Individual Point Percentage for 2011-12

Scott Reynolds
October 12 2012 09:38AM

Photo: Michael Miller/Wikimedia/CC BY-SA 3.0

The 2012-13 season was to have begun yesterday. It didn’t. That’s mostly extremely annoying, but one of the silver linings is an ability for me and get around to doing some things that we had perhaps intended to finish earlier. One of those things for me is calculating the individual point percentages for the 2011-12 regular season.

Individual point percentage is a calculation of the number of times an individual player gets a point (either a goal or an assist) relative to the number of total goals scored while he's on the ice. So, for example, if a player is on the ice for fifty goals-for during five-on-five play over the course of the season and he gets a point on forty of them, his individual point percentage would be 80%. I was first introduced to this statistic by Tyler Dellow who figured (quite rightly) that points alone don't tell us who's driving the play in the offensive zone, and that this little statistic might help.

A lot of the really good offensive players in the league should find a home near the top of the chart (which is exactly what you'd expect from a statistic that’s trying to tell you who’s driving the play in the offensive zone). But as with most things based on sample size of less than 100 (no one was on the ice for more goals-for at five-on-five than Steven Stamkos at 87), the year-to-year variation is quite high. Over a long period of time, the cream mostly rises to the top, but in any one season, you can get some funky results, similar to the kind of thing we see with shooting percentage.

The average individual point percentage for a forward who was on the ice for at least 25 goals-for (there are 258 players) is 69.1% and the median is 69.5%, while the standard deviation is 8.3%, so that should provide some idea of whether or not a player is doing well (i.e. those players above 77.1% or below 60.8%. The data I'm using is from five-on-five play only and the raw data comes from Gabriel Desjardins' behindthenet.ca:

As you can see, there are a lot of very good players on this portion of the list. Players like Henrik Sedin, Ilya Kovalchuk, Marian Hossa, Mike Ribeiro, and Ales Hemsky put up these kind of numbers consistently, so we can be pretty confident that they’re driving the play. Others like Logan Couture, Jordan Eberle, and Pierre-Alexandre Parenteau don’t have a long enough track record to say either way, but are obviously off to promising starts (though Parenteau’s number should give Avalanche fans pause about his new deal). Still others like Kyle Brodziak and Matt Halischuk probably just don’t belong here. Really, this isn’t much different than most statistics: a dollop of common sense will go a long way.

There are a few players here that we’d likely expect to be a bit higher. Steven Stamkos just missed the first cut-off, but others like Jarome Iginla, Daniel Sedin, Sidney Crosby, and Alex Ovechkin (to name a few) also feel like they should be in the top group. When I take a look at the longer term trends on Monday (combining the data for the last five years), we’ll have a better idea if this looks like bad luck for some of the league’s superstars.

There are a lot of pretty darn good players on this list, many of whom have teammates on some of the league’s top lines: Joe Pavelski was relying on Joe Thornton to drive the offense, Jaromir Jagr was relying on Claude Giroux, and James Neal was relying on Evgeni Malkin. Some of this has to do with role (Pavelski has generally been quite good by this measure, but wasn’t consistently paired with Thornton until 2011-12), but even with that caveat, I’d prefer to keep (and pay) the guy who’s driving the play.

Even if you’re not a very good player, you’ve had some bad luck to end up at the bottom of ths list. Antoine Vermette was having an awful season by the percentages in Columbus, and it got him traded. Rene Bourque was also moved, but he waited until arriving in his new destination before going into the tank. Dave Bolland didn’t get traded, and in fact, managed to maintain his sterling reputation as a defensive center despite seeing his IPP fall to the very bottom of the list.

This obviously isn’t the be-all end-all for analyzing which players are driving the play (or being impacted by luck), but I do think as one tool among many, it can be very helpful. As I mentioned earlier, I’ll take a look at the longer term trends on Monday.

