Putting the EVM controversy to rest

Doubts about the reliability of electronic voting machines will continue unless election results obtained through EVMs and voter-verified paper trails are matched through statistically rigorous tests

The 2006 Emmy-nominated documentary Hacking Democracy investigated anomalies and irregularities in electronic voting systems, particularly those made by Diebold Election Systems, during the years 2000-04 in the United States, especially in Florida. However, Diebold President David Byrd suggested that the documentary was “replete with material examples of inaccurate reporting”.

Most voters, like me, are technologically not knowledgeable enough to understand the security of electronic voting machines (EVMs), and would depend either on institutions or political parties to make up their minds in such a delicate issue. The Election Commission of India (ECI) has consistently maintained that EVMs in India are standalone devices not connected to any network, that the software is written onto a one-time programmable chip, and that these EVMs cannot be tampered with. The courts also have ruled in favour of the machines quite a few times. However, most major political parties have raised their voice against EVMs at some point or the other.

The EVM Hackathon Challenge organised by the ECI in June 2017 was, in fact, a litmus test for EVMs, but no expert from any political party could hack the machine. However, such allegations are not good advertisements for democracy. And the “common man” remains utterly confused. Can this be stopped by using scientific evidence?

Representative image

Now that we are almost into the era of 100 per cent Voter Verifiable Paper Audit Trails (VVPATs) in all elections, VVPATs can be used for macro-level verification as well. Some statistical application, specifically efficient sampling and probability calculations, might be enough to prove that EVMs have not been tampered with. In a Press Note dated May 20, 2017, the ECI stated: “The Commission will count VVPAT slips up to a definite percentage, which will be determined by the Commission.”

In the past, there was speculation about the possibility of a five per cent ceiling on counting VVPAT trails, and also mandatory counting of 5-14 polling stations in each constituency. There has even been speculation to the effect that 10-25 per cent of VVPAT slips would be counted manually. But, that is a huge proportion. Statistically, we really do not need such a large number of paper trails to be counted and tallied with EVM results. A much smaller number of VVPAT slips would be more than enough to ensure that EVMs have not been tampered with.

During the Assembly Election in Gujarat in December 2017, VVPAT slips and EVM results were matched in one randomly selected polling station in each of the 182 constituencies, and the Commission did not find any discrepancy. While it was an experiment involving 182 polling stations out of a total of 50,128, which is just 0.36 per cent, in practice it might be strong evidence of non-tampering with EVMs, provided proper statistical sampling design and techniques are followed in selecting the sample.

For example, if only two per cent of EVMs are tampered with, such a small sample size with proper sampling methodology will be able to identify a tampered EVM with 97.5 per cent probability. Let us try to understand how such a small sample size would work.

About 18 lakh EVMs are used all over India across 543 Lok Sabha constituencies. To check whether rice is properly boiled while cooking, we use a spoon to make the grains homogeneous in the rice pot, and then test only a few grains. That is sampling, enough to judge whether the rice is nicely boiled, and the technique can be used to identify possible tampering in the case of EVMs as well. However, “random” selection of EVMs should be executed by eminent experts in sampling, who could be appointed a priori for this job by an appropriate authority.

Some technical issues, such as the possibility of stratified sampling, selection of strata and determination of sample size within each stratum, should also be fixed a priori. For example, assuming that only 0.1 per cent of EVMs are tampered with, a sampling of one per cent will almost certainly identify at least one mismatch. If, instead, only 0.3 per cent of EVMs are sampled in this case, a mismatch can be identified with a probability more than 99.5 per cent.

Alternatively, we may prefer to assess the EVMs of each constituency separately, depending on its political and social nature. On average, a little over 3,300 EVMs per constituency are used in India, and the number of EVMs for which VVPATs are tallied should depend on the level of vulnerability and level of precision we need. For example, if we want to identify a “20 per cent EVM tampering” scenario in a constituency with 95 per cent probability, we would need to count VVPATs in 14 machines. On the other hand, a “three per cent tampering” scenario can be detected with 99 per cent probability with a sample size of 99 per constituency.

In both cases, nearly three EVMs that have been tampered with are expected to be found in each constituency. In any case, we would count the VVPATs of a very small proportion of EVMs, and yet that would provide enough scientific evidence. Also, there might be different levels of allegations of EVM tampering in different parts of the country, which might be an additional input for the formation of strata.

A “no mismatch” in the sampled EVMs in such a set-up might be very strong scientific evidence against the possibility of EVM tampering. Voters may also find it helpful to keep their faith in EVMs. Unless EVMs are put through such a clear scientific agni pariksha, grumbling about them will continue. Bolstering the common man’s confidence in the safety of the EVMs is no less important than making the EVMs themselves safe, if not more important.

The writer is Professor of Statistics at the Indian Statistical Institute, Kolkata