Counting the Baskets

Plausible Enough Win Projections

I put together some NBA Win projections, and had been just about to go on record with them when Jimmy Butler popped up with his demand to be traded away from Minnesota. I figured I’d wait since, otherwise they’d be obsolete before the pixels dried.

But, since no one knows what’s going on with the T-Wolves front office. I figured I would just move forward, and put my predictions out there.

This version of my predictions is using a combo of ESPN”s RPM and Jacob Goldstein’s Player Impact Plus Minus. Both use two years of ratings weighted two to one for the most recent year with additional adjustments for minutes played, changing teams and age. The two metrics are essentially weighted equally, though RPM has slightly more weight on defense and less on offense compared to PIPM. I then weighted each team’s strength by expected minutes and ran them through their schedule.

I solved the Jimmy Butler problem by removing him from the T-Wolves and replacing his minutes with a typical back up’s rating. If the Wolves get Josh Richardson, they will beat that estimate, but there is a significant chance they under perform that estimate, particularly given Thib’s love of 2018-19 vintage Derrick Rose. I did not add Butler to any team, so one lucky team should get a boost over these numbers (at some point).

Since I went a little lighter on obsessive line details and dropped my old Player Tracking Plus Minus, the following are my “Plausible Enough Win Predictions.” They are sorted by conference and predicted wins in descending order:

 

The projections have the race in the East as tight. Boston fans who think they’ve got a pre-paid ticket to the finals should be taking a couple of breaths.

The top of the west is projected as a two team race, but there’s a tight race projected for both the back end of the East and West playoffs. Interestingly, unless Butler moves to the Clippers, the expected wins for the eight seed in the East and West is almost identical.

New Celtics' Draftee Stats Compared to Peers

One way to get a sense of player's strengths and weakness is to chart them against their peers. The visual can give a more intuitive feel than numbers by themselves and it's much easier to take in information on a bunch of different players at a time.

I created one for the draft that groups the prospects stats in categories that are mostly independent and represent different aspects of my draft model's evaluation. All the stats are standardized in a z-score, with average players scoring at 0 and the spread between best and worst scaled by the standard deviation. That way the differences in each category  are set to the same visual scale.

The stats are then put in a box plot using position as the grouping variables, so a player's assists, for example is compared to guys at the same position, instead of comparing a power forward's assists to a point guard.

Below is an image of the rebounding category, with Celtics picks Semi Ojeleye and Jayson Tatum's mark in the chart highlighted with commentary.

The data  in the link is interactive so you can change the filters are hover over the screen to see more information on a particular data mark.

Among the highlighted stats for Jayson Tatum, he was:

• A very good scorer, among the best from the wing positions
• An above average rebounder at the three, but not great.
• Distribution was his biggest weakness, one the worst wings in that rating.
• Was above average for his position in blocks and steals rating.
• He was one of the youngest players in the draft, and played a very competitive college schedule.

For Semi Ojeleye highlights were:

• Very good rebounder as a three, combo forward potential
• One of the best wing scorers in the draft
• Weakness in blocks and steals
• Slightly above average for a wing in distribution
• Ojeleye is one of the older prospects in the draft.

 

49 Wins?

My Highly Plausible Prediction model this year projected the Celtics at 49 wins, a number above their 40 win performance last year, the initial Westgate line on them at 42.5 wins and the general consensus. I have been asked a couple of times whether I "buy" the model's number or not, and the answer to that is, of course, both yes and no.

 

The ambiguous answer comes from two parts. First, is my grasp of the model's strengths and weaknesses along with last year's model performance, which I will discuss below. Second, is the nature of what I am trying to do in making my predictions in terms of contributing an independent estimate to the general knowledge base.

 

The second point is interesting in this way;

 

If you want to make the most accurate prediction it is best to give significant weight to the wisdom of crowds, but the wisdom of crowds doesn't exist if no one is contributing their independent estimates. 

 

If everyone tries to free ride on the crowd, there is no wisdom generated. Similarly in the separate but related economic idea of the efficient market hypothesis, it is sometimes said that it can only ever be approximately true because the market needs traders attempting to beat the market to make the market function. Nate Silver has also pointed to pollsters tailoring their polls to fit the consensus as a factor weakening the performance of polling averages in recent elections.

