We have been discussing the CNN-IBN poll on Karnataka elections in these columns. The CNN-IBN poll prediction that Congress is expected to get majority in Karnataka has been widely criticized as “motivated”.

Now it is learnt that the punters have their own estimates and these estimates predict around 100 seats for BJP and around  60 seats for Congress with JDS around 20. This estimate is closer to what we thought would be realistic. 

Obviously the supporters of each party would try to say that their party will get the majority. It is the duty of the pollsters to raise above party loyalties and come to a close estimation based on public response. However it is necessary that the public responses have to be converted into seat estimates on a logical basis. We hope that the polling agencies have the confidence in their methodology to make their approach public. 

For the consideration of future pollsters, I am providing herewith a brief sample approach with reference to Karnataka for a pre-election opinion poll for the purpose of predicting the seats. I am placing this before the public for their comments so that like the evolution of an open source software, we may arrive at an “open source election forecast methodology”.  

With this sort of a transparent approach, the only variable part would be a survey which any field research agency may do and collate the results. Let’s call this as the Version 1 of the Naavi Election Forecast system (NEFS-v1) for the sake of reference.  

The prerequisites for this system are

a)      List of constituencies for the election with configuration with reference to earlier constituencies.

b)      List of constituencies in the previous election and party wise poll details

c)      Number of new voters

d)      Number of Deletions from voter’s list

e)      Total number of voters in each constituency and total. 

Methodology: A survey would be conducted in select constituencies based on a questionnaire. Responses will only try to estimate the “Swing” with reference to the previous election on select determinants and not directly try to ask “Which Party do you vote?” Reponses will be collated and put through the system of forecast explained below.  

  1. The State would be divided into some homogeneous zones with specific expected  vote patterns. Since the survey has to cover each of these representative zones, there is a limitation as to the number of zones selected because of cost and logistic considerations. For Karnataka, it is suggested that the following zones would be desirable.
    1. Bangalore District
    2. Rest of Old Mysore Area
    3. Malanad Area
    4. Costal Karnataka
    5. North Karnataka (Hubli belt)
    6. Hyderabad Karnataka

 If required, the survey can be restricted to four zones namely,  (a+b+c), d, e and f.   

  1. Survey would be conducted by covering each of the chosen zones atleast one constituency. The sampling size has to be at least 2,000 for each zone with as much spread as possible amongst men, women, new and old voters etc.
  2. The State has a unique position in this year’s election since this is the first election after the constituencies were re configured. Hence each present constituency represents a combination of a few earlier constituencies. This needs to be factored into the calculations. (Suggestion given separately)
  3. It is reported that there was a large scale revision of voter lists and nearly 40 lakh new investors have come into the fold and nearly 20 lakh names have been deleted. This is as significant number and needs to be factored in. (Suggestion given separately)
  4. Previous election vote shares would be taken as a base and this years swings are added or subtracted to arrive at the share of votes for each parties.
  5. The swing will be calculated on three factors namely
    1. National Factor
    2. Local Factor
    3. Candidate Factor

 On each of these factors, the survey will capture the relative change in the perception of the subject for each party on a scale of-5 to + 5. Of these -5 to-3 is taken as a negative swing, +3 to +5 is taken as a positive swing . The remaining is considered neutral.No question will be asked such as “Which party you voted last time? Or Which party you are going to vote this time”? 

The swing would be applied to the last known vote share to arrive at the estimated vote share for each party.

These vote shares would be converted into seats in each region on the following formula:

  1. At the first stage a seat forecast percentage factor (SFPF) would be calculated for each party based on the vote percentage forecast.
    1. Upto 10%, SEPF=0 each      
    2. 11 to 14 SEPF=1 each     
    3. 5-20% SEPF=2 each 
    4. 21-30 % SEPF =3 each    
    5.  31-and above % SEPF=4 each

[The figures above can be assigned on the basis of past data if available]

    1. Parties will be ranked in the order of  vote shares
    2.  The seat share would be calculated by applying the SEPF on the number of seats in the region first on the top party. Then on each of the subsequent parties in the hierarchy. The base for calculation would be the number of seats available. For  example , the base for the first party would be the total seats available. After arriving at their share, the remaining seats will be shared by the remaining parties . At this level their SEPF would be readjusted removing the first party from the pool. This process would be repeated for subsequent levels.

 A few examples of how the vote share forecast would be converted into seats under this scheme is given below. Example 1 

  Party 1 Party 2 Party 3 Others
Vote Share Forecast 40% 30% 25% 5
Total Seats for each party 28 13 9 0

    Example 2 

  Party 1 Party 2 Party 3 Others
Vote Share Forecast 30% 28% 25% 17%
Total Seats for each party 18 16 12 4

   Example 3 

  Party 1 Party 2 Party 3 Others
Vote Share Forecast 45% 43% 10% 2
Total Seats for each party 26 24 0 0

    P.S: Effect of Additions and Deletions: 

Effect of deletion  and addition from voters list will be computed as follows.

 a)       Deletion: The vote share first calculated ignoring the new voters and then the  adjusted (after deletion) vote percentage is calculated. For example, last election P1=35% P2=25% P3=20%.

b)      Deletions say 10%. Then adjustment due to deletions would be made on the previous vote share at 3.5%, 2.5% and 2 % Adjusted vote shares of the previous election will be used for further calculations. Adjusted vote share would be 31.5%, 22.5% and 18% respectively for the above.

c)      Additions say 20%: Ignored in the first level. Swings calculated. Let us say this gives swings for P1=+ 3%, P2 =-5% and P3=+2%.  The additional voters will be expected to represent a vote share of previous election adjusted for the swing. Ie at the rate 38%, 20% and 22% in the above example.  For the above case the swing adjustments would be increased by P1=+0.76%, P2=0.4% and P3=+0.44 %

 Effect of Delimitation 

Each present constituency will be expressed as a combination of earlier constituencies and the proportionate figures from the previous election would be incorporated to find the consolidated vote percentage for the new constituency. If a community of interested persons on the Internet can conduct an online survey at least for regions such as Bangalore, we can put the above method of conversion of vote share into seat prediction to test. We hope that the pollsters who hog TV publicity would make their methodology public so that we can at least be satisfied if there was any bias or not. 

Suggestions are welcome.


Be Sociable, Share!