SDI v2.1 - Shut Down Index 2012-13 - Part 2

So yesterday I posted Part 1 of this year's release of SDI (v2.1). Generally the gist of the re-introduction and re-calibration of the statistic revolved around the idea that I was switching away from Corsi QoC and QoT statistics on the Behind The Net site to Team CA20 and Opp CF20 statistics from stats.hockeyanalysis.com. The main logic behind this being that the latter site's data sets strip out ice time involving the player being analyzed and thus mitigate against the impacts they would make on the input values for their expected results.

Questions

The posting yesterday led to some questions in the discussion section and on twitter which I would like to address. Some wondered if SDI correlates more highly to subsequent seasonal data than a player's CA20 or Team CA20. The correlation for individual players for all 3 values are shown in the chart below - using data for defenders who played 3000+ mins of 5v5 TOI during a 3 year sample from 2007-10 and another 3000+ mins of 5v5 TOI from 2010-13.

As you can see, the correlation and predictive power for SDI is far higher than that for CA20 or Team CA20 independently. SDI is basically accounting for 40% of its own future outcome while CA20 and Team CA20 only give you 15-17% of the information you need to predict them going forward.

This isn't to say SDI is perfect - obviously it's still far away from telling you EXACTLY how a player is likely to perform defensively in the future, but it's an improvement over our current methods of assessing defensive performance on the basis of Corsi events against irrespective of usage.

Another discussion point raised the idea of whether SDI is a better predictor of Raw Corsi Against than CA20 is itself. The reason I am not responding with the relevant data in this case is because it seems illogical to me to assess the predictive power of SDI vs. a raw stat that is used as an input value to SDI.

The whole purpose of SDI is to assess defensive performance on a Corsi event against basis while factoring in player usage. If we compare it to results that again ignore usage, the meaning of the stat is in many ways lost. Obviously if a player's usage changes, their Corsi events against are going to change.

If they play with worse line mates or against tougher opponents or they get more Defensive Zone Faceoffs - they're GOING to have more Corsi events against. If their ice time conversely gets easier, they should see fewer events against. The question I'm trying to answer is - does a prior SDI result predict future SDI results - that is - do we know who should CONTINUE to outperform or under-perform defensively in comparison to their usage. Largely I feel the aforementioned correlations indicate the fact that SDI has significant power in this area.

The Fun Stuff

Ok - so all of the lead up discussion and responses to questions finally brings us to the part most of you are waiting for. The actual information on who is good and who isn't based on this measure. Before I throw up more charts and data I'd like to just present the equation I used to perform the analysis.

1.127443*(Team CA20) + 0.537060*(Opp CF20) + 0.137217*(DZFO%) - 15.949396

This resulted from a multivariate regression analysis using the three variables I mentioned in the prior posting. The co-efficients have been rounded to fit them into the posting - but if you wish to re-calculate the values I'm guessing that 6 decimal places should be sufficient.

The Top 20 Shut Down D Men

So here's where we get into it - the top 20 Active NHL Defenders in terms of SDI Score, who have played over 3000+ mins in the past 6 seasons, while being put on the ice for Defensive Zone faceoffs over 30% of the time in 5v5 situations.

This list contains a few surprises. For instance - Cody Franson is in the top 20 defensively? I personally have regularly taken issue with Franson's defensive efforts during his tenure in Toronto so far - but based on his usage he appears to be getting the job done effectively. It also makes one wonder how the hell the Leafs plan on surviving the possibility of trading him away following the arbitration case for Mark Fraser.

It also seems quite surprising that the Leafs are one of only two teams with multiple players on this list. This is a result of a few things. Firstly, the Leafs start in their own zone A LOT - like far more than most teams do. Secondly, the two guys on the list are really the only two notable and active Leafs D men that aren't well into the negatives over the past 6 years (actually Liles isn't that bad either, but he's not playing much these days). Carl Gunnarsson and Franson are our best "defenders" - and it's not really close. Dion Phaneuf has improved, and he does add offensive skills, but he's got a long way to go to improve on how horrible he was early in his career in Calgary. Having looked into this more - I've slowly begun to push towards the "trade Phaneuf to keep Franson" side of things... I'm not sure it's going to work out that way (in fact I doubt it will) but it's not as unbelievable an idea as I thought it was prior to this analysis.

All of that being said, I think it should also be pointed out that players in the range of 3000 minutes are still likely to regress quite a bit back towards the NHL average score of 0.000. The guys we can be more certain about defensively are the ones who are up above the 5000+ minute mark. The true stoppers of the NHL would be in this range consistently, and this list has 13 guys in that group. These defenders have effectively demonstrated that they are the guys to lean on in tough situations.

Which brings me to the another list for discussion and debate...

The Worst 20 Shut Down D Men

These are the worst 20 Defenders who have played 3000+ minutes of 5v5 TOI in the past 6 seasons while facing off in the defensive zone over 30% of the time. You would expect them, based on their linemates and usage to perform a certain way - they're all doing well worse than expected.

Again - some surprises confront people on this list. General perception is that the likes of Bryce Salvador, Dennis Seidenberg, Dan Girardi, and Zach Bogosian are solid defensively. Unfortunately (for their fans - and the GM's signing them to long term deals) they really aren't. Bogosian is an unfortunate case in particular as Winnipeg just extended him for another 7 years (OUCH!) with a contract that includes a No Movement Clause (double ouch!).

I have regularly argued that the likes of Seidenberg and Girardi are relatively over-rated, and this analysis bears that out. Similarly, Mike Komisarek and Ryan O'Byrne show up in approximately the same vicinity impact wise defensively - and yet nobody in New York or Boston would ever think twice about throwing these guys on the ice.

The other interesting point of discussion here is the fact that defense partners can obviously end up at opposite ends of the spectrum. Markov and Subban are a study in contrasts, likewise Girardi and McDonagh. The reality is, one of the two players in these pairings is getting the job done defensively, and the other guy isn't. Consider these tables food for thought.

Lastly I will link to the data most of you have been clamouring for. Please click on the links below to access google spread sheets that are searchable and sortable by individual player, team and each of the component statistics. Each workbook contains separate sheets for the seasons you'll be looking for so check the tabs at the bottom.

SDI v2.1 - 6 Year Data

SDI v2.1 - 2 Year Data

SDI v2.1 - 1 Year Data

Again - questions are welcome, so throw them in the comments section or direct them to me on Twitter. Discussion is encouraged.