Tweet
I usually try to avoid posting my own stuff on here, but it's the long dreaded off-season so I thought some content would be useful. I've put together a projection for next season based on WS and BPM from basketball-reference. It's similar to what I did last year with RPM but I think a little more thorough and using two different win approximations gives a nice range of likely records.
Obviously appreciate it if you click through (there's a lot of background and a little bit of proofing of the method in there) but I'll post the projections and a decent sample of the text here, enough I think to get what I'm going for here.
What do the Numbers Say? Projecting the Raptors' Win Total for the 2016-17 Season
2015-16 stats:
/cdn0.vox-cdn.com/uploads/chorus_asset/file/6849907/2015-16.0.jpg)
/cdn0.vox-cdn.com/uploads/chorus_asset/file/6850321/2016-17_predict.0.jpg)
Obviously appreciate it if you click through (there's a lot of background and a little bit of proofing of the method in there) but I'll post the projections and a decent sample of the text here, enough I think to get what I'm going for here.
What do the Numbers Say? Projecting the Raptors' Win Total for the 2016-17 Season
Basketball Reference captures an incredible range of basketball statistics, ranging from box scores to shooting percentages from different distances, to more advanced offerings of all sorts. But they are likely most widely known for their two popular catch-all statistics - win shares, a box score aggregator of sorts, that mostly measures production, and box plus minus (BPM), an attempt to measure a player's impact.
Win shares are easy to translate into wins - just add up the win shares for each player and the team total should be how many wins the team achieves. BPM can be translated into a wins over replacement player (WORP) value. Add up the WORP for each player on the team and you'll get how many wins better than a replacement team you'd expect. A replacement team, by my calculations (and based on the definition of "replacement player" by basketball-reference), is expected to earn about 15.5 wins.
Win shares are easy to translate into wins - just add up the win shares for each player and the team total should be how many wins the team achieves. BPM can be translated into a wins over replacement player (WORP) value. Add up the WORP for each player on the team and you'll get how many wins better than a replacement team you'd expect. A replacement team, by my calculations (and based on the definition of "replacement player" by basketball-reference), is expected to earn about 15.5 wins.
... taking the sum of the WS generated, we get a total of 54.5 wins, 1.5 wins below the actual record of 56-26 last season. Doing the same for WORP results in 41.0 WORP, or 56.5 wins. In this case WORP measured the season very accurately, within a half win of the actual total, while WS again showed roughly a 1.5 win error. For both seasons, an error of roughly 2 wins either way seems to be about the level of accuracy provided.
Here is the method I've come up with to remove myself (and my biases and assumptions) from the projections as much as possible.
1) Compile the current roster of players on the team.
2) From the previous season, take total minutes played, WS/48 and WORP/48 for each player.
3) Assume all players will play the same minutes they did the previous season.
4) Sum the total minutes played for the team. It should (but won't) be 19680 (48 minutes times 5 positions times 82 games).
5) Scale all players' minutes by the ratio of total team minutes to 19680. For example, if the sum from step 4 is only 17890 minutes played, all players will see their minutes from the prior season increase by 10% (since 17890 x 1.1 = 19680).
6) The new sum of adjusted player minutes will equal 19680.
7) Use the new minute assumptions for each player and multiply by their WS/48 and WORP/48 values to get total WS and WORP numbers.
8) Sum the WS and WORP numbers for the entire team to get the projected win total.
This process has no assumptions about roles, minutes, usage or fit. Its only assumption is that the best indicator of future role, minutes, etc, is the previous season. There are certainly cases where this is not likely. For example, perhaps Norman Powell or Lucas Nogueira will get a larger role this season than last. But perhaps not. Perhaps Lowry will see his minutes reduced. But perhaps not. In any case, this is the method I've chosen to remove myself from the process as much as possible.
1) Compile the current roster of players on the team.
2) From the previous season, take total minutes played, WS/48 and WORP/48 for each player.
3) Assume all players will play the same minutes they did the previous season.
4) Sum the total minutes played for the team. It should (but won't) be 19680 (48 minutes times 5 positions times 82 games).
5) Scale all players' minutes by the ratio of total team minutes to 19680. For example, if the sum from step 4 is only 17890 minutes played, all players will see their minutes from the prior season increase by 10% (since 17890 x 1.1 = 19680).
6) The new sum of adjusted player minutes will equal 19680.
7) Use the new minute assumptions for each player and multiply by their WS/48 and WORP/48 values to get total WS and WORP numbers.
8) Sum the WS and WORP numbers for the entire team to get the projected win total.
This process has no assumptions about roles, minutes, usage or fit. Its only assumption is that the best indicator of future role, minutes, etc, is the previous season. There are certainly cases where this is not likely. For example, perhaps Norman Powell or Lucas Nogueira will get a larger role this season than last. But perhaps not. Perhaps Lowry will see his minutes reduced. But perhaps not. In any case, this is the method I've chosen to remove myself from the process as much as possible.
Once again, like last year's projection, we have a minutes deficit, so once again players get a minutes boost. Right away we can say this is unlikely, so a little further down I'll allow myself to modify some stuff to avoid obvious problems like Lowry playing 3300 minutes (40 MPG).
Let's make those reasonable changes though. Lowry will almost certainly not play more minutes than he did last season, and nor will DeRozan. Meanwhile, we'd expect many more minutes for Carroll and Valanciunas due to lack of injury.
All told, the team total slides down to 54.7 wins (by WS) and 60.9 wins (by WORP). While having a minute distribution that is feasible.
These are still very nice projections. Considering that both WORP and WS have seemed limited to small errors (roughly two wins), both in the after-the-fact measurements and the sample projection we looked at, the simplest way to minimize the error for both measures is to settle nicely in the middle. Which means the final projected record for the team as constructed, with as few assumptions as possible on my part, is 58-24.
These are still very nice projections. Considering that both WORP and WS have seemed limited to small errors (roughly two wins), both in the after-the-fact measurements and the sample projection we looked at, the simplest way to minimize the error for both measures is to settle nicely in the middle. Which means the final projected record for the team as constructed, with as few assumptions as possible on my part, is 58-24.
Comment