APBRmetrics

Jose Martinez introduced me to this group and posted a few papers of mine he thought were interesting.

So I thought I would join group as a member and introduce myself as an old geezer with an interest in basketball stats.

A friend and I created a basketball measure some two decades ago, and actually ran the magicmetric.com site for four years.

Just for kicks I thought I would present the ratings of all the NBA players for last season.

You can find our rankings in the appendix of the paper Jose loaded here:

http://www.upct.es/~beside/Textos/MagicMetric_NBA.pdf

Hi Dick. Interesting to see your ratings again. Would be even more interesting to see them side by side with some other newer metrics. Do you have any thoughts about the new metrics and the various metric issue debates that have gone here (if you go back and look at them)? I assume you would see some added value in collecting Magic metric data for presumed counterparts (by game or from play by play) to get a bit more at individual defense?

Thanks for your interest. In the paper I compare the Hoopstat rating of the top thirty, but that is about it. Years ago, I made a pitch that the MagicMetric was a better player rating formula than most others, but I was too busy with my career to make spend any time marketing it. Now, being semi-retired, perhaps I can revisit things.

The MagicMetric does not measure every form of contribution, but I think it is hard to argue with the linear formulas at its basis. The result is that a different formula should be used for sixth grade players, than the NBA.

But my main interest is in the crazy coaching system, and the possibility that it might actually work at college and NBA levels. We had amazing success using it in Church and Industrial leagues, and I can't see any reason why it wouldn't work in the NBA. I think the Heat's rotation is too shallow, and it is not surprising to me that they tend to fade at the end in close games.

We creating the MM rating system thinking it was really good and worthy of note, and tried to promote it. The rotation system we just used to play ball for a decade. Looking back on the experience, the rotation system was the surprising success, and it would probably have worked just as well using TENDEX or any other player ranking.

Wow... very interesting paper. I haven't quite finished with it yet and I do have doubts about the rotation system transitioning to the NBA (perhaps the regular season but probably not the playoffs), but it's certainly a different perspective on rotations. I wonder if you could look at a team like the Spurs or Thunder (or Mavs/Lakers a few years back) who routinely play one of their top 3 players off the bench, and see if they aren't doing something similar to what you are talking about. I think that aspect of the system is the one that pro coaches will have the hardest time accepting.

The MM rating seems solid as well. I think people make too big of a deal about nailing down who the best players with numbers are when really all that matters is that you separate the top tier players from the 2nd tier and so on. Nailing down what specific rank players are at is probably more useful at the lower level when you have coaches looking at a summer league team or pre-season team and figure out who their 9th-10th guy might be... and even then sometimes coaches are looking for specific traits (rebounding, shooting) and not trying to decide who is really the better player.

agentkirb wrote:Wow... very interesting paper..

My heart soars like a hawk. - Chief Dan George, Little Big Man

I do think Scola is too low on the list. He is very solid post player, has great midrange game, and possesses great basketball IQ, durable and reliable. Finding him next to Boozer is just bad ranking. Evans and Jennings are players on the watch. Despite being so good in their rookie year, both declined since then. Ray is living on past accolades, should be around 90-100 in the ranking.

Taking a charge should be
valued at least as much as a steal as it results in a change of possession.

Why? One results in a dead-ball possession where the defense is set; the other is a live-ball turnover, where you're in a very favorable position to score. The alternative is the cost of either action, which isn't often reflected in the boxscore; an over aggressive steal attempt results in a wide open shot, and a bad charge attempt results in giving a foul. (And sometimes an And-1, worst possession type to yield!)

The game pace of one team might result in more possessions per
minute than another team. As the contribution tends to be related to the number of
possessions, a player on a fast paced team may have a slightly higher metric than if he
played on a team that featured more of a half court offense.

This seems easy to fix. Just make everything per possession rather than directly per minute. Unless you're making a subsidiary argument (that pace is somehow a positive thing, ie. playing more possessions/game is a good thing therefore you should not pace control)

The fact that a players per minute rating goes down as the minutes on the court goes up
seems indisputable.

