Shot-Adjusted Marginal Defense

[EvanZ]

I've been working on some scoring and defensive metrics the last few weeks. I've got the team-level shot-defense metric worked out now. The basic idea is that I categorize shots into 4 categories: INSIDE, MID-RANGE, 3-PT, FT. The first category includes dunks, layups, tips, and hooks. Mid-range includes all jump shots, except for 3PTA. The other two are self-explanatory. The formulas used for each shot type take into account the league average shot rate and (opponent) efficiency for each type, in addition to the team rate and efficiency.

The spreadsheet formulas are as follows, if anyone is interested in the nitty gritty details (all data is per 100 possessions):

Code: Select all

INSIDE=2*(IF(INSATT<28.5,INSATT,28.5)*(0.628-INSEFF)+IF(INSATT>28.5,(0.488-INSEFF),(0.488-0.628))*(INSATT-28.5))

MID-RANGE=2*IF(MIDATT<40,MIDATT*(0.386-MIDEFF),40*(0.386-MIDEFF))+2*IF(MIDATT>40,(MIDATT-40)*(0.624-MIDEFF),(40-MIDATT)*(0.386-0.624))

3-PT=2*(IF(FG3ATT<19.53,FG3ATT,19.53)*(0.54-FG3EFF)+IF(FG3ATT>19.53,(0.527-FG3EFF),(0.54-0.527))*(FG3ATT-19.53))

FT=2*(FTPOSS-11.62)*(0.498-0.76)

The "hard coded" numbers are league averages. Here is a chart showing the final results. Positive numbers are better.



What I like about this is that you can see in a nutshell the strengths of weaknesses of a team defense (well, everything except for the ability to cause turnovers, which is obviously important). I will do the same thing for offense when I get a chance.

[Crow]

That's a nice chart.

Looks like inside and mid-range defense by far provide more variance from average than 3 pt or FT defense. 3 pt defense is sort of in between but it is probably about 50% less likely to have a large impact as inside and mid-range defense. Looks like mid-range defense was a bit more likely to have a bigger positive impact (as opposed to bigger negative) than inside shot defense.



Defensive rebounding isn't included either (you sometimes treat it as a separate category).

[EvanZ]

Defensive rebounding isn't included either (you sometimes treat it as a separate category).

Right. Good point.

[Crow]

Of the top 8 playoff teams, the Lakers might possibly be considered to have been strong in all 4 defensive zones. Chicago and Boston were the only other ones which had 3 areas of significant strength. Dallas and Miami 2. OKC and Atlanta had just one while Memphis had none. This is by eyeball estimation of what seems like a sufficient length. A table of the team numbers might be a handy addition and allow more clear cut identification of size of strengths.

Miami's defensive strength was heavily based on interior defense but Dallas was a team that was 3rd lowest on shots at the rim in the regular season. Not a good match-up for Miami's defense.

Miami was 4th lowest on regular season shots at the rim. so Dallas' lack of strength there didn't hurt them. Miami was 3rd best FG% from 16-23 feet but 5th highest on % of shots from there and they are basically tied for being the worst shots in the game with the other mid-range distances. Dallas had modest strength defending the mid-range shot. Again a favorable match-up for them.

A playoff version of this data would be helpful. Series by series would be as well for understanding what happened.

On paper it looks like Dallas' defense was a good match-up for the OKC offense- no real strength inside. Likewise the OKC defense had some mid-range strength and wasn't likely to be hurt by Dallas' inside game. But it didn't work out. Would need to see more detail with respect to shot locations and efficiencies to piece together that story accurately, beyond what the summary stats explain.

Shot defense by shot type / distance should definitely be a regular tool used by teams and outside team analysts and not just a one-time demonstration. Thanks for helping highlighting / further opening up this path. It would add to the detail already available at the 4 Factor level by breaking the by far biggest piece out into 4 parts. Detroit was the only team -2 or worse from 2 distances (inside and mid-range).

[Crow]

I now see the table at your site. Thanks.

It looks like what I eyeballed as significant strength equates to about 1 point per 100 possessions as that standard yields the same analysis of the top 8 playoff teams.

Only the Lakers and Spurs got near a 1 point edge from "FT defense". No one else was over 0.5 though Dallas was close and in 4th place.

6 teams got 1+ point edge from 3 point defense, lead by Chicago and the Lakers. 11 teams got 1+ point edge from mid-range defense (or facilitation / encouragement), lead by Chicago, Orlando, Boston and San Antonio. 8 got got 1+ point edge from 3 point defense lead by Miami, Chicago and Orlando.

9 teams had 2 or more 1+ strengths (I included Dallas with a .99 strength). Orlando was the only team with fully two 2+ strengths. Boston and Chicago were close to that. Chicago's mid-range defense had the best mark of any distance at +3.88, Miami's interior defense second best at +3.38. There were 27 1+ strengths. That would be a simple average of almost one per team if evenly distributed, but actually 9 teams averaged 2.2 and the rest of teams averaged less than 0.4. No team with 2 1+ strengths had a -1 or worse weakness. All of the teams with 2 or more 1+ strengths (except the near case of Dallas) had 1+ strengths inside and mid-range. That apparently is the base rather than a variety of location matches. Pack it in. 3 of the 4 best defenses added a 1+ strength at 3 point range and the Lakers also added FT defense.

