With the end of the NBA Finals series last night, the national leagues that I mainly cover are over. With that I wanted to post a special piece that pays homage as well as explains what I had been focusing on since February and the launch of Advance Pro Basketball. To us analytics guys, aside from the statistics, and myriad of other qualitative and quantitative evidence that allows us to delve deeper into how a team wins and how players contribute to the “W”s; win probability calculations, lead tracking and changes as well as elements that relate to how win probabilities shift from one end to the other have been a focus for me.
What is win probability?
Simply put, win probability is a tool that helps calculate a team’s chance of winning a game at any given time, based in historical performance of similar teams in the same instance or situation.
Originally developed by Bill James for baseball and the “moneyball” notion it later on spread to other sports including basketball.
The current research which I too take part in involves the accuracy measurement of win probability estimations. That is, if an independent tool say estimates a 20% win probability because 20% of teams previously won in that situation/instance, do future teams win at the same 20% ratio?
The hard part is estimating from hidden data that makes use of testing/simulation tools like cross-validation.
While most prediction models involve analyzing frequency of past events, other models use Bayesian processing.
Since February along with other anayltics experts have been looking closer at models including a measure of teams’ strength coming into the game, while others assume every team is average. Adding in the strength factor estimates increases the number of probable states, and hence decreases an estimation power while possibly increasing its accuracy.
Recent research and Basketball win probability
Before discussing in depth on how current research is going, I wanted to give a bit more insight on the scope of work that has been done in recent times.
Every year at Sloan Sports Analytics Conference, we see more and more research papers that cover a wide range of topics. In 2012 one research paper that delved into the matter gave lots of weight on win probability calculations, here is a video to the paper explanation during the conference:
Win Probability Added
Within the realm of win probability calculations we also tend to look at players contributions individually and it is at this point that attempting to measure a players probable contributions before they take place are tricky and while most coaches consider this as delivering over-use of data and making their job obsolete. However to the contrary as analytics experts in basketball we can only estimate and calculate as events take place and while live solutions currently do exist taking on the decision making process of acting as a coach is not what we aim to do.
I should be clear that win probability added calculations are not the same as win shares. While both might indicate to a players overall contribution to the team stakes in winning games the win share attribute of say 0 pointing to no contribution whereas with win probability 0 points to the average.
Current Resarch and best field application tools
With regards to the work I do while I would love to share proprietary information my client relations do not allow me to do so however several key peers that do win probability calculations have information up on their websites.
Michael Beuoy by far out of most of our peers has the best site as well as probability calculator which can be reached through:
Misconceptions about win probability
With regards to this section of my post, I want to be very open and clear that there are many elements that go into calculating and graphing win probabilities and outcomes that indicate results are not definite. The countability of away games to home games, star players vs. high producing bench players, lead changes during games and possessions as well as pace are all factors that contribute to win probabilities. Taking away any or all of the above or omitting data does create probability miscalculations in my perspective.
Where will win probability and basketball analytics lead us?
Given the scope of current research and the progress that we are heading in the aim is to be able to have win probability calculations more commonly used in Europe as well as in Turkey soon enough. I should point out that at present given that with held data by local basketball government bodies across Europe might hinder the progress as play by play data as well as lead change data are crucial, furthermore the fact that historical data as mentioned earlier are also very important to take into consideration.
To close off here is a graph and table depicting the final NBA game of the season between GSW – CAVS: