Monday, May 14, 2007

How The Psephologists Missed The Wood For The Trees

WHEN I referred to the hypocrisy of the poll pundits in a previous post in the context of U.P. election analysis, some blogger friends could not understand whom I am against: voters, analysts, or parties/candidates? An interesting article in the HT edit page by sociologist Dipankar Gupta articulates the same point which I had made, albeit while examining why the psephologists failed to predict the scale of BSP’s victory. Although the author referred to psephologists, much of what he said, could well be true of the analysts, the pundits on various channels, and the so-called beat correspondents of various print newspapers. Today’s report in The Hindu, for instance, gives caste/community break-up of Mayawati’s Cabinet. One wonders what sort of reader interest is served by this break-up. I can understand if one section is completely left out of the Cabinet, or even over-represented rather disproportionately. The break-up of the numbers shows that the journalists are still trapped in the same mind-set which failed to foresee the BSP’s victory.
Gupta says: “None of the psephologists predicted that the BSP would get an absolute majority simply because they fractionated voters minutely by caste. These pollsters would have done better if they had asked on what grounds members of different castes coalesce politically. As the UP election has shown, jati loyalty is not the key. The emergence of a degree of caste correlation with electoral outcome is because economic, social and structural considerations bring otherwise hostile jatis together in caste blocks or clusters.”
Although I agreed with a blogger-friend, that we should not seek to restrict expression of such opinions in the media, saying they must have the freedom to fool the viewers and readers (because what they say and write do not reflect reality), I had also felt they were essentially playing a negative role, and must be discouraged. This is how Dipankar Gupta concludes: “But psephologists need to reinvent themselves in a hurry. By insisting on the pre-eminence of caste round the clock during election time they are not only wrong, but also dangerous. Incorrect though they are on every count, they succeed, however, in a somewhat devious way. They are successfully able to pander to popular prejudice by continuously harping on individual caste identities. It is in this sense that they play a negative social role that borders on the subversive.” What Gupta says about the OBCs being common adversary of both the Dalits and Brahmins in U.P. may be debatable, but on the issue of how pollsters went horribly wrong, one is tempted to agree with him.
Update: Mahesh Rangarajan offers a different analysis of the BSP's victory in Telegraph. He says OBCs are very much part of Mayawati's rainbow coalition, marking a clear disagreement with Dipankar Gupta. He says: " Contrary to the claims of some commentators, Mayavati has accommodated the Mandal classes. Her vote share among the lower backward classes, according to the surveys of the CSDS, is as high as 30 per cent. This makes her the head of a rainbow coalition." His article is here.
The editorial in Tribune today calls for retirement for pollsters till they improve their methodology.
Further Update: Mr.Suresh writes in the response section that I might have underestimated the problems of psephology in a first-past-the-post system, and that psephologists like Yogendra Yadav and his team are fully aware of the nuances of the Indian voters' behaviour, especially in t he context of U .P., and that Dipankar Gupta's article must not be relied upon to pass an unfair judgment on the entire class of psephologists. He also writes that YY and his team have consistently repeated in their programme on the CNN-IBN that their conclusions are only indicative, and that they should not be construed as making predictions. I agree with Mr. Suresh that the studies carried out by YY and his team have yielded rich data on Indian elections, and it is for this contribution, psephologists have to be admired. The 'predictions' aspect of psephology is important to sustain viewer interest in this era of 24-hour news channels, and therefore, the imperfections associated with this aspect may be condoned. However, I would still like to have a complete response from YY and his team to the issues raised by Dipankar and me (My grievance was mainly against journalists masquerading as poll pundits) especially with regard to the assumptions and premises underlying these 'indicators' (if the word 'predictions' may not be correct to describe their efforts).
In the past, YY had written a post-mortem of the opinion and exit polls done by his team to throw light on what had gone wrong, and draw useful lessons. I am looking forward to a similar exercise by him in a print medium of his choice. I shall also try to contact YY, as Mr.Suresh suggested, to throw more light and possibly dispel certain misgivings on the practice of psephology in India, in the context of the U.P. results. I am thankful to Mr.Suresh for this timely intervention. I have the highest respect for the scholarship and contribution of YY and his team, and my post was, in no way, aimed to disparage his methodology or analysis.


Anonymous said...

