Takeaways and Giveaways Going Rate

So yesterday over in the National Post, Michael Traikos had an interview with Leafs defensive guru Centre Jay McClement. They touched on a few concerns that many (particularly around these parts) have regarding the direction of the team and their style of play. I'll let McClement's commentary do the talking here (emphasis added by author):

"McClement, who had an even plus-minus rating, said the numbers are misleading. The Leafs were content giving up shots as long as they were from bad angles or far away from the net, he said, and coaches tracked advanced stats on everything from turnover-to-takeaway differential to where players were finishing their bodychecks that provided an accurate indicator of how well the team was playing."

- Michael Traikos, National Post, Aug, 27th, 2013

Now - I am unclear on how finishing their body checks in certain areas of the ice is relevant statistically to an NHL franchise - and I won't bother delving into that here and now as we've touched on hitting before and I'm not clear how anyone has assessed it's impact or role as an "accurate indicator of how well the team is playing".

But the turnover-to-takeaway differential discussion intrigued me. Not long ago, I ran correlations between the NHL's takeaway and giveaway statistics for teams, and compared them to win and point percentage. A few things cropped up in the analysis, and I think to some extent that was a flaw in the data we're provided with that the NHL tracks.

The most significant problem with counting hits, giveaways, takeaways etc. is that at either the team or individual level, they're counted horribly from venue to venue. As a result of disagreement between scorekeeper methodologies, there is bias in any player or team's home numbers, and the road numbers likely vary more than they should, due to differences in scheduling if nothing else. This is something we need to minimize as much as possible if we're trying to assess skill or value in these areas.

Another significant issue of concern is the way that these numbers are treated as counting statistics for comparison by teams, coaches, players, analysts and the average fan. Saying a team blocked more shots is nice, but that's only a good thing to know if we compare two teams that give up the exact same number of shot opportunities.

Saying a team had more giveaways is bad - losing possession is bad and we go on about how possession is important in the advanced stats community all the time. But again - if your team has the puck more than the opposition you'll inevitably end up giving the puck away more than a team that NEVER has the puck. So just counting up giveaways seems sort of ridiculous as a way of comparing players or teams. The exactly converse situation is true for takeaways. You'll likely get more of them if your team never has possession of the puck, purely as a result of the fact that you're constantly chasing around the other team.

So how do we account for or fix these two problems when trying to assess the value of these metrics. Well - first we ignore home stats. Home stats are just going to emphasize the bias of an individual scorer. At least with road statistics we can spread the error of one or two really bad scorers out across more teams, and in effect minimize for it as much as possible.

Secondly, we need to figure out a means of assessing who is getting the most takeaways or committing the most giveaways while controlling for the amount of time they do or don't have the puck in their possession. How can we accomplish this? Through an ever so convenient proxy for possession that we have at our disposal.... Corsi.

Studies have indicated that Corsi (and Fenwick) are excellent proxies for attack time and/or puck possession. This is one of the fundamental building blocks of advanced stats in hockey - so it makes sense to make use of it here if possible. Since we don't have actual zone time available, what we do have access to is Corsi For percentages at 5v5 in road games (aka 5v5 Road CF%), thanks to stats.hockeyanalysis.com, and the page also provides the total amount of time the team spent at 5v5 on the road throughout the season.

To get a proxy for time with the puck, I have multiplied the team's 5v5 Road TOI by it's CF%, to get a proxy for their opposition's time with the puck, I just use the remainder of the ice time (equivalent to 5v5 Road TOI * [1 - CF%]). Using the NHL's team RTSS data for the past 6 seasons, and working under the assumption that the majority of takeaways and giveaways occur at 5v5, I then calculated each team's Takeaways per 60 minutes without the puck, and Giveaways per 60 minutes with the puck.

Lastly I divided every team's TkA/60 Opp by their GvA/60 For to get a takeaway to giveaway ratio. Then I ranked every NHL team's season over the past 6 years and these were the results. I correlated the past 6 years worth of data to team Point %, Road 5v5 Corsi For %, and SF % and found the R and R^2 values shown below. The most interesting part to me is the correlation between this result and Corsi For percentage itself. That is to say, this value explains approximately 50% of Shot attempt differential at the team level.

