Chemistry Experiments:
Part II -- Passing Ratings

You often hear reference to a team having good or bad 'chemistry' and while the implied tone has more to do with character/leadership and the like, we like to think of it in more real terms -- the group dynamics forged from the varying skills and qualities of the five players on the court a given time.

What would be the perfect 5-man unit?


Also see: Part I of the "Chemistry Experiments" where we examined Shot Blocking.

5-Man Unit Passing Ratings

The first step is again to define how we want to measure passing ability. For this we will use the 82games Passing Rating that can be found on the player in depth stats page in the passing detail section -- see for example Steve Nash's stats page where he currently has a 16.4 rating.

Admittedly the rating formula is currently somewhat of a placeholder until we refine it further, but the gist is to reward passes leading to successful shots from close range at a higher rate than assisted outside shots, while including the rate of passing turnovers. Here are some players to give a sense of the range on this rating:

18.9 - Andre Miller 15.0 - Brevin Knight 11.2 - Allen Iverson 9.1 - Joe Johnson 7.8 - Mike Bibby 6.0 - Tracy McGrady

4.1 - Tim Duncan 3.0 - Dirk Nowitzki 2.2 - Antawn Jamison 1.4 - Tyson Chandler 0.7 - Channing Frye -0.8 - Samuel Dalembert

So to calculate a given unit's block rating, we just add up the block % for each of the five players on the court at that moment. If you had a lineup of Iverson-J.Johnson-McGrady-Nowitzki-Frye, you would score it as a 30.0 passing rating unit (11.2 + 9.1 + 6.0 + 3.0 + 0.7). Or to use a more realistic example, the Portland Trailblazers' two most used lineups are:

Blake-Dixon-Khryapa-Randolph-Przybilla -- good for a 23.5 unit passing rating
Telfair-Miles-Monia-Randolph-Przybilla -- with a 17.0 unit passing rating

We elected to use six groupings of units:

- 0 to 16.9 passing 'points'
- 17 to 19.4
- 19.5 to 21.9
- 22 to 24.4
- 24.5 to 26.9
- 27+

This represents a reasonably even distribution of the total units played in the NBA this season. Since we want to examine only meaningful minutes, we have discarded "garbage time" activity from the results.

NBA 2005-06 Season to 2/21
("Garbage time" excluded)

	Points per 100 Possessions			Offensive Stats
Unit Pass Rtg	Offense	Defense	NET	FG%	Blk%	Close Shot	Quick Shot	FTA/ FGA	Off. Reb%	T/O
0 to 16.9	104.1	104.7	-0.6	.440	6.3	33.7%	36.3%	.28	30.6%	17.1%
17 to 19.4	105.4	106.0	-0.6	.446	6.1	34.1%	36.8%	.29	29.1%	16.5%
19.5 to 21.9	107.2	106.6	+0.6	.451	5.8	34.0%	37.1%	.28	28.5%	15.6%
22 to 24.4	108.2	107.7	+0.5	.456	5.9	33.9%	36.1%	.28	29.5%	15.4%
24.5 to 26.9	107.5	107.5	+0.0	.453	5.8	35.6%	37.7%	.28	29.4%	15.8%
27+	108.0	108.1	-0.1	.453	6.0	35.5%	38.7%	.30	31.5%	15.0%

There are some interesting breaks on this table -- units with combined passer ratings of less than 19.5 have been significantly poorer in offensive efficiency and have been getting outscored by 0.6 points per 100 possessions. On the other hand the two middle groups of passer ratings (19.5 to 24.4) have been been winning combos in net points. Then the two highest levels of combined 5-man unit passer ratings have shown breakeven results, without any clear cut improvement in offensive efficiency over the middle groups.

Other trends evident at a glance include:
- units with poor passing ratings commit more turnovers
- units with the highest passing ratings take more shots from close range (but perhaps sacrifice some optimal efficiency in so doing)
- offensive rebounding effectiveness doesn't seem to correlate with the unit passing ratings

Since most teams feature a primary ball handler on offense though, it makes sense to look at the highest individual passer rating for a unit...

