‘Statistical methods need to be taken with pinch of salt’

“The doctrine, even tradition, of statistical significance in scientific research has come under some cloud for its veracity and the journal American Statistician published a special edition on this issue earlier this year. Critics allege that these tests are susceptible to manipulation in order to make desired results significant, and undesired results non-significant,” Das said at the Statistics Day Conference here.

In a recent paper which came under heavy criticism from the Economic Advisory Council to the Prime Minister, former Chief Economic Adviser Arvind Subramanian had used 17 high-frequency indicators that correlate with the GDP to suggest that India significantly over-estimated its GDP figures from 2011-12 onwards. Das on another occasion said that some important results may be discarded at the conception level itself just because they are highly unlikely. “Similarly, an opportunity to ‘cherry-pick’ variables is available. In other words, correlations can be conveniently extended to establish spurious causality,” Das said.