HS BOYS LSQRanking Jan 25, 2015 MN/WI/ND/SD

[greenway1969]

When someone questions your rankings, providing mathematical data that supports your ranking does not address the problem in your model. The identification of your variables, the number of variables and the weight assigned to them are not correct. You either need to add more variables or adjust the weight assigned to the variables. Just because you develop a mathematical model does not mean the construct is necessarily reliable or predictive. You have to look at the outcomes and see if they make sense. For example, Greenway defeated North Shore twice, one of the scores was 11-4, yet you have North Shore ranked significantly higher then Greenway. There are countless other examples like this. This should be a red flag that the model is not valid. I do applaud your efforts as it takes a huge amount of time and dedication to come up with a model.

[WestMetro]

LSQ

Something not right when I see your Elk, GR, and Bemid so out of sync with other polls at this point in season, including Mitch Hawkers own statistical based model which is tracking fairly in line intuitively.

I couldnt say what specifically needed, but some kind of fine tuning.

Also, would urge statistically based pollsters to more take into account head to head competition for teams in close proximity to one another, after all isnt that what its usually all about?

[LSQRANK]

LSQ

Something not right when I see your Elk, GR, and Bemid so out of sync with other polls at this point in season, including Mitch Hawkers own statistical based model which is tracking fairly in line intuitively.

I couldnt say what specifically needed, but some kind of fine tuning.

Also, would urge statistically based pollsters to more take into account head to head competition for teams in close proximity to one another, after all isnt that what its usually all about?

I think the weighting of the components is part of the issue here. Because the computer ranking system must use measurable attributes from a game, weighting each of these is critical. As of yet, I have not included code to systematically compared head to head outcomes to tweak situations as you mention between North Shore and the other team mentioned. Without digging into the data with a fine tooth comb, I suspect that a potential problem could in fact be that, say North Shore (Team A) has performed well against the bulk of their opponents, and Team B has not faired well against the bulk of their opponents, but for what ever reason Team B matches up well against Team A and beats them twice so far in the season. So another way of saying this is the (mathematical) dilemma that is being adjusted is that Team B gets placed at a lower rank than Team A because Team B has performed more poorly "within" network than Team A, even though Team B has beat Team A.

Another issue that I see is in games that have large differentials in shots. Quite often a team may win by only a few goals, but have a huge shot differential over the losing team. Basically I take this shot differential and divide it by 10, and add the product to the composite MOV. Because of this, the variance gets bumped up, i.e. increasing the plus/minus weight and allowing this observation to be adjusted more freely. Where as a tight game with almost equal shots on both sides has more importance. In one way it is good, but in another way it could cause the problems that you IDed with North Shore, and Elk river , etc. Curious, Do the section seedings look relatively correct, over looking the head-to-head exceptions?

[LSQRANK]

Combined network; white line are "with-in" league game and yellow lines are cross-league games.



Team distributions


White lines shown each state network.


Yellow lines showing games between networks

[elliott70]

Huh?

[greenway1969]

You keep rationalizing the illogical results of your model. Your response indicates that you have very little knowledge of the teams. I understand that a mathematical model is supposedly more "objective", but you insert bias with the weight you assign to the variables. It comes across as a someone who likes math but has very little knowledge of the game and teams. It's a lousy model and you'll continue to embarrass yourself until you fix it.

[boblee]

I would advise overhauling your system. These rankings are no good at best.

[Mitch Hawker]

I like the old "asteroids" graphics, but they would be much more informative with state boundaries added for visual reference.

Did this project involve getting the GPS coordinates for each high school with a hockey team?

[black sheep]

I would like to see a year end computer vs. people ranking compared to state tournament results. Maybe it has been done before?

I know in FB QRF is usually pretty heavily scrutinized but often times comes out really close.

Head to Head is really difficult because sometimes you match up well with a particular team or style, that may not fit many other teams. Or there is an injury, or illness etc...

[sticksave]

Thank God we keep score. The algorithms you are using definitely need to be fine tuned. They will never take into account the human element of the game and the fact that some teams just figure out how to win. Everybody who follows the game knows who they are. Your design will never be able to account for that. Maybe you should look to the Big Bang theory boys like Sheldon and Leonard for help. They are about as real as this attempt. Of course I need to add I am in no way capable of following your math. I am not even close to having the ability to begin to know how you put it together. This level of math always amazes me. Gotta give you that. The fact your system puts Elk River at 15th...... well.

Karl we should all send you a thank you note for the time you spend so we don't need to follow this nonsense.

[Bigcat99]

... The algorithms you are using definitely need to be fine tuned...

Karl we should all send you a thank you note for the time you spend so we don't need to follow this nonsense.

You nailed it right there!

[thestickler07]

Why are we looking at starcharts?

[Mitch Hawker]

I would like to see a year end computer vs. people ranking compared to state tournament results. Maybe it has been done before?.

PageStat and USHSHO rankings have been beating the coaches seeding committees at picking sectional games in Minnesota (and Wisconsin) on a regular basis for over 10 years.

[elliott70]

I would like to see a year end computer vs. people ranking compared to state tournament results. Maybe it has been done before?.

PageStat and USHSHO rankings have been beating the coaches seeding committees at picking sectional games in Minnesota (and Wisconsin) on a regular basis for over 10 years.

Yes, we love them. But you and Lee make (made) adjustments. This program needs to be adjusted.