You can download the analysis from here.
In the end of March 2021, the Anti-corruption Fund published a list of polling stations that were at a relatively high risk of vote purchasing or manipulation.
These polling stations were identified with reference to the results of the parliamentary elections in 2013, 2014, and 2017, and on the basis of the following three models:
The multicomponent abnormal behavior model helps identify polling stations which are outliers in terms of at least two out of three indicators, i.e., they stand out either with unusually high voter turnout in comparison with the voter turnout within the relevant municipality and/ or unusually high number of votes for the winner-party in comparison to the number of votes it received within the entire municipality, and/or with abnormally high numbers of invalid ballots or votes.
The second model helps identify deviations in the voter turnout in a specific polling station between two consecutive elections.
The third model helps record volatility in the political preferences of voters within a specific polling station between two consecutive elections — this can be a significant increase or drop in the votes given for a particular political party in that polling station.
On the basis of the results of the parliamentary election held in April 2021, the Anti-corruption Fund published an updated list of polling stations at a higher risk of vote purchasing or manipulation. ACF also provided an analysis of the dimensions of the votes at risk at national, regional, and municipal level. The methodology was updated by introducing a new variation with a higher outlier value threshold for deviations in voter turnout and for the results of particular parties.
ACF identified between 1,491 (high outlier threshold) and 2,101 (low outlier threshold) polling stations at risk, which received between 9% and 13% of the total votes in the country. The instant analysis consists of two parts. The first part focuses on the voting tendencies in polling stations that were considered at a higher risk of vote purchasing and manipulation during the April 2021 election. The second part presents the results of applying the methodology for identification of votes at risk of purchasing and manipulation to the data from the parliamentary election held in July 2021.