Except it's very disputable. There has been some research done showing players that play longer in games actually play better. There is anecdotal evidence to support this as well - a player coming in "cold" off the bench isn't accustomed to the game speed/etc., and takes some time to get settled.

Some of the evidence though is selective bias, ie. players that happen to play longer do so because they were playing well. Not that they were in the game longer so that caused them to play well. Likewise, short stints often mean multiple mistakes (or an array of fouls) which causes the performance to look bad. But it wasn't the short minutes that caused it; the poor play caused the short minutes.

From my own perspective, (playing and so on) getting fewer than 10mins/g was not nearly as good as playing 10-20, and the break even point was somewhere above that. After 25 or so, I'm not playing 100% at all, and resting/taking some plays off.

If anything, I think playing time is more like a C curve as far as individual effectiveness, rather than anything linear.

Kinda neat idea about the rotations...

Not bad, it certainly passes the sniff test for its rankings. However I don't love the theory behind the calculation. Eg. A stl adding 1 possession? It seems like because of gamblers, a player with high stls a lot of times doesn't help a team. Likewise a player's contribution to an assist may be worth a lot less than .9. Also the 2s and 3s are weighted weirdly. Also while I appreciate the rebounding coefficient behind as low as .65 to make up for the teammates effect, there's certainly going to be a varying effect IMO of whether a player was responsible for that rebound or not - some .1, some .9.

As a whole it seems solid but I trust PER more tbh... and I don't like PER that much. Reading how you came up with it two decades ago, that seems like it'd have been a really cool thing to have back then, but I think Hollinger and the analytics movement kind of drank your milkshake

" Hollinger and the analytics movement kind of drank your milkshake" - Dr. Positively,

LOL. John Hollinger certainly has had more success promoting PER. Our analytic did not catch on. I have never examined PER in detail so I am can not really comment on it. However, the hoopstat ranking uses Hollinger's PER, and I kind of like our top ten player ratings for 2011-2012 versus PER.

We have the same top three Lebron, K. Love, and Durant, and four more in common, Dwight Howard, Chris Paul, Blake Griffith, and Al Jefferson.

But our rating system puts Kobe Bryant, D. Wade, and Derrick Rose in our top ten.

Instead, PER has these players in its top ten.

Andrew Bynum, Paul Gasol, and LaMarcus Aldridge.

I think I would trade Bynum, Gasol and ALdridge for Kobe, Wade and Rose...

"Except it's very disputable. There has been some research done showing players that play longer in games actually play better. There is anecdotal evidence to support this as well - a player coming in "cold" off the bench isn't accustomed to the game speed/etc., and takes some time to get settled" - bobofittos.

Good point. I really do believe you can have a higher performance playing 10 minutes than just playing four. That is one of the reasons my player rotation systems don't provide for 2 minute substitutions. But at the other end of the spectrum, does it hold true? Would Lebron James Perform as well if he were on the court the entire 48 minutes each game instead of only forty minute? This is not easy to measure scientifically. But our best player was in his late thirties, early forties when we used the rotation for about eight seasons. He normally played 32 of forty minutes, but several times, when not enough players showed up for the game, he played the entire game. Jeff and I, (Jeff being the good player and co-inventor of the rating system) believe his metric/minute was negatively affected when he had to go the entire game, but we never measured this analytically. Jeff thinks the effect is about .o5-.1 per minute, and he had a per minute rating over.6 in his early seasons, drifting down to about .5 in his latter seasons.

But a strong case can be made that Tim Duncan, who still has a .6/minute rating playing only 28 minutes, should not be rated ahead of Kobe producing .57/minute playing ten minutes more per game. If Duncan had to do that, it is unlikely he would keep the .6 rating. And if he could, then the coach should be playing him more.

agentkirb wrote:
The MM rating seems solid as well. I think people make too big of a deal about nailing down who the best players with numbers are when really all that matters is that you separate the top tier players from the 2nd tier and so on.