Previously by Scott Reynolds

889688bd2efc9140bec34eeba34eacde
Avatar
#1 David Johnson
October 12 2012, 10:38AM
Trash it!
0
trashes
+1
0
props

It probably makes sense to multiply IPP by GF/60 (goals while on the ice, not individual goals, per 60 minutes of ice time) to identify which players are integral to generating a lot of goal production and not just happening to be the best player on the ice (sometimes the best of a weak bunch). This would drop out the Brodziak's and Halischuk's from the top of the list and push Stamkos up the list closer to where he should be.

Avatar
#2 Bob Spencer
October 12 2012, 11:45AM
Trash it!
0
trashes
+1
0
props

So you're saying we should massage the stats so they favor what "common sense" would dictate. Yep, sounds like a great plan. I'm not going to try to argue that Brodziak is as good as Malkin, and certainly to be > 2 SD above the mean takes some luck, but just because the numbers don't fit preconceived notions doesn't mean you can just fl dismiss them outright.

Avatar
#4 Jonathan Willis
October 12 2012, 01:11PM
Trash it!
0
trashes
+1
0
props

@David Johnson

I completely disagree.

A big part of this exercise - for me, at least - is to identify which players are riding an unsustainable IPP rate. A player's IPP over several seasons gives you an idea of their capability (as does common sense, for more recent careers) and so by organizing them this way we know it's likely that the Brodziak's of the world will see a regression in total points.

If you then multiply by goals for, you eliminate that. Besides which, there's a more obvious problem:

IPP = 5v5 Points/60 divided by 5v5 On-Ice Goals/60

If you then multiply that total by 5-on-5 On-Ice Goals/60... you just end up back where you started, at PTS/60. Which is a useful stat, but hardly a useful place to stop such an exercise at.

Avatar
#5 David Johnson
October 12 2012, 01:43PM
Trash it!
0
trashes
+1
0
props

@Scott Reynolds

Yeah, it does make the actual numbers somewhat meaningless but essentially what IPP is doing is identifying which players are better offensively than their line mates, not who the best offensive players in the league. By factoring in GF/60 you are asking who is better than his line mates (high IPP), and who plays with really good line mates (high GF/60) which in combination should do a good job of identifying the best offensive players in the league.

Avatar
#6 David Johnson
October 12 2012, 01:58PM
Trash it!
0
trashes
+1
0
props

@Jonathan Willis

"A big part of this exercise - for me, at least - is to identify which players are riding an unsustainable IPP rate. A player's IPP over several seasons gives you an idea of their capability (as does common sense, for more recent careers) and so by organizing them this way we know it's likely that the Brodziak's of the world will see a regression in total points."

I am not convinced that IPP is telling you that Brodziak's riding an unsustainable IPP rate. It's telling you that he is a better offensive player than his line mates and on his line, most of the offense goes through him. Last year he played mostly with Nick Johnson, a 26 year old playing his first full season in the NHL, and Darryl Powe, who is basically a 15 point NHLer. Compared to them Brodziak's looks like a very good offensive player. I suspect if Brodziak played with similarly weak players next season he'd probably have a similarly high IPP. Now, put him on a line with Heatley and Parise and I suspect his IPP will fall significantly and it will have nothing to do with his performance, his ability or luck but rather almost solely due to a coaching decision to move him up in the lineup.

Avatar
#7 Eric T.
October 12 2012, 04:14PM
Trash it!
0
trashes
+1
0
props

@David Johnson

I am not convinced that IPP is telling you that Brodziak's riding an unsustainable IPP rate. It's telling you that he is a better offensive player than his line mates and on his line, most of the offense goes through him.

I'm not sure whether you read the article by Tyler Dellow linked in this article, but when he introduced the concept, he presented this data to suggest that IPP is dominated by luck at the season level:

http://4.bp.blogspot.com/_YYb9OLgfeBg/SxAYEmeXPOI/AAAAAAAAAAc/itob2tFTP0k/s1600/boa1.bmp

Brodziak having a IPP% of 88.7% last year suggests that we should expect him to land at perhaps 71-72% next year. You'd have to argue that the overwhelming majority of people with a high IPP one year get stronger linemates next year and that the overwhelming majority of people with a low IPP one year get weaker linemates the next year to explain the observed very strong regression to the mean as being the product of talent relative to linemates rather than luck.