Nobody was -1 or worse on FT defense. Just 2 on 3 pt defense (the Cavs were -3.3. 9 were -1 or worse from mid-range, with 3 worse than -3 (Portland, Detroit and Toronto with the worst mark). 9 were also -1 or worse on inside defense with one worse than -3 (Washington's -3.75 mark here was the worst weakness at any distance) and 2 others close to it. 20 -1 or worse weaknesses. 5 teams had 2 -1 weaknesses each while the rest of of teams averaged less than 0.4. Toronto almost had 3. None of the 5 teams with 2 -1 or worse weaknesses had a +1 strength. All the teams with 2 -1 or worse weaknesses were weak from mid-range with 3 also weak inside, while 2 were also weak from 3 point range. 3 other teams worse than -2 overall got that way overwhelmingly via weakness at one location. More flexibility about how to be bad at defense.

[Crow]

Shot-Adjusted Marginal Defense says Chicago's biggest shot defense strength was in the mid-range. Defensive RAPM says their 2 biggest overall defensive impact were wings- Brewer and Deng. The multi-season Adjusted +/- Factor for Shooting ranks Korver and Deng as strong on overall shot defense.

Of course these two measures do not isolate mid-range defense but they do include it.

Miami was best on SAMD inside. 2010-11 Defensive RAPM says their biggest impact defenders were James, Bosh, Haslem, Dampier and Anthony. The Adjusted +/- Factor for Shooting ranks James, Dampier and Anthony highly.

SAMD says Washington and Detroit were bad defending inside. Defensive RAPM says McGee, Thornton,Howard, Yi and Seraphin were the biggest problem defenders. The Adjusted +/- Factor for Shooting says Thornton, Yi and Seraphin were problem shot defenders. Defensive RAPM also says Villanueva, Wilcox, Daye, Monroe and Maxwell were the biggest problem defenders for Detroit. The Adjusted +/- Factor for Shooting says Villanueva, Daye, Monroe and Maxwell were problem shot defenders.

That pretty good agreement in 4 cases between SAMD and what we have from RAPM.

The Cavs were the weakest on 3 pt shot defense. The fit with RAPM and the defensive factor for shooting is not as clear cut here with Sessions, Gee, Graham, Harris and Jamison weak on these measures while Davis, Gibson and Parker are not. And Hickson was the worst defender on the team.

Of course the Adjusted +/- Factor for Shooting is for all zones and more players on the Cavs might be bad on 3 point defense and more sharply than their overall shot defense rating.

Conceivably the Adjusted +/- Factor for Shooting could be computed for the 4 shot zones to compare with the statistical data. They won't be perfect matches as the statistical data is based strictly on counterpart match-ups while the Adjusted +/- factor is for shot defense impact of a player across the entire opposition. But it would allow another comparison of perspectives and might nudge one toward more confidence in the estimates if they agree or less if they don't.

[Crow]

If there was interest from a team, these charts could be done by position, starter and sub, month to month, by lineup, by strength of opponent (team level and individual), conceivably by defensive scheme (zone vs man to man or more detailed), by specific opponent, etc.

And subsets where 2 or more criteria are set a certain way.
And crosses of shot location and play type (as discussed here
http://thecity2.com/2011/05/22/quantify ... lay-types/ ) and possibly other things as well for both offense and defense.



if you wanted to mine the detail for potential strengths & weaknesses (keeping in mind sample size issues of course).

[Crow]

Correlation of overall SAMD with:

Defensive Rebounds 0.76
Steals -0.09
Blocks 0.09
Personal Fouls -0.48


Correlation of inside SAMD with:

Defensive Rebounds 0.50
Steals -0.20
Blocks 0.00
Personal Fouls -0.40


Correlation of mid-range SAMD with:

Defensive Rebounds 0.71
Steals 0.06
Blocks 0.31
Personal Fouls -0.16


Correlation of 3 pt SAMD with:

Defensive Rebounds 0.45
Steals -0.03
Blocks -0.05
Personal Fouls -0.38


Correlation of FT SAMD with:

Defensive Rebounds 0.36
Steals -0.18
Blocks -0.40
Personal Fouls -0.84

A comparison to rates instead of raw totals would have been better but this still should give some rough insight.

Overall SAMD has the most correlation with defensive rebounds and not fouling much.

Same for inside defense but steals become a bit more important and blocks strangely have no correlation with inside marginal defense.

Blocks though are the clear second most correlated stat for mid-range defense and positively so. 3 pt marginal defense correlations are lead in size by the standard big 2.

Steals, Blocks and Personal Fouls are all negatively correlated with FT defense. Effort for the first two or effort on shot defense all lead to some combination of personal fouls or passing and movement that leads to better shots from the field. Blocks slip into 2nd place on correlation past defensive rebounding.

It would be possible to also see the correlations of the SAMD for each scoring location with just number of shot attempts and with just shooting efficiency. One could look at the data for various sub-groups of the league as well to avoid only knowing just the league-wide correlations. One could also look at player strengths & weaknesses, overall pace and shot clock distribution, correlation with offensive performance, etc. If you had the data for multiple seasons you could attempt to break coaching impact on defense down for shot location (and the 4 factors level as well).

The size of the above correlations for marginal scoring defense steal and blocks are quite low- the R2s are below .01. They are not generally dependent and perhaps that separateness of the activities is not surprising. Defensive ratings that give a lot of individual credit for steals and blocks may be appropriate for those actions but blocks and steals do not look like good proxies for the rest of defensive impact.

Defensive rebounding appears pretty highly correlated with marginal defense- R2 of .58. Defensive rebounding and personal fouls combined to explain 81% of overall marginal scoring defense. Or make that effective shot defense or bad shooting or luck which allows a defense rebound and personal fouls. Back to that basic difference of perspective on the importance of the defensive rebound vs. what came before it.