Get Like Gregg: One Man's Quest to relate Coaching to Injuries.

Posted on Tue 06 December 2011 in 2012 Season Preview by Aaron McGuire

So, I was thinking. A dangerous habit, I know. Last year's underrated Cavs story was just how hard Byron Scott pushed the team in practices and off-the-court training. I'm talking suicide sprints after every loss, players throwing up in training camp, etc. Now, the 2011 Cavs were an awful team. But Byron Scott's "bad cop, crazy cop" routine made no sense to me. If your child is bad at a school subject, screaming at them and forcing them to do thousands of extra homework problems isn't going to do much of anything. Besides leaving them with crippling psychological disorders, anger management problems, and Samardo Samuels.

Overall? This had me rather worried about Kyrie Irving and the 2011 Cavs. Kyrie, as everyone knows, is coming off a injury-torn season where he played only 11 games of college ball. Not very pleasant, but he performed lights out when he played. What if Byron Scott's insane practices hurt him? The whole tangent got me to thinking about ways I could, perhaps, poke at a measure of coach-centered effects on injuries. As with my last big statistical post, this isn't an advanced model or a particularly advanced concept -- essentially, I'll be taking data from our pals at Brewhoop and repackaging it to describe coaches. Current coaches only, and for the majority of this post, only a smaller subsample of those coaches. Ones with enough seasons that I feel we can start to make some conclusions. There will be three parts -- an intro, some analysis of the big outliers, and a short discussion.

For my spreadsheet, check out the Google Doc. Let's get to it.

• • •

Part I: Introduction

For this part, turn to Sheet "GAMES LOST (sig)" in the spreadsheet.

Here are the results of my labors. In this spreadsheet, there are two main sheets -- a summary of all active coaches (plus one or two recently-inactive ones for completeness), and a summary of those who coached more than 3 seasons in the time period of this analysis (2001-2011). The most important of the two is the second -- the one that culls out the small-sample-size coaches and leaves us with some we can begin to make actual conclusions about, slim though they may be. To start, here's quoting the excellent folks whose post provided the numbers to do this analysis, summarizing how they arrived at the numbers (and it's recommended reading in its own right):

For this, I used the data from the excellent Basketball Reference. I started with the first team alphabetically by city (Atlanta), starting with the 2000-2001 season. I opened three BR tabs (no, not Bleacher Report) for the roster/stats of that year's team, the player transactions they made during that year, and finally the schedule of the team. I first looked at the transactions to see if any player movement occurred throughout the season for that particular team. Then, I looked at the roster and only examined the players who played an average of over 20 minutes per game, and only the top nine players if there was more than nine 20 minute players (the numbers will be different than Pelton's piece because he looked at all players, not just the top nine). Obviously players don't miss games just because of injury--suspensions and DNPs also factor in--but by restricting the sample to a team's top nine I tried to only look at rotation players who would be expected to play in every game. I then subtracted the amount of games they played from total possible games they could have played. If a player moved during the season, I looked at the schedule and dates of when they were traded to find out how many possible games they could have played with their team before being traded/waived, and subtracted from there. The same goes for if a team signed a new player somewhere in the middle of a season. If a player wasn't moved at all, I kept their total possible games at 82.

While it's relatively ad hoc (and does necessarily ignore suspensions, though it looks like he accounted for the suspensions having to do with the Palace brawl), it's a systemic and relatively robust way to account for lost games. The alternative is to spider through this particularly good NBA injury database, but honestly, given the already dubious assumptions I'm making in essentially "assigning" to coaches all injuries that occurred under their watch, I'm not prepared to spend a 5-6 hour period spidering that database and going through the massive hassle of linking it with my personal database. So that goes by the wayside for now. I took the data he presented in that post, then connected it with coaches -- if a coach only coached for a partial season, I pro-rated the number of missed games based on the percentage they coached -- IE, George Karl coached 40 games in 2005, so he is assigned a pro-rated 40/82 (48%) of their 74 missed games that season, and his # of seasons coached is adjusted accordingly.

Some basic stats from the analysis, to start -- according to the data, we have a baseline of 70 missed games a season for an average long-tenured coach, and 69 missed games a season among all coaches in the dataset. That may seem odd, at first, but think of it this way -- the fact that the averages are so close is actually a really good thing. It provides some evidence that there is an actual population mean, and that there's a fair chance that we're looking at a long-tail normal distribution (i.e., despite a true population mean, the data range is wide, spanning from far below the mean to far above the mean). It also gives us an effective midpoint. The average number of seasons coached by a coach in the dataset is 6 seasons among all coaches, and 8 seasons among coaches with > 3 seasons.

• • •

Part II: the Outliers

For this part, turn to Sheet "GAMES LOST (sig)" in the spreadsheet.

Instead of doing the workaday sort of analysis, the data here is odd and foggy enough that I feel a specific examination of some coaches significantly beyond the threshold from the average is in order. As a general rule, I'm going to look at coaches that are clocking in at an average of either 20 more games missed than the average or 20 games fewer missed than the average. Let's get to it.

GREGG POPOVICH, SAS -- 40 per year (11.00 seasons)

This one seems reasonable to me. Pop has never had a particularly bad reputation in terms of practice conditions, and (as Sean Elliot describes) hops through pretty large hoops to make sure his players don't play injured (and risk more injury). An ounce of prevention is worth a liter of cure, I always say. It doesn't really shock me that he's the best at managing injuries, though the magnitude by which he takes the spot surprised me. As did his consistency. Not only does he have the lowest raw average, he also has a standard deviation among the lowest in the set. Partially this is a figment of the relatively larger sample size, but it's worth noting -- Pop is essentially everything you'd want if you bought these numbers. Low average games lost, and (relative to everyone else) low uncertainty about the estimate. Plausibility: High.

