Thursday, November 05, 2015

Technology for Election-related Polling

There is some statistics behind the design of exit polls; but the state of the art is rather poor. Exit polls are giving contradictory predictions about the outcome of Bihar polls, 2015. They failed to predict the landslide in favour of the Aam Aadmi Party in the Delhi elections earlier.
I will offer a suggestion here for a new way of conducting exit polls. Any market research organization could easily try them out. My hope is that small experiments by students learning software technology or management would lead the way and lead to publishable R & D.

The basic premise is that a high percentage of voters have a cell phone. This is not entirely true, but we will return to that question later. Assume that the polling organization (call it pollster) offers during a national election handsome prizes like a hundred airline tickets to a holiday destination, a thousand gift coupons, and ten thousand brand-name pens. To qualify for a prize, you need to send an SMS within five hours of a poll closing, from the constituency concerned, using a cell phone registered in that constituency, giving the constituency name and the party you predict will win there. The polling computer will allow each cell phone to cast only one vote. Now comes the second premise: that you will most likely predict that the party you voted for will win. We assume that you did not waste your vote for a party you expected to lose. You reflect your own belief as well as those of the people who influence you by their conversations. We assume that political parties will not waste much effort in falsifying an exit poll as that confers no real benefits on them.

That is the basic idea. What would be cost-effectiveness of the idea, how can it be improved, can it be modified to serve other information collecting functions, etc., are good questions for students (and others) to examine. We also need to examine privacy issues involved and whether cell phone companies analyze the text messages they carry.

I would like to add another idea, to encourage voters to participate in the exit polls. The computer could randomly select a specified small fraction of the participants within a few minutes of receiving their text messages and send them a message like this: “You have been selected to receive a ----- if the party you have indicated wins in the constituency you have named. Prize winners are selected by a computer which uses random numbers to make fair selections”. This will increase the confidence of the participants that fair practices are in use.  

Finally, a word on my basic premise. Will the people at the bottom of the socio-economic pyramid really have proper representation in a technology-based poll? No. This is a matter worth considering.  

                                          - - - - - - - -  

1 comment:

Srinivasan Ramani said...

The results of the Bihar election came out on Nov 8, 2015. NDTV's Prannoy Roy's predictions, based on field work by a research agency, turned out to be wrong. There was chaos during the counting hours on Nov 8. Most channels were giving information which lagged behind announcements of interim counting data by returning officers. Only CNN-IBN (among the English channels I monitored) used faster methods of obtaining information and was giving accurate news.

Forget exit polls! Good polling technology should do a lot more for political parties. It should give them a continuous feedback of what the people feel. What bothers them, who are they pleased with, when do they feel let down, etc. If the technology is not good enough even to handle exit polls to give good predictions, how are we going to get real feedback on a continuous basis?