 

So, in markets, election predictions or sports predictions individual predictions are made better by pooling information from the consensus, but the consensus itself is made weaker if that behavior is too widely adopted. In this case my model is meant as an independent estimate, likely a best guess estimate would weigh it with something like the ESPN forecast that aggregates the opinions of ESPN writers and bloggers, but I think it is a strong and plausible estimate on its own.

 

The projection model itself last year was among the best publicly available analytic projections, for whatever that is worth. More concretely, the median error was 4 games and the average was 5.9 games per team, the larger average miss compared to median indicates there were a couple of outsized misses, like underestimating Atlanta and overestimating Minnesota and OKC, that pulled average number up. But in either case that is a good amount of expected error.

 

Celtics Specific Issues

 

But there are a couple of items specific to the 2016 Celtics prediction that could cut in in both directions. The first is schedule, according to my ratings the Celtics have the second easiest opponent schedule in the league with their opponent's average win estimate at 39.5 wins. Being in the Atlantic division is helpful, playing the Nets, Knicks and Sixers four times a piece is likely to pad the wins at least a touch. 

 

A good deal has been made of the Celtics' "mish-mash" roster and lack of talent separation among players this off season. The lack of certain skills and the possibility of too much skill overlap, not enough shot creation too many defensive oriented guards and wings, for example, could be a source of systematic error. This is reduced somewhat by enforcing rotation minutes that balance out positions, including, for example, increasing minutes for Evan Turner, a versatile point wing but also the lowest rated Boston rotation player. However, fit and overlap remain unknown potential sources of error.

 

The lack of separation is a double edged sword. Mentally assessing team strength it's easy to default to something like a playoff rotation model with all starters available and a shortened rotation for the bench. However, only nine players started all 82 games last year, and only 76 players, about half of the available opening day starting positions started at least 80% of the games. In that high frequency starter set the median minutes played was about 2,400 minutes, about 60% of the available minutes. In the next group starting over half the games but less than 80%, 52 players per basketball reference, the median playing time was about 2,000 minutes, or about only 50% of available playing time. In the regular season, at least, depth matters, though it is hard to predict before the season who it will matter for the most.

 

Lastly. the Celtics have a young group, with only Amir Johnson and David Lee on the wrong side of the aging curve. Their net aging curve, the amount of improvement or decline expected across the team weighted by minutes played is +1.99 wins. Some young players clearly won't improve, the age curve is a generalized trend, but having more young players gives the team more rolls of the dice on one of them having a breakout season.

 

Ultimately, I in terms of what I believe the Celitcs are most likely to do in the upcoming season, I would take the over on the Westgate line, but probably not the over on the Highly Plausible model's projection.

Terry Rozier Acceptance Phase

It is probably a good thing for Celtics fans that free agency in the NBA follows so closely on the heels of the draft, if only to distract them from their palpable disappointment. The collective draft let down stems from both the inability of the front office to move up in the draft from the sixteen spot and the resulting pick of Terry "Who Now?" Rozier out of Louisville with that pick. 

Rozier was projected by most mock drafts to be selected around the end of the first round, closer to the Celtic's 28th pick than where he was ultimately taken. His numbers weren't well liked by most of the analytic models in the public domain either, prompting Nylon's Nicholas Restifo to give the Celtics' an F as an objective and subjective draft grade. Restifo calculated that the Celtics left a good amount of value on the table with their maverick choice. Not great, Danny.

  

Acceptance Phase

 

First, it's important for fans and writers to remember that this was the team's choice to draft Rozier higher than expected, not Rozier's. All of the additional expectations should be on Danny Ainge, not Rozier.

 

However, there are a couple of reasons to think Rozier might be a better pick than the initial fan reaction held or that initial draft grades might have been a bit overly pessimistic on Rozier. Part of that stems simply from faith in the Celtics as a competent NBA organization, not a super all knowing, can't go wrong set of basketball geniuses, just a typically competent NBA organization. 

 

I applied that idea in my Benefit of the Doubt draft models the other day at Nylon Calculus, captured below using just the Celtics draft selections. The table shows the initial model rank and the rank in the Benefit of the Doubt models.

 

 

To be sure, the model would have estimated a higher value for a player at that spot who performed better in the initial, non-draft informed version of the model. But, balancing the judgement of teams with the analytic models does give better estimates and the Celtics seemed to move on Rozier because they thought other organizations would grab him before the 28th pick came up, so perhaps they weren't the only ones interested in him.

 

In addition, Nic Oza at the Georgretown Sports Analysis Business Research, had some thoughts on potentially overrated and underrated prospects in this year's draft. While Oza didn't mention Rozier, he did note that defensive skills are likely an underrated in the draft. A point I tend to agree with. Rozier is, of course, viewed as a defensive specialist indicating a reason he could have been an under valued prospect

 

Lastly, Rozier wasn't evaluated poorly by all of the analytic models. New Hampshire Math Professor Steven Shea's CPR rating had Rozier as the thirteenth best prospect. That in and of itself is not necessarily that interesting, any NBA worthy prospect can come out as well rated in one public draft model given how many there are now. 

 

But, it does get more interesting for two reasons. First, Shea's model rated all of the Celtics' first three picks well; rating R.J. Hunter as 11th and Jordan Mickey as 14th overall. Second, Shea's methodology focuses on the ten max performances by a prospect using game log data, then adjusted for class, suggesting those max performances may be more indicative than average performance of potential. What makes that interesting is that the Celtic's Director of Analytics, David Sparks, developed a game log metric back when he was a writer for Hardwood Paroxysm, so there's reason to believe the Celtics might be looking more at peak game performance as part of their analytic evaluation. I think there are pluses and minuses to that approach, especially if only a subset of games is used, however, I feel more confident in the Celtics selections thinking there was more than Danny's gut involved in the process.

 

As I said in the Benefit of the Draft model, all of the new rookies will soon have a chance to start proving themselves as NBA players and update our estimates of their potential.

 

The Celtics Lack a Two Way Player

With the Celtics twenty games into the season I collected the data for my Player Tracking Plus Minus (PT-PM) model, with a big assist from Darryl Blackport for the data scraping script.

With the data through December 15th, Tyler Zeller had been the Celtic's most productive player on the roster based on his extremely efficient scoring and solid rebounding for an estimated plus 1.8. Jared Sullinger is a distant second among rotation players at plus .45. Meanwhile,Marcus Smart was a tantalizing  plus 1.35 in minutes limited so far by injuries.

Below are the production rating for the whole team to date.

One clear thing to note is that none of the Celtics players are in the plus side on both offense and defense, though Zeller is close on defense. This reflects not only the Celtics' 'eveness' as Brad Stevens says, but also the challenge Stevens has in making line ups, since there are not enough all around players to make the mixes easier to fit together

I also have the standardized scores for each player by the model's elements (any score above zero means the player is above average in that category, negative means they are below average).

Obviously, with this year's Celtics one has to talk about Rajon Rondo. His line here is interesting, the passing catagory includes both points created per forty minutes and the assisted per pass, which are stable contributors to offensive efficiency in the model. Rondo scores over four standard deviations above the average player there, tops in the league. He also scores well in gaining possessions on the defensive end with his rebounding, but has been below average in scoring, extending and holding offensive possessions, rim protection and on court team defensive efficiency.

Here is the table above as a chart(click to enlarge):

In the chart you can actually see how polarizing Rondo is even statistically say compared to a guy like Bass, who scored near average in each category, Rondo has been great in a few things and rather poor in others with little in between.

How Much Does Preseason Matter?

The Celtics got off to a better start to their preseason this year than last year, the first under Brad Stevens.  After two twenty point point wins the optimism was running high in Celtic land, before it reverted after three losses in a row.

So, I wanted to quantify the weight one ought to give to preseason optimism or pessimism, as well as channel the things we most ought to watch for in preseason.

In the case of team performance, there is a definite positive, but weak relationship from preseason performance to the regular season. The website 82games did a study on preseason win percentage back in 2006, they found a .401 correlation (R² of .16) between preseason win percentage and the regular season.  

I did a similar regression using the seasons picking up from where 82games left off, 2007-2014, omitting the lock out season in 2011, which featured only 2 preseason games and a shortened training camp.  My results were slightly lower using both point differential and win percentage with a correlation of .367 and .319 respectively (R² of .139 and .102).  Interestingly the correlation for both in the last three non-lock out years has been even lower for both.  Three years isn't enough for me to take out my protractor and draw a straight line projecting out, but it's an interesting note.

