February 23, 2013 at 9:03 am #2616
I’ve wondered about how important faceoff success really is to winning in the NHL. There is some data that’s been linked around here about how well various things predict winning, with the core conclusion being that shots for/against in the best predictor, but it’s based on college hockey not the NHL. It’s also a few years out of date now. I’m after some quick simple testing of the faceoffs question in current NHL play (so under NHL rules with NHL players and the NHL’s competitive balance, etc.).
To do that I just dumped the last three seasons’ worth of NHL team stats into a spreadsheet and ran some correlations. Very quick explanation: a correlation is a specific standard formula which measures how well two sets of numbers predict each other. If one set of numbers — say the number of blog posts about whether shooting percentages are meaningful — exactly matches another set — say the number of times Laaarmer yells at his computer screen — then the correlation will be 1. (I find it more intuitive to think of that as 100% but the formal syntax is to show it as 1.)
Generally speaking scores of around .5 or higher (e.g. 50% or more) are considered to indicate strong correlations. Now correlation alone does not prove or disprove causation, but if two sets of data have say a .6 correlation then it’s sensible to think that the two things probably influence each other. Correlation scores in the .25 to .5 range are considered weak-to-moderate; correlation scores below maybe .25 are likely to mean that the two things don’t have anything to do with each other. (Different statistical authorities have somewhat different versions of those ranges but not drastically so.)
(Correlation scores can also go negative meaning that one set of numbers is predicting the opposite of the other: the correlation between my belching “Here Come The Hawks” at the dinner table and the wife inviting me upstairs to “relax” is -1.)
Okay so looking at the team stats for the last three seasons, comparing faceoff win rates to various other things, here are some correlations and what they seem to suggest.
One striking fact is how little variation there actually is among NHL teams in faceoff win rates over the course of a season: the best and worst team percentages in the league were 54 and 46 in 2012, 55/44 in 2011, 56/46 in 2010. The 30 teams are tightly distributed within those ranges, so for instance while the Blackhawks last year tied for 11th (50.6 percent) that was only one full point shy of the 4th-place rate (51.6 percent). So most of the NHL’s teams at any given moment are having pretty similar overall FO-win rates; if we graphed it we’d see a whole lot of points very close to 50 percent.
If the league’s teams are generally close to breaking even at the dot then it stands to reason that faceoff success is not going to be a _large_ variable in winning and losing over the course of a season, and that is broadly what the correlation scores suggest. But it doesn’t appear to be irrelevant either — in fact given the tight clustering of faceoff rates I was maybe a bit surprised at how _strong_ some correlations turned out to be.
Starting with the biggest picture, standing points, here are the correlation scores between the 30 teams’ FO percentages and their season point totals:
Averaging those gets us .35 which is a mild but noticeable level of correlation. [We’d need to look at this over a longer stretch to get a sense of whether that 2010 to 2012 decline is real or just a blip.]
One of the notions frequently stated on this subject is that faceoff success is “especially important” on special teams. But the data does not support that, if anything the contrary. Here are the correlations of team FO percentages with teams’ ratios of even-strength goals for and against:
Now here they are with team powerplay percentages:
and team PK percentages:
That last one is kind of funny: last season the NHL teams with _worse_ FO percentages tended slightly to do _better_ at killing off penalties? Umm…But really the keyword there is “slightly” as those PP and PK correlations are generally too small to suggest a relationship. The real takeaway here is that winning faceoffs appears to have more to do with each team’s success at even-strength play than on special teams.
As I mentioned there is some good analysis out there suggesting that at the NHL level the single best predictor of a team’s real level of play is shots: shots taken and shots allowed. (Hockey players and coaches at every level are reflecting this basic idea when they use “we are/aren’t generating chances” as shorthand for how well their team is playing regardless of how many pucks have yet gotten past each goalie.) When we compare faceoff win rates to shots per game we get the following correlations:
That 2012 one is a big outlier in all of this but even so, these scores suggest that faceoff win rates have a moderate but persistent relationship to how many shots a team tends to generate. Oddly that is not true for shots allowed:
Why would winning and losing at the dot have a decent positive correlation with generating shots on goal but essentially no correlation, if anything a very-slightly negative one, with _allowing_ shots on goal? No particular hypothesis jumps out right now, ideas are welcome.
So in summary, this quick-and-dirty look at the last three NHL regular seasons suggests that:
– NHL teams don’t actually vary a huge amount over the course of a season in how often they win/lose at the faceoff dot.
– the rates at which NHL teams tend to win faceoffs has a mild but persistent relationship to how many shots they average per game; it has no relationship to how many shots they give up per game.
– the rates at which NHL teams tend to win faceoffs has a mild but persistent relationship to teams’ goals for/against at even strength; it has little to no relationship with success on special teams.
– perhaps the ultimate punchline is that the rates at which NHL teams win faceoffs has a mild but persistent relationship with where they end up in the standings. Is it one of the _most_ important aspects of building a winning team? Probably not, so it arguably shouldn’t be very high on a priority list for roster and lineup decisions. But it does seem to matter at least a bit.February 23, 2013 at 10:24 pm #2624
Thanks for this, I’ve been wondering many of the same things lately! I took a quick look at a few seasons’ worth of numbers, but didn’t do fancy-schmancy spreadsheet things (largely because I’m not very good at that).
You must be logged in to reply to this topic.