Friday, September 4, 2009

Rebound Control

Another Behind the Net post I have been meaning to comment on is this comparison of rebound shot rates for goalie tandems. Gabe Desjardins posted the data more or less without comment, and since I think the numbers are worthy of being discussed in some more detail I'll take a stab at it here. There are a few things that can be done with this data, but the most obvious one is to use it to estimate the range of individual goalie rebound skill. There are also numbers for two teams that have often been discussed at length on this blog (New Jersey and Atlanta), which gives us more evidence of the relative skills of those goaltenders.

There is an average of about a 1% difference among goalie teammates in rebound percentage at even strength. The average shots faced was 845, so that means that a typical difference in the number of rebound shots faced between two goalies who split starts right down the middle would be about 8. Figure that about a quarter of those shots go in, and that's a 2 goal difference in a platoon scenario. That's not a whole lot of margin, and given the small sample size those numbers are likely to be subject to a high degree of randomness and luck.

The biggest gap observed was between Jonas Hiller and J.S. Giguere. Of the shots against Hiller, 7.9% were rebounds (shots within 4 seconds of the previous shot), compared to 5.78% of Giguere's. That's a difference of about 18 rebound shots or likely around 4-5 goals, assuming both would have a similar rebound save percentage. If Hiller is giving away 4 extra goals per season from extra rebounds and Giguere has a .920 save percentage, than Hiller needs to be at just .924 or better to make up the gap.

There are a few factors that need to be considered before we go ahead and claim that the goalie with the lower rebound shot percentage is better at controlling shots. We would expect to see the largest rebound differentials when comparing goalies with high save percentages to goalies with low save percentages (e.g. Hiller vs. Giguere in 2008-09). The reason is that the goalies with high save percentages make more saves, and in particular will tend to make more tough saves. Not every shot has an equal likelihood of a rebound. Every goalie in the league is able to deal effectively with easy shots. It is more difficult to control the rebound on a difficult save, and therefore rebounds are more likely to come from tough saves. If rebounds come primarily from tough saves, and good goalies are more likely to make tough saves than bad goalies, then we would also expect the good goalies to give up additional rebound opportunities after stopping pucks that weaker goalies wouldn't have touched in the first place. This seems to be supported by the data, as the goalie with the higher save percentage generally also has a higher percentage of rebound shots against.

This factor likely explains the Hiller/Giguere differential. Both play a similar goaltending style and are not noted for their ability to control rebounds. Another potential factor is whether the team adjusts its defensive coverage to reflect a goalie's rebound control skill. I doubt that was the case in Anaheim, but it might be a possibility for some other teams where there is more of a difference in goalie skill sets.

Finally, there is likely to be differing abilities to stop rebound shots. Dominik Hasek, for instance, was an innovator in dealing with second-chance opportunities, and would certainly have been better than average at getting in front of rebounds. The problem is that the rebound shot sample sizes are so small that we would likely rarely know with any certainty if one goalie is actually better than another. I also suspect that in most cases the differences in rebound saving skill is relatively minor.

Taking all aspects of rebound control into account (both the number of rebound shots against and the number of rebound goals against), the overall effect is likely to be worth only a couple of thousandths to a goalie's save percentage. Rebound control becomes a useful tiebreaker for goalies with very similar abilities, but most of the time it appears that the goalie who is better at making the first save should play. This is especially true if the goalie's defensive teammates are able to make a defensive adjustment to cover the areas where rebounds are likely to end up.

Rebounds also seem to have a very limited effect on the number of shots a goalie faces. If Hiller faces 18 more rebound shots than Giguere in half a season, that equates to an additional 0.4 shots per game, and that's the highest observed value. For most goalie tandems, the difference is more like 0.1 or 0.2 shots per game. Assuming that my estimate of +/- 1 shot as the typical range of goalie shot prevention is correct, rebound control is apparently a very minor variable in the shot prevention equation.

Let's look at New Jersey and the results for Martin Brodeur and Scott Clemmensen, two goalies generally seen as on opposite ends of the spectrum of rebound control ability:

Martin Brodeur: 629 SA, 32 shots within 2 sec (5.1%), 42 shots within 4 sec (6.7%)
Scott Clemmensen: 869 SA, 33 shots within 2 sec (3.8%), 44 shots within 4 sec (5.1%)

What we see, somewhat surprisingly, is that Clemmensen faced fewer rebound shots than Brodeur, by both definitions of rebound shots. I don't think this means that Brodeur is actually worse than Clemmensen at controlling his saves, but merely that there are other factors involved. It is possible that Brodeur made more tough saves that were likely to lead to subsequent rebound chances, but given that they had similar overall save percentages that is not likely to account for the difference.

What I would speculate is more likely to be a significant is New Jersey's defensive play. It makes sense that the Devils would take a more active role in clearing rebounds with Clemmensen in the net, knowing that there were likely to be more second-chance opportunities. Looking over the pairings in Desjardins' post, it seems that the effect of team context is generally larger than the effect of the individual goaltender, i.e. there are generally larger differences between teams than between goalies on the same team. The ability for the team to compensate could explain why there appears to be little margin in goalie rebound control.