It's possible that the quickening change in coaches and shorter playing contracts has made preseason more of a vehicle for discovery and experimentation or the growing talent divergence between East and West has made the largely intra-divisional preseason schedule even less indicative of the regular season (Hello Atlantic Division!).  Or it's just noise.

As a side note, preseason does retain some independent information even when paired in a model with the prior year information, either point differential or winning percentage.  Historically using both prior year and preseason information has explained ~ 38% of a teams winning percentage (correlation of .61) with the preseason providing a third of the explanatory power.  Of course, if you lose LeBron James or acquire 2008 vintage Kevin Garnett all bets are off, as the 2011 Cavaliers and 2008 Celtics were the two biggest outliers, which is why one is better off using the player metric methods like Nathan Walker or Kevin Ferrigan do.

Part of the reason, other than sample size and strength of schedule, preseason isn't a particularly good predictor of anything for teams is that rotations do not much resemble their regular season form.

So what to take from the preseason?

 

The above image is a diagram my informal model of the trade off between expert opinion and data analysis as more observations are taken.  Basically, the fewer observations one has the more weight opinions on process people with subject matter expertise, should get but with more observations we can weight the numbers more heavily, though subject matter experts continue to retain value. Preseason definitely lies somewhere towards the upper left of the chart with only seven or eight games and few extended outings by any one player.  

The main things to look at in preseason are the same ones I think most of the coaches are looking at; first, and foremost, whatever shiny new players the team signed in the off season, whether they are free agents or rookies.  Rookies, for example, tend to get much more run in the preseason than most of them will in the regular season and it's the first time seeing them against legit NBA opponents. 

Last year I created a visualization matching preseason individual stats with the regular season performance, which is here. In general the stats we generally think of a stable like rebounds performed well, while scoring was only marginally related to preseason performance. (Sidenote, that it's always important to understand that some of the 'stability' of stats like assists or rebounds is due the positional/role dependency; point guards will probably continue to have the ball and make passes and centers will continue to stand near the basket).

 The data gained from preseason is generally more valuable for any player that either has a little existing data like rookies or anytime there is a significant discontinuity between this season and previous seasons, and that's true of scouting or coaching observations too.  For example, last year building my win predictions, I found that the ASPM box score metric was less predictive season to season for any player that changed teams, which is similar to numerous other study findings on boxscore data.  Also, players like Derrick Rose or Kobe Bryant coming off of major injuries and essentially lost seasons are another example of a significant discontinuity.

We can also take advantage of the fact that we have other information about players, applying Bayesian principles. There's no inherent reason for preseason data to stand on its own, or for us to not incorporate that data with previous observations. Just as there is no reason for scouts to discard their previous film reviews or prior in person scouting every time the calendar turns over.

So, for example, looking at rookies preseason numbers, we can and should assume that the numbers will show some mean reversion to the long term rookie mean.  Using the linear box score metric Alt. Win Score (AWS) that I used for my draft model, I ran a simple regression with both the raw AWS and one stabilized with a reversion to mean.  The Stabilized AWS showed a consistent, though slight, advantage predicting the regular season AWS per 40 minutes.

For rookies with as few as 70 minutes played in the preseason and at least 100 minutes in the regular season I found a .44 correlation (R² .198) over the last three years (a period with very poor team win percent correlations from preseason play).

I am in the process of combining that preseason rookie data from 2010, 2012 and 2013 with their data from their pre-NBA seasons.  Using the sample matched so far I ran some regressions with both sets of data to predict their performances in the their rookie year:

• The R² in the complete set was .351, with sub-sample regressions R² ranging from approximately .26 to .44.
• The pre-season numbers explained 40% of the model's predictive power, with sub-sample regressions running from approximately 30% to 50%.
• Offensive rebounds appeared to be the strongest preseason predictor of any single stat, though that appears to be a function of the rookie success of a few outstanding big men like Andre Drummond and Athony Davis.

In addition to this general analysis, I plan to add to the cases and columns of data matched to date in order to try a more fully realized model and create a simulation run from that to apply to this year's class, as well as revisit a Bayes' theorum analysis I did.

Predictions are Hard, Especially about Three Point Shooting

I am in the middle of doing some modelling to predict a player's three point shooting percentages. Three point shooting is notoriously one of the most variable stats in basketball.  

