Login

locrianmidwest · 10-21-2023, 04:07 PM

A few days ago in the Citadelles Discord server, I made an offhanded comment during a discussion about the team's goal scoring about how ”The average qcc win is like 6-3“. [Image: sl6wy8p.png]

A few thoughts stuck with me long after this conversation ended, the main one being ”what would watching a Citadelles game look like?“ I don't mean that in terms of what the atmosphere or the arena would look like (and after the recent... ”upgrade“ to the ice surface design I'm not sure I want to give it any more thought) but more of what the hockey would be like. Would the Citadelles be a team that would be fun to watch? Frustrating? Painful, even? It's difficult if not impossible to turn boxscores into a full game picture. There must be a better way.

To even begin to create a metric for the most ”watchable“ SMJHL teams, we have to make some assumptions about what ”watchable hockey“ looks like. There are a couple things that I think are universally agreeable as fun to watch: scoring goals, not getting blown out, and winning (tanking team sickos need not apply). Some people prefer close games over blowout wins and vice versa, but any good metric for determining watchability should account for the criteria previously established. How do we measure how much a team gets blown out? Goal differential doesn't tell us much about a team's game-to-game performance besides negative differential being a bad overall sign and vice versa. A slightly better way of going about it would be to take the average of a team's margins of victories and losses (henceforth referred to as AMV and AML, better names would be GD/W and GD/L but I needed the excuse to name a stat after anime music videos). To calculate this for any given team, you first subtract the losing score from the winning score to see how much they won/lost by in each game. After that, separate the score differences into wins and losses and take the average of both pools. It's optional, but I round to the nearest hundredth because I don't like typing out long decimals.
[Image: 3uj5agz.png]

Above is an example of the calculation of AMV/AML for the Great Falls Grizzlies. Highlighted in green is margin of difference in wins High AMV indicates that a team tends to win in dominant fashion, and a high AML is a team that is often humiliated. Teams with low AMV keep their wins close, and likewise with AML teams in losses. This paints a lot more of a picture of how watchable an average game from a team is, especially when combined with goal differential; a team's AMV and AML show how they'll handle both victory and defeat regardless of how likely they are to win or lose. Before we get to the results, the data that we're using must first be addressed. The current season is being used, ending with the games played on January 10, 2040. There's nothing special about this date, it's just what the current games were when I scraped the schedule data from the API. This equates to somewhere from 45-50 games played per team. The standings at this time for the sake of knowledge are displayed here: [Image: JbG71dS.png]

And now, here are the results from calculating the AMV and AML for every team in the J:

[Image: irhAmuN.png]

The highest AMV team, surprisingly, is Yukon who are winning games by an average of 3.42 goals per game! To put this in perspective, the gap between Yukon and the second highest team AMV (the Citadelles) is about as large between the gap between QCC and the 7th place team (NL). The dubious honor of highest league AML belongs to Colorado, who loses games by an average of 3.58 goals. Another interesting result of this is that despite the Grizzlies being last in the standings, they actually have tend to lose by less than the Berserkers in 4th place! The reason for their woes from this statistical point of view is their league-worst AMV of 2.13. The gap between them and the second worst AMV team (Carolina) is *bigger* than the gap between Carolina and 7th place AMV team!

The simplest possible measurement of watchability using these new statistics is to subtract a team's AML from their AMV to create a simple, one number ”watchability score“ of sorts. *psst. That's just slightly adjusted goal differential but I won't tell anyone if you don't. Tongue

Besides redundancy, compressing AMV/AML down into one stat isn't a good idea because it doesn't account for teams that tend to keep their games close, let alone account for preferences in competitive games vs. blowout wins. For that reason I think our best option is to remain with two stats and try to represent each team in the league's AMV/AML on a graph and draw conclusions from that. That graph is here: [Image: bKO5HZD.png]