Mr. Venkatesan,

Since I know Rajeeva Karandikar (the statistician who works along with Yogendra Yadav) a little, I can assure you that neither he nor Yogendra Yadav himself subscribe to simple notions of voter behaviour like "a voter's voting decision is based on his/her caste and the caste of the candidate." They both are aware that voting decisions are more complex, though of course, jati plays a part in that decision.

From what I know, their work is solidly grounded in sampling theory. They basically sample the voting population taking due account of the diversity of that population. From the data that they collect, they analyse it to infer the way the population as a whole is going to vote. This part is relatively straightforward though not easy in practical terms: designing the sampling methodology and then carrying out the sampling involve a lot of hard work on the ground. But I know that YY and RL and their team do it fairly well.

The problem, as both YY and RL emphasize and continue to emphasize is one of translating the sample data into a prediction of the number of seats that will be won by each party. This is extremely difficult in our first-past-the-post election system with multiple parties. A small change in the voting percentage of one party can make a huge difference in the number of seats won by that party. If you analyse the past voting results, you will observe this phenomenon in a number of elections. Both YY and RL are therefore extremely careful in giving predictions about the number of seats and in the IBN program that I watched, both said that their prediction was just "an indicator." Indeed, RL said that he would not put any money on his own prediction!

I am not trying to defend RL, YY and their other counterparts. But I do wish you'd try to understand what they do before attributing harsh and simplistic views to them. Dipankar Gupta may be a sociologist, but I don't think he understands what was done by at least the more competent psephologists and the practical difficulties in giving a prediction of the number of seats.

As you are a journalist, I hope you won't mind my saying that you will get better insight of the psephology business by talking to people like YY and RL on their methodology. It is fashionable to criticise statisticians, but in the hands of a good practitioner, the data yields good and valuable insight.



Abi said...

I find Dipankar Gupta unpersuasive. After castigating the psephologists about their assumption about the primacy of caste, what does he have to offer? That this election's main message is that Brahmins and Dalits came together to vote against the OBCs!

The rest of his article is nothing but the now fashionable diatribe against the OBCs by painting them as the new oppressors in rural India. By overplaying the OBC ascendence, he seems to want to create the impression that the OBCs have well and truly 'arrived'. This line of thinking ignores evidence (which Satish Deshpande presents in his book 'Contemporary India') that shows clearly the huge lag between the upper castes and the OBCs.

I agree that caste is not everything. A smart psephologist like Yogendra Yadav would be the first one to admit it. And, in fact, he does!

So, in effect, Dipankar Gupta picks a straw man psephologist and castigates him for being such a straw man!

But, coming back to your own misgivings about caste identities, my question to you would be: what is wrong with it? If someone thinks of his/her caste as an important part of his/her identity, why can't we just say, 'so be it', and move on? In what way is the caste identity 'badder' than, say, class identity? Or, one based on language? Or, religion?

If caste is going to play a role in our politics, I would much rather have a situation in which we discuss it openly and dispassionately and collect data on it objectively. It's far better than treating it as a big no-no, when we all realize that it is possibly one of the elephants in our political living room.

V.Venkatesan said...

Dear Mr.Abi,
Thanks for the response. But I think we don't disagree at all. Where I thought I agree with Dipankar was on the question of how some poll pundits (His target was psephologists) assume that voters vote according to caste identity. I am not against caste identity as such. I do, like you, understand that it is the reality. I am only saying it is wrong to assume that voters belonging to a caste vote enbloc on the basis of their perceptions about their caste interests. Based on this flawed assumption, if the pundits make a forecast, it is likely to be a flawed forecast as well. I do not mean that caste identity is worse than religious or linguistic identities. I agree that caste is one of the significant means of political socialisation and even mobilisation. But it will be simplistic to assume that all members of a caste or community vote enbloc to a particular party or candidate, on the basis of narrow perceptions of who would better serve the interests of their caste or community. There is no data to suggest that it happens; still most pundits/analysts base their forecast or analysis on the basis of this flawed premise. Is it correct or desirable? Well, it is not real, so the question whether it is correct or desirable does not arise. The Election Law and the Model Code do say that it is not desirable to appeal to one's caste/religious/linguistic identity for electoral purposes - even though there is lack of clarity because of a lacuna which I pointed out in my earlier post.