Midway through the season, we have come up with some interesting results. Interesting enough at this point that we wanted to start sharing them with you.
Goals created
Anyone who has ever tried to argue for the value of an offensive player will know the strengths and weaknesses of goals and assists. Points, and the scoring leaders, give equal value to assists and goals, more frustratingly to second assists and goals.
Having watched my fair share of hockey, I can tell you I've seen second assists that were equally as important or more important than the goalscoring act in creating the open net; however, I think you'd all agree this is the exception rather than the rule. The same applies to first assists, though to a much lesser degree really (as we're finding).
Our desire to paint a clearer picture of goals created was born of this dissatisfaction.
Those of you who follow the new wave in NHL statistics will know that goals created has been tracked and analyzed before this. In most circles, Allan Ryder gets credit for doing the pioneering work on this. Tom Awad, who is now the darling of Behind the Net for his GVT numbers, also uses goals created as a way to discern offensive contribution. Hockey Reference, a more mainstream statistical site, has a column called goals created on its player pages as well.
In deciding how we would proceed with our own project, we did a careful search of the backstory on this statistic. Much of what we found was exactly in line with our own thinking – the reasoning for wanting to record and promote this statistic, for example. But there was a let-down, a gap if you like. All the methods that had been proposed in the past, and indeed, those being used in the present are based on factors. In all cases, these numbers were picked out of the air. Allan Ryder bases his calculations on the assumption that goalscoring is 50% of goal creation, with one or two assists making up the other 50%. Tom Awad has decided that a goal is worth 1.5 times an assist. Hockey reference, like Ryder thinks an assist is half a goal.
All of these assumptions are fine, and the calculations do drive a different list of offensive players to the top. And, since the list seems intuitive (at the top, anyway), the statisticians are pleased with their assumptions.
There are obvious limits to what it can tell you, though; especially in the cases where these calculations are just applied to every goal (unassisted to two assister) equally by using compiled scoring stats across the league. In fact, in these cases, I found that it told me absolutely nothing new. Once I got used to how the factors shuffled goals and assists, I could pick out who was the winner from the scoring leaders simply based on their goal:assist balance. It might as well be called G + 0.5A.
The Lions in Winter Approach
With our grasp on what we liked from the established models and what we didn't like, we took some (indirect) advice from Allan Ryder and began to sort out our new method:
"A good statistic captures a great deal of data without destroying too much information."The first condition in our new approach is that we would be recording everything about each goal. This included when it was scored, at what strength, against whom, and who contributed. In addition, we recorded (in parallel) how the NHL scored the goal (number of assists).
That was the easy part. The next part was the important bit. We needed to create a way to actually break free of the NHL's record to add more data, more richness of information. The way we do it is by scoring the goals individually, by allocating points (an equal amount are available for each goal) to as many players as are involved. Importantly, this is a collaboration, so this isn't just what Topham saw, but what multiple people saw (averaged to give the score).
Now there's an obvious objection to this method: that it's subjective. It's something that can't be denied. But I have some thoughts on that:
1) The NHL already uses subjective statistics (hits, shots, giveaways, etc.) and we're perfectly happy with these, because we can appreciate the value while recognizing their limits.
2) Scientific research quantifies qualitative data like this all the time. Frankly, if it's good enough for the scientific papers I edit for healthcare decision makers, then it's good enough for a fun statistic in hockey.
3) There are many people who want more than goals, assists, PIM to spice up their reading and arguments. The league is already recording most "factual" events (with the notable exception of possession time), and sites like behindthenet.ca are mining just about as much as can be mined. Our goals created adds another category (albeit only for the Habs for now). Those who are satisfied with scoring leaders won't have that taken away from them, they will just have an extra stat to hold up beside that.
In taking things in this direction, we have taken that good advice seriously by losing nothing from what we had already, but have also added richness by capturing how a panel of evaluators thought contributions to each goal were made.
This gives our finding novelty (nobody has done this in this way before) and relevance, in that we are recording what happened, not what the NHL recorded multiplied by an imagined factor.
By products of the LIW approach
Strangely, our method also has something for those who promise to be it's biggest critics. By attaching value to goal contributions we have also accumulated a wealth of data on the average value of a goal and an assist. What's more, because we have recorded the data diligently, we can further sub-divide this by situation and category. So, for those who want to stick with a factor, we will at least be providing a better factor. We couldn't have designed the study much better had we only wanted to look for the value of assists, to be honest.
In addition, our approach can also be translated much more simply. For example, we can see who we rewarded extra assists to on a goal (beyond two) and we can see those assists that were just fluke touches and didn't enter into our contribution tally. We can see how many goals a player made a contribution on, and how many they didn't. We can see how many they were the major force behind a goal and how many times they were bit-part players.
So you see, even if don't believe in the number beside Cammalleri or Plekanec (though be open-minded, you can), you can believe in the value of the data...
Goals created on the Canadiens
Without further ado then, a taster on this data. I promise we will publish more, and do more with this as the season goes on...
Goals created leaders (Games 1-47)
| Player | G | A1 | A2 | LIW GC | % team total |
| Cammalleri | 20 | 14 | 3 | 18.4 | 16% |
| Plekanec | 10 | 19 | 17 | 15.8 | 14% |
| Gomez | 6 | 15 | 9 | 11.7 | 10% |
| A Kostitsyn | 12 | 9 | 4 | 10.6 | 9% |
| Bergeron | 10 | 4 | 9 | 8.6 | 8% |
| Metropolit | 10 | 5 | 5 | 7.1 | 6% |
| Gionta | 10 | 3 | 5 | 7.0 | 6% |
| Hamrlik | 5 | 8 | 3 | 5.1 | 4% |
| Moen | 7 | 1 | 2 | 4.5 | 4% |
| Markov | 3 | 6 | 4 | 4.5 | 4% |
As you can see, Cammalleri is a beast. And this looks less impressive after his mini slump these last few games. He and Plekanec account for 30% of all offensive production together. On game winning goals, they account for 45%. Incredible to think where we'd be without that signing and the accompanied turnaround.
Habs fan contributions
As we've said this is a collaborative project, so we are always open to more collaborators. If you are interested in helping us with ideas or by inputing your assessment of goals, get in touch via liwblog@gmail.com.
No comments:
Post a Comment