BPM stat sensitivities

[Crow]

Daniel, I was wondering if you would help me with some estimates of BPM stat sensivities.

How much on average would BPM move if efg% moved .05 on typical shot volume? Or 3pta fg% by .01? Or 1 assist or 1 rebound per 100 possessions?

I have some thoughts in development that would be improved by decent estimates on these things.

[DSMok1]

Hello, Crow,

Here is a spreadsheet that has the calculations. You can save a copy and play with the numbers.

https://docs.google.com/spreadsheets/d/ ... =307166562

Please let me know if that works for you!

[Crow]

Thank you very much.

So use the BPM coefficients? Can you confirm the right way to do that with "adj pts"?


Also, I spy 4 factors (I think) of BPM with the pre 74 historical players... (or are 3 of them net factors?) Is there an explanation for why that isn't part of the BRef display? Did / do they not want to provide the real estate for it or is there more to it than that? Are 4 factors of Defensive BPM possible to break out, available anywhere in this file or elsewhere? Weakening the "black box" dismissal of BPM (or any all in one roll-up metric) has long been an interest of mine.

[Crow]

Altering a player's line in a copy of the file to get a recalculated BPM is cool and should be very helpful.

Would a similar file with NBA 20-21 data be available to copy? How about 20-21 for all NCAA or just top draft prospects?

[DSMok1]

Thank you.

So use the BPM coefficients? Can you confirm the right way to do that with "adj pts"?

Or can I alter a player's line and get a recalculated BPM?

Also, I spy 4 factors of BPM with historical players. Is there an explanation for why that isn't part of the BRef display? Did / do they not want to provide the real estate for it or is there more to it than that?

--

I now know I can use / alter a copy file player's line and get a recalculated BPM. That's helpful. Would a file with 20-21 data be available to copy?

Alter the player's stat line to get a recalculated BPM.

I do not have data assembled for 20-21. It would take a bit of work to update that spreadsheet, because of the several tables needed for the team. I originally had that set up to automatically scrape BBRef for a given team, but then Basketball reference changed their HTML and Google Sheets can no longer access much of their pages.

[Crow]

I caught on about using a copy of the file after a bit. Jumped in before digesting your post fully.

Too bad about a 20-21 file not being immediately available but I will work with what you generously provided. If it arises in future, I'd remain interested in opportunity to use.

Thanks again.

[Crow]

Do you have formal estimates based on historical data or knowledgeable guesstimates about the college to NBA league level transition in general or for last year of college to first year NBA?

Which of these is true: NCAA BPM estimates are completely based on an NBA model, based on a mostly NBA model with some modifications or is similar to NBA model but built specifically for NCAA?

In comparing a specific player's NCAA BPM to future NBA BPM estimates, is it basically talent in NCAA / specific team and role context - league quality adjustment and then +/- specific NBA team & role adjustment and whatever "development" & personal style change?

[Crow]

I recently saw someone say 1% more usage was worth x points but I'll have to find it again as I don't remember what x was, where said and if it was based on recent research or old stuff.

Do you have such a rule of thumb based on BPM and is it fairly consistent with other such estimates or notably different?

[DSMok1]

Do you have formal estimates based on historical data or knowledgeable guesstimates about the college to NBA league level transition in general or for last year of college to first year NBA?

Which of these is true: NCAA BPM estimates are completely based on an NBA model, based on a mostly NBA model with some modifications or is similar to NBA model but built specifically for NCAA?

In comparing a specific player's NCAA BPM to future NBA BPM estimates, is it basically talent in NCAA / specific team and role context - league quality adjustment and then +/- specific NBA team & role adjustment and whatever "development" & personal style change?

NCAA BPM estimates use the NBA coefficients. There is, in my opinion, no way to construct a stable enough RAPM basis with the constraints of the NCAA dataset to provide meaningful insight into any differences in coefficients from the NBA. NCAA stats do have a lot more "outlier" data points, so BPM 2.0, which is linear, is much better at coping with those types of players than BPM 1.0.

I have only briefly looked at NCAA BPM to NBA BPM translations. The other huge factor is age/class. As you know, being good when young is a very good indicator of NBA success.

There are certainly other factors. It seems easier for big men to dominate in college, so they are somewhat overvalued when looking at a straight transition to the NBA. That is particularly true of big men that play against weaker opponents.

Here are the BPM stats over the past 12 years or so: https://barttorvik.com/playerstat.php?l ... nd=all0501

Here's a quick adjustment to try to rough out some numbers:
Freshmen and Sophomores only
Subtract 2 from Sophomore BPM
Subtract 2 from 7 footer BPM vs. 6 footer BPM
Subtract 10 from all players to make scale look better.

https://docs.google.com/spreadsheets/d/ ... ingle=true

[Crow]

Thanks for the new file. May work with it harder later to try to get at nuance.

I started with a simple college to 1 yr NBA adjustment of -11 BPM (-5 on offense, -6 on defense), so we were similar.

For many players I assume a roughly +2 improvement to peak BPM.

To hold off part of the on average major decline of NCAA BPM to NBA BPM, players need to retain usage, hold onto ts% and not lose assists or rebounds from role adjustment. Mostly on offense. Haven't thought that much about defense yet. But most of the decline does appear inevitable and based on league quality level.

The conventional way of thinking is player development / improvement but retention of early relative advantage / lesser decline than average from league change is another perspective using these BPM numbers and thinking about the factors that go into them. There is improvement potential but decline thru league change is the bigger movement and perhaps deserves highlighting / more attention.

What is easier or harder to retain? Need to think on it. May not be simple, general answers.

[DSMok1]

I recently saw someone say 1% more usage was worth x points but I'll have to find it again as I don't remember what x was, where said and if it was based on recent research or old stuff.

Do you have such a rule of thumb based on BPM and is it fairly consistent with other such estimates or notably different?

You can work this out from the coefficients, but "usage" is not directly used. It's part of the "offensive role" position assessment. Note that the penalty for FGA is different (lower) for Creators than Receivers. The difference is 0.22 points per FGA between max creator and max receiver.

[Crow]

Can the "league adjustment" on offense be equated mostly to impact of usage change, ts% change and assist changes? Aren't these the main things it has to be mediated thru?

[DSMok1]

Can the "league adjustment" on offense be equated mostly to impact of usage change, ts% change and assist changes? Aren't these the main things it has to be mediated thru?

I am not sure what you are referring to. There is no league adjustment in BPM.

TS% context is adjusted for at a team level. Shooting is measured relative to the rest of the team.

At the end, there is the overall team adjustment, which simply slides everyone on the team up or down an equal amount so that they sum to the team's rating.

[Crow]

"League adjustment"... implicit in NCAA to NBA BPM estimate transition is what I meant.

Face players who form an offense who collectively on average are about 5-6 pts / 100p better in NBA than in NCAA and who form a defense which is 5-6 pts better on average.

[Crow]

If a lot of a player's pre-NBA value is based on super-sized usage, ball control and / or grab rebound & go after others box out, they will have to hope or force those things to continue. Retain or quickly restore. Politics. "Play my game."