Advanced Stats & Metrics

[Deems]

Call me old-school, but this generations obsession with advanced stats & metrics is blown out of proportion. What are y'all thoughts on it?

I believe nothing is a better judge of a player than the eye test (observing a players basketball IQ in different situations, their shot/play making ability and most of all... their character, hustle, willingness to win.) The advanced science hoopla makes coaches, GM's, and even fan's overthink the game.

[DanH]

Call me old-school, but this generations obsession with advanced stats & metrics is blown out of proportion. What are y'all thoughts on it?

I believe nothing is a better judge of a player than the eye test (observing a players basketball IQ in different situations, their shot/play making ability and most of all... their character, hustle, willingness to win.) The advanced science hoopla makes coaches, GM's, and even fan's overthink the game.

The eye test is full of bias (even moreso than the admittedly problematic numbers we have).

Also, as fans in particular, we aren't able to watch every game from every player. To be able to make evaluations, we need tools at our disposal.

Stats solve both issues. Simple ones and advanced ones.

Of course, it is important to know what stats tell you - that is far more important than any attack on the objective quality of a statistic. For example, PER. PER is seen as a garbage stat by many, but is also one of the most used stats. It has issues (rewards volume over efficiency, doesn't account for quality of opposition, or workload). But all in all, it's a very good box-score aggregate for describing star power - the guys who are most likely to end up in an all star game, for example. Teams need high PER players - guys who can provide heavy production in various aspects of the game - be it rebounding, scoring, etc. But if used as a catch all player evaluator, it falls flat. So it is on the user to know how to evaluate what the stat is actually describing.

Win shares (and wins produced) are similar to PER - box score aggregates. But they rely more heavily on efficiency (though volume is rewarded to some degree). And wins produced in particular weight rebounding very heavily (I would argue far too heavily - I prefer WS as a player evaluator tool).

Then there are team "advanced" stats, which really aren't advanced at all. In the same sense that hockey advanced stats are not advanced at all. It's just raw stats with adjustments to level the playing field. For example, offensive rating and defensive rating. They are built using box-score and play by play data, but the end result is essentially just points scored and allowed per game, adjusted for pace of play (ie a team running up and down the floor in a 7 seconds or less offence scoring 110 points is actually a pretty bad offensive team). These team ratings are also calculated for players and lineups, based on the time they are on the floor, just to judge how the team performs with the player on the floor. All very simple, nothing much advanced to them. But they are among the best impact evaluators we have (impact being different from production, a topic I've opined at length about).

Then we have the family of adjusted plus-minus stats, the only really advanced set of stats in the bunch. They are an attempt to take those on-court ORTG ad DRTG's for players above (pure, clean data, but with lots of context and noise), and find out how much of those performance numbers the player is actually responsible for. Versus how much is just their teammates, their opposition, etc. They are incredibly imperfect, these stats, and should be taken with a great deal of caution. There is the most pure form of them, APM, adjusted plus minus, which is a model of hundreds of variables (each player in the league's offensive value and defensive value) all balanced against eachother, so when you weigh each player's time on the court with each of their teammates and each of the opposing players they have played against, every equation balances, and you are left with each player's true offensive and defensive impact. To say the least, this is incredibly unstable and noisy, and players can have wildly different values from year to year.

So advances are made in the APM approach. Multiple year models are done, to stabilize the result (2-year APM). A ridge-regression technique is added to further stabilize the results and eliminate extremes (2-year RAPM). But more required data means no current day stats - no measure of how a player is performing right now. So introduce the box-score based and box-score prior (a way to bias the regressions to require less data to be stable by assuming a baseline value for each player based on their box-score data) APMs - the most popular being BPM (box-plus-minus, a box score regression with multi-year APM results) and RPM (real-plus-minus, a regression model with box score priors, based on the 2-year RAPM results). Both are once again reliant on box score data, but are tied into the work done on APM which was totally box score independent. The result is a stat no one really understand well. Except that it is a box score influenced attempt to judge the parts of the game that are not captured in the box score.

