The 2005 Miami Heat: a puzzle

[Harrison Brown]

Hi all,

I've been playing around with modeling the relationship between Dean Oliver's Four Factors and offensive/defensive rating. It turns out that a straightforward linear model fits the data quite well, with R^2 north of .98 and what look like close-to-normal residuals, etc.

There is one very noticeable outlier when you plot the residuals for the offensive model -- the model predicts an offensive rating almost .8 standard deviations above the actual offensive rating. (The next-biggest residual is something like -.48 s.d.) I sort of gave it away in the subject line, but that outlier is the 2004-05 Miami Heat. That Heat team did pretty well, actually -- 5th in the league with a 110.2 offensive rating -- but according to the model it should have done a lot better.

So why did the '05 Heat "underperform"? Their stars were pretty much healthy; none of their four factors are anything extreme; they were actually worse in their subsequent championship season. I have a theory on this, but I'm interested to see if anyone else can come up with something better.

Hope everyone is handling the offseason well!

Best,

Harrison

hrbanalytics.wordpress.com

[exergosu]

Because they were built to cheat this model. Unbalanced offense with Shaq leading it. And Shaq is screwing everything free throw related. I don't know why you think there were no extremes, at least they lead the league in ratio and were last in percentage. May be it's not all time numbers but pretty rare combination for sure and should be at the top off all time list. High FTA ratio affects 3 of four factors. So all that reflected on Shaq and Wade's rating, Shaq splits of Non FT/FT would have insane difference, but at the time everybody already knew that fouling him is good way to stop 60%FG and all offensive rebounds from his misses. But difference between that season and their Championship run was supporting cast. Their first team had very bad distribution of scoring, to have adequate average stats for modeling. Both Joneses enjoyed playing with Shaq too much, posting insane numbers. And understandably subs weren't able to duplicate that, and when Wade played without Shaq he would run into trouble. So if you split their lineups that year, there should be some horrible ones, covered by overachievers who Shaq fed with open shots. Shaq rating has celling though, because of his FT% and huge amonts of them among total possesions. But Wade's rating was the catalyst. It jumped in Championship year, but his improvement wasn't so big (stats wise). Shaq played 4 minutes less that year, so if it was same supporting cast he would drop even more, but instead he was better without Shaq. All because they had 8+ man rotation and all guys had good eFG and added something more than just shooting open shots. Even Haslem started hurting his production, by shooting further (his famous baseline jumper) and getting less offensive rebounds and less rim attacks, just to clear the lane for Wade and his P'n'R with Shaq. And that little SvG stamp on way NBA is played still forgotten. Even Di Antoni is praised now, but not the guy who made "1 center and 4 perimeter guys" playstyle successfull and well known. And that's why I once again repeat, that this is the problem with linear modeling and using of these type of metrics. Sometimes it's bad by desing, sometimes same player do less stats wise to make his team better. And that outlier is great example of this. It's not underperforming, it's boosted by model. And I'm sure their team next year was better, despite what stats say. 8 man deep balanced roster in playoffs is much better (latest finals as proof) and you don't have a headache of how to split time between uneven 5 man units with least amount of damage. Ability to sustain injuries to key players is not in any model, but it's safe to assume that 04-05 one wouldn't sustain injury even to Eddie or Damion Jones that well, since it was so dependent on their key 5man units. Understanding of this just allowed GSW to demolish NBA, and Spurs before them. I know it's tough to build team like GSW (many lucky breaks) but value of versatile team is huge. So in these terms model showed underperforming in versatility of lineups, while 4 factor data was victim of Shaq effect. Pretty sure that his splits difference that would kill anything he had in his prime years in Lakers. As additional evidence of that, his net +/- was his lowest to date but his On Court rating was 2nd highest in his career, while he played 6 less minutes per game. Lots of garbage time to screw the average data you used.