APBRmetrics

[BobboFitos]

Not sure if others have worked on their projections or whatnot, but seems like this thread should start soon. (Reason I thought of this was http://sports.espn.go.com/nba/news/story?page=SummerForecast10-EastStandings - ESPN released their expert picks) I wont have mine up for a little while, but just testing the waters for whenever people want to make their picks.
_________________
http://pointsperpossession.com/

@PPPBasketball

[back2newbelf]

I think the list from ESPN is quite solid. I don't think it'll be easy to make a substantially better predition. My (very early) guesses where that list can be improved the most:

Milwaukee(-)
New York(-)
Washington(-)
Cleveland(+)
Portland(+, might be a gamble though)
San Antonio(+)
Houston(-)
Phoenix(+)
Sacramento(+)
Golden State(+)

[BobboFitos]

[schtevie]

Well, let me try to enliven in some small way these summer doldrums and be a bit mean to approximately half of "ESPN's best basketball minds". Looking at the forecasted Eastern Conference standings, I see very little wisdom coming from the crowd in expecting the Heat to win only 61 games. And this is not just to quibble with the predicted value (I take the over, by the way.) What I am having trouble imagining is any plausible reasoning that led the more pessimistic half (approximate) of their 93 experts to an expected win total that must have been what? 55 games? 57? And more generally, it is a bit saddening to contemplate the implication that there is little statistical reasoning whatsoever that influenced this crowd.

The average over the last 20 years (a reasonable reference, I think) for the highest regular season win total in the NBA is approximately 64 games. So, straight off, for perspective, what we are being told, is that a team that was 47-35 last year and then acquired possibly the greatest crop of free agents, in the prime of their careers, in the history of the league won't really even approach the average best record in the league. Even a "naive" prediction, for those pessimistic, simply applying the historic second best record in the league (60 games), say, would surely have raised the overall estimate.

Instead some reasoning was used that drove down the average. But what could it have been? I am going to suppose that the basic consensus among the pessimists was the same as the optimists, that as a result of the transactions, Miami would be no better on defense but considerably better on offense. The difference being that the pessimists believed that offensive improvements would only add 10 to 12 wins, max.

It is this implication that seems completely daft. And on this point, neither APM voodoo nor Wins Produced magical regressions need to be invoked. If one believes that the current collection of offensive talent on the Heat is greater than that of last year's Heat roster as well as that of the Cavs, Raptors, and Wizards (and everyone believes this, no?) the inference is that 64 wins should be an approximate lower bound.

Fleshing out the point, LBJ, Bosh, and Miller all had a TS% of about 60%. And collectively they used 50% of their teams' scoring possessions (FGA + 0.44 FTA). Miami, by contrast, jettisoned half their roster, players that utilized 49% of last year's scoring possessions at an average TS% of 53%. As a first cut then, one should expect that half the offense should improve by 7%age points. In terms of Pythogorean Wins, on a base of last year's Miami points scored and allowed, this yields 64 wins (64.2, if it matters).

Now, one can argue that the Juwan Howards of the world are going to get some shots, driving down the average, against which is the more persuasive argument, that with LBJ, Bosh, and Miller on the court, that they and others are going to see their TS%s increase.

The bottom line however is that approximately 64 wins is a huge difference from what the ESPN pessimists were predicting.

What is the larger point?

It is not unreasonable to expect, what with the ubiquitous internet sources of basketball statistics, and public discussions dating back close to twenty years, that all NBA-related media would avail themselves of what is on offer and thereby improve their craft.

Perhaps I am reading too much into a silly poll, however. Maybe the participants weren't really paying much attention, for the anonymitiy? How about ESPN shows the predictions of each participant? And to further incentivize excellence, a modest proposal: How about mimicking the organizational structure of English Football and dividing the 93 experts into three leagues (or more). The experts that best predict the Heat record can form the Premier League, the following 31 would be in the Championship League, followed by League One, or whatever. Then each year thereafter, the bottom and top of each table would be relegated and promoted accordingly. That should better focus the imagination.