Avatar
#8 Jonathan Willis
October 12 2012, 05:00PM
Trash it!
0
trashes
+1
0
props

@David Johnson

Brodziak was actually blended through the lineup - he spent nearly as much time with Heatley as he did with Powe.

Avatar
#9 David Johnson
October 12 2012, 09:25PM
Trash it!
0
trashes
+1
0
props

@Eric T.

From Tyler's article:

" It certainly doesn't seem too - see the chart at left, although there's a caveat to that, in that it seems to me that a lot of the players who are recognized as "star" players seem to have some repeatability in this department. I suspect a lot of the real goons fall into the same category."

There is a reason for the above. The stars always play on the top line, are always better than their line mates, and are more likely to have stability in their line mates. Conversely, goons are always among the worse players on their lines.

Now, for the more middle of the road guys, they probably see their roles change more often, move from team to team, play with different line mates, etc. One year a second liner may be an injury fill in on the first line and be the worst player on the line. The next year when the regular first liner is back and healthy he drops back to the second line and is now the best player on his line. This sort of thing is not conducive to year to year stability in their IPP, but has nothing to do with their talent level or necessarily their luck level. Certainly luck may be a factor, but I am not convinced that luck is necessarily the driving force in season to season variability.

Avatar
#10 David Johnson
October 12 2012, 09:28PM
Trash it!
0
trashes
+1
0
props

@Jonathan Willis

Yeah, he played some with Heatley, but not enough to offset how much he played with weaker players. Nick Johnson 687:17, Darroll Powe 463:01, Heatley 305:32, Clutterbuck 297:49

Avatar
#11 Eric T.
October 13 2012, 12:32AM
Trash it!
0
trashes
+1
0
props

@David Johnson

So I take it that means you _are_ arguing that the overwhelming majority of people with a high IPP one year get stronger linemates next year and that the overwhelming majority of people with a low IPP one year get weaker linemates the next year to explain the observed very strong regression to the mean as being the product of talent relative to linemates rather than luck?

His chart shows the group of people over 0.86 (average 0.883) and the people below 0.41 (average 0.374) ending up within 0.05 of each other the next year. That's not just a little bit of movement towards the mean; it's an almost complete return to the center. If you want to argue that IPP isn't predominantly luck at the single season level, that that return to the center is the product of changes in circumstances rather than luck, it isn't sufficient to say that some of the middling players change lines -- you need to argue that really systematically, almost everyone with a high IPP moves to a better line next year and almost everyone with a low IPP moves to a lower line. I would be skeptical of that claim.

It may be true that some people can sustain above- or below-average IPP%. That doesn't preclude the single-season data being heavily influenced by luck; it doesn't mean you have to argue that someone showing up with a high IPP% in a single year should be expected to do so again if he plays with the same linemates again.

Malkin had a 75.1% IPP last year (when he spent 27% of his time with Crosby and the rest was Cooke/Kunitz/Talbot/Dupuis/Letestu/Asham/etc) and 88.8% IPP this year when he played with Neal 81% of the time -- is Crosby really _that_ much better than Neal? Ribeiro was 86.4% this year (playing with Ryder and Eriksson) and 76.1% last year (playing with Morrow and Benn) -- is Benn that much worse than Eriksson? Thornton had 83.8% this year (with Marleau and Pavelski) and 59.7% last year (with Marleau and Heatley) -- is Pavelski that much worse than Heatley?

I'm not saying there's no talent here. I believe there is, but that there's also an awful lot of noise. So I'd be awfully hesitant to try to invoke complex causal explanations for someone popping up or down the list from one year to the next.

Avatar
#12 David Johnson
October 13 2012, 08:15AM
Trash it!
0
trashes
+1
0
props

@Eric T.

"...is Crosby really _that_ much better than Neal."

