The Election Commission of India does not like Opinion or Exit Polls. The latter a bit more than the former according to that 1998 order; this is true especially for elections that stretch across a few months. But if the recent recommendations of the Commission are to be considered, the Election Commission’s view of Opinion Polls is no less disapproving.
An essential aspect of political messaging is to imply winnability. That’s part of the reason why large rallies are sometimes advertised and reported as being large. Rarely does a neutral voter go to a political rally and come out making one’s mind up upon listening to partisan rhetoric. The size of the rally though lets the neutral voter guess the value of voting with that given block — it provides both validation for thought a priori in terms comfort in numbers and hope that one’s vote will not be meaningless after it’s been cast. On that count it’s difficult to argue with EC’s diagnosis: a well received/trusted poll by itself will move the reality of the landscape that poll was measuring in the first place. But should that be a reason to not poll at all is a separate question that isn’t answered by a mere diagnosis.
What happens to the above question if Opinion Polls peddle partisan agenda or just bad methodology? Let’s consider the recent Opinion Poll that was called India TV-Times Now-C Voter Survey. No where does the poll give out its methodology in any significant detail. That makes it hard for any one following the poll to trust its absolute numbers. However, this is the only poll in recent times that has published successively. Therefore, the one good thing to follow from this poll will be to ignore the absolute numbers and just follow trends benchmarked to the previous poll by the same organisation. For even a bad methodology cancels itself out to some extent in the next iteration when all we want is the incremental movement. The other Opinion Poll on the national level a few months ago did a fantastic job in giving readers/viewers the methodology; however it hasn’t yet followed up with the next iteration and therefore leaves us guessing what that trend might be.
Another recent poll by the AAP — though just for Delhi Assembly Elections — has done the spectacular: they’ve released raw data for analysis by anyone. If not for anything else, and even if their methodology upon analysis comes up woefully inadequate, they should get the vote of everyone with faith data science. Just for trying.
Even if we aggregate the above polls and every other such poll and track them over time and adjust each for ‘house effect’ and other individual quirks, the sample size in terms of actual number of polls in India is too little for anyone to attempt what Nate Silver did. That, and the small complication that we live in a Parliamentary system with First Past the Post where converting vote share to seats won is enormously complicated and not a strict science.
In this scenario, a polling idea that is most suited for India appears to be what the RAND Corporation did. A fixed set of people, of considerably lower sample size, were picked. And the same people were polled every week. This might work far better in a complex and complicated country like India than attempting to perfect the holy grail of random stratified sampling across the multiple fault lines in our society. This allows the errors to largely get benchmarked against themselves as we then get to track the trends in a far better manner. Why no one has yet attempted to do that is baffling — not only will it be more accurate but one imagines, a lot less expensive too.