Adjacent Wikipedia votes had a 1.10 percentage point daily agreement excess under a voter history model in 2004–2013

We tested whether voter support tendencies account for adjacent vote agreement in a Wikipedia record. We used 109,602 adjacent binary vote pairs for 2,290 candidates in 2004–2013. We estimated each voter's oppose odds from votes on other candidates. A conditional sign model kept each candidate's daily support and oppose counts fixed. Observed agreement was 80.705 percent. The model expected 79.600 percent. The excess was 1.105 percentage points, with a descriptive 95 percent candidate bootstrap interval of 0.930 to 1.285 points. A uniform reference gave an excess of 1.120 points on the same pairs. Weighting raised expected agreement by 0.015 points, with an interval from −0.006 to 0.037 points. The excess also remained with one-hour counts and same-year voter histories. The specified voter history model therefore changed the earlier result only slightly. The result does not identify social influence. Candidate-specific relations, common information, source errors, and model errors can remain. We provide exact reference code, aggregate results, checks, source hashes, and repeat instructions. Raw vote records remain external.

collective decisions · Wikipedia · public votes · voter history · conditional Bernoulli model · temporal dependence · reproducible analysis