13 min read

Why It's Called March Madness

My notebook and game recipe was formed off of the classic sports saying, “offense is the best defense”. I believe that the fundamental thought that constantly scoring will result in a win, will apply to March Madness. The predictors I will use are effective fg %, true shooting %, rebounds, only offensive ,three point %, free throw %, point per game true shooting % and points per possession. I added free throw % and points per possession to the recipe because I think it matters if teams can capitalize off stopping their opponent. Finally I added offensive and defensive efficiency due to team needing to stop their opponents while still scoring.

The biggest reason I believe my model did not predict correctly, other than a 15 seed to the elite 8. My model was predominately an offensive/shooting predictor and the teams that shot well all season failed to do so in the tournament.

The predictors
A mix of efficiencies, ratings and recency bias.
variable role
team_rolling_OE predictor
team_rolling_free_throw_factor predictor
team_rolling_orb predictor
team_rolling_true_shooting_percentage predictor
team_rolling_eFG_pct predictor
team_rolling_turnover_pct predictor
opponent_rolling_OE predictor
opponent_rolling_free_throw_factor predictor
opponent_rolling_turnover_pct predictor
opponent_rolling_orb predictor
opponent_rolling_true_shooting_percentage predictor
opponent_rolling_eFG_pct predictor
TeamResult outcome
By: Alex Kopf

As far as mess ups, it is difficult to narrow it down to one. I guess first I would like to see why my model loved Norfolk State so much, sending them to the elite 8.

East Regional: Round 1
I was sure Baylor would lose and Kentucky was Final Four bound.
Team Prediction Win Confidence Opponent
Saint Mary's W 91.8% Wyoming
UCLA W 87.6% Akron
Kentucky W 69.9% Saint Peter's
Texas L 49.3% Virginia Tech
Purdue L 43.4% Yale
North Carolina L 33.8% Marquette
Murray State W 64.3% San Francisco
Baylor L 18.2% Norfolk State
By: Alex

My model had Norfolk State as a lock against Baylor primarily due to three players on their team shooting 70% on the season. The same problem occurred with South Dakota St, regarded as one of the best shooting teams in the country. The Jackrabbits were shooting 44% from 3 point range all season, against Providence they only managed to shoot 7-23 or 30.4%.

MidWest Regional: Round 1
I was sure SDSU would wipe Providence
Team Prediction Team 3pt % Opp 3pt % Opponent
Iowa W 35.23548 33.56452 Richmond
LSU L 32.75161 32.76774 Iowa State
USC W 35.35806 34.76774 Miami
Kansas W 36.26774 33.38065 Texas A&M-CC
San Diego State W 36.65357 30.86000 Creighton
Auburn L 32.80000 38.01613 J'Ville St
Wisconsin L 31.38667 40.03438 Colgate
Providence L 35.23571 44.19412 S Dakota St
By: Alex

As the highlighted cell displays, SDSU shot significantly worse than they had the entire season. I think a majority of the teams that I predicted to win just failed to show up on the big stage and there is not much to predict regarding that. However, next time I would add more of a defensive outlook to counter the offensive heavy model.