This is how we felt about our rating system. We were only trying to reward better players with more playing time, and nothing more than that. But we created the magicmetric site after using it for a few years because we thought it could catch on, but we ran out of gas keeping the website running for four years without any interest from some larger sports news group. Maybe we should have kept at it!

DickMays wrote:Jose Martinez introduced me to this group and posted a few papers of mine he thought were interesting.

So I thought I would join group as a member and introduce myself as an old geezer with an interest in basketball stats.

A friend and I created a basketball measure some two decades ago, and actually ran the magicmetric.com site for four years.

Just for kicks I thought I would present the ratings of all the NBA players for last season.

You can find our rankings in the appendix of the paper Jose loaded here:

http://www.upct.es/~beside/Textos/MagicMetric_NBA.pdf

Hey Dick - I remember Magic Metric back in the day. I also remember you doing it for college players - and for a while ESPN had the Magic Metric player score & rank on each player's page. How in the world did you even connect with them back then with that. Outside of a few basic conversations with DeanO - I've never had any major site ever email me back about my rankings... what kinda "magic" did you spin to get that attention?

It's nice to see you check in at the old APBR.

Statman wrote:
Hey Dick - I remember Magic Metric back in the day. I also remember you doing it for college players - and for a while ESPN had the Magic Metric player score & rank on each player's page. How in the world did you even connect with them back then with that. Outside of a few basic conversations with DeanO - I've never had any major site ever email me back about my rankings... what kinda "magic" did you spin to get that attention?

Dan,

Sportsticker picked up the MagicMetric and began using it out their site. I believe they sold it to ESPN. I contacted Sports Ticker, trying to get a small deal for them to us the stat, but they blew me off, and then I had my lawyer send them a letter and they stopped using it. No sales skill here, lol. I believe Dwayne Wade had the highest NCAA rating, higher than Carmelo as I recall. We were a bit before our time.

I'm curious about the coefficient 0.9 for assists. Does a player who makes the field goal only get 0.1 pts? Otherwise, how do you deal with the double counting issue?

Dick, I developed a player metric a few years ago called ezPM (I never say MagicMetric, but I think it's a similar line of thought that I used):

http://www.d3coder.com/thecity/2010/12/ ... valuation/

EvanZ wrote:I'm curious about the coefficient 0.9 for assists. Does a player who makes the field goal only get 0.1 pts? Otherwise, how do you deal with the double counting issue?

Dick, I developed a player metric a few years ago called ezPM (I never say MagicMetric, but I think it's a similar line of thought that I used):

http://www.d3coder.com/thecity/2010/12/ ... valuation/

Evan,

We have a paper, I think there is a link to it, where we explain how we solve for the coefficients. The player rating coefficients could be anything. The average player in the NBA has a MM/min rating of.4, but we could have set that anywhere, say to 1.0, as player ratings are relative things. But we decide to have the overall metric approximate the points scored, so for that reason, a 2 point FG comes out to being worth 1.8. A 3 point FG was slightly higher that 3, as I recall, but we rounded it off to make it simpler.

The biggest factor affect the relative values of contributions is the value of a possession. A sixth grade boys team will score less per possession, so scoring should be valued higher, and missed shots (and rebounds) are of less value. Doesn't make sense to use the same rating system for a youth team as college or the NBA. Assists are arbitrarily given the value of being worth half as much as a 2 point field goal. I believe we estimated that shots where an assist would be given, are made about 50% more frequently than shots where no assist is rewarded. Avg FG% was 46% and our estimate was made be watching lots of games. You could argue its relative worth up or down a bit without any argument from me. But that is how we got .9. Because a 2 point FG is worth 1.8.

Interestingly, the total points scored during a game does approximate the total metric, but the point differential tends to be about half as much as the metric differential. Not always, but its a good rule of thumb.

Dick