The EVM-VVPAT case judgment is disappointing

“The ECI and the Supreme Court cannot proclaim ‘all is well’ with the present egregiously wrong sample size and opaque audit protocol”
| Photo Credit: S.S. KUMAR/The Hindu

“Statistical thinking will one day be as necessary for efficient citizenship as the ability to read and write.” — H.G. Wells

In February 1897, the House of Representatives of the U.S. State of Indiana passed Bill no. 246 that sought to legislatively establish the ‘true value’ of pi (π) as 3.20. This strange Bill was introduced at the behest of a certain Edward Goodwin who claimed to have solved the ancient geometrical problem of “squaring the circle” thanks to a revelation from god, but it required changing the value of pi to 3.20! Fortunately, the Bill was stalled by Indiana’s Senate due to the efforts of a mathematician called C.A. Waldo, and Indiana was saved from embarrassment.

My point in narrating this anecdote is that in the fields of science, mathematics and statistics, truths cannot be established by fiats of the executive, parliament or the judiciary. Just as Parliament cannot legislate water to run uphill, the Supreme Court of India cannot arbitrarily mandate that a uniform sample size of “5 EVMs per Assembly Constituency” is good enough for Voter verifiable paper audit trail (VVPAT)-based audit of electronic voting machines (EVMs) for all Assembly Constituencies across the country. With due respect, its sample size does not conform to fundamental principles of statistical sampling theory.

Voter verification of VVPAT slips ensures that the votes have been “recorded as cast” but it is no guarantee that they have been “counted as recorded”. There is always some risk, however small, of EVM malfunction or manipulation. To ensure “counted as recorded”, we should tally the EVM count with the manual count of VVPAT slips for a statistically significant sample size of EVMs drawn at random from a suitably defined ‘population’ of EVMs.

A typical case

The VVPAT-based audit of EVMs is a typical case of “lot acceptance sampling”, a statistical quality control technique widely used in industry and trade. If the number of defectives found in a randomly drawn, statistical sample is less than or equal to a specified ‘acceptance number’, the entire lot (or ‘population’) is accepted; otherwise, the entire lot is rejected. We define a ‘defective EVM’ as one which has a mismatch between the EVM count and the VVPAT count due to EVM malfunction or manipulation. We specify the acceptance number as ‘zero defective EVM’.

The Supreme Court did not specify the ‘population’ of EVMs to which its sample size relates and the ‘next steps’ in the event of a ‘defective EVM’ turning up in a sample. The Election Commission of India (ECI) has also chosen to leave them vague. Both these are important because in the event of one or more ‘defective EVMs’ turning up in a sample, the entire ‘population’ from which that sample was drawn should be ‘rejected’. In other words, manual counting of VVPAT slips should be done for all the remaining EVMs of that particular ‘population’ and its results declared based on VVPAT count only.

Statistical sampling theory tells us that the probability that the Court-mandated sample size will fail to detect a ‘defective EVM’ is 95% if ‘EVMs deployed in an Assembly Constituency’ are defined as the ‘population’, and 70% if ‘EVMs deployed in a Parliamentary Constituency’ are defined as the ‘population’. This defeats the very purpose of introducing VVPAT.

Reasons behind the ECI’s claim

The ECI has claimed that there was not a single instance of mismatch between the EVM count and the VVPAT count in all these years. Leaving aside the fact that this is not true, there could be three possible reasons for the very few mismatches: the EVMs are in fine fettle; the prescribed sample size is erroneous and fails to detect a ‘defective EVM’ most of the time; both. The correct answer is the last one — both. The ECI and the Supreme Court cannot proclaim “all is well” with the present egregiously wrong sample size and opaque audit protocol.

The Court’s judgment in the Association for Democratic Reforms vs Election Commission of India and Another (2024) is disappointing because it did not compel the ECI to make public how it has defined the ‘population’ to which its sample size relates and its ‘next steps’ in the event of a mismatch.

The Court also did not clarify these points on its own after seeking expert opinion. Even more disappointing was the ADR not pressing for these vital clarifications. Instead, it demanded a return to paper ballots or 100% verification of VVPAT slips. The Court was right in rejecting both these demands.

Other critics of the ECI have been guilty of demanding arbitrary, non-statistical “percentage samples” for EVM audit under the mistaken belief that a bigger “percentage sample” guarantees greater accuracy of results. Congress leader Kamal Nath sought a “10% sample” and lost (2018). Then Andhra Pradesh Chief Minister N. Chandrababu Naidu sought a “50% sample” and lost (2019).

The Supreme Court rightly rejected these demands but the uniform sample size prescribed by it in 2019 was equally arbitrary, non-statistical and wrong.

What needs to be done

We do not know and we do not need to know the various ways in which an EVM may fail or be manipulated. What we need to do is implement a statistically sound VVPAT-based system of EVM audit which can detect instances of mismatch with 99% or 99.9% accuracy. The matching exercise should be done at the beginning of the counting day. Not at the fag end. Where there is a perfect match, the results should be declared based on the EVM count. Only where there is a mismatch, should manual counting of VVPAT slips for all the remaining EVMs of the particular ‘population’ be done and its results declared based on the VVPAT count. This statistical sampling-based, ‘management by exception’ approach represents the golden mean.

K. Ashok Vardhan Shetty is a former IAS officer of the Tamil Nadu cadre and a former Vice-Chancellor of the Indian Maritime University, Chennai