Context In Statistics (Monta Ellis Edition)

Sports statistics is a fascinating business. We have already seen the stats-based revolution in baseball, which has already had a major impact on many front offices. While basketball has seen some significant change in terms of player selection and personnel moves, there are a few different reasons that it has not had the same penetration into management as baseball.

One of these reasons is the nature of who has and gains GM positions in the two sports, yet a larger factor is that while baseball is a game of largely independent events, basketball is entirely personnel and situation-dependent.

Incidentally, there may be no better example of this incongruence than Monta Ellis.

As was discussed this summer with Kevin Durant, one of the issues with stats like +/- when it comes to basketball is that it can be heavily affected by elements like the quality of backups, substitution patterns, and statistical noise.

On Thursday, ESPN writer John Hollinger responded to a question about whether Monta Ellis should warrant All-Star consideration in his chat that the Warriors ?play dramatically better when he?s off the floor; that's a pretty damning statement there.?

In one way, Hollinger is right; the statement is pretty damning. However, his problem is that it is a prototypical example of how ignorance of the context surrounding statistics (whether deliberate or accidental) can lead to misinterpretation and misuse of the stats in play.

What makes the case of Monta Ellis different than Kevin Durant is that many of the suspicions I had about elements that biased his +/- that were argued for Durant have been issues I have seen in person with Monta thanks to having a media credential for the Warriors frequently this season.

The first issue centers on usage. Any argument about +/- inevitably compares the times the player is on the court with the time they are not. In some situations, there is some statistical value on both sides of the equation. However, the minutes Monta Ellis has played are more than atypical.

For the season, Monta has played 1,697 of the Warriors? 1,930 minutes (they get an extra 10 because of two overtime games). That?s a whopping 87.9% of the available minutes, but it gets even more insane. In the month of January, Monta has played an unbelievable 95.6% of Golden State?s minutes. Over the last 30 days, Mr. Ellis has been on the floor for less than 40 minutes twice- once for 39 minutes in a win at Minnesota and once for 35 in a loss at Memphis. That?s it. As such, any emphasis on +/- almost completely ignores the last month of the season and also provides little to no relevant data for December when he logged 43.3 minutes per game.

The other major issue is that the rare times that Monta sits are not randomly distributed. When looking at the games where Monta plays 35 minutes or less, a remarkable trend emerges. There are nine games that fit this restriction and eight of them were Golden State losses (and yes, that is below their typical winning %- .111 in these games compared to .355 otherwise). Furthermore, the average margin of defeat for the Warriors is 20 points in those games. From a practical perspective, what that means is that Monta sat out a lot during games that the Warriors got blown out of. This makes intuitive sense as someone who goes to a ton of Warriors games and watches the rest on TV.

From a +/- perspective, this also creates one of the most logical flaws for statistical biases: If Monta plays during the portion of the game when the other team?s core guys are killing the Warriors then sits down, he is not on the floor for when the winning team pulls their core and it is garbage time players on each side. In those periods, it is not remotely surprising that the Warriors hold their own as a team with great guys for that situation coupled with the fact that this garbage time is being juxtaposed (in the context of Monta) with a blowout loss.

As such, a vast majority of the minutes Monta has spent off the floor this season that could contain any semblance of statistical significance were garbage time in blowout losses that contains even more bias than the few minutes he is off in a game he plays 44+ minutes in. Amazingly enough, Monta has played 44+ minutes in 24 of the Warriors? 40 games, going the distance in ten games thus far.

Sadly, I still do not have access to minute-by-minute substitution patterns to completely justify this conception, but it has been backed up by prodigious personal experience this season. Should anyone have that information, I?d love to see it whether it supports this or not.

In essence, the problem with statistics is that they can easily be either misinterpreted or manipulated in order to justify a point. They still hold an absolute ton of value in the right hands, but the context surrounding the stats must be taken into consideration, whether that is because the guy plays on a fast or slow paced team, the quality of the talent around him (and how they play), or the time a player on the floor and when he is not being dramatically different. Unfortunately, there will be times when writers and analysts do not flesh out the story and that creates the importance of answering the ?why? question when a stat either does not make sense or does not reflect the readily-available reality.

I am incredibly interested to find out which shooting guard has had the third best season in the Western Conference thus far (behind Kobe and Roy, naturally) if it is not Monta Ellis.

Also see my piece on John Hollinger and the perils of fatally flawed analysis from last summer.