Now THIS is interesting! Something that I noticed while putting the graph together is that you can classify teams on this graph in four different ways. In the bottom right of the distribution are the ”Goldilocks“ teams. They trend towards the top of the standings, they win by a lot and keep their losses close. If you're watching a game from a Goldilocks team, chances are you're gonna have a good time. This is good hockey. On the complete opposite end of the spectrum lies our ”Hell“ teams, whos wins are all close and losses are noncompetitive. Watching a Hell team is an endurance test, a masochism in 3 acts. The victories are hard-fought and fleeting, the losses are heavy and brutal, an anvil falling from a great height onto an unsuspecting hockey fan. Avoid teams in the top left of the graph all costs. Here's where it gets interesting; there are two more team result types beyond good and bad. ”Cardiac“ teams are defined by both losing and winning by the thinnest of margins. Each and every game is sweated out, and anything can happen. You can find them in the lower left of the graph. Last but not least are the ”Volatile“ teams, who comprise the gloriously inconsistent upper right of the graph. At times they are worldbeaters, at times they look like they would get pummelled by a team of 6-year olds. If you love not knowing what team you're going to get on any given day, they're the ones for you. As to how to classify the 14 teams into these 4 categories, I have somewhat arbitrary rules I used to sort each team into their respective buckets, decided on mainly by looking at where the big gaps were on the graph and demarcating accordingly. I classify Cardiacs as being teams who win and lose their games by 2.5 games or less, Hell as teams who lose by more than 3 goals and win by less than 2.5, Volatiles are teams that win by more than 2.5 and lose by more than 3, and the Goldilocks zone lies beneath 3 goals per loss and over 3 goals per win. Sectioning off leaves us with these groups:

[Image: 7LvguZM.png]

Cardiac:

Volatile:

Hell:

Goldilocks: Armada

First, the sanity checks. 5/7 of the top teams are Goldilocks and 2/3 of the bottom-feeders are in Hell, which is to be expected. As for the outliers, let's start with the 2/7 of the top half of the league that aren't perfectly watchable. Regina and Newfoundland both fail the qualifiers for being Goldilocks teams by quite a bit but are both different types. Both are still watchable as evidenced by being good teams and not in the Hell category (granted, the two are probably mutually exclusive...) but have different appeal. For competitive tight games, the Elk are for you. If *entertaining* neutral fan observation hockey is more your speed, you may have a second home in Newfoundland. Onto the outlier from the bottom three, Kelowna. The Knights are by *far* the team that embodies the Cardiac label, being the furthest team in the bottom left of the graph. The Knights actually have the lowest AML in the league, their second to last place in the standings may belie poor puckluck or just **grit**. As for more generalized observations; the data here inadvertently supports bandwagoning. The most “watchable” teams as measured this way are the ones in the top half of the standings for the most part. While unsurprising, the writer finds herself disappointed that there aren't more top-half teams in the Volatile/Cardiac zones and more bottom-half teams in the Goldilocks zone. In that regard, shoutouts to Yukon, Newfoundland, and Regina for defying expectations and producting interesting hockey regardless of place in the standings! It might be possible to get a single numbered “watchability score” from this graph using some proper form of graph analysis outside of pattern recognition (or even better, a third dimension to the AMV/AML graph that adds wins so you can measure a sort of “appeal to neutral fans“ score in 3 dimensions!) but this is getting unwieldy and I don't have any 3D graphing software on hand so this is left as an exercise to the reader.

So, what did we learn here? In terms of the question I *initially* set out to answer, Citadelles games are mathematically some of the best in the league to watch, so you should go do that. If we're measuring the success of this analysis in terms of what fuels my ego, this was unparalleled success. Jokes aside, this... *thing* is a melding of my love for statistics and writing and producing something as big as this is is h ugely satisfying considering my lack of creative output in years past; I hope I would have done SIBR proud. In terms of new observations, managing to even come up with an answer of quantifying something as ridiculous as ”watchability“ for a hockey simulation or even the types of team performance as we did is quite impressive! The 4 classifications is a far more interesting way of evaluating team performance than just “did team A score more goals than team B” like has been done by analysts since time immemorial. Real sports has lots of stories of imperfect but storied teams: teams that go on miracle runs, teams that choke, teams that are a screaming tirefire the whole way down. Enough of this ”did we have a good year/bad year“ stuff, that's for nerds. We should shove more of them into lockers. Even if you want to look at our results in a purely black and white sense, we did establish the 5 or 6 most consistently watchable teams in the Goldilocks zone The initial reason for researching this was resolved, we found a new way to evaluate teams, and I feel good about myself. I'd chalk this up as a complete success. SHL Site Team: will be looking for AMV/AML on the standings page starting tomorrow. Wink

[Word Count: 1829]

Special Thanks:
Glumbaron for serving as a consultant for SillyStats 01 (as of yet unreleased, far bigger undertaking than this) & 02's basic ideas and scope
@RAmenAmen for Excel assistance and moral support
The SHL API and its developers
SIBR & Jon Bois for fueling my love for sports data silliness
Net Man #DoItForNetMan