Value of advanced stats? (PER/TS%/Win Shares etc)

[dn0774]

For the purposes of this topic I am mainly referring to the most commonly used advanced stats, namely PER, TS%, Win Shares and Defensive Rating.

How indicative are these stats of actual ability versus the standard stats points, rebounds, assists and wins? Which specific advanced stats do you guys feel actually hold weight?

[mike-]

For the purposes of this topic I am mainly referring to the most commonly used advanced stats, namely PER, TS%, Win Shares and Defensive Rating.

How indicative are these stats of actual ability versus the standard stats points, rebounds, assists and wins? Which specific advanced stats do you guys feel actually hold weight?

Advanced stats are like any other stats. They may show something, but most of the time it isn't necasarrily the conclusion that people are inferring from looking at the stat.

For example, PER doesn't tell you who is a better basketball player, but you have people misuse it all the time to show that.

[illiance]

All stats are bad without context and actually watching games/players. Advanced stats are similar. They are a lot better than PPG/APG/RPG but they still are not completely reliable.

[doctorfunk]

personally I like TS%
if you'd look at Brent Barry stats who shoots 41%FG and you'd think he's mediocre scorer at best,
but then you look at his TS% and it turns out that in fact he's an elite scorer. Stats never say
the whole story you'd have to watch the game to judge a player but TS% tells you more than FG,3pt and ft, it not only combines it but weighs it as well, that's why I prefer ts% over traditional fg 3 and ft listing.

[Bgil]

The real question is what is the value of a box score. Most of the advanced stats talked about on this forum come from box scores and box scores are extremely basic and highly flawed game tallies.

The stats kept at Sports Synergy (I think that's the name), 82games.com, and basketballvalue.com are much better than what you find at basketball-reference.com

A level above that it actual X's and O's breakdown of every play. Nothing is better than that.

[mopper8]

The real question is what is the value of a box score. Most of the advanced stats talked about on this forum come from box scores and box scores are extremely basic and highly flawed game tallies.

The stats kept at Sports Synergy (I think that's the name), 82games.com, and basketballvalue.com are much better than what you find at basketball-reference.com

A level above that it actual X's and O's breakdown of every play. Nothing is better than that.

Synergy Sports gets as close to this as is possible in a quantifiable way, I believe.

As for PER, its an incredibly flawed stat. I basically use it as a catch-all way of saying, "so and so filled the box score better than this other guy," and even then its only really useful when the differences in PER are large. The difference between 29 and 30, let alone, 30 and 30.3 or what have you, are basically meaningless IMO in a stat that takes a composite of so many different things.

Bgil is right about asking what the value of a box score is, because PER basically compiles and standardizes box score production. While that does have some very limited value, again, I think we all agree that what is and is not tracked in a box score is largely arbitrary. For example, there is no reason that charges drawn and deflections can't be tracked in a box score. Assisted FTAs could be tracked, rather than just assisted-FGs. Heck, assisted missed FGs could be tracked, a sort of "assist attempt" to parallel a field goal attempt and free throw attempt.

Meanwhile, PER compiles a bunch of offensive statistics, but it throws in some defensive stats as well (Drbs, steals, blocks), even though it is universally acknowledged (even by the author of PER himself, John Hollinger!) that those are inadequate measures of defensive performances.

So, you compile point and assist and turn over production, total possessions used, offensive rebounds, team pace and minutes played and usage % and all that stuff to get what probably would be a pretty good compiled "offensive score" that would probably be very useful, and then...throw in 3 admittedly-inadequate defensive statistics to make it an "all-in-one" stat.

Just dumb.

An OPER I think would be pretty good actually, and using it conjunction with Ortg and Points/100 possessions +/- data could probably give you a very robust total picture about a guy's offensive value. But, we don't have that right now unfortunately.

[Agenda42]

Advanced stats are the best quantitative tools we have. That's not to say that they're "good", and they're certainly not definitive, but they're better than basic box score stats and you can learn useful things by studying them.

[Nivek]

In combination, stats reveal a lot. But, you need to look at an array of stats (and understand what the numbers are actually telling you) for advanced stats to mean much. Some of the advanced numbrers are merely better ways of measuring things that the "old-fashioned" stats measured.

For example, eFG is FG% while accounting for the value of the 3pt shot.