From a repeatability standpoint, using year over year data, the R-value correlating one season to the next came out to 0.4567, which means a single season of data needs to be regressed 54% to the mean (which over 6 seasons worth of data works out to 0.8548).

All-in-all I think this provides some further insight into why tracking these types of stats MAY be statistically meaningful - but at the moment they're poorly counted by the league and frankly not all that reliable - even as road stats.

For those of you that don't want to click on the links - here are the highlights:

Only 6 of 180 teams since 2007-08 have managed to have a TkA60/GvA60 ratio above one and produced a seasonal point percentage below .500. Those four teams are shown below - and something might stand out to you about them:

From the looks of things crappy shooting or goaltending in a given season is probably outweighing whatever impact reducing giveaways and increasing takeaways is having on a team's results.

Now you're probably wondering how many teams actually produced a takeaway/giveaway ratio over 1.0? Surprisingly only 35 teams out of 150 have managed to do it over an entire season, and 8 of those were from this past season (I'm guessing over the course of a full year some of the teams on the edge would have dropped below 1.0) So that means that 83% of the teams that managed to produce a Takeaway/Giveaway Ratio over 1.0 using this analysis produced over a .500 pt percentage, and in fact the average point percentage of these teams is a whopping .598 (98 point season).

Here is a look at the top 10 teams by this metric over the last 6 years and their point totals:

Ok - so apparently Chicago is REALLY good by this metric, and LA has done very well the past two years. What I find most interesting is generally speaking these are teams that are winning without a huge amount of luck. Chicago this past season is the only one of the top 10 to post a PDO over 1000, and that likely would have regressed over the remainder of the year. Something tells me the other numbers probably wouldn't so much.

Now - here's where I bring the discussion back to the Leafs. It's really cute how they try to keep convincing us they're doing good things defensively because Randy Carlyle is a great coach and they have these amazing defensive systems. The problem is, it seems like EVERY time we explore them in more detail, the things we're being told don't seem to pan out exactly as described. We actually started an entire series to dissect some of those mis-conceptions that are promulgated in the Toronto mainstream media (and we think we've made a dent in how people perceive Carlyle's defensive guru-ness).

It seems like we might have stumbled across another one thanks to McClement letting the proverbial cat out of the bag. You see - here are the Leafs results using this TkA60 / GvA60 metric over the past 6 years:

You see - the problem is - the Leafs have NEVER been particularly good via this metric (at least over the last 6 seasons). But under Ron Wilson's first two seasons they were improving in this regard - then the bottom fell out in 2010-11 (I'm wondering if trading Tomas Kaberle to Boston and handing more minutes to the likes of Luke Schenn and Dion Phaneuf made an impact in this regard? It seems logical). 2011-12 was again slightly improved, but then - Randy Carlyle came along with his systems and we're worried about giveaways and takeaways a lot - but... uh oh... we had our worst season by this measure in the past 6 years?

If Randy Carlyle is placing an emphasis on this skill - why is his one season as head coach the WORST one amongst the Leafs last 6? This could in fact be pointed at as a contributing factor TO the Leafs atrocious shot differential. I'd just like to point out that the 2012-13 Leafs number of 0.6602 ranks 160th out of 180 team seasons over the past 6 years - that's the 11th percentile folks - in other words 89% of NHL teams in any year would be better at this than the Leafs (9 out of 10 if that helps).

So is this just a recent problem for Randy? well let's look at how his prior teams have fared in Anaheim, and how they've done since he left:

It seems like they were better than Toronto, and they've gotten better in each of the last 6 years, but they've never been a team over 1.0, and they've always had some shooting and goaltending on their side. No team Randy Carlyle has coached over the last 6 seasons has posted a TkA/GvA ratio over 1.0, so I don't see it as likely anytime soon. Thus - I think it's a tad disingenuous for players on the team to assert that Carlyle's defensive prowess in this regard is paying dividends.

This seems to be right up there with fights winning games, blocked shots winning games, etc. It's a lot of smoke screen to cover up the fact that the team shot a ridiculously high percentage last year and got good goaltending. Don't expect these Leafs to morph into Chicago anytime soon. Now - that being said, players like Bolland and Clarkson are coming from teams that are regularly good in this area, so maybe they will have a positive impact?

I don't honestly know, but it's interesting nonetheless. Let me know your thoughts.