CS Squirt Rankings

CS Squirt Rankings

Joined: July 1st, 2014, 9:46 pm

November 18th, 2014, 10:05 pm #1

Quote
Like
Share

Joined: March 31st, 2014, 6:35 am

November 20th, 2014, 3:06 am #2

Interesting results. I don't normally keep track of CSDHL because it involves a lot of manual data entry (whereas NIHL is easy copy-paste and I can get 100s of game results in seconds), but I did input entries for Squirt CSDHL to compare. I got some slightly different results, but the two at the top actually remain in the same order.

Team - Power Rating
1) Bruins - 1.58
2) Falcons - 1.53
3) Hawks - 1.19
4) Blues - 1.18
5) Ice Dogs - 1.05
6) Affton - 1.04
7) Huskies - 0.95
8) Chesterfield - 0.94
9) Sabres - 0.93
10) Glenview - 0.92
11) Winnetka - 0.28
12) Rockford - 0.26

The power rating I'm using is my own formula. It is an attempt to guess the points per game a team will have at the end of a season if they played against each team in their division 1000s of times with no change in team abilities. Right now, the Falcons are at 2.00 in actuality versus their 1.53 rating, so my system expects them to drop games somewhere along the line. The Bruins are currently at 1.82 in actual rating versus 1.58 in power rating, so it's expected that they will lose some more games too. The Blues are at 0.9 in actuality versus my 1.18, so I expect their record to improve over the season.

A difference of under 0.30 rating is expected to be a close game. So we're looking at #3 Hawks versus #10 Glenview as being a good matchup. The only real outliers are the Bruins and Falcons on the top (but can still be challenged by Hawks and Blues), and then Winnetka and Rockford on the bottom seem like they don't belong in CSDHL this year.

If you'd like to recreate my system, here's how my system works:
Each game, I give out 2 points just like normal. However, I distribute those points differently from normal (normally 2pts for win, 1 for tie, 0 for loss). In this system, I start both teams with 1 point and then add 1/3 of a point for each goal they win by or subtract 1/3 of a point for each goal they lose by. I cap it at 6 goal differential, which gives 3 points for a 6-0 win, and -1 points for a 0-6 loss. If you win by 3 goals, you get 2 points and the opponent gets 0 (like normal). If you win by 1 goal, you get 1.333 points and the opponent still gets 0.667 points.

I then take your adjusted point total and divide by games to get your adjusted average points (this number will be between -1 and 3). Next, I need to add in the opponent strength. I take the average of all the adjusted average points of the opponents you've played and call that the opponent strength. So if you played a team with an adjusted average of 1 and you played another team twice that has an adjusted average of 2, your total opponent strength would be (1 + 2 + 2) / 3games = 1.667.

Finally, I weigh the adjusted average more heavily than opponent strength and combine them together to get the power rating. Power Rating = (adjustedAvg * 2 + OppStrength) / 3.

A team can technically get a power rating over 2.0 or below 0, but those are rare outlier cases where they're getting blown out (or doing the blow outs) every game. Also, as mentioned before, any difference below 0.30 is pretty minor. The difference between a 10-6 record and an 8-8 record is 0.25 and I'd expect an 8-8 team to be able to beat a 10-6 team on any given day, let alone the really big upsets that sometimes occur.

Admittedly, I'm not completely satisfied with my system because it doesn't give enough credit to strong defensive teams that consistently win 2-0, 2-1, 3-2, etc. without blowing out teams. And then rewards teams too much that score a lot of goals against bad opponents, but has questionable defense.

I simply haven't cared enough to mess with the formulas to see if I can find one that is a better predictor of future results. And even if I did make it better, there will always be upsets of the big and small kind. So, take my power rating for what it is, not as some perfect predictor of the future.
Quote
Like
Share

Joined: July 1st, 2014, 9:46 pm

November 20th, 2014, 5:24 am #3

GCER,

Thanks for outlining your methodology and giving me the resulting data dump. It's great stuff! I may actually try to recreate your system some time, because I find the predictive ability of it to be unmatched really. I think I tried to do it once before but my brain melted. Which of the formulae do you run as the iterative calculation? As to your strong defensive team problem, I'm thinking you could use a nested IF statement to pull off a weighting ie a formula similar to the one I used that caps goal differential at 6 in my spreadsheet. That's only step one obviously, not sure how to automate finding the consistency ...
Quote
Like
Share

Joined: April 30th, 2013, 5:23 am

November 20th, 2014, 10:19 pm #4

Understanding and acknowledging that this is a bunch of 9 and 10 year olds playing which probably doesn't merit this level of analytical rigor, I do find the exercise academically interesting. GCER - are you applying a Monte Carlo analysis to get your theoretical 'thousands' of games?

The most interesting aspect of both systems (and I wonder how myhockeyrankings addresses this - I doubt that they do) is that neither takes into consideration head-to-head results. At the end of the day it's pretty tough to argue that the Bruins are 'better' than the Falcons when they have actually played one another and the Falcons won (by 3 goals - not really a close game even).