MIKE D'ANTONI, NYK/PHX -- 46 per year (7.74 seasons)

Now that's a bit of a surprise. The coach who oversaw Amare's microfracture is somehow among the lowest in missed games per season? Bonkers. Until you think about it a bit more. D'Antoni is a lot of things, and he does run his teams at a fast pace, but the correlation between pace and injury rate is shaky at best and fallacious at worst. D'Antoni spent years coaching for the Suns, an organization renowned for their best-in-class medical staff and their very own resident player-friendly health expert in Steve Nash. His loss numbers aren't actually up in New York yet, though, and that's what makes me wonder if (perhaps) D'Antoni didn't just take his offense from Phoenix -- he may also have taken their best-practices coaching and knows how to best keep his players from getting injured. Far be it from me to assert that the D'Antoni Knicks are going to be impervious from injury or anything like that. But might he have picked up some institutional knowledge from the Suns that he spread in the Knicks organization? Plausibility: High.

STAN VAN GUNDY, MIA/ORL -- 49 per year (6.25 seasons)

I actually was expecting this, even though I don't know if many people would. Three reasons. First, SVG is a loudmouth, but he isn't actually a mean coach -- his players love him, almost without exception. SVG doesn't leave his players out to dry, and more often than not, criticizes himself and his game plan above his players. When he complains about playing on Christmas day or about the number of back to back games the Magic play in a season, he's not doing it solely for his benefit, he's doing it for the players. Second, he's coached Dwight Howard. Big men are injured more than virtually any other position in the league -- if you cull through injury reports, you'll find proportionally more big men than you'd expect. Dwight, though? He puts in 36-40 MPG and has missed a grand total of 7 games in his career, many from suspensions. If you play Dwight, you've necessarily got fewer minutes available for big men, and that means less chances for your 7th man stiff of a big man to tear his ACL or shatter his knee. Third? He's an extremely good, underrated X's and O's coach who strikes me as one to make practices more about learning his system than simply running his guys to the ground. So it didn't really surprise me that he was low. Plausibility: High.

• • •

And now, for the other end of the spectrum -- coaches whose averages were 20 games HIGHER than the average. Two of them.

MIKE BROWN, LAL/CLE - 99 per year (5.00 seasons)

This is the only result that really shocked me. Why? I honestly thought Brown was going to turn up pretty well out of this analysis. I've never really heard anything rough about his practice schedule, and he got to coach LeBron at his prime, where he barely missed a thing year after year. Then again. Big Z, Ben Wallace, Antawn Jamison, Shaq, Delonte -- the Cavs had a lot of injuries and turmoil during his tenure, and while I think he's better suited as a teaching example for why this sort of analysis is inherently flawed (is it really his fault that these players got injured?) it makes sense that he's this high. He also has fewer seasons than anyone else 20 games outside the average, so you'd think if it isn't an accurate representation of his coaching talents, it'll even out by the time he has as many seasons under his belt as Pop, Sloan, or Scott. Plausibility: Low.

BYRON SCOTT, CLE/NJN/NOH - 99 per year (9.62 seasons)

Scott was the impetus for this post, sure, but even I didn't expect the numbers were going to work out this well for my case. Scott's system is a meat grinder for young players and especially guards -- I have a friend in New Orleans who honestly thinks Scott might've damaged CP3's knees permanently with how hard he played him, and while I don't know if you can blame him for 100% of that, you can certainly blame a system where he punishes losing with suicide sprints and snidely putting down his players to the press. Well, okay -- you can't blame that second part, that's just a dick move on his part. Regardless. Scott's teams don't actually play all that fast, and as I said earlier, the research I've seen seems relatively lukewarm on the idea that pace has a real impact on how injured your players get in the first place. But of all the coaches in the sample, there's not one coach's numbers I buy more than Scott's. A lot of seasons, a lot of variance, but nothing about it seems off to me. And everything about it seems depressing to me, knowing that he's going to be coaching a team chock-full of young talent. Someone get me a stiff drink. Plausibility: Through the Roof.

• • •

Part III: Conclusions

Honestly, having said all this, I think there's enough material here and in the spreadsheet for you all to draw your own conclusions. I would just like to offer a few big caveats to this post before you do. First is the most important one -- I wouldn't be comfortable presenting this to my boss. I'm a statistician, and these numbers (while honestly extremely interesting) aren't nearly rigorous enough to be anywhere close to, say, an academic paper or even a predictive model.

The fact is, injuries AREN'T endemic to the coach, and it could be very reasonably argued that all the findings of this post are a factor of luck more than anything else. I wouldn't necessarily argue against that, even. However, I do think that it's within the realm of possibility that certain coaches run their practices and their training camps in a way that makes their teams more injury prone. And I think it's within the realm of possibility that others do the opposite. I wouldn't claim that these numbers "prove" anything -- I just think examining the outliers in an analysis like this and seeing if they pass the smell test (which, I'll note, 4/5 of them did) is a valuable way to look at this sort of stuff.

In any event, I hope you found these injury trends interesting. I won't be using this data for my pre-season win prediction model I'm working on, because I don't think it's nearly robust enough for that, but I do appreciate the fact that the analysis challenged my preconceptions about Mike Brown's coaching and confirmed my darkest fears about Byron Scott's coaching. Well, no, I'm not ACTUALLY that appreciative about that, damnit. But at least I know what to expect. As always -- any questions, let me know. I'm a really busy person lately, but I'm always happy to spend a bit of time answering questions or clearing up inaccuracies. Best way to reach me is either through the comments here or through twitter, @docrostov. Stay frosty.