State-wise Prediction


Everybody is enjoying the fun of Exit polls now. Some have taken them seriously, and are either rejoicing or are expressing their remorse. A few others are watching the whole drama dispassionately. Some are trying to connect the figures with the stock market bulls and bears, some are getting so paranoid that they are framing various conspiracy theories and worse, they have began believing their own fictions.


Exit polls are banned! Yes, there is no REAL exit poll data anywhere, all what the channels are showing are opinion poll survey data at its best. Anybody who is aware of the fundamentals of exit polls, how it was done before 2012, would know that Exit Polls used to be mock polling with ballots and boxes conducted among them who had already voted in the real poll. The booths were chosen randomly or maybe not so randomly but what we call in the language of statistics, sampling with replacement among the already stratified population. I am not going into the theoretical details here, but one thing was sure then that the anonymity of the participants was stringently maintained. The surveyors were unknown faces, not representing any channel or house, no on-camera interviews were recorded, and the participants too were random people, just anybody seen coming out of the booth. However, this format of the survey was banned in 2012 by the SC and the ECI. What they do now is a completely different thing where the trail of the informant is always visible; either a phone number, or the physical address, or at least the name. The surveyors too are just too busy advertising their sponsors. The fundamental principle of Exit Poll is therefore gone.


We are a very small, but hi-tech organisation. The entire survey data was deposited in our server via a password protected page on our website that opens on any smart phone if you know the correct passwords. The volunteers are all political workers who don’t have any direct stake in the polling system. My friends who know my political affinities would understand who they are. They all work under cover, and very few know that they have any political affinity at all. However, we didn’t survey the whole country, nor was our intention there to present an alternative prediction. The objective of our survey was slightly different, and the results I am going to present below is just a by-product.


  1. Our study was limited to only three states, UP, Bihar, and WB that survived the longest ever polling season with having to be on the toes for all 7 phases.
  2. Since ECI implicitly suggested that these states are the most violence-prone (H0) and our belief was that violence was a lame excuse to give NDA some added benefits by scattering the resource allocated (H1), we tried at first a simple chi-square test to validate either H0 or H1.
  3. However, we split our target area in 7 distinct geographical regions, rather than going by the standard Constituencies list. We have an explanation why did we do this.
  4. The zones are Upper (Industrial, incl. cottage industries) UP, Sugar Cane UP, and Rest of UP (mostly producing food grains), Gangetic Bihar, and the Bihar on Plateau, North Bengal, and Rest of Bengal. We believe that these 7 are 7 different Cultural Zones that may or may not accommodate violent kind of politics by default.
  5. We divided each zone in 12 sub-zones, which implies we studied 84 sub-zones in total. 30 volunteers worked to cover the whole area, talk to random people. Each of them talked to almost 60 random people in those 7 days (April 4- April 10).
  6. We carefully avoided the urban locations, as we believe that rural India has a bigger stake in the election results. The urban people would somehow manage irrespective of who runs the govt., but rural people often do not have that luxury without migrating from their roots.
  7. The random respondents that were chosen all have had one thing in common, which adds to the bias. They are all active political workers; third or fourth level leaders who actually operate at the booth-level. These are the people who bring voters to the booth, and operate the rigging machinery if required. Chances are there that we have talked to more BJP people than anybody else, because BJP has outnumbered anybody in deploying these booth-level leaders. In UP alone, there were approximately 40 lakh such people.
  8. We had a short questionnaire of 10 questions, out of which 6 questions were about the national politics, and 4 were about the local issues.
  9. We have had 1764 respondents in total. You may say, this is the sample size, but since all these people actually talk about 1000-1200 voters in his booth, we have the pulse of 1.8 lakh voters through them.
  10. This sampling is done with 95% confidence level, and 40% Confidence Interval, which means – 5% error tolerance is granted, and we assume that each respondent have answered at least 3 questions correctly.


After processing all the data, we find that H0 is wrong – that is there was absolutely no need for this 7-phase suffering of polling in the fear of violent pockets. No single zone or subzone was ever more violent than others. Hence, H1 was correct, that this design was implemented to give BJP an extra edge. But did it work? No!

Our survey covered 162 seats. In UP, there is hardly any viable Left candidate, and in other two hardly any viable INC candidate. So, if we can write the formations in coordinates like {NDA, INC/LEFT, Non-NDA} – then what we find in these 3 States are the following formations:

{22, 6, 52} out of 80 in UP

{27, 2, 11} out of 40 in Bihar

{7, 3, 32} out of 42 in WB

In total, {56, 11, 95} is the trend in the whole area. If this trend is projected on the whole country, assuming a homogeneity there, NDA gets 191 (+/- 5%) seats.

It’s a test for my methods too, whether this result tallies with the actual (assuming no significant EVM tampering happens) result to be declared on the 23rd.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.