If we look at how my Regressed Fenwick Based Win Probabilities Model worked out for January, the results were less than impressive. I had the Leafs going 6-7-2 for the month, and their Expected Points for the month using the calculated Probable Win% values were 14.146. They ended up going 9-5-1 for 19 points, so I was off by quite a wide margin. My record in Predictions for January ended up at 4-11... So not even close.
On top of the results predictions failing overall, individual games were completely topsy turvy. The model had them going 5-0-1 through the first 6 games of January, and in fact they crashed and burned early, posting a 2-4-0 record with both wins occurring via shootouts. Then the model had them going 1-7-1 to close out the month, but instead this unpredictable and amazingly volatile (from a shooting perspective at least) team completely flipped that around going 7-1-1.
So suffice it to say, that made THIS article by Michael Grange look sort of wrong. I'll chalk it up to Murphy's Law and just acknowledge that predicting anything in sports on a game by game basis is amazingly difficult. Reality is 8 of the 15 games had Probable Win% values between 0.425 and 0.575 - in other words, virtual coin flips... and while I'm trying to make a call on those games, the idea that we should EXPECT those to go one way or another is probably not reasonable.
Be that as it may, I'm going to keep trying this for the rest of the year, and see where we end up when all is said and done. So far this season my predictions total up to 14-17. So obviously as I've said in the past - I'm not recommending BETTING on the basis of any of this work - and it's all a work in progress that I'm "testing" going forward. I've already got some ideas on how to improve the model based on the work by Nicholas Emptage at PuckPrediction.com. For instance this quotation from the linked article:
"The implication here is that, in small samples, we really know nothing empirical about team performance, and would tend to regress nearly all the variability out of each team’s statistics if we’re doing things right. But, clearly, this becomes less true as the season goes along. As a team gets closer to 82 games played, we should become more confident that their performance is sustainable, for the simple reason that it has fewer opportunities to regress back to average."
- Nicholas Emptage, Jan 6th 2014
So with this (and other learning) in mind, let's move on to my attempt at February's record for the Leafs (don't worry, there aren't a lot of games this month so it won't take long!)
|GP||Date||H/A||Opponent||Leafs P(Win)||Opp P(Win)||Log5 P(Win)||W-L-OTL|
|59||06/02/2014||@||Tampa Bay Lightning||0.418||0.587||0.336||1-2-0|
|61||27/02/2014||@||New York Islanders||0.418||0.552||0.368||1-3-1|
Ok so there you have it, the slightly modified model predicts that the Leafs will go 1-3-1 in February. Just to re-iterate, I'm calling anything over .525 a Win, anything under .475 a Loss, and anything in between (i.e. the Canucks matchup) an OTL. Realistically you could just say that game is too close to call probabilistically speaking. The Expected Point Percentage for the month is 4.298, so the 3 point prediction shown in the record obviously would require either another overtime or shootout loss in this stretch, or the listed overtime loss to turn out as a win.
Oh and one last thing that is on my mind and seems to be coming up more and more recently...
If you're interested in more detail on prediction and the work behind modelling this type of thing, I strongly suggest you take a look at the work of the previously mentioned Nick Emptage at PuckPrediction, Eric Tulsky who now writes Outnumbered here for SBN, Josh Weissbock who writes at NHLNumbers.com, and a wide variety of other people doing excellent work to try and get at what drives results in an NHL season, for both individual hockey games and long term outcomes.
There are a lot of people (beyond those just mentioned) who have been working on this type of analysis for a while, despite some of what is floating around Twitter via writers for The Star, panelists for TSN, Sportsnet, etc. While this field is growing quickly, there IS a pretty sizable body of work behind what's being done these days.
Alan Ryder has been publishing Hockey Analytics work since 2004 and once wrote for the Globe and Mail. I know Tyler Dellow of mc79hockey.com, Vic Ferrari and the guys at Irreverent Oiler Fans were writing by at least 2005-06, Gabriel Desjardins has been posting at BehindTheNet.ca since 2006-07 at least, and David Johnson at HockeyAnalysis.com has been around for about that long also. I wrote my first stats based posting for a hockey blog during the 2005-06 lockout - before writing the Leafs section at Leafs.HockeyAnalysis.com, and then migrating my way over here to PPP. I'm certain I am leaving out many great writers and researchers at many great sites (apologies if I have).
This "fancy stats" stuff in hockey has been percolating around the internet for going on a decade now. The field isn't brand new, and while coming into it with the perspective of trying to break new territory is admirable, it often leads people to try and re-invent the wheel. I would never discourage anyone from trying to apply their interest and energy to working on this sort of stuff - I do it for fun myself - but please try to explore what's already out there and see where you'll make the best use of that energy. Informed research and work is far more useful than the uninformed.
Thanks for reading and try to find out more as you go!!