Election forensics are methods used to determine if election results are statistically normal or statistically abnormal, which can indicate electoral fraud. It uses statistical tools to determine if observed election results differ from normally occurring patterns. These tools can be relatively simple, such as looking at the frequency of integers and using 2nd Digit Benford's law, or can be more complex and involve machine learning techniques.
Election forensics can use various approaches. Some approaches include looking at data distribution, particularly voter turnout, to look for outliers. Other approaches can include comparing the observed distribution of the digits themselves to typical digit distributions (Benford's law). Other signs of fraud are overrepresentation of round numbers rather than those with decimals, or overabundance of numbers that are a multiple of 5 (e.g. 50%, 70%, 75%). More recent and statistically advanced approaches use machine learning, as machine learning can incorporate a large volume of data and use several different statistical models instead of a single one.
Between 1978 and 2004, a 2010 review concluded that 61% of elections examined from more than 170 countries showed some signs of election fraud, with major fraud in 27% of all examined elections. Since the early 2000s, election forensics has been used to examine the integrity of elections in various countries, including Afghanistan, Albania, Argentina, Bangladesh, Cambodia, Kenya, Libya, South Africa, Uganda, Venezuela and USA.
Compared to other methodsEdit
Relative to other methods of monitoring election security, such as in-person monitoring of polling places and parallel vote tabulation, election forensics has advantages and disadvantages. Election forensics is considered advantageous in that data is objective, rather than subject to interpretation. It also allows votes from all contests and localities to be systematically analyzed, with statistical conclusions about the likelihood of fraud. Disadvantages of election forensics include its inability to actually detect fraud, just data anomalies that may or may not be indicative of such. This can be addressed by combining election forensics with in-person monitoring. Another disadvantage is its complexity, requiring advanced knowledge of statistics and significant computing power. Additionally, the best results require a high level of detail, ideally comprehensive data from the polling place regarding voter turnout, vote counts for all issues and candidates, and valid ballots. Broad, national-level summaries have limited utility.
- Stewart, Charles (2011). "Voting Technologies". Annual Review of Political Science. 14: 353–378. doi:10.1146/annurev.polisci.12.053007.145205.
- Hicken, Allen; Mebane, Walter R. (2017). A Guide to Elections Forensics (PDF) (Report). University of Michigan Center for Political Studies.
- Mebane, Walter Jr (2006). Election Forensics: The Second-digit Benford’s Law Test and Recent American Presidential Elections (PDF) (Report). Cornell.
- Zhang, Mali; Alvarez, R. Michael; Levin, Ines (2019). "Election forensics: Using machine learning and synthetic data for possible election anomaly detection". PLOS ONE. 14 (10): e0223950. Bibcode:2019PLoSO..1423950Z. doi:10.1371/journal.pone.0223950. PMC 6822750. PMID 31671106.
- Lacasa, Lucas; Fernández-Gracia, Juan (2019). "Election Forensics: Quantitative methods for electoral fraud detection". Forensic Science International. 294: e19–e22. arXiv:1811.08502. doi:10.1016/j.forsciint.2018.11.010. PMID 30527668. S2CID 54481752.
- Noonan, David (30 October 2018). "What Does a Crooked Election Look Like?". Scientific American. Retrieved 10 August 2020.
- "Notes on Election Forensics, Exit Polls, and Baseline Validation". CODE RED-Computerized Election Theft. 2018-08-08. Retrieved 2020-11-28.