APBRmetrics

[BobboFitos]

I asked this friend/acquaintance of mine if he minded if I posted this, got the green light, so figured I'd go ahead and do so... (essentially the perils of the box score; there have been posts about scorekeeper bias before)

[erivera7]

Interesting story. Thanks for sharing.
_________________
@erivera7
I cover the Orlando Magic - Magic Basketball

[jmethven]

Fascinating... and frightening. I was curious, so I looked up the Van Exel game the author is referring to (1/5/97) on B-R. According to the box score he assisted on 62.6% of the Lakers' field goals, while committing just 2 turnovers. His 23 assists that night rank as the 12th highest single-game total in NBA history as well as his personal career best.

I had always assumed that most statistical errors would even out over the course of time. But that was assuming there was no bias in the discrepancies. The existence of Van Exel's '23 assist' game suggests otherwise. Intentional errors, especially if widespread, could create a lot of problems.

I wouldn't jump to drastic conclusions regarding the validity of box score-based metrics, but it's clear that there needs to be a greater push for consistency. Perhaps a new APBRmetrics game charting project is in order?

[DraftGuy]

rofl! statistics are used right if u ignore blips and use lots of data. i might change the stats if deke glared at me!

[socialpsychologist]

If this account even has a shred of truth to it, then it could explain some behaviors, such as Mutumbo's finger wag. He could be talking to the scorekeeper as much as anybody else.

[Harold Almonte]

If all of this isn't an exaggeration, then we must say it is negligible a proffesional league manages that in a so informal way (taking account a pair of assists on records might mean extra dollars in a contract year- or might not mean anything given the bad decissions GMs eventually make). We could even say this might be a kind of "steroid-ed" stats NBA scandal.

[BobboFitos]

[Harold Almonte]

Scorekeeping can't affect the game outcome (scoring is not subjective), but it can affect the market, not other players directly, since counterpart stats aren't recorded (blocks against are actually recorded).

[Mike G]

[devin3807]

On the flip side to inflating your players stats for coverage, what if a cash-strapped team made it harder for their contract year starting point guard to get assists?

[kjb]

[mtamada]

[IrishHand]

First - I am the author of the quoted material in the original post. Some interesting replies and feedback that I'll respond to.

[kjb]

By the way, maybe scorekeeping is much better of late, but the claim of 20 unintentional errors per game is exaggerated in my experience. Over a 3-season period, I hand-tracked at least a couple hundred games as part of my defensive charting project. I tracked both home and away games that involved the Wizards, as well as a smattering (10-20) of games in which the Wizards didn't play.

In charting defense, I recorded the games on TiVo and jotted my notes directly on a printout of the play-by-play. I did find errors, but nowhere close to 20 per game. Even the least accurate play-by-play I used did better than 20 errors.

I did find evidence of bias -- generally in the awarding of steals and assists. I never found a case like the one this scorekeeper describes (ball being deflected out of bounds and then awarded as a rebound to the player closest to the ball).

Probably the strangest stat award I saw was when Kwame Brown received an assist for setting a screen. It was a bone-crunching screen, but I've never encountered another screen-assist.

There are subjective plays that could reasonably be scored one way or another. For example, a player gets stripped of the ball while he's bringing the ball up to shoot. I've seen that kind of play scored a steal or a block, and on most of the plays I think the scorekeeper got it right (or at least arguably right).

But an average of 20 errors per game strains credulity. Unless, of course, my 200+ game sample just happened to catch each scorekeeping crew at their most accurate.

[IrishHand]

I'm using a very liberal definition of unintentional error - basically any time the game log deviates from what I'd consider a "perfect" game log. In my experience, the most common tend to be things like:

-wrong player awarded stat (rebound, turnover, steal are easy to mess up w/ tips/deflections in each - catalyst for stat should get it, not ultimate beneficiary)
-missed event(s) - anytime there is a cluster of events (mutiple missed offensive rebounds/putback attempts...turnover/steal/turnover/steal sequences, that sort of thing) there is, in my experience, a much higher chance of unitentional error

Keep in mind also that when I say 20 errors, I'm counting something like (ball stripped from player going up for shot on drive --> game log shows block, rebound) as four errors. Two bad stats awarded, two actual stats not recorded. I consider every "bad" or "missed" event an error. That may be double-counting, but I was concerned with both ends at the time.

Also, from an analytical perspective, that "20/game" number is soft and really only for the sake of discussion (as is the 'double that in intentional errors'). I'm estimating it ~12 years after the fact based on a fairly small and uneven sample size (perhaps 12-15 games, mostly involving teams I cared to watch). If I'd known that the original e-mail was going to end up being scrutinized by people who cared about the underlying numbers, I would have been a lot more conservative (or tried to find the spreadsheet that I had back then with the actual numbers - flawed though they may have been). Also, those numbers are based on a pre-NBA (meaning I hadn't done an actual game), ivory-tower view of stats. I had college experience, which was pretty pure. I had the NBA rule book and related guidance on stats, and I had game tapes.

Perhaps scorekeeping is much, much better now than 12 years ago, but given the consistently strong human element, I tend to doubt that current game logs would be down to <5 errors per game. That's basically screwing up one or two sequences out of several hundred recorded events.