Joshua Cortez, a member of our Data Science Team, looks at using analytics to predict sports leagues outcomes.
You are half way through a particular sport season and you want to know where your team will finish. Here is a statistical way to determine that. This framework was inspired by a blog by Daniel Weitzenfeld and a paper from Baio and Blangiardo. Originally, they modeled football games and I have now created a generic framework that can be applied to any sports league. In this case I applied it to the current Philippine University basketball league.
Using Bayesian methods, I predict that DLSU is most likely to win the league and that UE and UP will finish 7th and 8th. Below is the heat map representing the probabilities of where each team will finish.
Scores of games from a season of Philippine collegiate basketball. Format is double round robin and half of the season’s schedule is finished.
Framework for Modelling:
Scored points depend on each team’s offensive and defensive strengths. A team’s offence increases their own total points, while defence decreases the total points of the opposing team. An additional factor is considered, home court advantage. But in the Philippine collegiate basketball setting, courts are neutral, so home court advantage has zero effect.
The mathematical model relies on total points scored per team - no need for other statistics such as steals, rebounds, blocks etc. We can estimate offensive and defensive strengths based only on points scored.
Visualisation of Framework:
Alternative visualisation of framework
After fitting a Bayesian model to the data, the underlying offence and defence attributes (median) of each team are uncovered. We can see that DLSU’s dominance in the season is due to its offensive dominance.
Teams ranked by offence. There is a 95% chance that the true value lies within each bar.
Teams ranked by defence. There is a 95% chance that the true value lies within each bar.
Using the uncovered offensive and defensive ratings for each team, we can simulate the second half of the season. The darker squares have higher probability.
It’s clear that DLSU has the highest chance of finishing the tournament in 1st place. UE and UP have the highest chances of finishing in either 7th or 8th place.
This generic model can be applied to any sports league to predict the likelihood of where you favorite team will finish.