APBRmetrics

Hi All,

My latest post looks at ranking the performance of NBA coaches this year. The performance of coaches is a very subjective subject as we all know but I try and tackle the problem with as much data as possible. This is never going to be a perfect model and is really just a first crack at it using public available data from a few sources.

There are 3 key indicators I measure coaches on to give an overall rating out of 100 and allow for injuries.

Check it out at:

http://www.zigzaganalytics.com/home/nba ... f-the-year

This is a pretty great effort. Thanks for sharing.
Right now, it looks like a lot of bad teams are tanking -- giving away their talent. Those teams' coaches may not be in the running for COY, but other coaches will look better if they snagged some of that talent.

You've used the Vegas pre-season predictions vs the 538 current projections. These give you a projected over-achievement of 14.5 wins for the Rockets.
Vegas said they'd win 43.5, and only one of our 20 guesses was that low (mine).
http://www.apbr.org/metrics/viewtopic.p ... &start=165

Comparing the b-ref.com projection (56.0 wins) to our 20-guess average for Hou (46.8), they're just 9.2 wins over predicted.
That's a lot closer to what Was (+8.4) has done, by the same standards.

Currently, 'our average' error (and RMSE) are a bit better than Vegas', though the gamblers have recently closed ground.

Thanks for the feedback mike. The predicted wins was a tough one, hindsight I maybe should had gone with 538, but Vegas is always pretty accurate. I'll plug in 538 numbers and see what I get, I can see brooks maybe sneaking ahead. Don't see the top 3 changing though.

I had a thought on adding an "x's and o's" component where I look at conversions out of a timeout, it would play a part in the "pressure" component already.

I would need to play with the weights if I added X's and O's, still like wins in the 50-60% range and would split the other 3 over the final 40-50%

I don't know if you want to, but have you thought about trying to syndicate this article (or another version of it) to somebody (nylon calculus / fansided / the step back, the ringer, bleacher report, whomever)? Or maybe do an interview with nylon calculus, dunc'd on, locked on or ? The topic should get more exposure.

One issue is to what extent a coach's reputation is baked into the pre-season projection and whether it is close to accurate. It might be in Vegas estimate. It probably isn't in a numbers based projection separately (though it is implicit in the numbers from last season if he coaches it and the level of player changes isn't real large).

With actual vs. projected wins and player improvement aren't you in most cases mainly measuring possible change in coaching quality not coaching quality itself?

Did you consider using actual wins vs. expected based on point differential in addition to the other criteria?

Performance against top 10 (an eventual higher goal)?

Any targeted attention to performance of biggest minute lineups?

I have contacted nylon calculus in the past when Seth was running it, I would always be happy to syndicate and I agree this is a topic that is rarely discussed in the analytics word and really it should be. It's kind of seen as 2 hard, I'll keep trying to build on it from here. The timeout conversion is certainly 1 attribute I will add.

In terms of projected wins and a bias, I have no doubt Dallas for instance had somewhat of an inflated win projection just because they are Dallas and traditionally always make the playoffs (and I think because Carlisle always does a great job with very little, especially the past few seasons)

Wins vs +.500 teams and point differential are good thoughts To explore as I expand it. Just want to ensure I look at measurements with as much of an influence by the coach as possible.

Another I will look into is timeout management down the stretch in games. You find a lot of coaches burn timeouts early and dont have enough in the bank for critical moments in the final seconds of a game. It's a critical part of game management for coaches.

Time management of players, sub pattern is another, I will explore. But what is right and what is wrong is the question here.

When you add (point differential) and (pressure coaching, close games, timeouts, x's and o's), don't you pretty much get Win% ?

The same coach with the same team can have good luck in close games one year, then bad luck the next year. But probably not over many years, just one or the other.

We should get a feedback loop going, wherein coach ratings are added into our pre-season win predictions; and those are used vs. teams' ensuing wins, to rate the coaches.

They all contribute to winning yes, but if you just take win % then what are you finding out about a coach? Do they just have a loaded team? Could they be even better if a coach was making the right decisions down the stretch? Could they be better if the coach had better sub patterns, timeouts used wisely? Better inbounds plays? Maybe there is a certain area they could really improve upon to become a better coach. Take a look at the warriors winning % this year when I think it's 3 mins to go and within 5 points they are 0-7 If I recall. There must be a gap here somewhere with what they are doing, identifying what that is and rectifying might see them win more games.....especially in a tight playoff game

... this year when I think it's 3 mins to go and within 5 points they are 0-7 If I recall...

Very interesting. Around this time last year, their results were the opposite. The last 1/3 of the season and the playoffs, their luck was gone.

This is just the opposite of their Finals games: They were 6-1 in the middle of 2 Finals, 1-5 over the first/last.
So if luck evens out, Dubs cruise.