The biggest election in history is聽under way. Machines have the responsibility of counting the votes in India, but how should the聽results be checked? A neat statistical test could be the best way to guarantee that everyone can trust the outcome.
Polls opened earlier this month and, with 900 million eligible voters, the election will take more than a month to complete. Since 2004, people in India have cast their ballots in national elections using electronic voting machines (EVMs) by pressing a button next to their chosen candidate鈥檚 name.
However, everyone from opposition parties to security researchers have raised concerns over how reliable these devices are. Local media has reported large numbers of faulty EVMs having to be replaced during the first phase of voting, and previous research has suggested that it is possible to hack the machines.
Advertisement
As a safeguard, every EVM is now connected to a device that prints a slip with the voter鈥檚 choice, which the voter can view before it drops into a secure box. Some machines will have their tallies and slips compared.
How many should be checked is contentious though. The electoral commission planned to audit one randomly selected EVM in each of聽the 4125 electoral divisions, but opposition parties petitioned the Supreme Court of India to up this to half of all the machines. The commission said this would delay the result, so judges settled on five聽machines per division.
Fixing digital elections
But another option, a technique called a risk-limiting audit, offers a compromise between speed and reliability. An RLA can provide very high confidence in a result or聽indicate if something may be amiss and further investigation is needed, after just a few thousand ballots, much less than other methods. And Vishal Mohanty at the Indian Institute of Technology Madras and his colleagues have recently shown how the method could be used with EVM slips.
With an RLA, first you must decide how confident you want to聽be of the result. This, along with the margin of victory, determines how many ballots need checking. Random ballots are pulled out and a statistical test sees if the votes fit聽with the result called by the EVMs. If it doesn鈥檛, more ballots are checked until the chosen confidence level is reached.
The benefit of an RLA is that it is聽self-correcting. This means that initial counts are generally small, but if these don鈥檛 fit the expected vote distribution, the approach automatically triggers an extended count until the desired confidence level is reached.
One drawback is that you don鈥檛 know how many votes need to be checked until the results are in, hindering planning. And major disparities will trigger a full recount that could delay results.
But when democracy is at stake, the right answer is worth waiting a little longer for.
Edd Gent is a science journalist based in聽India
