Bihar poll: Pollsters may get vote share predictions right, but here's why their seat projections will be wrong

By Nikhilesh Rao In the past several elections, opinion pollsters have gotten pretty good at getting the vote share projections right. For example, in the 2014 May General Election in Uttar Pradesh, Hansa_-NDTV_ and Lokniti-IBN had a only a 3.3 percent variance from the final vote share. However, by the time, they converted the vote share projections to seat share, Hansa underestimated BJP performance by 17 seats and Lokniti gave 24 less seats to the BJP than the final tally. However, every pollster after the election always laments: “Look, we got our vote share estimates right. Don’t blame us if the seat projections come out wrong”. And that is going to be the same story after the Bihar elections. The below table and data from Ogmer shows a comparison of vote share & seat projections by various opinion polls: In every single poll, there is not more than a two to three percent difference between the NDA combine and the JDU-RJD-INC alliance. But, if you look at the seat difference, it is all over the map. For example, in the Swarajya/543 poll, a three percent vote share difference between BJP and the JDU-RJD-INC alliance has a difference of only five seats. While a two percent difference in the India Today-Cicero poll has a 19 seat difference. Similarly, c-Voter’s Sep Eighth Poll has a three percent margin in favor of JDU-RJD-INC alliance which gives them a 22 seat advantage. These differences are because each polling firm has their own secret sauce in converting vote share to seat projections. They could use the cube rule (where the seat shares have a multiplying factor with every 1 percent change in vote share), or the swing method (by comparing swings from previous elections) or any number of hybrid black magic methods. It is well understood that minor changes in vote share could lead to disproportionate impact on seat shares. This was the case in the UP general elections of 2014, where 3.3 percent vote share led to differences of 17 to 24 seats. Similarly, in the Bihar elections, a one percent vote share variance between projections and actuals will lead to a multiplier effect when the votes are converted to seats. Therefore, in close elections like in Bihar, the risk to pollsters will be extremely high because not only do they have to get the vote share exactly right but their vote share to seat projections should have no margin of error to not risk looking like being completely wrong on the final seat projections. The odds of all of these lining up perfectly are extremely small. Therefore, the Bihar elections will be close from a vote share perspective but will not seem that when the seats are finally counted and pollsters will be left defending their models once again. The author is a US-based data scientist Data courtesy:  Ogmer, which provides detailed data-driven research on consumer-oriented topics