Image Courtesy: bookadda.com
The tittle and blurbs of Indianomix: Making Sense of Modern India might give an impression that the book is about economics or more specifically about how economics could help us understand modern India. The book indeed has many examples related to India. However not all examples are confined to economics. The authors try and make sense of a range of topics from the surprising verdict of 2004 Lok Sabha elections, Nehru’s (and India’s) surprise when China attacked in 1962 and how MSM (Main Stream Media) in India confuses opinion with facts.
In one of the chapters, the authors describe how in Delhi economic incentives are used by Delhi Police to improve pedestrian safety and in Mumbai an NGO is using behavioral psychology to prevent deaths at unmanned railway crossings. These two examples show different applications of behavioral economics, a branch of economics closer to psychology than economics. The NGO in Mumbai succeeded in reducing the death toll at railway crossing whereas the experiment in Delhi seems still to be a work in progress. Different chapters in the book explore themes like randomization or luck, how traditions gel with modernity (the most interesting example in this chapter is from China) and how MSM (Main Stream Media) in India prefers sensationalism over fact checking.
The chapter on randomization or luck had two election results (2004 Lok Sabha and 2012 UP Vidhan Sabha) as examples among others. Based on the electoral data and opinion polls/exit polls the authors make two interesting points. First being that close elections are like a coin toss and second (a corollary of the first point), it is a matter of luck to correctly predict the election results. So the authors suggest that saying that NDA lost because of India Shinning campaign and BSP lost because UP voters voted in favor of development politics makes little sense.
I am a student of Indian Politics and hence this chapter was of special interest to me. The authors conclude that in a close election even data (opinion and exit polls in this case) would not be of great help in predicting the outcome. They arrive at this conclusion by comparing the relation between vote share and seat share at macro level i.e. for the whole of India while analyzing 2004 Lok Sabha elections and for entire state of UP while analyzing 2012 UP elections. Lets us look at the case against polls. Yogendra Yadav, India’s most famous pollster and a fox has been conducting exit polls since 1996 Lok Sabha elections and announced his retirement from polling after 2o12 UP polls. A fox is someone who is pragmatic, not ideological and open to discard a cherished theory according to Philip Tetlock, author of Expert Political Judgment: How Good Is It? How Can We Know? where as a hedgehog is a single minded devotee to a particular big idea or concept, which they stick to through thick or thin. According to Indianomix such people exude confidence, boldly tell you without mincing words whats going to often and are more often wrong. Yogendra tried explaining 2004 debacle (of exit polls) by suggesting that there were three main reasons why polls can go wrong: sampling error, response bias and difficulty in translating vote share to seats.
The explanation given for sampling error in 2004 is that urban voters were oversampled and since BJP has most of its supporters in urban areas, most exit polls were off the mark. However a closer look at election results of 2004 would reveal that BJP under performed in urban areas. It did not win a single seat in Delhi and Mumbai. In 1999 NDA had won every seat in these two cities. In Maharashtra, the second biggest state in the country, NDA won 25 out of 48 seats and UPA won the remaining 23. The 23 seats won by UPA include 9 urban seats of Mumbai, Pune, Nashik and Nagpur. NDA could win only 1 urban seat in the state. In fact in the Vidarbha region, the agrarian suicide capital of the country in 2004, out of 11 seats, NDA won 10 and the one seat that NDA could not win was the lone urban seat in the region i.e. Nagpur. In Maharashtra, NDA won more rural seats then UPA and UPA won more urban seats than NDA. Outside the state too, NDA lost most of its seats in urban areas. So this argument of sampling error does not hold much water. Response Bias as explained by Yogendra Yadav, where voters purposely do not divulge correct information, is very similar to the Bradley Effect in US. A closer study of polls in India would reveal that hardly has any poll correctly predicted seat share and vote share correct at the same time. Now if pollsters want to explain this by citing response bias then it would be better if they stopped conducting polls altogether.
The third reason i.e. difficulty in converting vote share to seats has some merit. In a multi-cornered fight, it is almost impossible to predict seat share. Even when the contest is mainly between two parties, it is not easy and we would soon see why.
