I'd like to model Tennis results in a logit regression. How can I include head-to-head data?
Simply taking the share of won matches between two players isn't ideal, because it doesn't take the number of matches into account. E.g., a H2H of 20-1 would be worse than 1-0.