After last night’s epic upset, I decided it would be interesting to take a look at the last month’s comeback victories. I’m defining a comeback as being down any number of points at halftime and coming back to win the game. Of 585 NCAA games that I have data for since the end of February, 155 of those games resulted in a comeback win, or 26.5%.
Here are some interesting statistics on those comeback victories. Note these data only include games from late February until the beginning of March Madness.
I wanted to also understand which aspects of a team’s performance were mostly correlated with mounting a comeback. Besides points, what other statistics matter the most in winning a game when down at the half? Using a feature selection method in the caret package, I ranked all of the variables both at the half and end of the game (besides straight up points and FGM). The input variables are just the differences of the stats for each team e.g. HALF_REB.TEAM1 – HALF_REB.TEAM2. Surprisingly, being at home mattered the least in coming back to win a game. Final and half defensive rebounds, final free throws made and final assists were at the top of the list, while blocks and steals were at the bottom.
I’m sure a lot of brackets are busted at this point, but I’m sure Wisconsin is thinking: