A Substitute for War

Basketball philosophy

Not so fast: Major discrepancy in quoted “clutch” performance.

with 3 comments

This is a quick post in response to some more of the details in Henry Abbott’s last post on TrueHoop about Chris Paul‘s clutch performance. 

Here’s what he wrote that blew my mind:

Over the last five years, in the final 24 seconds of games his (Chris Paul’s) team trailed by a point or two, or were tied, the Hornets have scored 102 points on 86 possessions (as of a few weeks ago). That’s an offensive rating of more than 118 points per 100 possession.

Remember that number. 118.

Now, consider that most of the NBA is below 85, and 27 teams are below 100. That’s a blowout.

Only the Magic and Blazers are even close (at 107 and 104, respectively). The Cavaliers had LeBron James most of that period, and come in ninth, at 96. The Lakers are 14th at 83. The Celtics rank 20th at 78. Steve Nash‘s Phoenix Suns are way down at 28th on the list, while the Rockets are dead last, with an offensive rating just about half of the Hornets’.

When people brought up Paul’s team’s performance in the clutch before, I went with it, but the idea that Paul’s Hornets’ in the clutch have an offensive efficiency rating of 118 and Nash’s Suns had an ORtg near the bottom of the league more than 40 points per 100 possessions worse just seemed crazy. So I looked up what 82games had in that same time period (last 5 years), here’s what I found:

As you can see, according to 82games, Nash’s teams have actually a clearly superior clutch ORtg when Nash is in the game, compared to Paul’s respective performance.

TrueHoop is using a metric based on the last 24 seconds of a game while 82games is using a metric base on the last 5 minutes, so it is not necessarily a problem with the data that they deviate so much.

There is however a question of credibility imho in the 24 second stat. We’re talking about a swing in ORtg between the two on the order of 50, and this is a stat where the difference between great and terrible is about 10. Either we’re dealing with a gap in performance so huge to be absolutely revolutionary…or we’re dealing with sample size too small to be meaningful on the part of the 24 second stat.

I’m going to have to see more data and analysis before I really let this influence my opinion.

NOTE: Edited with a better description of the two systems used to describe clutch performance.

Advertisements

Written by Matt Johnson

February 21, 2011 at 11:24 am

3 Responses

Subscribe to comments with RSS.

  1. The first thing I thought when I read that today was that the sample size borders on a joke. We are always hindered by small sample in these clutch studies, but 80 possessions. 80!? That’s less than a game.

    It doesn’t make much in 80 possessions to have a 60-point change in ORtg — BR won’t list O/DRtg in their team game log page for whatever reason, so we can’t calculate a STDEV easily, but it’s not that unusual for a good offensive team to have an ORtg of 75 on a bad night and then a 130 on the next night.

    ElGee

    February 21, 2011 at 6:06 pm

  2. Yeah this is just one of those cat-in-the-matrix moments. I had just kind of been assuming that the data people were quoting was large enough to be meaningful, so I wasn’t looking at things too closely.

    In +/-, we lament that the 8000 or so possessions in a season aren’t enough to make causal statements, and people’ve been using something with 1/100th of the sample size. Ouch.

    Interestingly though, the Kobe-clutch case is so open and shut, that even if this were the only piece of evidence suggestion he was human, it would still hold merit as a statement against certainty.

    However, this narrative people have been exploring about Paul being the true clutch king I think probably has to be thrown out.

    Matt Johnson

    February 21, 2011 at 6:18 pm

  3. Not too mention the fact that 24 seconds is an awfully small window with very different situations that are likely to produce very different efficiency outcomes. Getting the ball back off a defensive rebound with 4 or 5 seconds left and no timeouts is a very different situation than having :20 with a couple of timeouts. With such a small sample size, the clock/timeout/deficit scenarios aren’t going to be normally distributed among all teams.

    Nat

    February 21, 2011 at 8:59 pm


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: