I want to start this post by sharing my plans for this season in terms of media. Over the past season, I have deeply involved myself in the statistical side of LG Media. Everything from creating spreadsheets to cool trading cards. I plan to continue most of that this season as well as some new concepts. All of these ideas and models will be presented in a weekly thread post just like this one. Here I will display all of the stats I keep track of on a week-to-week basis. The majority of the data types will remain constant, I may just add in some unique stats from time to time based on where we are in the season. So without further ado here is the Week 1 Stats Update: First, I will start off with the piece that I have spent the most time on by far. I decided to create a Team of the Week when I was looking through old Player of the Week threads and realized that the numbers didn't make a ton of sense. As well as that there was a missed opportunity by the site to do a Team of the Week style post. I combined two of the things I love, data and photoshop, to create some cool-looking cards that also display the player's accomplishments during the week and make it easy for the average Joe to read and understand which stats mean what and their importance in the video game. Below, ill break down how the ratings were calculated as well as some flaws the model may have. How it works: I took every stat available in a single game and ran a correlation on each one to find out which stats contribute the most to winning a game. I also broke it down by position in case there were any discrepancies between positions. From that data, I chose 9 stats for skaters and 6 for goalies on which I would base my rankings off. For skaters, I divided them into 3 categories; Offense, Defense, and Team Play: Forwards - Offense: Goals For per Game, Points per Game, Marginal Goals Created per Game - Defense: Goals Against per Game, Shots Against per Game, Marginal Goals Prevented per Game - Team Play: Win %, Plus-Minus per Game, Opponent Win % Defensemen - Offense: Goals For per Game, Points per Game, Marginal Goals Created per Game - Defense: Goals Against per Game, TwP per Game, Marginal Goals Prevented per Game - Team Play: Win %, Plus-Minus per Game, Opponent Win % Goalies No categories: Win %, SV%, GAA, SO, Goal Support, Shots per Game *The goalie model I am not as confident in because the correlations were much lower than the skater's stats* Overall: I then take those 9 stats ratings for skaters and 6 ratings for goalies and average them together to get a composite Overall that ranges between 50 OVR and 99 OVR. So now that the math is out of the way, here's a breakdown on the best part, the cards: Now that you should have an understanding of how the cards work, here is this weeks TOTW: LINE 1: LINE 2: LINE 3: I apologize if it is hard to read the cards. I forgot that Tris only allows images of 1MB at a time, so I had to downscale them quite a bit. Also below are spoilers with EVERYONE'S ratings in them per position. Spoiler: LW Overalls Spoiler: C Overalls Spoiler: RW Overalls Spoiler: LD Overalls Spoiler: RD Overalls Spoiler: G Overalls The next segment is on ELO. Many of you may be familiar with the term ELO from the drafts discord. And while that is called ELO, it really isn't the same thing. ELO was originally designed for 1 v 1 chess matches. So when you try to adapt it to a team setting like 6 v 6, there are some flaws. What I did was, I took the original ELO formulas and adapted them to a team setting. I then tinkered around with some models to create the Fleury ELO model that roughly can track who is performing and who isn't in LG. Disclaimer, I am not a statistician by any means, so for all I know this could be completely invalid, but I will say when I've run it for previous seasons, it has somewhat lined up. I won't get into the specifics of the formulas but basically, Team 1 is expected to score X amount of goals on Team 2. Team 2 is expected to give up X amount of goals to Team 1. Those two stats together give us an expected margin of win/lose for both teams. If Team 1 wins but does not win by the expected amount, they will not gain as much ELO as if they were to beat them by more than the expected goal differential. I then add a game stat side to show how much each individual player contributed the Win or Loss and ELO is assigned accordingly. Here are the top 10 gainers and losers for Week 1. Note: Since it's Week 1, everyone starts with 1200 ELO. Also, defensemen typically gain less / lose less ELO than forwards because of how I grade scoring. So try to compare ELO inside the 2 position groups. Spoiler: LW ELO Spoiler: C ELO Spoiler: RW ELO Spoiler: LD ELO Spoiler: RD ELO This will be the last section of statistics. Strength of Schedule refers to the quality of opponents a player has played over the course of the week. I calculated this by adding the total wins and total losses up for each individual opponent, to create a composite opponent record in which I can then get a W%. The list is sorted alphabetically for ease of search. It's much easier to view as a sheet, so I am pasting a Google Sheets link below: https://docs.google.com/spreadsheets/d/1rsnMyuiU_ifo4MCUtUC2_zXnM8hKPBG0RLu5IjlPHqc/edit?usp=sharing Finally, I'm going to plug my discord. I used it for support on my Bidding Spreadsheet and I plan to keep it somewhat active with discussions on a wide range of topics. You can join here: https://discord.gg/yeJBpDmy
Took me forever to try and figure out what you were saying. Yeah google sheets had a brain fart with the logos for C overalls.
I find these interesting, and also a bit of a testament to how much the stats potentially don’t capture. For example, how to truly capture “the best C” in the league. I say this because I am puzzled at how I was the “12th overall” considering the amount of mistakes I felt I made. It would be so interesting if there was some system or time even to have someone scrape some form of analytics from streams, I’m curious how it would compare to these stat based compilations (I don’t mean just this one in particular). edit : Hit enter too soon. Nice work, hopefully we see more throughout the season! Cheers
Overalls right now are more line-based because it is only week 1 and we have 3 games of stats to go off of. If you want to look to more of individual contribution, id look at ELO, you are slightly ranked lower in the Centers for ELO compared to your overall for Centers. Again it's hard because it's only week 1, but it's a rough estimate. I'm curious what kind of stat you may be looking for. Right now I pull everything game by game and get just about every stat available. Ultimately the best "stat" is the eye test, but for obvious reasons, there's no way to measure that.
@Fleury l8l this is great stuff. Like how for the G position you have incorporated advanced stats (e.g. goal support). TOA of the opponent could be another useful data point. Likely requires some sort of SQL join query to produce because the data resides in separate tables. This data may in a directionally correct way speak to the quality of shots faced in a game .
It wouldn’t be too hard. I do everything with python so I’d just pull the other table. I’ll look into it for next time.