If the argument from Timothy Ferris' 'The Science of Liberty' is to be believed, the collective wisdom of the average citizen is a good predictor of results. That would mean that any opinion poll would be able to correctly predict the result of a general election before hand. If this is considered to be true then one would assume that the BJP would clearly lead in the 2014 polls. But this may not necessarily be the case. The reason is due to unreliable opinion polls.
Why are opinion polls so unpredictable, particularly in India ?
Timothy Ferris in his book The Science of Liberty tells an interesting finding that occurred in 1906, during the annual 'West of England Fat Stock and Poultry Exhibition', in Plymouth, England. The polymath Francis Galton observed a very intriguing statistical phenomenon. One of the many stalls at the exhibition was holding a competition where a visitor could, after paying a six pence fee, guess the weight of a slaughtered and dressed ox. The closest guess would win the meat. Galton wrote down the guess of around eight hundred people or so. He found that though no single person guessed the exact weight of the ox , the collective average of the eight hundred visitors was remarkably close. The crowd estimated the weight to be 1,197 pounds; the exact weight of the ox was 1,198 pounds.
This was in 1906. Since then a number of similar examples have shown how accurate the average prediction of large crowds are. In the Who Wants to Be a Millionaire game show, the consultants who are telephoned by contestants to advise on the right answer, get the answer correct 65% of the time. The studio audience gets the right answer 91 % of the time. ArcelorMittal, the world's largest steel maker uses internal markets to predict the price of steel. Future Markets for orange juice concentrate predict Florida Weather better than the National Weather Service of the United States. Prediction markets -Websites where thousands of people invest in predicting election results and other unambiguous future events -routinely outperform the experts. The Iowa Electronic Markets bested 596 opinion polls in predicting presidential elections from 1988 to 2004- at least 74% of the time.
There are however some conditions, or ( as Timothy Ferris calls them) precepts. Firstly, the outcome must be unambiguous, meaning that the result can only be limited to a this or that and not a maybe. Guessing the number of jellybeans is an unambiguous result. So is the prediction of an election. There can be no grey areas in a result. Only black and white.
Secondly, the participating members must function independently of each other, making their own choices, and not being influenced by others who may or may not be participating in the same exercise.
Thirdly, it helps if the participants are invested in the process. The people at the weight guessing stall for example had to pay six pence. They had something to lose if they guessed wrong. People who vote for a leader have their economic futures in mind. I assume.
Finally, group predictions are increasingly accurate if the members of the group are more diverse; socially, ethnically, sexually, and intellectually.
Well, if the above idea is to be believed then it is quite likely that the opinion polls predicting the BJP coming on top, and The Congress doing dismally in this year's Lok Sabha elections, might be correct. India's voting citizenry enjoys the above mentioned precepts completely, and the opinion polls show that the BJP will be the leading party in the elections.
However, no BJP supporter, pining for a BJP victory, can afford to relax just yet. Looking at the previous two elections in 2004 and 2009 the pre-poll data compiled estimating the number of seats won by each party and alliance versus the actual seats won shows some incredibly stark differences. The party and alliance that actually came to power(The UPA) in the 2004 elections never once came up on top in any of the pre-election opinion polls. The difference in the estimated and actual results is particularly startling when one looks at the data from the 2004 elections.
| 2004 Lok Sabha Elections | |||
| Predicted | Seats | Actual | Seats |
| Congress | 95-105 | Congress | 145 |
| BJP | 190-210 | BJP | 138 |
| UPA | 105-115 | UPA | 275 |
| NDA | 330-340 | NDA | 185 |
Source: INDIA TODAY ORG MAG, A.C. Nielson
Even in the 2009 elections where the opinion polls predicted the leading party and alliance correctly, It's estimate of the number of seats actually won by the INC and UPA was way off the true value.
| 2009 Lok Sabha Elections | |||
| Predicted | Seats | Actual | Seats |
| Congress | NA | Congress | 206 |
| BJP | NA | BJP | 116 |
| UPA | 215-235 | UPA | 262 |
| NDA | 165-185 | NDA | 159 |
Source: CNN-IBN CSDS
Assuming that the precepts mentioned by Timothy Ferris hold true, then why should there be a stark difference between the opinion polls and actual poll data ?
The problem arises due to insufficient amounts of collected data.The problem is that opinion polls do not capture the opinion of all eligible voters. No single firm has the resources to gather the opinions of the whole voting population of any country, let alone India. It's just too difficult. So to come to any conclusion, data must be obtained from a representative sample. And here is the source of the problem. Making a sample representative can be hard, especially if a population is as large and as diverse as the Indian voting citizenry. The sample in any of these pre-polls is often too small, which means that not enough people are being represented. You are missing the views of enough people such that if they were to be added into your sample, they would significantly alter the results you would come up with.
Another problem,one that I feel is more true for Indian opinion polls, is that the sample is too narrow. This means that you are not representing people from enough diverse backgrounds.( The background could be geographical, societal, socioeconomic, etc). I think this narrowness of samples is the main problem because India is so diverse in almost every possible way that to accurately represent every diverse background in a sample is not possible.
So seeing the information from the past two elections, how much can we trust the results of this years opinion polls which calculate the BJP led NDA coming on top ? Making an educated guess from the past results, I would say not alot. I don't think anybody really knows what's going to happen in this year's election.
No comments:
Post a Comment