Quote:
Originally Posted by critiakx2
Hockey pbp and shift data back to 2007 is available at: https://hockey-data.harryshomer.com/
I don't have a ton of confidence in the xGoals predictions that are derived from this (think you need shot speed/quality in addition to type and distance) but it's gotta be enough to crack face-offs, right? Maybe the historical line matchups can predict who will open (scoring line 1 or a defensive line) and player-specific rather than general R/L edges.
MLB has been great but looking like there won't be much other than football Nov-Dec.
Thanks for posting this. Getting data is often harder than many people think so it's great to have in this thread for the data guys to look over.
The problem with FO's isn't data IMO. We have a ton of faceoff win/loss data. It's trying to figure out who will show up in the opening faceoff that is the problem. Also it's a small group who can maintain a 53.5%+ FO win rate (breakeven at -115). Then when that person doesn't even show up in the opening FO, your EV gets wrecked. It doesn't help that despite being the most exciting 2 seconds in sports betting, the actual players and coaches seem to treat it less seriously than the bettors.
Also, I have my doubts about how effective a Log5 type approach to calculating FO WL works especially as we get further from 50/50. For example, if a 55% guy faces off against a 45% guy, what do you think the fair odds should be? What about a 60/40 matchup?
In my experience, xGoals as it's currently used is a really bad metric. When I tried substituting xG for Goals it ended up penalizing the star players and rewarding scrubs. It took me longer than it should to realize this because star players make up the majority of player prop over/unders and I kept taking the best players under. It's when I was messing around with DFS that I realized xG was boosting poor shooting players that I made the obvious connection. I now never use it and prefer to just use a regressed shooting percentage. That's leaving a lot of context on the table which could definitely be improved upon. If I were to create my own metric, I'd use some type of KNN grouping and try to come up with an adjusted xG stat based upon talent level. Maybe people are already do this I haven't really kept up.
Even pbp is missing a lot of context since I feel like so much more is happening besides whoever has the puck. Soccer guys probably already have it all figured out and it just needs to be translated into hockey.
I am not an expert though--it's just my impression from my experience tinkering and reading about what hockey nerds were working on.