The attempted rollout of the Results Transmission System (RTS) during the general elections has driven increased interest in the role technology plays in elections and the technical process of casting votes and vote tabulation.
As the Election Commission of Pakistan (ECP) looks to introduce another technological solution, this time the i-Voting system for absentee voters in upcoming by-elections, what should the National Database and Registration Authority and the ECP expect from new technologies in elections?
What do we know about why and when they break down, and why and when they succeed?
Here, I discuss three types of technological innovations employed in elections and their use, successful and unsuccessful, in combating election fraud and malpractice. Additionally, I consider the role that surveys can play in ensuring the fair conduct of elections.
Results transmission systems
First, do systems like the RTS reduce fraud?
In theory, such systems mitigate aggregator malpractice, where those in charge of auditing elections are able to manipulate reported polling station returns before they are publicly reported.
For example, in the step between the polling station and the central election commission, a returning officer often gathers the polling station results and aggregates them into one file.
It is at this junction that returning officers could easily manipulate the various polling station level returns just enough to get their preferred candidate into office.
Related: Who rigs polls in Pakistan and how?
An interesting example is the case of Russia, where allegations of electoral fraud are widespread.
Research by several political scientists — summarised here — demonstrates that Putin’s United Russia party received votes that totaled exactly 50, 55, 60 and 65 per cent of the vote in far more polling stations than one would expect by chance.
This indicates that returning officers were simply rounding numbers when choosing what share of the votes to allocate to Putin’s party.
Other forensic methods, imported from the field of accounting, uncover that patterns of electoral returns in Russia are consistent with numbers that are human-generated rather than genuine vote counts.
While these forensic tools help us detect aggregator fraud, can RTS-like systems prevent them in the first place?
In Afghanistan, two researchers, Michael Callen and James D. Long, sent letters to randomly selected polling stations, telling the polling officers that photos would be taken of the local results before they are aggregated.
They then measured the difference between the total number of votes reported at the polling station (akin to the Form 45s in Pakistan) and the results reported by the national aggregation center (akin to the Form 48s).
They find that this monitoring system reduced votes for the politically connected and powerful candidates by a whopping 25pc.
Therefore, even the threat of rapid transmission of results from polling stations, let alone the actual transmission of results, could reduce the malpractice connected politicians can engage in to entrench their positions.
However, such a system is unlikely to have the desired effects if it breaks down, or if fraud is shifted to other methods, such as double voting.
Biometric verification machines
Do biometric verification machines improve election administration by reducing double voting and voter rolls with too many registered voters?
Many election commissions are concerned with double voting, where one person is able to cast multiple votes.
To combat this, countries have turned to biometric verification machines at polling stations. Biometric verification works by scanning voters’ fingerprints at their polling station and being checked off the voter list before they vote.
While it may restrict double voting, this approach faces many hurdles: for example, can the government easily create a dataset of citizens’ fingerprints?
Even if this were done, what about people without clear fingerprints that might be difficult to read for biometric verification machines on election day (or even in the initial dataset)?
How has this fared in practice? We can look at the recent trial by the ECP in the NA-120 by-election, where 12pc of fingerprints could not be read.
The ECP smartly decided that this error rate was too large and the technology would not be ready for implementation in the 2018 general elections.
Also relevant in this context is the case of Ghana, where biometric verification machines have been deployed for three consecutive elections.
While it is likely that the incidence of double voting and irregularities on the voter registry list have been limited, there have also been widespread accusations that the biometric machines are being manipulated.
In research conducted during the 2012 elections in Ghana, Miriam Golden, Eric Kramon, George Ofosu and I found that biometric verification machines were 50pc less likely to break down in polling stations where we had randomly assigned an election monitor.
It is difficult to study what the effects of this breakdown are, but the fact that the presence of election monitors reduces machine breakdown implies that machine breakdown is not random and may be part of more systematic attempts at the manipulation of elections.
Therefore, while the introduction of such biometric technology may be well intentioned, it can still be circumvented by miscreants through the strategic breakdown of the technology, if need be.
Third, does electronic voting make the electoral process easier and hence reduce rejected votes?
While there are many tools meant to ensure the validity of each ballot, these tools can also make voting difficult for some and completely inaccessible for others.
Indeed, even the process here in Pakistan can appear unnecessarily complicated at times. This could result in high levels of invalid or rejected ballots being cast.
One mechanism commonly used to combat the complexity of ballots is to make voting electronic.
Take Brazil, for example, where electronic voting was introduced in 1998 in its largest municipalities.
