Regressing to an imaginary mean

There have been recent allegations that opinion polls are up for sale. Typically suggesting they could manipulate aspects of the poll such as favorability of a specific candidate for PM if not the actual vote share.

In that context, the CNN-IBN/CSDS poll for Maharashtra deserves some attention. The first thing that anyone would notice about the poll is the dramatic fall in Narendra Modi’s support for choice of Prime Minister. The poll shows a 9 percentage point drop in a 1 month period — a steep enough fall in a short enough time period for any presenter to make this the primary object of discussion. However, if one listened to the panel discussion on Television instead of reading their summary tables, one’d think Modi were still on the upswing. The anchor even went on to say things like “it’s Modi all the way.”

The next interesting and never discussed fact was that the NDA camp had only one PM choice in the summary tables. The NDA has obviously a declared candidate and hence there is only one. However, the UPA has not officially declared its candidate. And it’s reasonable to assume the choice of Rahul Gandhi, Sonia Gandhi and Manmohan Singh are all interchangeable. If one adds their three numbers together along with their ally, Sharad Pawar, it comes up to 30%. That’s just a one percentage point difference between the two major camps according to the poll; Narendra Modi polled 31%. Yet, the entire discussion on Television was how a “majority” of the people wanted Mr Modi as PM and how this was driving the BJP. That could mean only one of few things — either the panelists had not read the poll numbers or don’t know how to or worse, have been asked to toe a certain line as alleged by Caravan.

If one assumes the benefit of doubt goes to the panellists and they were merely ignorant on how to read successive polls, one then is left with the poll itself. As anyone who has watched polling numbers can attest, nine percentage point movement in one month is phenomenal. So it has to come from somewhere in this zero sum game. If it were a natural variation in the data, one’d then expect the numbers to probably be the result of one person catching up with the leader. As it usually happens in Gallup Polls in American primaries. But instead, in an odd looking table of comparison, each person in the INC camp — Rahul Gandhi, Sonia Gandhi and Manmohan Singh — have all risen exactly two percentage points. The probability of everyone having the same percentage point raise is, well, remote. Add to this, the table also means Manmohan Singh has doubled his popularity as a choice of PM in the past month. This after expressly declaring himself out of the race and not being seen in public in that period. Baffling is a major understatement.

If one were to assume this to be a freak result of normal variation in data, what’s even more interesting is that in the latest instalment of the Maharashtra poll the choice of PM question is shown in the data region-wise. While in the previous month’s poll it wasn’t. So there isn’t any chance of seeing where the catching up has been for such a uniform two percentage point increase favouring Sonia Gandhi, Rahul Gandhi and Manmohan Singh. Especially Manmohan Singh.

Finally, the panellists on the show may not be statisticians. But they are literate enough to understand 2009 actual result has no meaning as a base value when the past month’s poll exists. Yet they insist on looking at the absolute number instead of the direction of the poll movement incrementally, given the latter is self adjusting and thus probably a far a better reflection. Strange are those who are wrong once and knew they were. Especially when they are in the business of predicting the future and so do not know what the right answer was. Or is.

Read more here –


Enhanced by Zemanta