Posted: Wed Sep 19, 2007 11:09 am Post subject: Offensive and Defensive PER
At various times people have asked for a breakdown of PER into its offensive and defensive components. I'm not the biggest fan of PER, but I definitely see how it could be more useful if it wasn't just an all-in-one "lots of things that capture offensive contributions pretty well plus a few things that capture a small part of defensive contributions" statistic. Since I already had a spreadsheet that calculated PER, it wasn't that hard to create one that split it into OPER and DPER. I've posted it on the Yahoo group and it covers the last five seasons (it's called "offensive and defensive PER.xls):
OPER is the offensive components of PER - 3PM, Ast, FGM, FTM, TO, missed FGA, missed FTA, and ORB. DPER is the DRB, Stl, Blk, and PF parts. The breakdown isn't perfect since it treats all fouls as defensive fouls.
A difficulty arose in deciding how to standardize OPER and DPER. I wanted to be able to fairly compare ratings from different seasons (which would mean separately standardizing both OPER and DPER), but at the same time I thought it would be interesting to see how much of a player's single-season PER came from his offensive stats and how much from his defensive stats (which would mean just standardizing total PER and then splitting it up). I think both ways of looking at it are useful, so I included both in the spreadsheet. The second method yielded OPER and DPER, which always add up to a player's PER (I just took the ratio of non-standardized OPER to non-standardized DPER and used it to break up PER). In the seasons I looked at the average breakdown of OPER to DPER was around 12 to 3 (totalling the average PER of 15). However, because the league average ratio of OPER to DPER changes every year, comparing OPER's or DPER's from different seasons can be misleading when there's no standardizing (it wasn't a big deal in the five seasons I looked at, but if one were to go back a few decades there might be significant effects). So I also listed OPER15 and DPER15, which are standardized around 15 just like PER (i.e. each season the average OPER15 and DPER15 are set at 15). I could have used a different number (and there's no reason both OPER and DPER had to be standardized around the same number), but 15 seemed to make some sense given people have some familiarity with it representing the average in PER. You'll notice that the greater variance in DPER's leads to DPER15's that are much greater than anyone's PER or OPER15.
Enough description - if that didn't make sense maybe some examples will clear things up. Here are the best OPER15 and DPER15 seasons from the last five years.
Code:
Year Player OPER15 OPER/DPER
----- ------- ------ ----------
02-03 McGrady 32.47 26.38/3.89
02-03 O'Neal 30.87 25.07/4.42
05-06 Bryant 30.01 24.80/3.20
06-07 Wade 29.13 24.15/4.80
05-06 James 29.04 24.01/4.06
Here are the average OPER's and DPER's by season (they always sum to 15) - you can see that the ratio hasn't changed that much, so standardizing isn't vital when comparing players from nearby seasons:
After I did all this I remembered that Hollinger had done something like this before. Looking back, I was able to find his description of Offensive PER on page 12 of his first book (Prospectus 2002), as well as in his comments on Steve Nash in Forecast 05-06 (page 229), and in one Insider article ( http://insider.espn.go.com/nba/playoffs2005/columns/story?columnist=hollinger_john&id=2041492 ). In describing it in his first book, he says, "simply break down the PER by eliminating the defensive categories (blocks, steals, personal fouls, and defensive rebounds) and it leaves the league leaders in offensive PER." However, the numbers he lists for offensive PER in all three sources differ from mine because he also leaves out offensive rebounds. I'm not sure whether that was intentional or a mistake, but for my version offensive rebounds count toward OPER and defensive rebounds count toward DPER. It also looks like he chose not to standardize separately and thus arrived at an offensive PER that when summed with defensive PER will equal PER (though he'd have to include offensive rebounds in his defensive PER, or have three breakdowns into offensive PER, rebounding PER, and defensive PER). That means the figures he listed don't allow for unbiased between-season comparisons like my OPER15 and DPER15 do.
This is all a work in progress so I'm open to any comments and suggestions, especially when it comes to dealing with the standardizing issues. Another way of presenting it that I considered was instead of having an OPER and DPER that sum to PER, just have a representation of a player's ratio of offensive to defensive parts of PER. So Camby's 05-06 season could be represented as a DPER15 of 51.24 with a 55%/45% O/D split (compared to the league average of 83%/17%), meaning 45% of his PER total came from his defensive contributions. The ratio could also be represented as 1.2-to-1 offense-to-defense, compared to the league average of 4.8-to-1. I also thought about a different way of representing standardized OPER's and DPER's. Instead of centering both around 15, I could have OPER12 and DPER3, which would allow for between-season comparisons while also retaining a sense of how much each contributed to a player's PER (since each is standardized around a figure near the league average OPER and DPER broken down from PER). But that would be just a rough estimate since 12 and 3 are approximations and the real numbers change every year.
For those interested in seeing how the calculations were made, I left the formulas in all the cells of row 2 of each sheet (02-03 Abdul-Wahad and 02-03 Atlanta), and all the necessary columns are still there even though many are hidden (between Y & AL, AN & AU, and AU & AX). I deleted the formulas and just left the values in all the other cells to cut down the file size, and I hid a lot of the columns so that those who just want to see the end results don't have to wade through all the clutter. _________________ Eli W. (formerly John Quincy)
CountTheBasket.com
Thanks very much. This should be interesting to check out. OPER15 and DPER15 are excellent followthough contributions that I have wanted to look at. Your database will also allow the by position sorts and potentially position adjusted OPER15s and DPER15s.
I thought the offense representation might be 3-1 over defense, but 4-1 exceeded that expectation and solidly demonstrates that roll-up PER (lacking any shot defense plug) is mostly an offensive metric.
Have any interest in calculating team PER and offensive and defensive splits and comparing the results to the efficiencies and point differential? I might try further on that later with what you've kindly provided.
That could be an additional cut. Raw player 4 factor data is another approach.
Looks like big man offensive PER a bit more than 10% less than perimeter but defensive PER is a bit more than 20% higher.
Players could be grouped in to groups that conform with positional bias on which side of PER is higher or not. Lineups could be analyzed in this way too.
When playing Fastbreak Basketball I used to split my PER calculations into "on the ball" (shooting, assists, TO) and "off the ball" (rebounds, steals, blocks, fouls)... I always meant to do this with real basketball and see if the numbers were meaningful but never got around to it. _________________ No books - no articles - no website.
Just opinions.
You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum