No abstract

We present a collection of 7582 real-world elections divided into 25 datasets from various sources ranging from sports competitions over music charts to survey-and indicator-based rankings. We provide evidence that the collected elections complement already publicly available data from the PrefLib database, which is currently the biggest and most prominent source containing 701 real-world elections from 36 datasets [53]. Using the map of elections framework [76], we divide the datasets into three categories and conduct an analysis of the nature of our elections. To evaluate the practical applicability of previous theoretical research on (parameterized) algorithms and to gain further insights into the collected elections, we analyze different structural properties of our elections including the level of agreement between voters and election's distances from restricted domains such as single-peakedness. Lastly, we use our diverse set of collected elections to shed some further light on several traditional questions from social choice, for instance, on the number of occurrences of the Condorcet paradox and on the consensus among different voting rules.1 An election is single-peaked if there exists a societal order of the candidates and each voter ranks candidates that are closer to its top-choice according to the societal order above those which are further away.