After a few months of accumulating data from several sources, we have finally tabulated the prospective and retrospective performance of the Advanced Bracton statistic for the 2017-18 season.
A comment – one of the reasons it takes us so long to cull all of the relevant data is that we are a) not skilled at web scraping data b) even if we were, the statistics we need to garner may not be available from the NHL API (we assume, since they are not available anywhere else on the internet in the format we desire, published by people who seem to be skilled at scraping said data).
In this regard, it is important to note that, in our opinion, that there is considerable black hole in hockey analytics, especially within the philosophy of mistake minimization, which is what we are trying to capture. We think that players that don’t hurt their teams (what we attempt to quantify), are as valuable, if not more so, than players who “help” their teams (much easier to quantify, and therefore where real life decisions are made). Good managements find guys who are at least neutral throughout their entire line-up (and do not over pay for the privilege). Bad teams who sabotage themselves play golf April 8.
Why do we think this? First, we think we have shown, admirably, on a prospective basis, that even though we have ZERO raw data of the type we need to build or model (due to its lack of availability) for roughly 25% of the players who make NHL teams at final cut time in October, our extrapolations are generally acceptable to us. We have made a strong argument over four entire seasons that mistake minimization could assume a predictive posture if better data at junior, college, and other ranks of hockey existed.
For example, on a prospective basis in 2017-18, we produced a 0.39 Spearman’s rank sum correlation for our eastern predictions and a 0.18 in the western conference. Given the low sample size, this calculation is subject to an immodest skew for outliers. In this season, our results would have been far better had the actual performances of MTL and NJ been swapped in the east (would have resulted in 0.55) and VGK and EDM been reversed in the west (would have resulted in 0.50). Interestingly, our results are almost exactly the same as those published at sportsnet.ca by Corsica Hockey and Hockeyviz. Moreover, USA Today predictions produced a 0.43 in the east and a 0.01 in the west. It was that type of year.
We have shown in the past that rank sums for statistics such as Corsi and Fenwick have not outperformed, and have actually slightly under-performed, our rating system both prospectively and (especially) retrospectively. While we are, candidly, the beneficiaries of the groundbreaking forays of each into the hockey analytic world, we are not convinced that the emphasis the hockey community places on shot based metrics has much to do with winning hockey games. Sure, there may be a function of shots taken versus goals scored, but if these analytics were capturing the veracity of that function optimally, they would be as retrospectively accurate versus actual results as the Advanced Bracton is. In fact here is an interesting article pertaining to scoring chances as opposed to shots taken.
Why do we say this? For the fourth year in a row, our retrospective calculations of mistake minimization as a philosophy have been are near exact reflection of actual results. What we do is simple; we compare the actual conference standings results with our metric, tabulated by player, to amass a total score for a team. We then rank the teams, by conference, by score on the Advanced Bracton method. Lastly, we compare our retrospective ranks with actual results, and run a Rho calculation comparing the two.
This year, as was the case in all the other years we have been doing this, we produced a 0.95 score in the eastern conference and a 0.90 in the west. Yes, this includes the Las Vegas experience (about which we will have MUCH more to say…..if for no other reason that the performance of the Golden Knights is the single biggest story in the history of professional sports. Yes, you read that right. Las Vegas, hold on to your hats, had the highest AB score in the West (+34) during the regular season (with NSH a somewhat distant second at +27). As a result, their run to the Cup Finals was no accident, at least according to our model. It is with a VERY heavy heart that we did not have the time to publish this before the playoffs started……we could have bet Vegas, in Vegas!
The management of sports in general, and hockey in particular, should be changed forever. As we have been saying, and have yet to be proven wrong, keeping the puck out of your own net is EXACTLY THE SAME as scoring a goal against another team. In fact, our logic is precisely correct when comparing the AB score to a rank of the goal differentials of teams versus actual results. The results cannot be differentiated. A calculation of the difference of goals for versus goals against is almost exactly the same as Goals not against versus goals not for (which is in essence the AB score).
Please stay tuned for a far more active summer than the past few months have been. As mentioned above, we are penning something about Las Vegas that we think will be as much of a watershed piece as was our original missive regarding our inspiration for our work on mistake minimization in the NHL.
Oh and BTW, Darren Helm has finally passed the torch as patron saint of our endeavor. There is a new sheriff in town…….and his name is Jake DeBrusk.