APBRmetrics

an article from daryl morey:

http://blogs.hbr.org/cs/2011/08/success ... _data.html

any thoughts on this?...

I agree, sort of.
Analysis is limited by data, therefore as data increases, possibilities in analysis increase (exponentially, I would assume).


However -- to "improve" upon the data would require an improvement in analysis, no?

bbstats wrote: However -- to "improve" upon the data would require an improvement in analysis, no?

I think what he's saying is that given the same data set, many analysts would end up making the same types of conclusions using similar kinds of analyses (and that analysts probably wouldn't agree with him). So the best way to get a leg up on the competition isn't to get another analyst but to get more or better data than the other guy. With better data even an ordinary analysis will provide information that other teams don't have.

Flawed but interesting.

Me in 6 months:

Every analysis leads to desire for more data.

One of my main discomforts with the article is that it is missing any mention or discussion of the importance of effective utilization of the analyzed data. Unless you have a strategy and implementation that successfully utilizes the findings of the analyzed data you have no value "thru-put". Search for knowledge, extract it, then ship it efficiently. At least three main steps, not just two.

I don't think basketball coaches have fully tapped out knowledge of player, human and group behavior at basic levels and how to do behavior modification effectively. Most players could be better even within the bounds of the existing data and existing analysis and their physical capabilities and existing skills if their ability to use knowledge and management direction was improved. Often the goal is using that guidance more precisely and fully, but sometimes it could also mean using it more supplely and contingently as the final intelligent agent of action.

I don't know the scale of potential values still to be found in data, analysis and implementation and the relative proportions, but I think / assume they are all still large enough to justify new, more and better efforts in each area.

I have some doubt that all potentially useful new compilations of contingent, complex data from existing databases have been fully createdm analyzed and utilized. Just counting an activity does not fully realize the value. You still have to figure out how to encourage the desired actions in the right way without creating new issues or disrupting other positives. "Just do it" doesn't cut it.

If one organization really believes that new data is "the key edge", at this point, for them and in their view in general over the other parts of the cycle and acts accordingly, ok, then after that determination / choice they'll need to show better performance results compared to the competition to support their contention. If they still heavily invest in analysis and implementation then the point of pitting the m against each other becomes less clear.

The further the part of the cycle is from the game the more opportunities for loss of value thru-put from the other parts of the cycle not given the same elevated rank of importance. Then again the value you can pass on is dependent on the potential value you had passed on from an earlier stage. I would think it best to try to maximize the value gain at each stage and to assume each has untapped frontier.

In a league without a cap on spending of data gathering, data analysis and knowledge application efforts, I probably wouldn't want a strategy that relied primarily on new work in one area for edge and to some degree signaled acceptance of current efforts in the others. If a team feels their team has reached or nearly reached its potential value production from data analysis from existing data, then one could try data analysis by new analysts who might find different methods and perspectives than applied to date. How many analysts do you try before you have "done enough" and accept a limitation on what is possible from data analysis or data analysis within existing data resources? Any team that were to limit themselves to just 1-2 analysts or just 1-2 high level analysts (as opposed to data assemblers) might be self-creating the limitation of returns from data analysis by that choice. How many years do you try 3,4 or 6+ instead of 1 or 2 to try to get edge? How much rotation in staff and approach do you try over time before you accept limits of analysis?

What is the typical % of revenue spent for R & D of a $100 million revenue business and how does the R & D budget of a typical NBA franchise and leading edge NBA franchises compare? A quick check suggests R & D expenditures in the general business world are affected by industry, sector maturity and technological considerations and firm size but I got the impression that average R & D expenditures for firms of this size was about 4% of revenues and in some cases was up to 10%. If one counts traditional basketball operation staff as partially R & D the amount spent will be higher than if you don't and basketball teams may also spend on sales R & D, but is there any team or more than a few that spends even 2% of revenues on what could be called dedicated basketball R & D? I'd guess that the average team spends well less than 1% on data collection and analysis beyond the traditional video efforts. I don't think add-on data analysis has hit a responsible ceiling many places.

What is the biggest R & D investment that a NBA team would make at the margin to win one more game or advance one more playoff round? How much are they willing to spend overall on R& D to win a title or even have a 1 in 4 shot? The willingness to spend for results will vary. I am not sure how good a sense teams really have about the margin return on the dollar in wins and yes actual revenues too from analytic investments but ultimately it is their call.

A team could also try new efforts to increase the effectiveness of knowledge application. New basketball coaches provides a new opportunity for different knowledge application practices and hopefully better results but one could still focus on enhancing that knowledge application thru monitoring, analysis and bringing new knowledge and teaching advice to the coaches.