Brodeur was not only outperformed by Clemmensen, he was actually among the leaders in highest percentage of rebound shots faced. Could it be that his rebound control is somewhat overrated, possible team effects notwithstanding? It is possible that his teammates have done a good job over the years of allowing easier shots against (including fewer power play shots against) and clearing pucks and made him look better than he actually is. Or it might be possible that his age is having some effect on his reflexes and reaction speed. Or maybe Brodeur just had an unlucky season, or the numbers are a result of making more tough saves or his team's defensive play. Either way, I think rebound control has little to do with Brodeur's success in assisting his team's shot prevention.

The numbers for Atlanta are also worth discussing, given that the Thrashers employ the two goalies with the largest shots against gap in the league over the last few seasons (Kari Lehtonen and Johan Hedberg).

Lehtonen: 1068 SA, 40 shots within 2 sec (3.7%), 66 shots within 4 sec (6.1%)
Hedberg: 621 SA, 22 shots within 2 sec (3.5%), 38 shots within 4 sec (6.1%)

The Atlanta duo had almost identical numbers, which does not support the theory that the reason that Lehtonen faces more shots against per game is because of additional rebound shots. Lehtonen's numbers actually appear to be quite strong, given that his save percentage is so much higher than Hedberg's.

Looking at the big picture, these numbers seem to confirm my contention that rebound control is an overrated skill. Because of its visibility it is something that observers often focus on, but the actual rebound shot frequencies don't suggest that it is important as many claim.

This analysis does not consider any possible indirect effects of rebounds controlled or allowed, such as for example whether rebound control might be something that impacts a shooter's decision to shoot from a bad scoring location, or whether some goalies have an effect beyond the 4 second mark by doing a better job of directing pucks to their teammates instead of putting rebounds back out in areas where the other team can get to them and maintain possession in the offensive zone. I wish the NHL still tracked the time the puck spent in each zone, which would allow us to investigate possible indirect effects. I'd guess that these factors may have a slight impact but not enough to make much of a difference in overall terms.


Ryan said...

I'm not even sold that there's anything special going on here at all, though.

Here is a quick check I did. The columns labeled "av" are averages for all goalies, the columns labeled "sd" are expected standard deviations assuming that rebounds are binomially distributed, and the columns labeled z are z-scores.

The point is that the stdev of z-scores, for both 2- and 4-second rebounds, is near (and below!) 1, which is basically what you'd expect if this were all just randomness and goalie skill played no part at all in rebound control. (More accurately, if the deviations in goalie skill between the goalies in the sample played no part.)

Of course, that doesn't mean there's nothing to see in these numbers. It just means that you probably can't see whatever is worth seeing. In other words, if I used a computer simulation to make up the number of rebounds for each goalie and gave you those numbers side-by-side with real ones from a different season, you wouldn't be able to tell the difference.

overpass said...

Multi-year numbers would go a long way in providing a better sample size. Without that, we can't really have much confidence in any of the statistical results for individuals.

But yes, if there's anything to take away from these numbers it's that "rebound control may be an overrated skill", as you said.

Bruce said...

Interesting post, CG. That 2- and 4-second data is valuable and interesting in their own right, but cannot possibly tell the whole story of rebound control.

This analysis does not consider any possible indirect effects of rebounds controlled or allowed, such as ... whether some goalies have an effect beyond the 4 second mark by doing a better job of directing pucks to their teammates instead of putting rebounds back out in areas where the other team can get to them and maintain possession in the offensive zone.

Aye. What happens within 8 seconds? 15? 30? 60? Those to me are intriguing questions, albeit with diminishing returns.

I share your frustration in the NHL's decision to drop valuable information from their statistical package, however play-by-play still provides much context, namely what* happened next, where it happened, and when. (* = tracked events only of course)

Chances are, if the next event occurred in another zone, the goalie and his teammates successfully dealt with the rebound. Chances are, if the next event occurred more than 15 seconds later the goalie and his teammates regained control of the situation if not the puck. As for the what, there are a range of results of what could happen next within the defensive zone: goal against, penalty against, shot (opportunity) against, icing against, all indicating the other guys regained/maintained control and pressed their advantage. Perhaps there could be a surfeit of hits in the aftermath of saves, suggestive of more 50/50 pucks. One could sort out how many saves resulted in faceoffs, and whether those faceoffs led to better or worse results than plays kept alive.

It would, however, be an exquisitely difficult task to parse that information into meaningful conclusions. Hockey is a flow game which often unfolds simply through everyone doing their jobs, all of which will put 100 dB of noise on the centreband, leaving the signal (if any) to be teased out from the margins. Even then, only if a sufficiently large data set exists. (But so it is already with thing like Save POercentage, where "differences" between elite 'tenders are measured in thousandths)

I wish I had the computing chops to stripmine play-by-play data in search of such effects. It might be a gold mine or it might be a goose chase, but I'd bet on the former.

Of course at the end of the day, there would still be the entrenched argument about whether all these were goalie effects or team effects. As usual, and as the saying goes, the truth lies in the middle.

Host PPH said...

You are right. Numbers don't say anything without an interpretation. It needs to be interpreted to have some meaning.

yanmaneee said...

giannis antetokounmpo shoes
curry 7
kd shoes
kevin durant
nike off white
golden goose
nike lebron 16

baliski said...

get more click here to read click this site Dolabuy Valentino additional reading straight from the source