Over at Nylon Calculus, Darryl Blackport did a great analysis that indicated it takes 750 shots before half the variation in a player's three point percentage is explained by their shooting percentage rather than noise.  By contrast in a similar analysis in baseball strike out rate took only 60 at bats to stabilize and walks 120 at bats, though on base percentage was 460 at bats and batting average was 910 (but that stat is not recommended).

My modelling work is designed to help that along a little by looking at other measures related to a player's shooting ability in order to stabilize things a bit quicker.  One of the data sets I have available is from my draft models, so I decided to do a little exploration with that data.  

Whenever we talk about a player's shooting potential coming into the draft our natural inclination, or at least mine, is to look at the their three point percentage in either the last year, or over their career and project them from there.  The problem is that very few players take anywhere near 750 three point attempts in college or European basketball before declaring for the draft, and, in fact, in the first few years of their career in the NBA there aren't many player's with that kind of shot attempt numbers.   

But, it turns out there is other data in that stat sheet that is just as helpful projecting their numbers forward as the three percentage, if not more so.  I ran a number of models and, so far, none of them has gotten to the .50 R², but the models have clearly been a significant improvement over just running the pre-NBA three point percentage, which explains no more than a small percentage of variation in any of models.

In general there is a significant information conveyed by the player's free throw percentage, the frequency of three point attempts and then the three point percentage.

Using the three point percentage of third and fourth years or using years two through five as the target variable weighted by three point attempts the model explained  about 30% of variation in three point percentage.  The formula for the third and fourth year three point percentage using pre-draft stats is below:

NBA 3 Point % = .175 + .128 * Free Throw Percentage + .00449 * 3PTA per 40 + .163 * Three Point Percentage

The three stats carry approximately equal amounts of information as measured by the beta coefficient.  Though running numerous random data sub-filters most consistently indicated that free throw percentage is the best indicator. Here is how I view the variables in relation to predicting downtown shooting numbers:

• Free Throw Percentage basically conveys information about the player's shooting stroke without the 'noise' that can effect any particular three point attempt from shot clock, pass placements, or opponent contests.
• 3PTA per 40 is a variable that I think one needs to be a bit careful interpreting. In essence, it conveys something of the confidence both the player and their coach have in their three point stroke.  On the other hand we can't necessarily interpret as a causal variable, Josh Smith and Charles Barkley spent their careers proving that jacking up threes doesn't necessarily cause your accuracy to rise.
• Three Point percentage, of course, mimics most closely the variable we're looking for: three point accuracy in the NBA.  But the high variation and relatively low number of attempts in pre-NBA shooting make that number unreliable as a predictor.

An alternative model that uses only the player's three pointers made per 40 minutes and their free throw percentage actually explained somewhat more of the variation, about 34% with its simple formula below:

NBA 3 Point % = .22 + .01571 * 3pt Made per 40  + .1389 * Free Throw Pct

• 3PT Made per 40 conveys some of both frequency and accuracy, but has the same interpretation issues as three point attempts.

With all that said, I thought I'd apply the models to some of the notable rookies from this incoming class. Below I have both models one and two, which give largely the same result.

•  Marcus Smart comes out much better than would be expected from his three point percentage only due to his decent free throw percentage and confidence taking the long shot. As such he's a decent test of the three point volume variable.
• James Young doesn't project much better than Smart from deep, which is concerning given that's his primary selling point.
• Doug McDermott and Nik Stauskas project the highest, unsurprisingly, which is like the LeBron test for a player metric, your model doesn't think McDermott is a good shooter you may want to re-check your model.
• Neither Elfrid Payton or Aaron Gordon project well going forward as shooters.  Gordon due to his poor free throw shooting and relatively low volume from beyond the arc, and Payton due to a combination of all three.
• Noah Vonleh probably isn't going to hit 48.5% percent of his three pointers in the pros.  

Three point shooting is definitely one of the areas that I think scouting comes most into play, especially for the player's with more unusual profiles. Players like Vonleh or Smart or Bogdan Bogdanovic where the volume and shooting percentages don't match up, given the variation at play, context is necessary to add to stats modeling's reach.  Did they get pushed into poor shots as each tried to carry their team, or is their volume of three pointers something to dismiss because of no effective options on the team?

In any case, next year I will try to avoid mentally projecting a player's NBA three point prowess looking simply at his percentage this year.

