Breaking down what FiveThirtyEight's statistical model says about the Rockets

The ESPN-sponsored data journalism site FiveThirtyEight recently presented the world with their NBA player projections system, playfully backronymed as C.A.R.M.E.L.O, which stands for career arc regression model estimator with local optimization.

You can read up on its origin and methodology in the introductory article on FiveThirtyEight, but basically, it is a fancy way of saying, "we look at a bunch of players like [insert name], and we guessed how [insert name] will perform based on how those guys performed." ESPN's Calvin Watkin already has written a preview for the Rockets on FiveThirtyEight based on CARMELO, so consider this a supplementary piece.

Using the individual projections of every player on the team and an estimation of minutes based on analyst-predicted depth charts, they were also able to come up with team projections, complete with win-loss records. According to the projections, the Rockets are pretty good. CARMELO projected the 2015-2016 Rockets to go 52-30, which while not explicitly stated, should place them in the middle of the western conference playoff picture in the range of 4th to 7th seed.

Upon first glance, this seems to be a fair projection. Actually, consider the difficulty of being a Western Conference team, some may even call it optimistic. However, dig a little deeper, and we'll find that CARMELO might be more pessimistic about the Rockets than what the 52 win prediction presents.

On top of simply presenting the number, CARMELO also gives players categorical designations based on their individual projections. Here is the Rockets top 10 players and their designations according on CARMELO

Name	Category
James Harden	MVP Candidate
Ty Lawson	Average Starter
Trevor Ariza	Average Starter
Dwight Howard	Key Role Player
Terrence Jones	Up and Comer
Patrick Beverley	Rotation Player
Corey Brewer	Rotation Player
Donatas Motiejunas	Rotation Player
Marcus Thornton	Scrub
Sam Dekker	OK Prospect

The good news is, Rockets have eight solid NBA players in their rotation, which is more than a lot of NBA team could claim. The bad news is, like the real Carmelo, CARMELO is not very impressed with the Rockets' roster. Beyond the amazing James Harden, the rest of the roster seems to be collection of unspectacular options who nonetheless get the job done, which is not too far from the truth. The Rockets are purposely built with a top heavy approach because Basketball is a sport where a single great player's impact can outweigh a collection of good players.

However, according to CARMELO, Houston might be less talented at the top than they originally planned, as Dwight Howard is only categorized as a key role player. This stems from the fact that Dwight Howard missed significant time last season, and accrue significantly lower WAR than his career norm. History is not kind to big man coming off an injury-shortened season at this stage of his career. In fact, despite predicting a little bit a bounce back in the upcoming season, CARMELO see it as the beginning of the end for a dominant Dwight Howard.

This is not all that surprising. Dwight Howard's game, predicated greatly on athleticism and agility, has never been one that people expected to age gracefully, so it is not entirely out of line to predict a quick and sudden crash. Injury woes only exasperates the concern. He may not necessarily get worse, but simply playing less dips his production just the same.

It's important to highlight, however, that Dwight Howard's career arc seems to be fairly unique. Of his top 10 comparables, seven have similarity score in the 20s, which means they are not that similar at all. So it is possible that CARMELO is not able accurately projecting Dwight Howard given that he has very few comparable peers. Howard is, nonetheless, an important part of the Houston machine, as the projected team contribution +/- has him almost single handedly keeping the Rockets' defense above average.

Looking into some other player's projections, Donatas Motiejunas stands out, not so much for how well he is rated, but how average he is in the eyes of CARMELO. Motiejunas does not stand out in any of the key stats that CARMELO evaluates. Motiejunas is so average, his number one comparable is Reggie Johnson, a man who seemingly averaged 0 WAR over his short 4 year career. He is not even historically unique, as many of his comparables posted similarity scores over 50, which qualifies as extremely similar.

