[InternationalVillager]

Originally Posted by Enoch Root

Looking at HD chances for and against, as a specific example of how a stat can be wrong, as well as being just not very useful:

1. Small sample sizes can be problematic (in this case, one, or a few games)
2. Small sample sizes of stats with very few data points can be very problematic (typically a player will have between zero and 6 to 8 HD chances for or against, in a game)
3. Converting very small numbers into percentages, is just flat out dangerous (example: player has 3 HD chances for, and 4 against in a game, so they are 43% - sounds like they had a pretty terrible game)
4. The actual events being counted are flawed - a tip is an HD chance, regardless of how weak the situation was. Conversely, Coronato's goal against the Kraken was not an HD chance because he was off to the side of the net, even though that chance was literally the best scoring chance in the game (open net play)
5. There is no ability in the stat to consider times and space, which, IMO is even more important than location (which is essentially the only variable considered). As an example, on the 3rd VAN goal last night, the opportunity wasn't all that dangerous, except for the fact that the shooter had all day to look at the goalie, pick a target, and get a good shot off, towards that target. The primary factor there was time, not location (as it usually is in hockey)
6. There is no ability to factor in the quality of the player and their shooting ability, or their ability to get a shot off quickly, or their ability to create open space (and thus limiting another player's ability to generate HD stats)
7. Bounces (randomness). A perfect play can be thwarted by bad ice. Conversely, a great chance can be the result of a lucky bounce. When the sample size is only a few events, random bounces can literally be the primary, or even only, determinant.

When you factor all these issues in, and then you consider that in any given game, a player is only on the ice for a few HD shots for and against, it is extremely likely for the numbers to be misleading - especially when presented as percentages.

As an example, a line has a pretty dominant night, having possession in the offensive zone for several shifts in a row. And because it is a play that has been working, they keep setting up the winger with one-timers from the side, none of which count as HD chances. They also set up multiple shots from the point, none of which get through and/or get tipped on net. So they only actually take 3 HD shots over the course of those shifts (despite prolonged domination, and lots of actual chances). Then, one bad turnover, and the other team has possession for 5 seconds, and gets a not-very-dangerous HD shot off (rushed), followed by a couple of (not very dangerous) rebounds, then a weak shot from the point that gets tipped, and suddenly the line that has been completely dominant, is sitting at 43% in HD chances for and against. "Hmm, I thought they played pretty well, but I guess I'm just biased because the stats say otherwise."

A lot of these nuances are addressed in public vs private models. So unless you have access to private models or pay to access some, it is always best to look at multiple data sources.