Ottawa Senators

By thewordbird on Aug 3, 2008

That is very extensive and a level-headed approach. I love it and Volchenkov!

By thewordbird on Aug 3, 2008

oh yeah, lookin’ forward to reading those a great deal.

By Gerald Norton on Aug 3, 2008

Sorry about the mess, I’ll figure it out, lol. Please ask, and criticize away, it will only make it better to get other views.

By David Johnson on Aug 3, 2008

We can combine the two stats by creating a +/- type state by calculating Prev - Prod.

3:31 Lidstrom
-12:34 Campbell
-13:35 Chara
-19:09 Kubina
-19:34 Kaberle
-23:41 Meszaros
-24:54 Redden
-32:49 Bouwmeester
-35:50 McCabe
-42:33 Volchenkov
-50:38 Finger
-54:10 Phillips
-92:12 Smith

The problem with this stat is it isn’t considering comparable offensive and defensive stats as it considers points a player scores vs goals the other team scores. To be comparable it should use goals the players team scores vs goals opposing team scores. Making that modification is probably advisable, partly because being responsible defensively can allow the rest of ones team mates take more chances, and thus likely get more goals, in the offensive end. A player may not get credited with a point but may have directly or indirectly helped his team score. In the player rankings I developed I looked at goals scored while a player was on the ice and not just the players point totals.

By Gerald Norton on Aug 4, 2008

David,
Those are some great observations.
As for the goals for, I spent a lot of time thinking about that too. My rationale is that there are already (usually) 3 points assigned on every goal, making this a pretty robust record of influence on goal scoring, whereas there is no record for influence on goal allowing, another indicator of how stats are production oriented. As I only used Dmen, I “assumed” they played a role in any goal against, as it is their primary role to prevent this occurance. Providing attribution for all goals scored for seems a bit too prone to over attribution (as is attrtibution for all goals against, but there exists no alternative). I wish there was direct attribution for goals against. As it stands, the PROD side of the table is more robust, I’d rather maintain that, then water it down, to match the less robust PREV side of the ledger.
BUT, you do raise a very valid point that how you play can encourage offense, or discourage offense. This can be evidenced by the number of SH goals for Ottawa has scored. You say you already do this (goals for on ice, goals against on ice)? I will post a caveat explaining how the PREV side is based upon unattributed goals against.
I will also seriously consider an equally unattributed PROD chart, with the same caveat, then do the plus minus, as this would be a very interesting stat (do you do this already?) I’d be interested to see how much it changes the appearance of productivity for forwards, I suspect it won’t. But, it would have a huge impact on defensive players like Kelly and Phillips who would instantly benefit from un-attributed offence.
Thanks for the observations, the more the better.

By David Johnson on Aug 4, 2008

I have already created a rating system (see http://stats.hockeyanalysis.com/) that I feel is as good as any in evaluating the true benefit a player offers his team. What I do is look at how many goals are scored for and against when the player is on the ice vs when he is not on the ice. But I don’t do it in a traditional +/- way. What I look at is pairs of players and see how well they do with each other and without each other.

Let’s take Phillips as an example. See http://stats.hockeyanalysis.com/200708players/player0102.php. That webpage shows his even strength ice time with his team mates and against opposition players as well as the goals for and against when the players are on the ice together and when they are apart. In Phillip’s case, he played most with Volchenkov (894:53) and in that time the opposition scored at a rate of 0.603 goals per 20 minutes (GA/20). When Phillips wasn’t playing with Volchenkov opponents scored at a much higher 1.006 GA/20 pace and when Volchenkov didn’t play with Phillips opponents scored at a rate of 0.878/20. So clearly they were much better defensively together than apart.

When I take into account all teammates and all opponents for goals for and goals against I can get a clearer picture of who are the better players. The better players are the ones who help their team mates score more goals and give up fewer goals than they do when they aren’t playing on the ice at the same time and who hold opponents to fewer goals than normal and score more goals on an opponent than normal.

Make sense?