A Little More Green Space

(The Celtics Have More Shooters Than They Started with Last Year)

The clock is ticking down on the NBA off season and cap space and roster spots are drying up like the California reservoirs.  That means that the odds of significant changes to the Celtics roster before the start of the season are also shrinking.

So, I thought I'd apply my shooting style analysis to what we're currently working with in terms of the Celtics roster.  I did something similar for the Celtics last year in a post that ended up being titled, "No Threes for Us, We're Celtics."

Long distance shooting isn't the only valid playing style, but a number of  studies have indicated that having long distance shooters on the court is helpful to offensive efficiency even beyond the extra point on the three.  The general thought these days is that an effective line up needs to have mixture of shooting styles with at least a few players able and willing to shoot from three. 

There has actually been a significant turn over in the Celtics back court since the start of last year, and we've gained a year of data on last year's rookies.  In addition, there's also been an evolution in Jared Sullinger's case in his first year under Brad Stevens with a not always successful addition of the three point shot to his arsenal.  So that's no longer an exactly accurate description of the roster.

Quickly, the following table shows the results of a statistical data mining technique to group data with similar cases called Cluster Analysis on the shooting styles of players.  The Clusters were run separately on front court players and back court players to give a better opportunity to finely analyze playing styles.

First the guards and wings:

In contrast to last year there are three back court players that qualify as Stretchers based on their shot selection, where Keith Bogans, who effectively ended up not playing at all last year, was the lone Stretch player in the back court.

In addition, there are the incoming rookies Marcus Smart and James Young.  Smart took over 42% of his field goal attempts from three, and 35% at the rim, according to Hoop-Math. Though his conversion rate on threes was only 29.9%  Young even more clearly has Stretch potential with 52% of his shots from three and only 22% at the rim.  Young had a decent 34.9% conversion rate, but the Celtics front office seems to think that number is below his expected long term shooting ability.

The front court also has a couple of players who adopted a 'Stretch' shooting mix, though the percentage of threes taken for a big to be considered a 'Stretch' player is lower than for guards and wings.

Sullinger, Kelly Olynyk and Vitor Faverani all took enough shots from three under Stevens to rate as Stretch Bigs, while Brandon Bass remained a mid-range specialist.

Of course, shooting and making the three are not the same thing, though there is a relationship. In order for a shooter to provide effective spacing he has to be guarded, meaning the defense has to want to discourage him from shooting.  The final column on this table with the points per shot from three gives a very rough idea of the player's immediate 'stretch potential' based on their shooting percentage from three last year. (I actually have another project in development looking at predicting three point shooting year to year, if I ever get around to completing it).

 

Shot Selection Avery!  Based not only on last year, but over his career as a 36.6% shooter from three Bradley has the best potential as a Stretch shooter, which would not only increase his efficiency but add provide more space for other player's to get to the basket. 

Chris Johnson, Marcus Thornton and Jeff Green under performed their long term three point shooting, so should be added as potential Stretch threats.  Although Green is probably more valuable as a Slasher over the long term, but in certain line ups he could take the Stretch role.

Phil Pressey needs to be able to shoot an NBA three to remain in the NBA beyond this year.

And the same for the front court.

Obviously, Olynyk looks like the most legitimate Stretch big threat on this team, while there's probably only one more year on the Sullinger from three experiment unless he starts hitting at a higher rate. I also wouldn't mnd seeing Bass expand his range at leas to some corner threes if he sticks around. It would be good for both his trade value and next contract if it works.

There isn't really a knock down shooter on this roster, but there are at least a couple of willing and marginally capable three point shooters, which is at least a step in the right direction. (Now if only Avery would take two steps back).

The Upside Down Celtics

 

Over at the Nylon Calculus, I noted that the Celtics have sometimes been accused of having an upside down line up, in the sense that the back court players can't shoot and the front court players can't play defense. In that post I was looking at a study using a stabilized version of plus/minus data called RAPM that indicated that, while the traditional stereotype was in some ways true there may be an important caveat.

To put it succinctly, guards, on average, contribute more on offense and less on defense,while for centers and power forwards the opposite was true. But the interesting bit is that the spread between the best and worst center on defense and the spread between the best and worst point guard was largely the same.  

