brooks
Thunder
- Messages
- 3,849
Well hello there. Welcome to my new attempt at using statistical indicators to predict the quality and ability of each of your teams. In the past, I had generated my own composite ranking that was, well, not all that great. It was heavily biased and relied on a lot of tweaking on my end in order to provide representative rankings. Not to shit on it too much, as it was fairly useful in identifying the elite/great/good/crappy/atrocious teams in the league.
This time around is different. All of the results shared below are calculated by the means of a Partial Least Squares (PLS) regression. I'm not going to get into the specifics of how a PLS regression works, as that would be time consuming and really... not worth it. Please refer to the screenshot from our helpful friend Wikipedia.
Ultimately, a PLS model is great at identifying linear regressions in data sets that contain a significant number of variables. As we all know, FBB produces and shares A LOT of team statistics that can be used in this manner.
In order to generate a PLS model, you need to have calibration data (or a training set) to leverage. Thankfully for us, we have plenty of data that we can utilize in that manner. What I've done is used all of the finalized team statistics and wins/losses from the 2001-2005 seasons for this training set. You can find a screen shot of the prediction plot below... this is what demonstrates our effective calibration curve, if you may.
As you can see, we get a fairly linear response using this data! In fact, its actually quite good at predicting the total wins for a team by simply using their team statistics. There are other statistical details we could get into, such as outliers and overall leverage on the model, but again, that's beyond the scope of this post. Maybe in the future.. who knows.
Aaaaaaand here's Sim 1:
*Please note that the "Measured Wins" was calculated by extrapolating each team's win percentage to a full 82 game season.
Key Takeaways:
This time around is different. All of the results shared below are calculated by the means of a Partial Least Squares (PLS) regression. I'm not going to get into the specifics of how a PLS regression works, as that would be time consuming and really... not worth it. Please refer to the screenshot from our helpful friend Wikipedia.
Ultimately, a PLS model is great at identifying linear regressions in data sets that contain a significant number of variables. As we all know, FBB produces and shares A LOT of team statistics that can be used in this manner.
In order to generate a PLS model, you need to have calibration data (or a training set) to leverage. Thankfully for us, we have plenty of data that we can utilize in that manner. What I've done is used all of the finalized team statistics and wins/losses from the 2001-2005 seasons for this training set. You can find a screen shot of the prediction plot below... this is what demonstrates our effective calibration curve, if you may.
As you can see, we get a fairly linear response using this data! In fact, its actually quite good at predicting the total wins for a team by simply using their team statistics. There are other statistical details we could get into, such as outliers and overall leverage on the model, but again, that's beyond the scope of this post. Maybe in the future.. who knows.
Aaaaaaand here's Sim 1:
*Please note that the "Measured Wins" was calculated by extrapolating each team's win percentage to a full 82 game season.
Key Takeaways:
- Sonics are damn good. Statistically they are the best of the best. I'll be curious to see if any of the Heat, Bullets, Timberwolves, or Cavs catch up to them.
- Lakers are completely a red herring. Don't trust them. All stats indicate that they'll be fighting for a playoff seed come Sim 6.
- Pacers and Spurs are doing their jobs effectively. The predictions insinuate that the Spurs are in a class of their own for being terrible, but we'll have to see as we move forward!