By comparing municipalities that were just large enough to receive the electronic voting machines to municipalities that were just not big enough, Thomas Fujiwara demonstrated that the electronic voting increased the share of valid votes from 78pc of votes to 89pc of votes.
The results indicate that the introduction of electronic voting drastically increased the number of people able to effectively cast their ballot.
Fujiwara also demonstrates that the rollout of these machines had positive policy consequences for those likely disenfranchised by complex voting processes.
While the introduction of electronic voting also introduces the problem of easier manipulation of votes by internal or external forces and the same problem of “breakdown” that plagued the biometric machines in Ghana, they seem to have a clear positive effect on improving the accessibility of elections.
While technology is often viewed as a panacea, it is not sufficient on its own, as some of the case studies above demonstrate.
Therefore, additional solutions that are independent of the formal electoral process are necessary to ensure that elections proceed as expected.
One such solution, while not perfect, is to improve the quantity and quality of pre-election surveys about vote choice.
Rigorous surveys, held to high academic and statistical standards, provide important benchmarks along which the public can judge election results.
Election 2018 survey results: And the winner is...
If the results of the election are mostly consistent with a large quantity of high quality pre-election polls, then at the very least the level of fraud was low, in that it did not massively change the overall distribution of votes.
Let’s take the recent 2018 elections. There were three major polls carried out in June and July by major survey firms, the Sustainable Development Policy Institute (SDPI), Institute for Public Opinion and Research (IPOR) and Gallup.
If we take the simple averages of the expected votes across all three parties, we can compare those predicted values to the resulting numbers.
For the Pakistan Tehreek-i-Insaf (PTI), the surveys reported an estimated 28pc vote share. The PTI ended up with 32pc of the vote.
The expected percent vote for the Pakistan Muslim League-Nawaz was also roughly 28pc, and they ended up with 24pc of the vote.
For the Pakistan People’s Party, the expected vote share was 16pc, and they received 13pc of the vote.
Because all of these numbers are within four percentage points, and that the trend in the surveys over time was towards the final election results, we should be more assured that massive election day rigging was unlikely.
There are, of course, several caveats.
First, many of these polls were carried out well before the Sharif family received their conviction and sentencing, which most likely had an impact on voter sentiment.
Second, many of these polls had large numbers of undecided “swing” voters that could go one way or another.
Third, it is not entirely clear whether these polls used a simple random sample of voters, or whether they tried to over-sample likely voters.
This last point is of significant concern, and led to some erroneous state level polls in the 2016 United States presidential election.
While the overall pre-poll averages accurately predicted Trump and Clinton’s vote shares to within roughly two percentage points, at the state level, many of the pollsters failed to accurately model which of their respondents were likely to end up voting.
This skewed the predictions, as while the accurate national-level results were encouraging, the state-level results matter more due to the Electoral College.
Read next: The prospect of electronic voting
Therefore, for surveys to better serve as clear benchmarks for the validity of election returns, they need to be of high quality and their methods need to be transparent.
Furthermore, there is always a fear that surveys will be taken out of context and misrepresented in the media and by political parties.
Indeed, analysts often overinterpret the results of each and every poll and parties have incentives to highlight certain results while dismissing others.
It is important to convey that all surveys carry a large degree of uncertainty, and that the results be explained as clearly and as fairly as was done by Ali Cheema, Sarah Khan, and Asad Liaqat in their analysis of the pre-2018 survey carried out by SDPI.
The ECP has been fairly progressive in incorporating technology in the elections process. They attempted to roll out the RTS system and had previously discussed rolling out biometric verification machines as well.
As mentioned previously, the ECP’s decision to not implement their use in the general elections due to their high 12pc failure rate was a responsible choice.
In the case of the RTS, however, this piloting was not done with enough time, and they were not able to learn from their mistakes.
As global experiences demonstrate, technology in elections can be applied effectively to improve elections administration.
However, technology fails and may be abused; fail-safes as well as ample piloting should be used when it is implemented.
The ECP should be applauded for its attempts to implement new technologies, and should be given both the capacity and the bandwidth to pilot new technologies and implement back-up systems for when they fail.
To support the ECP in its efforts to ensure the validity of elections, the media and interested election observers should support additional independent, rigorous surveys, especially as the election draws near.
The government of Pakistan should facilitate such survey firms in their operations to help lower costs and administrative barriers to surveying, thereby encouraging research which adds to an ecosystem of well-informed voters, parties, and a transparent electoral process.
Are your researching electoral reform? Share your insights with us at firstname.lastname@example.org