But then again, the rant aside, maybe I am just plain wrong about the Heat. How does one get to 58 or so wins with that roster, though? What is the argument?

P.S. And how does the ESPN average for the middling improvement of the Heat coherently relate to the predicted collapse of the James-less Cavs?

[Crow]

If one was trying to make the expert panel's average a better predictor, I'd think it would help over years to vary the weights given to individual predictors or even knock some out based on performance as you did. Your structure might well be helpful- market just "the Premier League" prediction average or a variably weighted blend of the leagues.

Or there might be a more flexible or complicated or just different theory driven structure that would be even better that other trained stat experts are more able to suggest, describe and possibly apply.

I haven't really thought much about what my best guess for the Heat would be, at least yet. At first blush though I am not particularly bothered by an average prediction of 61 and I'd probably move fairly cautiously if I moved up.


Royce, do you feel you could post (this time or in the future with the notice and consent of the voters) some fairly high roll-up of the team prediction distributions for the predicted wins? Say the number of votes for over and under the average? Or the number of votes for 1 to 5 wins under or over the predicted average and the same for greater than 5 over or under the predicted average? You could even use large bands for reporting the results- say less than 10 votes in that win prediction range, or 10-30, or more than 30 or whatever to give some detail but not put anyone on a small island. I'd find that both interesting and helpful with regard to the Heat and other teams, if it were considered acceptable.


back2newbelf, can you clarify what your + and - signs mean? That ESPN is too high or low in your opinion or that you personally would adjust the prediction up or down as that sign indicates? I am not clear.

[Royce]

I might be able to provide more info later, when I have more time.

As for the Heat ... a quick count reveals that 26 voters predicted fewer than 61 wins ... another 35 predicted fewer than 65 ... just to provide a couple of benchmarks that suggest in this case it's a little "easier" to pull the mean down than to lift it ... which is fairly obvious.

High: 71

Low: 50

Average prediction: 62.4

Adjusted average: 61.2

_________________

Royce Webb
NBA Editor | ESPN.com

[erivera7]

I'd like to know who predicted 50 wins for the Heat on the ESPN panel. For the record, I was one of the panelists and predicted 65 wins.

50 wins? That's ... a little absurd, even from the most pessimistic view.
_________________
@erivera7
I cover the Orlando Magic - Magic Basketball

[Kevin Pelton]

I'm trying to come up with a technical term that explains why projections have to be regressed to the mean, but I can't think of such a thing. Suffice it to say that just because we know a team will win 64 games doesn't mean that whoever the best team is should be predicted to win 64 games.

[Crow]

Since 1980-81 there have been 41 teams at or above 61 wins, 11 at or above 65 wins, 6 at or above 67 wins.

[DSMok1]

[Crow]

If the Heat get to 67 wins that would at least tie for 3rd best out of 800 some team seasons in the last 30 years or within the top 0.5th percentile. 65 will at least tie for 9th or close to top 1%. 61 would only be about top 5% but if they hit 65 you only miss by 4 games, which is still good for any one prediction.

If one of the big 3 missed 15+ games more than expected or if they collectively missed 15+ more than expected that might knock a couple games off the predicted based on individual data, or worse than average luck in close games could, or perhaps other things (chemistry on offense and / or defense or whatever) might.

Going to 63 - 65 for Miami instead of 61 and being right would have a modest impact on the overall average error contest. It would knock about 0.1 off the average error. But if you are trying to get to an average error of 7 or significantly below 7 every 0.1 is valuable.

[back2newbelf]

I don't think the Heat care about more than having home-court advantage and might rest some players once they feel they're "safe" in #1. That would result in

max(# of wins of all other teams) + 2 or so

the "+2" is obviously debatable, since they're so good, they might even win 60% of their games when two stars are resting. Miami's strength of schedule for the last ~10 games might be important here for the final projection

[mtamada]

[schtevie]

I am hoping that Kevin and Mike will expand a bit upon their remarks. In particular, I don't see what "regression to the mean" or "loss functions" have to do with the issue at hand...at least the issue I raised.