Now, the numbers on those links are just even strength numbers but I also use power play and short handed numbers and then aggregate them all to form an offensive, defensive and overall player ratings as well as an overall contribution (i.e. http://stats.hockeyanalysis.com/200708/northeast_playerrankings.php for the northeast division). The ratings are time on ice independent while the contribution rating includes a players time on ice (it’s not a straight rating*TOI calcuation but it is not far off).

By Steve on Aug 4, 2008

I understand the desire to eliminate goaltending and shot type/distance from your analysis of defensemen, but I disagree with you in considering them a non-factor in goal scoring.

You say that you don’t consider there to be a robust assessment of their relevance, yet at the same time you are damaging the robustness of your own assessment by excluding their impacts.

Obviously team-mates have a huge impact on these stats as well. If they had no impact on a player’s defensive stats, then a players performance should be independent of the team he is playing for. I would think analyzing performance for a player on different squads within the same season would confirm whether or not this is the case. Have you considered comparing the numbers for Brian Campbell both before and after his trade from Buffalo to San Jose last season? Or perhaps the play of Brad Stuart with Detroit and with Los Angeles? I know the sample size varies drastically, but I would think some trends would be indicated.

Interesting work though Gerald, and no worries about the chart issues. Keep up the solid work.

By Gerald Norton on Aug 4, 2008

Goaltending and shot quality have HUGE impacts on scoring, I would never discount that. What I do discount is the ability to accont for it without injected a variable that will actually confuse the data more then clarify it. This is where it’s up to the reader to put the stats into context, and why stats alone may be interesting, but can never tell the whole story. I do find that the more you try to tell with stats, the murkier the results become, not the other way around. By no means do these rational groupings tell the whole story, but they do tell a story, and I’d rather that, even when imperfect, the have a situation where it is too difficult to translate the results. Maybe what I’ll do is also provide some shot data, and SV% data, to lend some context.
As for the teammate effect, for sure. But again, so much can come into play, from health, to system style. The variables are unending, so I try NOT to account for them, as odd as that sounds. I dislike “all in one” solutions, or I should say, I yet to like one, if one could exist, and be robust, I’d absolutely love it!
Out of curiosity I’ll try to find some time to look at Gill, Corvo, Campbell and Stuart, and see what apparent impact team has to play.

By Gerald Norton on Aug 4, 2008

David, I like your look, it’s another great tool in focusing the picture. I personally like to keep things simple, as I find this does a better job of presenting a specific occurance. It would be a good idea to look at combining a few snapshots, then blowing them up to show more detailed takes. It wold be awesome if each blogger took responsibility for certain stats, because it is an onerous job compiling, sorting and rationalizing all the data!

By David Johnson on Aug 4, 2008

In my opinion, any tool to evaluate players that doesn’t take into account who the player plays with and against is a flawed tool. The most common argument I hear when someone mentions an offensive or defensive stat is ‘but he plays against the opponents best players’ or something of that sort. The problem is, no one is really any good at instinctively knowing how much that is true and what effect it has on a players stats. It is really difficult to accurately ‘put stats into context’ with hockey.

By Gerald Norton on Aug 5, 2008

I agree completely, that is why the context (teammates, opposition, ice time, etc) needs to be added, but not, I argue, in the form of “incorporated” variables, but rather identified modifiers, that can then be used when looking at a particular statistic, or even added, one at a time. There is tremendous value in examining the situations you present.
It is my intention to develop a stat that is based upon very few variables, thus producing a more attributable result. Even something as simple as points represents a stat with hidden variables that can be easily accounted for, but are not, and end up hiding some significant facts. No stat tells the whole story, and the more stories it tries to tell, the more variable interference it incorporates. Using a direct stat, then adding in variables, allows the evaluator to mix and match variables and more clearly see the impact of each, and more readily identify anomalous variables, or irrational groupings that have significant impact in altering the “impression” of that player. Basically, it’s a show your work mentality. The more stripped down it is, the more clearly you see what it is saying, then the evaluator doesn’t have to attempt to deconstruct the result in order to determine if a specific variable had an overwhelming impact on the result. It’s just a different way of looking at it. Nobody is right or wrong. I’m loving this discussion thou, it is making me re-think a lot of my rationale.