2018/09/11 by Kellie Ottoboni, Ottoboni, Kellie, Philip B. Stark +5 · 1 citation
Computer Science · Economics, Econometrics and Finance · Social Sciences · #Applications (stat.AP) #Electoral Systems and Political Participation #FOS: Computer and information sciences #Game Theory and Voting Systems #Internet Traffic Analysis and Secure E-voting #Methodology (stat.ME) #Sports Analytics and Performance
paper · pdf · doi:10.48550/arxiv.1809.04235
openalex publication_date 2018/09/11 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Risk-limiting audits (RLAs) offer a statistical guarantee: if a full manual\ntally of the paper ballots would show that the reported election outcome is\nwrong, an RLA has a known minimum chance of leading to a full manual tally.\nRLAs generally rely on random samples. Stratified sampling--partitioning the\npopulation of ballots into disjoint strata and sampling independently from the\nstrata--may simplify logistics or increase efficiency compared to simpler\nsampling designs, but makes risk calculations harder. We present SUITE, a new\nmethod for conducting RLAs using stratified samples. SUITE considers all\npossible partitions of outcome-changing error across strata. For each\npartition, it combines P-values from stratum-level tests into a combined\nP-value; there is no restriction on the tests used in different strata. SUITE\nmaximizes the combined P-value over all partitions of outcome-changing error.\nThe audit can stop if that maximum is less than the risk limit. Voting systems\nin some Colorado counties (comprising 98.2% of voters) allow auditors to check\nhow the system interpreted each ballot, which allows ballot-level comparison\nRLAs. Other counties use ballot polling, which is less efficient. Extant\napproaches to conducting an RLA of a statewide contest would require major\nchanges to Colorado's procedures and software, or would sacrifice the\nefficiency of ballot-level comparison. SUITE does not. It divides ballots into\ntwo strata: those cast in counties that can conduct ballot-level comparisons,\nand the rest. Stratum-level P-values are found by methods derived here. The\nresulting audit is substantially more efficient than statewide ballot polling.\nSUITE is useful in any state with a mix of voting systems or that uses\nstratified sampling for other reasons. We provide an open-source reference\nimplementation and exemplar calculations in Jupyter notebooks.\n