I have always thought that he is less likely to receive a significant contract due to his fairly pedestrian counting stats, but Tristan Thompson just got 18 million a year, so what do I know? Since Motiejunas is already 25, he is not very likely to significantly improve, but it is not without precedent. Both Cliff Robinson and Mehmet Okur are in Motiejunas' top 10 comparables with fairly high similarity scores in the 50s, and they both experienced significant uptick beyond this point of their career as a function of greater role/playing time. Can a post-injury Motiejunas received enough minutes/play well enough to warrant those minutes?

New addition Ty Lawson is another interesting case. He is categorized as just an average starter, not even a good starter, but his projected WAR is quite high at 4.4. Lawson's defensive drawbacks are obvious, even in the preseason we have seen it manifest on the floor. He ranked nearly at the bottom of every defensive stat that CARMELO considers and posted negative defensive +/- for the last 3 seasons.

For the Rockets, it is a necessary trade off since Lawson's elite assist % and low turnover % are something the Rockets have very desperately needed for a while now, and he will surely add more diversity to a rather monotonous offense. It is no surprise that Lawson compared favorably to many great passers, and most of them were effective players everywhere they went.

CARMELO does not exactly paint a flattering picture for the backup wing positions, as both Corey Brewer and Marcus Thornton have negative +/-  projections on both side of the floor. Luckily, James Harden and Trevor Ariza will account for the majority of the wing minutes, with Corey Brewer mopping up most of what's left.

Despite the negative ratings, Corey Brewer is another perfectly average player that CARMELO predicts to be worth at 1.3 wins above replacement (give or take approximately 2 wins). What is odd is that Corey Brewer only has 1 top comparable that played mostly in the 2000s, which I find is fairly rare after sifting through about 20 some odd players on the interactive tool. Corey Brewer apparently is a true throwback to 80s Basketball.

Marcus Thornton played rather poorly the last couple years, which earned him the designation scrub. He is the only non-rookie on the roster to have a negative WAR projection, although it is not far from replacement level at - 0.1. His veteran status will likely earn him more playing time than he deserves, but he is likely the Rockets' weakest link according to CARMELO. His top comparables are a regular who's who of smh hall of fame with Shannon Brown, Anthony Morrow, and Nick Young leading the way, and the similarity score are extremely high at around 60s, scores of which CARMELO describes as "separated from birth." When "way past his prime" Jason Terry is projected have a better WAR, there is cause for concern.

A glaring omission from the CARMELO projections is the presence of foreign prospects, as CARMELO just have no idea what to do with them without the rich historical data set that NBA and NCAA provides. They do a baseline projection based on age, height, and draft position, but nothing more sophisticated, which is why everyone's favorite draft steal Clint Capela is severely underrated in these projections.

Clint Capela could very well be the extra "oomph" to the Rockets roster that takes them to another level. Affectionately nicknamed "baby deer" by TDS commenters because he started his NBA career playing a lot like this, but since then, he has looked a lot more like this (just so you know, Swiss Wendigo is an available nickname).

Ed. Note: I have clearly failed in my attempts to make Swiss Roll happen.

The 21-year-old Swiss burst onto the scene in the playoffs last season, exceeding everyone's expectation by playing remarkably well. He has continued the momentum in the preseason, taking advantage of Dwight Howard's extended hiatus by putting up double doubles in 3 of his 5 preseason appearances. Simply replacing Capela's CARMELO projection with a reasonable facsimile, like say this guy

and the Rockets roster suddenly seem a lot more complete with another key role player in tow. Clint Capela had a TS% of 65% on 15% usage and a 17% rebounding rate last postseason, and he has done fairly well in the preseasons so far, with 57% TS% on 15% usage and 20% rebounding rate. A Brandan Wright-like contribution is very much attainable in the upcoming season, and because Clint Capela is 21 and not 29, his career arc is more likely to go up rather than down like Wright.

All in all, CARMELO provides an interesting perspective using historical data to project player performance, and it is surprisingly an uncommon approach when it comes to basketball analytics. Nate Silver will be the first to tell you that this system, like any other system, has its flaws. Nuance of context are often ignored in analytics, it's our job to add the context back. In the end, it is just a projection, something fun to look over and discuss.

You can find the interactive tool here. play around with it and express your opinions below.