Highest Passing Rating player of the unit

So this time we'll group the units by the rating of the best passer of the five, rather than the combined score of all the unit. We'll also narrow it to five groupings:

- 0 to 6.9 passing 'points'
- 7.0 to 8.4
- 8.5 to 9.9
- 10.0 to 11.4
- 11.5+

	Points per 100 Possessions			Offensive Stats
Highest Pass Rtg Player	Offense	Defense	NET	FG%	Blk%	Close Shot	Quick Shot	FTA/ FGA	Off. Reb%	T/O
0 to 6.9	104.7	105.1	-0.4	.440	6.8	33.5%	35.6%	.29	26.4%	16.8%
7 to 8.4	106.3	107.3	-1.0	.451	5.8	34.4%	37.5%	.28	29.7%	15.9%
8.5 to 9.9	106.0	107.3	-1.3	.444	6.1	33.8%	36.5%	.28	31.8%	15.7%
10 to 11.4	109.4	107.4	+2.0	.461	5.6	35.7%	36.7%	.30	31.6%	15.8%
11.5+	107.0	106.2	+0.8	.451	5.8	34.4%	38.8%	.27	28.8%	15.4%

The highest player looks are somewhat troublesome since the extremes can be too swayed by one player. For instance in the 11.5+ category you get such heavyweights as Nash and Billups whose teams are doing fabulously well. On the other hand, the inclusion of a Brevin Knight has a dramatic effect on the "Net Points" number since he at this moment carries the 8th worst raw plus/minus total in the league (albeit with a slightly positive 'Roland Rating')!

Still, the table has a pretty strong division between the units with good leading passers and those without such high quality floor leaders -- the under 10.0 top rated player groups are all in the red, while the 10+ units show considerably better offensive efficiency, positive net points, and surprisingly no real drop off in defense excepting the lowest passer level group.

Does this make the top echelon passers that much more valuable? Or are we just wandering into the minefield of using a biased data set [the passing ratings are determined from the same data that we then query on, rather than say using last year's player ratings to this season results or an in-season A/B sampling]?

If you believe in the former, then here are the players of note:

10+ -- Miller (18.9), Arroyo (17.6), B.Davis (16.6), Nash (16.4), Knight (15.0), Paul (14.8), Billups (13.5), Kidd (12.9), Wade (11.4), Iverson (11.2), Francis (11.0), Ridnour (10.9), Blake (10.8), J.Williams (10.8), Livingston (10.3), A.Johnson (10.3), Marbury (10.3), Watson (10.3), Claxton (10.0), Diaw (10.0)

Almost -- Calderon (9.7), L.James (9.7), Nelson (9.7), Cassell (9.5)

What happens you may wonder if a team has an excellent second passer in a unit, or perhaps no one beyond the main ball handler who can "deliver the dish"?

Second Highest Passing Rating player of the unit

So now it's the second highest passer in a unit that we're tracking, using this grouping scheme:

- 0 to 3.9 passing 'points'
- 4 to 4.9
- 5 to 5.9
- 6 to 6.9
- 7+

	Points per 100 Possessions			Offensive Stats
2nd Highest Pass Player	Offense	Defense	NET	FG%	Blk%	Close Shot	Quick Shot	FTA/ FGA	Off. Reb%	T/O
0 to 3.9	105.1	106.3	-1.2	.446	6.0	34.1%	36.7%	.28	31.0%	16.9%
4 to 4.9	106.7	105.7	+1.0	.452	6.0	33.8%	37.2%	.27	29.9%	15.9%
5 to 5.9	105.9	106.2	-0.3	.449	6.5	34.8%	37.0%	.27	28.7%	15.8%
6 to 6.9	108.0	107.7	+0.3	.452	5.8	35.0%	36.7%	.29	29.1%	15.2%
7+	107.9	107.6	+0.3	.451	5.6	34.4%	37.9%	.30	29.7%	15.4%

This presents more of a mixed read, but the two highest level groups for second best passing of the unit do show a positive net and significantly stronger offensive efficiency. It's not however the FG% that changes so much as reduced turnovers and increased free throws.

Turning then to the question sent in by a hefty percentage of the readers of Part I of this "chemistry analysis" series -- what the heck does it all mean! -- it's again a matter of acknowledging that by isolating such a small component of the game any conclusions we try to draw from the numbers are problematic. What we think we see may be more a reflection of some other causative factor, and we're dealing again with limited sample sizes, arbitrary groupings, inherent bias by reusing data, etc.

Nevertheless, it appears that an exceptional passer in a five man unit (as judged by the "placeholder" 82games current passing rating) is a very good thing, although surprisingly having a combined unit passing rating that is far above average does not seem optimal compared to one that is average/slightly above average.

Could a Marbury-Francis backcourt work (as has been a hot trade rumor of late)? Well, there haven't been many teams trying such an approach this season, and the slim evidence suggests it's unlikely to succeed, but from a research perspective we would sure like to see it happen!