Friends at a company play pick-up basketball during their lunch hour. Since there are no established teams, players can be randomly matched up. But these folks happen to be engineers/data-nerds, so they keep detailed track of games, teams, players and win/loss. The diagram view of the data is shown further below.
Question: How can we determine which player pairing is the most successful?
Since players are randomly teamed up, are there combinations which when teamed up have an unusually high winning percentage?
Application: This would naturally extend to other sports, but I believe may also apply in many non-sports scenarios, where items are paired up somewhat randomly (or by design) and we want to know how effective those pairings are.
Thanks to Kirill Perian (basketball nickname K-Real), an attendee of one of our past webinars, who sent us the dataset and posed this question. Dataset has been simplified to showcase this scenario and anonymized to protect the identity of the losers 🙂
File can be downloaded here.
First we will address the easier scenario of creating metrics for individual players, using the many-to-many pattern. Next we will take on writing measures to compare performances of pairs of players.