All of that to say - a player's basketball IQ is only of value if it leads to results - results that help his team win. That would show up in APM data. A player's shot making ability shows up in their efficiency metrics, including the catch-all stats like PER and WS. Their character, hustle, and willingness to win are of great value - but only in so far as they actually lead to their own team outscoring the other team in one way or another.

Advanced stats don't tell us to value anything differently than what you are saying - they don't tell us to devalue a player's hustle, or their ability to shoot. They are tools for better understanding the value of those things in the big picture, and better evaluating just how good a player is at those things. No one says to disregard the eye test - especially the eye test of qualified evaluators like NBA head coaches and GMs. But more information is better information, and those who make their assessments with data AND their eye test will usually outperform those who rely on only one.

Uhh... Those are my thoughts on it.

[Deems]

^^^
That's the most educated thought on a topic I've seen here. I guess my question to you would be, could team management be swayed by these advanced metrics to the point of creating a bias, just like an eye-test could?

For example lets say you have someone like Kobe Bryant, who in the advanced stat's world I don't think seems like a top 3 all time greatest player; and then someone like Kevin Durant; let's assume both of em are the same age in their first real breakout season. A GM proposes a trade of Durant for Kobe Bryant, statistically speaking Durant is the better player, but I think 90% of us agree who we would like to lead a team, defend the best player, and take the last shot in the game. Would this new found focus on advanced stats skew a team's perception of talent?

We can have smaller examples such as Lucas Nogeuria, advanced stats seem to think he's a defensive force; but watching him play tells otherwise. Draymond Green seems to be average offensively, yet he's the main force that drives that Warriors machine.

My biggest issue is not these new metrics, which I do believe are important and useful; it's more so the obsession with them.

[DanH]

^^^
That's the most educated thought on a topic I've seen here. I guess my question to you would be, could team management be swayed by these advanced metrics to the point of creating a bias, just like an eye-test could?

For example lets say you have someone like Kobe Bryant, who in the advanced stat's world I don't think seems like a top 3 all time greatest player; and then someone like Kevin Durant; let's assume both of em are the same age in their first real breakout season. A GM proposes a trade of Durant for Kobe Bryant, statistically speaking Durant is the better player, but I think 90% of us agree who we would like to lead a team, defend the best player, and take the last shot in the game. Would this new found focus on advanced stats skew a team's perception of talent?

We can have smaller examples such as Lucas Nogeuria, advanced stats seem to think he's a defensive force; but watching him play tells otherwise. Draymond Green seems to be average offensively, yet he's the main force that drives that Warriors machine.

My biggest issue is not these new metrics, which I do believe are important and useful; it's more so the obsession with them.

I'd take Durant over Kobe 10 times out of 10, for the record. I might not be the intended audience with that analogy.

I agree with you on the Nogueira scenario. Therein lies the need to understand what the advanced stats are telling you. He has a lot of context in his stats (purely a backup, huge shot block percentage that skews the regression hard) that muddy the water.

I don't think anyone who is serious about stats has this obsession with them. Those who truly understand this stuff (and I wouldn't really count myself among them, I'm speaking of the great minds that are developing these models and end up working for NBA teams) always emphasize the need for context and consideration of the eye test. Even the guys who write stuff for stats websites and things like that, tend to be the first to say they watch at least a subset of plays or games that they are analyzing, so they know what happened that the stats are describing.

I guess I don't really know what you mean by the obsession with these stats. I've rarely seen anyone make sweeping declarative statements based purely on stats - typically they are used to support a position already gleaned from watching the games.

[Deems]

What I meant by the league's obsession of the advanced stats and metrics is the game is being driven to be played in a certain style that analytics show is the 'optimal' way to play. It's just a direction I'm still not used to I guess.

[DanH]

What I meant by the league's obsession of the advanced stats and metrics is the game is being driven to be played in a certain style that analytics show is the 'optimal' way to play. It's just a direction I'm still not used to I guess.

Oh, well I think if analytics don't expose truths about the game that aren't obvious in the first place, what's the point of them? I think it is great that the game is evolving in ways we can really appreciate now with these stats allowing us to better judge what's happening out there.