Thoughts?
Quote
Like
Share

Joined: July 1st, 2014, 9:46 pm

November 20th, 2014, 10:55 pm #5

Fair points, especially when it comes to arguing about who is "better". That is one reason that college hockey people gravitate towards the KRACH system and round robin winning percentage. However, much MUCH greater analytic rigor is required to do the KRACH method (not to mention insano math skillz yo) and I much prefer the speed and simplicity of the SRS for cranking out blogs. Besides that, I think both GCER and I would dismiss one game's results and conclude that the Bruins and Falcons just haven't played each other enough yet! These predictive systems are designed to give you the information to analyze the probability of a team winning their next game or of where they will finish the season in the standings. So rather than saying the Bruins are better, say instead that the next two games between the Bruins and the Falcons are a toss up but you are watching the top two teams in the division play.

Myhocekyrankings does not address the head-to-head factor:

"How does MYHockey compute its ratings?

MYHockey 's ratings are computed mathematically, with no subjective weighting or human determined values. MYHockey rates teams based upon how well they play against other teams and how good those teams are. These two factors are "AGD" or "average goal differential" and "SCHED" or "strength of schedule". AGD is currently calculated by accumulating the goal differential of each game, to a maximum of 7, and divides it by the number of games played. Using a max game goal differential of 7 does "penalize" teams that blow out opponents, but this is common in rating systems as it takes away the incentive for teams to run up scores and allows them to have a bad game by capping the mathematical damage. The strength of schedule is computed by averaging the rating of each game opponent. AGD and SCHED are added together to compute a team's rating."
Quote
Like
Share


Joined: November 22nd, 2014, 8:41 pm

November 23rd, 2014, 5:20 pm #7

Interesting results. I don't normally keep track of CSDHL because it involves a lot of manual data entry (whereas NIHL is easy copy-paste and I can get 100s of game results in seconds), but I did input entries for Squirt CSDHL to compare. I got some slightly different results, but the two at the top actually remain in the same order.

Team - Power Rating
1) Bruins - 1.58
2) Falcons - 1.53
3) Hawks - 1.19
4) Blues - 1.18
5) Ice Dogs - 1.05
6) Affton - 1.04
7) Huskies - 0.95
8) Chesterfield - 0.94
9) Sabres - 0.93
10) Glenview - 0.92
11) Winnetka - 0.28
12) Rockford - 0.26

The power rating I'm using is my own formula. It is an attempt to guess the points per game a team will have at the end of a season if they played against each team in their division 1000s of times with no change in team abilities. Right now, the Falcons are at 2.00 in actuality versus their 1.53 rating, so my system expects them to drop games somewhere along the line. The Bruins are currently at 1.82 in actual rating versus 1.58 in power rating, so it's expected that they will lose some more games too. The Blues are at 0.9 in actuality versus my 1.18, so I expect their record to improve over the season.

A difference of under 0.30 rating is expected to be a close game. So we're looking at #3 Hawks versus #10 Glenview as being a good matchup. The only real outliers are the Bruins and Falcons on the top (but can still be challenged by Hawks and Blues), and then Winnetka and Rockford on the bottom seem like they don't belong in CSDHL this year.

If you'd like to recreate my system, here's how my system works:
Each game, I give out 2 points just like normal. However, I distribute those points differently from normal (normally 2pts for win, 1 for tie, 0 for loss). In this system, I start both teams with 1 point and then add 1/3 of a point for each goal they win by or subtract 1/3 of a point for each goal they lose by. I cap it at 6 goal differential, which gives 3 points for a 6-0 win, and -1 points for a 0-6 loss. If you win by 3 goals, you get 2 points and the opponent gets 0 (like normal). If you win by 1 goal, you get 1.333 points and the opponent still gets 0.667 points.

I then take your adjusted point total and divide by games to get your adjusted average points (this number will be between -1 and 3). Next, I need to add in the opponent strength. I take the average of all the adjusted average points of the opponents you've played and call that the opponent strength. So if you played a team with an adjusted average of 1 and you played another team twice that has an adjusted average of 2, your total opponent strength would be (1 + 2 + 2) / 3games = 1.667.

Finally, I weigh the adjusted average more heavily than opponent strength and combine them together to get the power rating. Power Rating = (adjustedAvg * 2 + OppStrength) / 3.

A team can technically get a power rating over 2.0 or below 0, but those are rare outlier cases where they're getting blown out (or doing the blow outs) every game. Also, as mentioned before, any difference below 0.30 is pretty minor. The difference between a 10-6 record and an 8-8 record is 0.25 and I'd expect an 8-8 team to be able to beat a 10-6 team on any given day, let alone the really big upsets that sometimes occur.

Admittedly, I'm not completely satisfied with my system because it doesn't give enough credit to strong defensive teams that consistently win 2-0, 2-1, 3-2, etc. without blowing out teams. And then rewards teams too much that score a lot of goals against bad opponents, but has questionable defense.

I simply haven't cared enough to mess with the formulas to see if I can find one that is a better predictor of future results. And even if I did make it better, there will always be upsets of the big and small kind. So, take my power rating for what it is, not as some perfect predictor of the future.
I think you're being generous with Rockford and Winnetka.
Quote
Like
Share