APBRmetrics

[HoopStudies]

I did a little side project recently on how teammates fit in the sport of frescobol (aka fresco ball, beach paddleball, smashball). I got inspired while playing on the beaches of Brazil when I, as a good general athlete who could chase down a ton of balls, still wasn't a great teammate with some people.

It inspired a bit of math, but mostly through examples and not theory. And it inspired a paper, which I'm making available in its draft form here.

It's only kinda relevant to basketball. I don't think it does any calculations relevant to basketball (which is way too complex for the type of analysis in the paper), but the concepts are things I think we think about.

If you have comments, you know how to reach me.
_________________
Dean Oliver
Author, Basketball on Paper
http://www.basketballonpaper.com

[Mountain]

I skimmed the article and probably will read it more carefully later. I can see it is a useful example for talking about the concept of fit and tracing out its implications on a competitve league.

The statement near the end: "When there is little player movement, player value estimates can be influenced by context. Our league studies showed that player values estimated through regression led to sometimes misleading (and often optimistic) numbers ..." should be considered in the basketball world.

The paper also tends to support my belief that restricted trade action tends to leave teams less optimized than they could be. I generally support aggressive team trading and easing of trade rules in the NBA.

I have seen fresco ball or some loose version of paddle ball on the beach. At the very beginning you say "Our discovery of the game was on the beaches of Brazil, where good players hit the ball hard and straight, keeping the ball alive for up to about a minute." Playing straight and fast may be the way the players enjoy the game the most, played casually for a high level of action and excitement but I wonder if the point was duration and the competition was serious whether a more badminton like slow arching style would do better? Would slowing the pace improve efficiency?

Do the players of fresco ball need to be matched by pace preference? If you analyzed play and drops with a stop watch and shot count and calculated pace of action might you find the correlations of speed and performance for different players and player pairs significant? If so maybe the 2 factor model of athleticism and consistency could be modified to incorporate pace preference? Adapting to pace of partner may be part of consisistency and athleticism right now. Breaking it out separately sounds promising to me. Pace preference is sometimes significant in the basketball world for determining quality of teammate fit as well.

[HoopStudies]

[Mountain]

"I describe two forms of the game in the article, one where long rallies are best and the other where you maximize number of hits in a given time period. I analyze the former."

Right, I saw that. I followed your selection of the former.

I wasn't criticizing a 2 factor model as much as just noting a possible enhancement - a pace factor- which happens to garner attention in basketball (whether that be too much, about right or not enough).

[HoopStudies]

[Mountain]

Carrying the topic back into the NBA a bit further, a GM evaluating player performance for a potential acquisition should be evaluating goodness of fit in prior situation and its impact on performance. Players from good fits have that advantage on players from bad fits but some expectation of adaptability / ability to fit can keep part of the responsibility on the player. Transferability of performance (overall or if the focus is particularly on scoring, rebounding or passings) will depend on the goodness of fit on the new team. Truly similar fits should be easier calls. Different fits offer more of a challenge. As much as every player is different I would think that having a player typology and drawing upon knowledge of asserted "similar" players and similar contexts and the results would factor into the decisionmaking. GMs may have an informal typology and carry much of the data in their head but analytics can add depth and value on the handling of fit.

[Roland_Beech]

Brilliant.

Conceptually as you mention it has a lot to do with basketball, although we have the added complexity obviously of having to optimize the team makeup for a variety of opponents aka "the matchup".

How would you approach Tennis with something like this? That seems like it could be conducive to the next step up: you could have consistency-athleticism-placement-power as four player characteristics overlaid on a variety of shots (serve, return, forehand, backhand, approach, volley, lob etc) along with surface I suppose (which interestingly you didn't discuss in reference to frescobol, but surely the nature of the sand could be a significant factor with sand that's heavy and hard to move on perhaps placing a premium on consistency, etc)

Anyway I found it really really good and thought provoking although the mind reels some at how many 'skills' one might come up with to describe basketball player characteristics...

