March Mathness 2024 - The picks
Last year, we showed off our psychometrics team by using their model to pick a bracket (March 17, 2023) for the NCAA Women's Basketball Tournament. The Rasch model (the same model used for ARRT exams) scored more points than both Brandon's top-seed strategy and Zachary's mascot-based approach (April 28. 2023). We heard a lot of positive feedback on the contest, so March Mathness is back again with a new challenger.
The Rasch model treats the basketball season like an exam. Teams (candidates) play a series of games (answer items), and the number of wins (correct answers) corresponds to their ability. Since not every team plays the same set of opponents (takes the same exam), one schedule (exam form) might be more difficult than another. Thankfully, we can use common opponents to link the schedules together and put them all onto a single difficulty scale (just like an item bank). Once that's done, we calculate each team's ability and pick the higher score to win each matchup.
This year's challenger is Classical Test Theory. It's an older method, but it checks out. We start with the win rate (percent correct) for each team (candidate), and then adjust our estimate based on the win rate of the opponents (items) they faced. Teams earn the most credit by winning against opponents that also win a lot. After all, you must have a strong team if you win where nobody else can. Once again, we select the higher adjusted ability to win each tournament matchup.
Check out our picks below and look for a follow-up post after the tournament to see how well we did. Admittedly, these methods were never designed to pick basketball winners, so we will also discuss some difficulties we encountered while setting up this demonstration.