But again this reason would sound valid if at least pollsters could get vote shares for parties right. Predicting the vote share correctly is a rarity for Indian pollsters and mostly happens when fight is between two main parties. Ironically, Yogendra Yadav who correctly predicted seats for Samajwadi Party in 2012 UP elections, could do so because his vote share predictions were spectacularly wrong. He overestimated SP’s vote share by a massive 5%. In UP, where vote share difference between top two parties was just 3.4%, this error is huge. However, Yogendra Yadav was widely praised for correctly predicting the UP polls as most people including astute political observers (foxes and hedgehogs included) seldom look at the details of elections or polls.
So are outcomes of close elections really like a coin toss as the authors of Indianomix seem to suggest? If we only look at polls conducted in India, we would have to agree with the author. However, if we look at the second biggest democracy, a math geek has cracked the code of predicting elections. Nate Silver not only correctly predicted the final outcome of US Presidential Elections (that Obama would win) but his prediction was correct in all fifty states. And this was a close election with less than 4% vote share difference between Obama and Romney. Its equivalent feat in India would be to not only predict the winning party or alliance in a Lok Sabha election nationally but to also correctly predict the winning party/alliance in each state of the country (if not the actual seats in each state). Needless to say, no pollster has been able to achieve this in India since the first national opinion poll conducted in 1980 by Prannoy Roy and Dorab Sopariwala for India Today.
Nate Silver is not a pollster. His methodology is very simple. A poll of polls where each poll is weighted by its past performance, sample size and other attributes. A blogger who goes by the pseudonym Albatrossinflight, performed a similar exercise before Gujarat election results and predicted 130 seats for BJP. The actual tally for BJP was 115. So does that mean the Nate Silver approach would not work in India? It most probably would not if surveys are conducted in the same way as they are conducted now in India. Most surveys only cover a handful of constituencies and extrapolate the results to the remaining constituencies. Even Lok Sabha elections in India, are an aggregate of 543 constituencies. An urban voter in Mumbai voted differently in 2004 (where BJP lost all seats) from an urban voter in Bangalore (where BJP won both seats of the city). Nate Silver could predict the results correctly in US, because his sample size i.e. the opinion polls were better in quality than the polls in India. His algorithm had multiple state specific opinion polls at its disposal. Indian pollsters assume that an urban voter in Nashik would vote in the same way as an urban voter in Nagpur. It hardly matters to them that MNS is a strong factor in Nashik and may not win the seat but still impact the result whereas in Nagpur MNS has no major impact.
In short, because of quality and quantity of the polls, Nate Silver was better placed to predict elections than an Indian Nate Silver would ever be. Indian pollsters would not only have to conduct polls in every constituency but also get the sampling right in each constituency. This may seem difficult to execute but this is exactly what C-Voter, a survey company tried in Gujarat this time. For their exit poll, they collected samples from each constituency in the state. They may not have got the seat share correct but were very close (within 1% ) in predicting vote share for BJP and Congress. A more detailed study of the poll results of C Voter would give a better idea of its performance. So by 2014, if we have more C-Voter type survey companies, who poll in every constituency, predicting the election result would not be an outcome of coin toss. Since any election is an aggregate of results of its constituencies, the authors are most probably right in asserting that making sweeping generalizations about election outcomes at a macro level makes little sense (like NDA lost in 2004 because of India Shinning or SP won in UP because people voted for good governance).
My favorite chapter in the book is News from India. As the title suggests, this chapter is on Indian Mainstream Media. The authors cite several examples ranging from HIV, child malnutrition, violence against women etc and show how our media in general either does not understand data and logic or choose to take the easy way of rhetoric and sensationalism. Not only this chapter but the entire book should be made for compulsory reading in all journalist schools in the country.
The book may not focus exclusively on any one field and has a range of topics but the theme that binds all the topics is that data and reasoning should be given prominence over popular belief. In one of the chapters, Nehru is quoted after the defeat in 1962, “We were living in an artificial atmosphere of our own creation”. Whether it is a Prime Minister, an NGO working towards saving lives of people who cross railway tracks, someone trying to predict outcome of an election or someone who wants to understand factors affecting women safety after the horrific Delhi gang rape, we all need to be open to reasoning & logic and not swayed by rhetoric. This is the first India centric book which makes this point and hence is a must read for all.