If guards really weren't important to defense we would expect the spread between the best and worst point guard to be much much smaller than for centers since it wouldn't matter, for example, if Aaron Brooks, ones of the worst PG defenders or Chris Paul, one of the best, was on the court.  The RAPM data and a more complicated version that combines box score data called Real Plus Minus (RPM), via ESPN, indicate otherwise.

So, in theory, this data indicates that it wouldn't necessarily be bad that the Celtics were  'upside down' on defense.  One would always prefer to have five good defenders on the court and five good offensive players, of course, but it is not inherently less valuable to have an elite defensive guard replace an average guard than an elite center replace an average center.

Focusing on the defense, I put this table together with the defensive RPM scores of the Celtics' current roster, excluding the rookies.  It has each player's raw D-RPM, then adjusted for position and their percentile rank for their position.

 

So, how upside down is this roster?  Outside of Olynyk and Bradley, and perhaps Marcus Smart who projects very well as defender, the Celtics roster is actually not all that upside down.

Based solely on raw D-RPM one would be lead to believe that Joel Anthony, Jared Sullinger, and Tyler Zeller were the C's best defenders, after Gerald Wallace.  However, Sullinger, Anthony, and Zeller are all just average for the their position, indicating they're adequately filling their defensive role but not little more.  

Based on the Position Adjusted column, however, the clear difference makers on defense for the Celtics were Gerald Wallace and Avery Bradley, both of whom are above the 90th percentile and rate 2.4 points per 100 possessions over position average on defense. (Side note, Bradley is listed as a point guard by ESPN, despite manifestly NOT being a point, though for defensive purposes guards are ranked together here as there was no significant difference on defense).

On the other side, while Rajon Rondo and Kelly Olynyk both rate near the bottom on raw scores. Rondo, despite his reputation, scores a little below average for his position,but Olynyk is much much worse when position is taken into account.  A little of that can be attributed to listing Olynyk as a center, where his more natural position is power forward, but he would still be last on the team by a wide margin.  

Olynyk scores poorly on Rim Protection using SportVu data, allowing opponents to shoot well above average on contested shots near the basket for a big man at 54.6% (average being 50%). But, the team as whole faired no worse protecting the rim with Olynyk on the court, according to NBAWOWY data.  However, Olynyk fouled too often when he did try to protect the rim, a stat that shows up in even the raw on/off data for the team, with the free throw rate jumping from 28% to 32% for opponents when Olynyk was on the court.

Olynyk as a big man that can shoot from outside and pass, has the potential to be the Celtics most complete offensive player, but in order to impact the game positively he will need to learn passable defense.

This is the part of the post where I add the caveats and disclaimers about RPM as a stand alone metric.  It is not perfect, with a single year there are multi-co-linearity issues, which adds to noise.  There are always factors on defense that emerge from the combination of players that can't be completely isolated to individuals.  But RPM does have a decent track record, with its predecessor xRAPM, in terms of prediction.

Like any metric or stat we expect some reversion to the mean next year, adding to that an aging factor, the safest predictions are that Olynyk's defense will improve and that Bradley's and Wallace's, especially, will get worse.  Making the defense that much less upside down, though not that much better.

Passing Efficiency on the Celtics

(Avery Bradley is Still not a Point Guard and an optimistic note on Evan Turner)

Over at the Nylon Calculus, I unveiled a little exploratory analysis I had been dong with SportVu data compariing those stats with offensive efficiency as measured by a stablized plus-minus system for offense called O-RAPM.  The SportVu measure that looked like the largest contributor was a measure of passing efficiency that divided the assists, hockey assists and free throw assists by the number of passes the player made, first developed by Nick Neuteufel and Aqeel Phillips.

In that vein, I thought I would increase my research efficiency a little by taking a look at how Passing Efficiency measure applies to the current prospective Celtics line up.  Sure, it's August, the NBA is in full off season mode, and the Celtics have a roster that feels likely to change.  But, there are no guarantees that it will change significantly.  Plus, applying the new measure to a prospective roster helps put it in context and often helps me understand the measure better.

The first thing of note that came out of the analysis for Celtics fans, is that Rajon Rondo is indeed an efficient passer as well as being a ball dominant passer.  Rondo rated as the 18th most efficient passer in the league last year but had the highest Passing Impact in the league due to how many passes he threw per 100 possessions, passes per one hundred possessions was also positively correlated to offensive efficiency in the study.