The one criticism I would have is that I know it's in keeping with academic paper style, but all the p11 stuff makes it a lot more cumbersome to read versus just spelling it out in words: "% chance to hit the ball from the sweet spot" ..."% chance to return the ball to the opponent sweet spot" etc

[tpryan]

I've glanced over the paper briefly, which is all I have time to do at the moment. My only comment is a very minor one. Fienen listed as "Visiting Scholar" caught my eye since that is generally not a formal university title. Presumably the intent is just to describe what he is in general terms and if that is acceptable to JQAS, where I see the paper has been submitted, then that's fine. I'm sure that some journals would want his formal title to be listed, which I believe is Postdoctoral Research Fellow. Again, a minor point.

[HoopStudies]

[Mountain]

I think the focus on teammate fit could be brought back to basketball, using the paper it as inspiration. Here are some rough thoughts on one way:

Look at passes that lead to a shot (from tape), first from PGs (but it could be done for others). It could be just passes that directly lead to a shot or more liberal, covering all shots, recognizing passes that lead to drives.

Divide the court into zones, perhaps using NBA hotspots.
For each shooter categorize the hotspot as a good shot or a bad shot for them using some criteria perhaps league average eFG%. Or lower. If the study were very detailed you could take into account time left on shot clock and grade more shots as good shots. Shot attempts could be split between "open" and contested".

Deliveries of a pass that lead to a shot attempt would be compiled and scored (for passer and shooter) by hotspot, open/contested, good/bad quality based on experience (this season/last or career?) and made or missed.

Players can be typed simply at outside shooter, inside scorer, both or neither.

You could describe how well the passing to the shooters appeared from an expected points perspective vs actual.

Not sure how to fairly divide "blame" for departure from expected but maybe you put it mainly on the shooter for an open shot and perhaps more evenly shared for a contested one (for a pass that maybe shouldnt have been delivered?).

The main praise or blame for the passer would be based on relative performance on delivery to right player, right situation, right zone.

Shooters with good zones that are underutilized certainly could be stressed by coach to players for their discretion but it also could guide playcalling. Shooting too often from bad zones are either a problem of the shooter, passer or coach or some combination.

Teammate "fit" would be a passer/shooter combo who get the ball to zones and make shots in an efficient manner. The performance of a PG with the various shooters on a team could be compared and guide which PG they play with in the rotation. Going beyond rollup shooting percentage or team +/- and making heavier use of the shot chart. Trades could be evaluated on basis of player ability to make shots from spots where a passer can get the ball and vice versa.

Just a sketch to start and perhaps build on.

The rest of the paper's discussion about strategy for fit and success in a competitive league would seem within reach if the fundamentals were developed fully.

[Mountain]

Where the PG can get the ball might not strike as that important to study for itself compared to makes & misses but where he can get ball and to whom and open or contested essentially grades the offensive prep system and his running of it.

The fresco ball study would also prompt me if I were working for a team to track the location of turnovers for passer and receiver (and the relationship involved even if the fault appears on a single party) and I would adjust the offense if necessary if there are persistent trouble spots where passing is not being made safely enough or handed safe enough thereafter.

Now as for actual new use of Markov chain method in basketball (beyond what was presented at that past conference)? Maybe look at offensive play sequencing. Does what you called in 1-3 previous plays (track location of the eventual shot and the player who did it and whether it succeeded) affect the success rate of the new play?

[HoopStudies]

[Mountain]

I think the questions underlying my brainstorming are these:

Are player and team shot charts optimized?

Do the PG and shooters fit together in terms of where the shots are coming from, the quality of shots and the results with consideration of turnovers experienced?

Is the playcalling optimized?

I ended up not really needing the frescoball Markov chain method for the first 2. As I said it mainly provided inspiration to focus on passing and successful response to passing and the teammate passing / shooting fit.

Maybe it fits the last question, maybe not.

Sorting and looking at the data doesn't necessarily have to a focused question at the very start but focused questions are good to have at the start or to find along the way of course and I think these 3 might be the main ones I'd organize research efforts around.