Rebounding before 1974 & inflation-adjusted stats (Patreon)

There's a legendary poster on realgm (who has worked for NBA teams) with the handle TrueLAFan -- as in, a Clippers fan, not a Lakers fan -- and many years ago he put together a guide on estimating rebounding percentage (TRB%) before 1971, when the league started officially tracking opponent rebounds. 

https://forums.realgm.com/boards/viewtopic.php?f=344&t=955514

I had done some work converting Dennis Rodman's rebounds to 1960s figures way back in 2005, so the most I ever did with his post was applaud, then plug in Russell and Wilt and compare them to modern players like Rodman. 

Until now. :)

The missing variable for knowing the percentage of total boards a player grabbed before 1971 is simply opponent rebounds, and he has a quick but accurate method for estimating them. Given all the work I've done on pre-merger stats, I actually have a few more tools to improve on this and generate offense and defensive rebounding rates. 

I'll outline the method below -- and how confident we should be in it -- because I know some of you are interested in the minutiae. If you're not, you can jump to them in the CLASSIC section of the historical sheet labeled "rDRB" and "rORB." These are each player's offense/defense rebounding percentages relative to the league average in that season. (Strategy has changed offensive rebounding tactics wildly over the years.)

Adjusting stats for Inflation

My latest YT video is about adjusting stats for inflation. The method discussed in the video takes yearly averages and shifts raw stats based on them. It's a nice extra layer IMO to have environmental context for some of these numbers -- personally, I'm most interested in scoring environments -- so I'll add a few numbers to the insider database sheet in a yearly-adjusted section on the far right:

1. You'll see a player's per 75 scoring, adjusted to 2019 numbers

2. You'll see the per game numbers converted to 2019 for every player

Calculating old rebounding rates

OK, here's the method for figuring these things out and how confident we can be in these numbers. 

First, I wanted to use my expanded toolbox to see what I could do with opponent rebounding estimates. Turns out, using estimates of pace and offensive rating generates a pretty darn accurate estimate -- 80% of teams I tested it on in the 1970s were within 2.4 rebounds per game of their actual opponent rebounding number. From there you use the formula to calculate TRB%. 

How accurate are the TRB% estimates you'll see in the sheet? So accurate it feels weird calling them estimates. (Technically, Oliver's formula is an estimate too.)

-99.4% of players I tested in the 1970s were within 1.5% of their actual TRB%.

-98.2% were within 1%. 

-89% were within 0.5%. 

That means for ~9 in 10 players, the number you see will be +/- 0.5% from their actual number. I'm also not particular concerned about degradation in accuracy here going back in time. Even back in the '50s, were I think the estimates might be slightly less accurate (1958 to 1960 saw some radical changes), they should still be extremely close to the actual TRB% of a player because of how simple this calculation is.

You might think, "awesome, but there's no way we would know which of those boards were grabbed on offense and defense, could we?"

Well, we can do that too. 

Starting with some complex stuff I did for estimating pace for basketball-reference, I created a fairly accurate estimate of total offensive and defensive rebounds based on a player's statistical footprint (eg height, position, free throw rate, etc.). 

From there, I still needed to calculate how many offensive and defensive rebounds were available in a game. For this step, using a team's pace, total rebounds and shooting accuracy created a strong estimate of a team's DRB/ORB split. I'm guessing that shooting accuracy serves as some proxy for estimating the number of available offensive rebounds for that team.

Then, simply take the estimated total offensive/defensive boards for a player and plug it into the Oliver formula, with those team estimates standing in for available offensive/defensive rebounds, and we get nice individual estimates for ORB% and DRB%. DRB% is more accurate, but I think some of that is just the much small sample (and higher variance) around ORB numbers. How accurate are these numbers?

For DRB%:

-99% of players were within 3% of their actual DRB% on players tested in the 1970s

-97% were within 2%

-88% were within 1%

For ORB%:

-97% were within 3%

-86% were within 2%

-59% were within 1%

And that's it. Have a great week!