The Mavericks idea of a stat coach on the bench and with the team every day, every minute is another strategy that could also be considered. In addition to being a resource person on hand to offer reminders about the data or the data analysis or the agreed upon ways to utilize that knowledge for the Coach to use as they want (the way the position sounded to me based on the few articles about it), it could / should / probably does have another side of plan monitoring for management enforcement of the plan with the coaches and modification of the plan as suggested by on-going data and data analysis. A team could have a very detailed plan (building from existing and new data into data analysis and then into decisions about what knowledge is to be applied and preserved against loss) that is prescribed to the Coach from the beginning by directive. There would be a range available for how much of the efforts of a stats coach would be devoted to each role. If it is too much or too politically difficult for one person to do both roles to the satisfaction (and perhaps comfort) of both, then having some form of always on the scene, separate plan enforcement monitor might make sense. There are different ways to run an organization and those choices about plan monitoring and enforcement will probably affect total value realization.

Now, it is of course true that teams spend on knowledge application thru their coaching staff and traditional support staff, but how much of that spending, time and action is really consistent with "the data" and data analysis and decisions from it? It often sounds like the coaching staff generally has its own views, goals and methods and that the input from data analysis is grafted on only partially where it appears the most promising or easiest to understand or do or most consistent with what they wanted to do anyways. Overall it sounds to me like the task of knowledge application from data collection and analysis still has a frontier with plenty of room for improvement in data, analysis, practice and results.


From the article: "The answer is better data. Yep, that's right. Raw numbers, not the people and programs that attempt to make sense of them."

Given all that is involved, I am not too justified with this simple statement or convinced by the presentation.

Better analysis often gives better understanding of what "better data" is needed.

Also, isn't it obvious that you would want data for the entire league? If your current analysts don't realize this, sure, you probably need better analysts.

Hockey assists, potential assists, good picks, ball saves, tips, block-outs, intimidated shots, help defense that stopped a shot or an open shot or a drive, FG% on open and contested shots, bailout fouls avoided and other and even smaller stuff, there is a lot not in the boxscore that is worth knowing about one way or another. But as for counting them, my main reactions are 1) what took so long, and 2) while they add plenty of useful information, these datum are not stand alone, the end of the road or independent of the need for data analysis (just as the boxscore was not) and efforts to improve knowledge utilization efficiency are still needed. Highlighting the value of new data gathering did not require denigrating data analysis or ignoring efforts to improve implementation of strategy and tactics consistent with the raw data and the best findings of the data analysis.

After you count a good amount of the uncounted activity, one still has to assign it value. There will probably still be uncounted activity with impact not covered by the new counting. Adjusted +/- tries to encompass it all. Adjusted +/- actor analysis adds some specificity and conceivably one could push further with data collected from video. I'd rather have both sets of data and lots of comparing and evaluating of the findings than just the counts.

If the article accurately reflects where the NBA really is right now, ok, go get that missing, overdue raw data. It is easier to get than it was before. But continue to address the rest of the responsibilities and opportunities as much as possible during and especially after the data gap is closed.

If missing raw data moves Coaches into action more than mildly or very sophisticated data analysis that says something about people and how to achieve change and not just the data. There might be ways to enhance receptivity to such data analysis. There could be edge to gain from finding those ways, whatever works. It may not necessarily have to come from explaining them in detail. It might help more to show that recommendations work (assuming they do) than to explain them much. It may help if they are simple and clear enough to be digested, remembered and used even if some fine points are left out in the process.

i just thought it funny/ironic to say that "...If talented analysts are becoming plentiful, however, then it follows that analysts cannot be the key to creating a consistent winner, as a sustainable competitive edge requires that you have something valuable AND irreplaceable..."

how are analysts any different than players (or coaches, GMs, etc)? if every team had 5-10 analysts how is that any different? every team has 12-15 players yet every year one team wins the title and one team has the worst record in the league...

new kinds of data is always good, like 82games.com, synergy, statscube, the various websites with +/- and adjusted +/- data, etc. but there is nothing "wrong" with box score data - it still tells you more than any other data set, and there is still alot of information you can extract from it...

kind of reminds me of this article:

http://blogmaverick.com/2010/10/03/buil ... ast-rites/

and this quote "...Until you can quantify coaching and chemistry, you can not use the numbers to build a team. Period end of story. You can use them as partial input along with scouting and other elements, but there ain’t no Moneyball solution for the NBA...". i don't believe there is any team that looks at the numbers and sees them as the only thing of importance. but it's funny how more teams than ever are now taking a hard look at the stats compared to previous years, and those teams are for the most part seeing varying levels of success...

... Until you can quantify coaching and chemistry, you can not use the numbers to build a team. Period end of story. You can use them as partial input along with ..

It's funny when someone says, "end of story" and then continues with the story.

Did Cuban write that?

Yes Cuban wrote that.

As more teams adopted a new layer of data analysis, he and a few other early and large adopters understandably scanned for new sources of edge over trying to catch up to them and new things to point to and maintain their public position as innovators.

Understanding coaching and chemistry impacts better than others are natural though challenging next level targets of study. By direct and probably pretty sophisticated statistical analysis of basketball data and probably by using other forms of social science knowledge and analysis. And taking a new step in bench staffing to try to improve the linkage between data discoveries and analysis and on the court implementation. I see the new data as an extension of current existing efforts rather than a real departure from them.


Morey's article referenced the value of new data to Amazon's success but it is hard for me to separate the value of that new data from the intensive process of extracting it and understanding it and figuring out what to do with it.