Applying the coefficients from the study to both Passing Efficiency and Passes per Possession, I came up with an estimate of Passing Impact, which is the estimated plus/minus of each player's passing on offense. [1. Measure is calibrated so that every player's Passing Impact is positive outside of turn overs, which was a separate measure in the study]

Below are the three passing measures applied to the Celtics on the current roster other than the incoming rookies and Chris Babb who didn't play enough possessions to be included in the study.

To put some perspective on the numbers here, the Passing Efficiency average of players in the study was .097.  In other words, for the average player just under 10% of passes lead to an assist of some kind.  The number of passes per player was 71, though that varied greatly by position.

• Not surprising that Rondo and Phil Pressey land in the top spots.
• Evan Turner comes in third, one thing to watch here is that Turner played in the Sixers' very fast paced, if sloppy, offense.  There may be some dual causation in the measure with pace, so we'll have to see if he maintains that kind of Passing Efficiency in a new system.
• Avery Bradley IS NOT A POINT GUARD!!!!!!!! He might look like one, he might on the small side and quick, but, no, he is NOT A POINT GUARD!!!!!
• Marcus Thornton had decent efficiency but was an infrequent passer, showing how much he played off of the ball.
• Kelly Olynyk had good Passing Efficiency for his position.

Position Adjustments

Passing Efficiency, like assists, is highly correlated to position, as is Passing Impact.  Here's the table with averages by position:

 

Using those we can adjust the roster numbers based the level of Passing Efficiency typical of the position, via Basketball Reference, except for Bradley who I changed to a two guard, see above! Any player with a position adjusted rating above one rates as above average for their position, anyone below one is below average for their position:

• Turner comes in as rated the highest over his position of anyone on the prospective roster.  If we considered him a two guard, where he played part of his minutes last year, he would still rate as above average.  Turner's more traditional distribution metrics assist to turn over ratio only rate him as average, possibly showing the influence of the fast but chaotic Philly offense with the worst turnover rate in the league.
• Olynyk is second in position adjusted passing efficiency, though listed as a center.
• Rondo comes in third in the much tougher point guard spot.
• Bradley, even as a shooting guard has well below average passing efficiency.
• Jared Sullinger is just a bit above average.

There has been some discussion over at Celtics blog on building a roster around a non-shooting point guard as Rondo is reputed to be, and whether or not, in fact, Rondo actually deserves that reputation.  I come down on the side that his reputation may be slightly over done, but there's no escaping that he's been a below average shooter from distance and improved mid-range shooting is only a partial salve to the line up spacing problems.

The answer to building around Rondo's shooting, if the Celtics decide to do that, is, of course, to surround him with shooting.

It's a little less clear what the synergies are with passing, and at this point there probably isn't sufficient data to do much with line up analysis, at least using the SportVu measures.  That said, my starting assumption is that having additional play making on the court doesn't crowd out the efficacy of the point guard's passing.

At this point, I want to bring in the more traditional assist measurements, in this case assists per 100 possessions and assist to turnover ratio.  Both are fairly well correlated to Pass Efficiency, though there may be some diminishing returns for both.

We can also bring in the rookies, Marcus Smart and James Young using their college numbers via Draft Express.  Smart had 5.5 assist per 40 minutes pace adjusted, approximately 7.6 per 100 possessions, and an assist/turn over ratio of 1.78.  Young had 2.1 assists per 40 minutes, just under 3.0 assists per 100 possessions, and an assist to turnover ratio of .89. I am somewhat more encouraged by Young's good assist numbers than discouraged by his meh asst/tov ratio given his age.

If we look at a possible Zeller-Bass-Green-Bradley-Rondo starting line up, that is four average to below average playmakers on the court with Rondo, with no real secondary play maker.  One also would have a fairly lackluster rebounding front court.  Even without roster moves, I think that either Sullinger or Olynyk end up getting the starting nod over Bass or Zeller. 

The issue with the Celtics roster, as one would expect without real stars, except, perhaps Rondo, is that they don't have any complete offensive players.  The shooting guards have limited playmaking skills, the one decent shooting wing, Green, also has limited playmaking abilities.  Meanwhile the better playmaking wings aren't shooters.  The closest players the Celtics have to complete offensive players are Olynyk or Sullinger, if Sullinger's outside shot starts to fall.

Here's hoping Olynyk, Sullinger or Smart have break out years.