Ummm, yes. Until playing with Malkin Neal had never had more than 55 points. Crosby over his career has average 1.4 points per game. Crosby is light years better than Neal. But also Crosby is a similar player to Malkin. Malkin and Crosby are both puck carrying centers and playing them together moves Malkin to the wing and deprives Malkin of being the main puck carrier on the line. Playing with Neal allows Malkin to play center and be the main puck carrier on the line.

"I'm not saying there's no talent here. I believe there is, but that there's also an awful lot of noise."

I am not trying to say that there is no noise either. Clearly there is noise and we know goal data is highly random over short periods of time so that is no surprise. What I am question is the relative importance of noise in year to year fluctuations as well as the value of IPP once noise is eliminated/accounted for. Is IPP a better indicator of luck/noise than other metrics (i.e. the percentages)? Does a high IPP automatically mean the player is lucky? Also, if we are able to rid ourselves of the luck component of IPP, what remains? Probably not so much an indicator of talent but more an indicator of talent relative to talent of line mates which is far less useful. That fallout of that is it may mean good third liner has a similar IPP to a good second liner and thus even noise eliminated IPP won't tell us anything useful (other than players that might deserve a promotion up a line or demotion down a line).

There may be something here with IPP. I just don't think there has been near enough work done to know when and how to safely and correctly draw conclusions from IPP be they conclusions about luck or conclusions about talent.

Avatar
#14 David Johnson
October 13 2012, 12:42PM
Trash it!
0
trashes
+1
0
props

@Scott Reynolds

"I'd say that a very high IPP (something over 85%) does automatically suggest that the player has been somewhat lucky since no one can really sustain that. The same would be true of an established NHLer with a very low IPP (something under 50%)."

Probably true. I'd be curious to see what Crosby's IPP is over the past 5 years. I suspect he is probably at or near the top of the list so his IPP should give you a sort of theoretical maximum. I think it is important to identify what is a normal long-term range for players before we start drawing conclusions about how lucky they are. If I get a chance I'll dig into some numbers myself but kind of busy with real work right now.

Avatar
#15 Eric T.
October 13 2012, 02:05PM
Trash it!
0
trashes
+1
0
props

@David Johnson

"...is Crosby really _that_ much better than Neal." Ummm, yes.

So it's your position that Crosby is so much better than Neal that the difference between a guy's linemates being Crosby 27% + Neal 0% and his linemates being Crosby 8% + Neal 81% is what would explain him going from 75% IPP to 89% IPP?

Even if Neal had no impact on his IPP, had no more of the offense go through him than the Cookes and Kunitzes he was replacing, arithmetic says that Crosby having this impact would mean that Malkin's IPP% without Crosby would be 95% and with Crosby 25%. Assuming Neal is a better offensive player than Cooke would only increase that spread.

It's just nonsense to argue that Crosby has that large of an impact. You're trying to make the data fit your position instead of fitting your position to the data.

I am not trying to say that there is no noise either. Clearly there is noise and we know goal data is highly random over short periods of time so that is no surprise.

So having acknowledged that the single season data is full of noise, why are you trying to argue that Malkin's total jumping up 13% is a product of a small change in how much time he spent with Crosby, or that Brodziak having the second-highest percentage in the league is the product of playing with bad linemates? Why do you look for narrative to explain fluctuations in a measure that you know has a lot of noise in it?

Does a high IPP automatically mean the player is lucky?

Perhaps not, but if the noise component is much larger than the talent component, then trying to craft narrative around single-season IPP is a fruitless exercise.

If you take a list of high IPP and say "these players were all good (or better than their linemates)", then you'll be getting a lot of false positives. If you try to guess which players are the false positives and argue which players are lucky and which ones are good, then you're not really using IPP at all -- you're just reasserting your pre-formed opinions and selectively accepting IPP when it matches those opinions.

Also, if we are able to rid ourselves of the luck component of IPP, what remains? Probably not so much an indicator of talent but more an indicator of talent relative to talent of line mates which is far less useful.

This is an assertion which may or may not prove to be true. Fortunately, Scott has articles looking at this planned for Monday and Tuesday, so we'll learn more then. I'd hold off on making conclusions about what it means until we see the data.

Comments are closed for this article.