When thousands crossed over from Pakistan to India on the advent of Partition, Delhi had a great problem accommodating the refugees. The authorities in charge did not have any idea how many people they were looking after. No one had the time or the facility to count. Pandit Jawaharlal Nehru asked Prasanta Chandra Mahalanobis, the great statistician, how could they estimate the numbers. Mahalanobis asked him to find out how much salt was being used in feeding the people. Once he had that information, he gave the number of refugees in Delhi. It proved a very accurate estimate. All he had done was to divide the total amount used by the average per capita normal consumption.
That is the beauty of statistics. It can come to a conclusion about large numbers from small numbers. It also infuriates many people, especially politicians, if the answer statistics gives is not to their liking. This happens each time poverty estimates are published. Politicians denounce the results and, thereby, the honest people who work for the National Sample Survey which produces the results.
Shooting the messenger when you don’t like the message is a normal practice of tyrants. It is also an outcome of the lack of ability to understand how such methods work.
This is obvious in the new clamour for banning opinion polls. The signs are clear in poll after poll which has been released by virtually all the TV news channels since summer of 2012.
The Congress/UPA is unlikely to get beyond 110-120 seats. The BJP/NDA numbers were modest at first—around 130-150. Now they seem to have moved to 160-175. When the BJP/NDA seat estimates were low, the Congress and everyone else used them to pronounce that the numbers proved that Narendra Modi could not win. The BJP needed someone more ‘secular’ like Lal Krishna Advani.
Now that the numbers look worse, the Congress wants opinion polls banned. That illustrates their panic levels. The spurious argument is that they are not accurate. Of course, they are not. They are sample estimates. They carry confidence intervals, plus or minus x%. No one counts or needs to count