Second, statistical tools for election forensics are always based on probabilistic tests. The patterns they flag as anomalous are unlikely to occur randomly according to their presumed model of accurate elections, but are not impossible. If the same statistical test is applied to a large set of elections that were all accurate (according to the underlying model), we should expect some of the elections to be flagged as suspicious, because suspicious patterns do arise by chance. Similarly, if many different tests for different types of irregularity are conducted for the same elections, it is likely that one of these tests will show a “false positive” (Leemann and Bochsler, 2014).