Rebounding percentages are an improvement over rebounds per game, or even per minute. And so on. Each tool has its value, but there is no Holy Grail of basketball stats. Despite what some of the adjusted plus/minus people (Wayne Winston) and some Wins Produced people will say.

[hasslinghoff]

their value entirely depends on the person interpreting them. as someone already mentioned, they can't explain context. lots of people like +/- data, but it's generally a pretty inaccurate stat as it requires a huge sample size ( several years ) to minimize errors. still infinitely more useful than per, which is nothing more than a messy fusion of common box scores.

[Nivek]

In addition to sample size, a problem with APM is that the assumptions made in developing the stat are poorly understood. It's a "black box" stat, meaning that unless you're doing the calculations you don't know what's in it. For example, most APMers give extra credit for "crunch time" however they define it. Leaving aside how it's defined, there's a tacit (but unproven) assumption that crunch time is more important than other parts of the game. By logic, giving extra credit to a play that happens late in a close game discounts a clutch play made earlier in the game.

I like +/-, and I do look at APM data. But I don't think it's as valuable as its proponents do.

[Doctor MJ]

For the purposes of this topic I am mainly referring to the most commonly used advanced stats, namely PER, TS%, Win Shares and Defensive Rating.

How indicative are these stats of actual ability versus the standard stats points, rebounds, assists and wins? Which specific advanced stats do you guys feel actually hold weight?

Typically advanced stats along these lines are things that take basic stats, and combine them for nuanced information. If they are combining basics to tell a fundamental ability, it's typically an improvement. If they are combining basics to give a general piece of information, then there is no better or worse, just different - and the ideal analyst should use everything available to him.

Of the stats you mention:

TS% - this is a case of an improvement. Using this is an absolute must.

PER - the best single stat to indicate per possession box score impact.

WS - the best single stat to indicate total box score impact.

DRtg - tread really carefully here. Make sure you really understand it before you use it, and know ahead of time that even Dean Oliver (the inventor) doesn't think that highly of it.

Additionally: +/- stats are huge. They don't have any inherent causality in them, so you've got to be really careful - but they are the proof-in-pudding stat. If a star goes several years in a row without getting a strong adjusted +/- score, you really have to question whether this guy is lifting his team up like he should.

Generally, the single best reason to me to use advanced stats, is the recognition that they are competing with basic stats, rather than against the eyeball. Basically all defensive accolades are a joke, and it's because even the best minds around (coaches, GMs, etc) will side with woefully inadequate basic stats over raw subjectivity.

[raptorforlife88]

[
Synergy Sports gets as close to this as is possible in a quantifiable way, I believe.

There's also Project defensive scoresheet, which involves someone watching each game and breaking down each play, tallying up stops, defensive possessions and what not. On the Raptors board there's a poster doing this for each game. As far as quantifying defense into a box score goes, this is the closest you can get.

viewtopic.php?f=32&t=1059243
viewtopic.php?f=32&t=1058824

[Nivek]

Defensive scoresheets are great. I did modified defensive scoresheets for Wizards games for the better part of 3 seasons. You learn a lot tracking games. You learn even more when you go deeper and track stuff like offensive play type. The funny thing about tracking -- the more you do it, the more stuff you see that could be tracked.

[Shinyhubcaps]

I like Win Shares as being a "most valuable player" argument and Player Efficiency Rating as a "most outstanding player" argument. The problem with PER is that sometimes one player is on the same plane as another who you would take 9 times out of 10 if you were actually building a team, but it's great because it does not seem that one's stats are cannibalized as much by his own teammates.

One thing I find fun is typing in random stats and seeing how many players achieved them, and advanced stats make that fun because PGs can be ranked alongside Cs, for example.

So let's look at NBA players to have seasons with 30.0 PER, 15 Win Shares, 60% True Shooting %, and 75 games.

http://www.basketball-reference.com/pla ... i?id=gRz3F

Sure, it just confirms what you already thought to be true, including 3 MVP seasons, but it gets a more interesting look at it than simple stats.

[numbers]

All stats are bad without context and actually watching games/players. Advanced stats are similar. They are a lot better than PPG/APG/RPG but they still are not completely reliable.

I partially agree with what you are saying but the problem is that most fans are not qualified (or should I say knowledgeable enough) to interpret what they are seeing on the floor. The preceding comment is obvious based on comments you see on the general board all the time.

In that case, stats are valuable because they are the one true objective measure of on court production.