1) Shouldn't empty-net goals be excluded from the model? The whole notion of shot quality presumes a goaltender on the other end attempting to make a save. For 2005-6 (based on PBP files) I show 7,428 goals, with 178 empty netters. That's around 2.5% enough, to tweak the margins a bit.
2) Should penalty shots be split out as a descriminating factor? Certainly the scenario is somewhat different than the rest of in-game action, and I show 35 penalty shot goals on 85 shots, reflecting a shot quality that is higher than mere distance and shot type would predict.
3) What to do about Missed Shots? While they shouldn't factor into goaltending evaluations, I would think they should certainly count against an offense. There are lots of defensemen with big slapshots that go booming off the endboards, and I would think those should be assigned a zero or near-zero Shot Quality. If you have two players with equal numbers, but one took twice as many Missed Shots as the other, I would think that should reflect in their Shot Quality. You could also give credit to a team defense for higher levels of Missed Shots (not giving opposing shooters enough time & space to shoot well).
4) When looking at Shot After Turnover, should a time constraint be used to ensure that the shot is somewhat related to the turnover? Realistically, after 5 or 10 seconds, the short-term effect of the turnover has pretty much passed as the rest of the players react. I would also add Hits into that filter (i.e. a Takeaway, Hit, or opponent's Giveaway followed by a shot within 10 seconds). For instance, right in the first game of the 2005-6 Regular Season, plays 2 & 3 show a hit by Patrice Bergeron of Boston, followed by a Hal Gill wrist shot 9 seconds later. You could argue that the Hit led to the shot just as surely as a Takeaway would.
Overall, this is a nice update to a very useful piece of analysis by the gang at Hockey Analytics. I've got a slightly different spin on the anlaysis which I hope to post here next week.