Incentives in Best of N Contests: Quasi-Simpson's Paradox in Tennis

Simpson's Paradox is a statistical quirk where an apparent correlation of variables is reversed when the variables are combined. We investigate tennis' unique “best of N” scoring system for evidence of Simpson's Paradox among certain players who win more points than their opponent but lose the overall match. Using data from more than 61,000 professional tennis matches over the course of 21 years, we find about 4.5% of matches exhibit such characteristics, providing insight into incentive effects derived from tennis' scoring system, (sub-)conscious strategic decisions by players, and player-level traits that may be relevant to match outcomes. Practical implications in connection with sport integrity and narrow aspects of the sports gambling market are also discussed.