in Madhya Pradesh, Rajasthan and Mizoram.

In Rajasthan, out of 10 agencies, seven are predicting a BJP victory. In Madhya Pradesh, out of 10 agencies, the verdict is split with five each giving BJP and Congress victory or lead. In Chhattisgarh, there is consensus amongst nine agencies of a Congress victory.

In Telangana, there is unanimity of Congress causing a big upset defeating BRS out of the seven agencies. In Mizoram, the verdict is split, two agencies predicting a hung house, two each an MNF and ZPM victory.

Additionally, many agencies give big ranges like Axis My India 80-100 for BJP in Rajasthan, C-Voter 113-137 for Congress in Madhya Pradesh (MP). Now a 20/24 seat range in Rajasthan/Madhya Pradesh is 10 percent of house strength of 200/230. This is bizarre and shows how vote share to seat share conversion is so difficult in Indian context. Conversely, such a wide range also allows these agencies to boast that they got the elections right.

Exit polls are based on the assumption that voters are saying the truth during PIs (personal interviews). Many could deliberately lie, others (poor/minorities/marginalised sections) may be scared of speaking the truth.

Further, since exit polls are done mostly outside booths, people may give politically correct answers where confidentiality of information is low.

Exit polls normally have a margin of error of 1-3 percentage points. Every state is witnessing a kaante ki takkar (nailbiting fight). The vote share difference in 2018 in Rajasthan and MP was less than 1 percentage point.

Cost-cutting and a lack of the proper research depth hampers data collection. Channels have tight budgets which can impact quality. Many new polling agencies spring up every election season, with not much information on the background of their promoters. They may lack a solid/demonstrated research methodology.

Agencies are usually under pressure to turn around findings by evening leading to errors. Many agencies are adopting computer assisted telephone interviews which may not necessarily gauge the true pulse on the ground.

While there have been technological advancements, apps, online questionnaires, geo tagging etc, but there is still the human element in the sampling. Often field resources have a tendency to sample convenient booths like urban booths, which can lead to distorted results.

In India, given the diversity, population growth, number of first-time voters and inconsistency in polling station coverage, historical elections data is far less reliable than say in Europe, the Americas or Australasia increasing the complexity for pollsters.

Exact caste data of no constituency is available, as the last caste census was done in 1934. Socio economic class-wise data, income levels of voters are also not available, which pose challenges.

Women voters have emerged as kingmakers over the years. The sample size of women in most surveys is 25-30 percent despite their share in the population being close to 50 percent. Women voters are in more numbers than men in three of the five states.  Normalisation of samples can result in errors.

Let’s wait and see which pollster gets it right this time around.