The stable of ESPN writers were asked to estimate the mean of the ex ante distribution of wins for each team in the NBA. That 64 out of 93 came up with a number for Miami that was 64 or below is a bit disturbing, as it would seem to betray a lack of familiarity with or interest in statistics and statistical reasoning.

What can explain this result? It cannot be based upon what Crow seems to suggest, that one of the Heat stars misses an unexpected number of games. Because....that is unexpected, part of one tail of the distribution and not the mean.

Unless someone has some inside info, I see no evidence whatsoever that last year's numbers (in terms of games played and productivity, as a baseline) shouldn't be the basis for next. Opinions on the shape of that earth shouldn't differ.

Unless.

Unless there is really a widespread belief that the coaching staff of the Heat stinks. Or, put more kindly, that James and Wade are basically the same player necessarily implies that there is no way to have them on the court at the same time without significant decreases in their productivity (compared to previous years). And that no teammates will see better shots for their combined presence on the floor. But there is no such widespread belief.

And a final point. Estimating the Heat's record matters. Not in the sense that the ESPN poll matters. It matters because it provides as good of a test as one could possibly hope for. A test of one's understanding of the game of NBA basketball, offensive basketball in particular. There is no pretense of defense being a big part of the equation, so no black box there to worry about. And there are no age issues to inject additional uncertainty.

It will be an interesting season.

[Crow]

What specific assumptions are you making about injuries? The injury impact distribution will vary by team and the Heat could have more performance reducing injury impact than the average team.

James looks heavier to me and he is older. These things are both documented to increase injury risk. 2 of the last 3 years he has missed or passed on 6 and 7 games, the most of his career. 10 or more wouldn't surprise me. Bosh has missed 12 or more games 4 of the last 5 years. Wade has missed 20+ games in 3 of 7 years for a career average of 15. On quick inspection, that seems to add up to more than an average injury impact risk for leading players to me.

Diminishing return impacts for likely usage cuts for the big 3, shared rebounding, etc. can reasonably vary. Your in / out TS% change estimate for the offense is pretty rudimentary. How good will they be as a team from 3 pt land in quantity and FG%? None of the big 3 is anywhere close to even average on 3 point accuracy for their careers. That could matter in a way that overall TS% doesn't pick up exactly. They have other 3 point shooters but how many will they get up and how will the team averages turn out?

How well will they accept their roles and get along on the court? I can see a number of possible frictions that might hold performance somewhat below what a simple projection of past stats would predict.

Each additional win after 55 and 60 is probably increasingly tough (likely needing to win more against good teams to achieve this). A linear win estimate system might well predict too high.

Defense is hard to predict. Counterpart data at 82 games suggests Wade and James to be a little better than average 1 on 1 when facing a lot of starters, Bosh moderately worse. Another tool out there for individual impact estimates, but considering both 1 on 1 and help dimensions, was defensive 4 factors and it shows 2 of the big 3 to be weak (40th percentile or less) on 3 of the 4 defensive factors and 4 of those 12 ratings were in the weakest 16% of the league. It is hard to say how good the team defense will be with them in together at the core with a bit less room for defensive assistants to help than the Cavs or Heat could use as desired or needed with James and Wade. It is hard to say how quick the defense gets to its maximum or how consistent they will be.

How intensely focused and successful will they be in Miami-South Beach with all they want to do and will be asked to do off the court and with the constant media scrutiny? No other team will face this level of scrutiny and potential distraction or harm. It is also not unreasonable to expect that almost every team they face will try very hard to keep up or beat them, perhaps more so than against just any routine preseason #1 or #2.

The average wins of the two recent super teams (Boston and LA) who have won titles over their 2 peak was 62.5.

The average for the best team in the 11 seasons post the Jordan championships and the lockout was 62.7 actual wins. 5 times the leader was at 61 or less. 4 times at 65 or more. 62 was the median.

61 or 65. They are different. I am not sure yet which is better bet but I don't think it is so easy and obvious that 65 is better. It would take more a detailed presentation to convince me that 65 is clearly the better estimate and better than 63 as well. 63 seems a bit safer at this point to me on the surface but it would take very little in a detailed analysis to swing the estimate 2 games either way.