Election forensics
English: A portable Tableau forensic write-blocker attached to a hard disk drive (Photo credit: Wikipedia)
A recently published virtual issue on election fraud and electoral integrity [Ines Levin and Michael Alvarez Publishers] assembles recent research on the use of mathematical methods to detect fraud in elections. The papers are made available for free download during a limited time. Under the appealing title of “election forensics” the four papers included in the first part of the virtual issue deal with the recent use of methods based on mathematical laws such as Benford Law to search for “patterns that deviate from an expected distribution” of figures. Such deviations are considered to be indicators of possible errors or manipulations of election results. Such methods are being applied to data from past elections in several countries. Researchers check whether the employed method produces results consistent with allegations of fraud or statements of correctness coming from other sources such as classic observation at the respective election. The three papers of the second part introduce methodologies for “election forecasts“. Such methods could also affect the development of techniques for evaluating the integrity of elections according to the authors.
Mathematic laws as election oracles?
The use of mathematical laws alone to predict or evaluate outcomes of highly complicated social choices such as political elections has always intrigued me. By contrast the application of mathematical methods to predict or evaluate results of a vote in a specific place based on data collected from several previous votes and elections that took place in that same place over years is something I can more easily understand.
In this sense some general conclusions of the virtual issue on the recent use of mathematical methods in the electoral field are very interesting. The first conclusion is that it is necessary to have an explanation of the relationship between the electoral behavior and the output produced by the forensic indicators. In other words a theoretical foundation for their application in the elections’ field is still missing, at least for some of them. Another one, obvious as it may seem, is that scholars applying these tools should have adequate knowledge of the substantive context of the election, such as knowledge of which regions of the country are more vulnerable to fraud, and what specific types of fraud to look for. According to the publishers, more theoretical development about when and where we might expect to see election manipulations attempted is necessary. Scholars should also work on the development of criteria for distinguishing between deviations that are caused by fraud, and deviations that are caused by other factors. Data availability and data quality are also important issues.
Election forensics and internet voting
The Swiss federal legislation on political rights contains several provisions that deal with detection and reaction to errors and fraud allegations. The use of specific forms for the transmission of results, especially in complicated procedures such as the proportional elections for the federal Parliament, allows the authorities to detect possible counting errors. Vote recounts and the holding of a new vote are foreseen in specific cases. For the rest it’s up to the cantons to decide on the kinds of checks they conduct. Some of them, Zurich and Geneva for example already use statistic methods to detect possible anomalies in voting and elections results.
More detailed provisions on fraud are to be found in the federal legislation on internet voting (art. 27a ff. Federal ordinance on political rights). One of the conditions to obtain the federal authorization for using internet voting during a federal vote or election is for the canton to prove that all technical and organizational measures to exclude systematic fraud are adopted (art. 27d). If the electronic voting results are contested, their plausibility must be proved. Three tests are foreseen in this context (art. 27nbis): checking the results of the control ballot box, comparing results (for example percentages of yes and no votes) obtained through different channels as well as comparing results with log files. The use of checks based on mathematic laws is not specifically foreseen but, as always, cantons are free to introduce such checks.
That’s what the canton of Geneva, a pioneer in this area, has done. Methods based on 2D Benford Law are being used in combination with other statistical methods that refer to the voting profile of communes over the years to detect possible anomalies in internet voting results. Geneva regularly publishes a summary of the tests and their conclusions at each voting event. An example from the last federal vote of 25 November 2012 can be found here.
Election forensics’ role
Once election forensics has shown that possible errors or fraud occurred, the question is what comes next? And in case claims of fraud based on statistical tests results are brought before a Court, what is the legal value of such tests?
As mentioned in the virtual issue, the use of methods in the electoral field is very recent. In the case of Geneva, 2D Benford Law is used in combination with other statistical measures. So far it never occurred that the combined result of such checks indicated anomalies. In case this happens authorities will most likely have to further investigate such indications in order to confirm/deny that errors or fraud happened. The competent authority can and must correct errors. If fraud is confirmed the competent authority has to evaluate its impact. If such evaluation shows that the manipulated results are capable of changing the outcome of the vote, the cantonal authority must organize a new vote.
Another question related to election forensics tools is their relationship with (individual or universal) verifiability tools. Indeed, to the difference of other voting channels such as postal voting and ballot box voting, internet voting systems can (theoretically at least) allow citizens to control that their vote has been transmitted and counted as intended. Work in this area is ongoing. If and when this is the case, do we still need forensic tools in addition to verifiability tools? Deeper investigations in election forensics and presumably some adaptations of political rights legal provisions (with respect to the legal value of forensic methods) may be necessary for these methods to be used more extensively in the future.
