Using STDEV to create player scores. Should I separate forwards from defense?

[OceanMon]

So I normally rank my players with a score based on a sum of the standard deviation in each category.

I am still unsure if I should use the system with all forwards and D lumped together...or if I should have it done on forwards and D separately and then lump them together. With the latter, top D get super high scores. For example, Subban would be the second most valuable player. However, if I use the former system, most of the top 20 players are forwards.

Can anybody help shed some insight on this matter?

[Sulla]

Whenever you do work like this I think it's important to consider what team structure and format you are working too (fantasyhockeygeek tool is pretty good with this). If it's a league where D points carry extra or you need to dress a lot of d men then it probably makes sense to separate and re combine?

[Eskimo Brother]

I think it makes sense to separate, then you can play with the numbers so the D and Fs line up as you wish, and then re-combine.

[JimmyMac]

You would want to separate I think. The issue is that when taking a standard deviation you normalize by the mean. For the players in your list the forwards are likely to have a higher mean for say goals and assists than the average defender. So when you separate you get a more accurate picture of how defencemen compare relative to other defencemen. Now, the scaling will be off so that forwards and defencemen are no longer directly comparable because of the different normalization. So you'll want another method that allows you to compare how relatively valuable a defenceman is compared to a forward.

[Eskimo Brother]

You would want to separate I think. The issue is that when taking a standard deviation you normalize by the mean. For the players in your list the forwards are likely to have a higher mean for say goals and assists than the average defender. So when you separate you get a more accurate picture of how defencemen compare relative to other defencemen. Now, the scaling will be off so that forwards and defencemen are no longer directly comparable because of the different normalization. So you'll want another method that allows you to compare how relatively valuable a defenceman is compared to a forward.

That's exactly what I was trying to say, but it obviously didn't come across as well. Thanks for that!

[JAMZ]

You would want to separate I think. The issue is that when taking a standard deviation you normalize by the mean. For the players in your list the forwards are likely to have a higher mean for say goals and assists than the average defender. So when you separate you get a more accurate picture of how defencemen compare relative to other defencemen. Now, the scaling will be off so that forwards and defencemen are no longer directly comparable because of the different normalization. So you'll want another method that allows you to compare how relatively valuable a defenceman is compared to a forward.

This. And the other reason I would separate is based on the roster structure. The scarcity at each position is based on the number skaters eligible to play in that roster spot. So if you have to populate D slots with a D, and F slots with a F, then you should be separating the skaters into forwards and defense. The marginal player eligible to fill a 6th D slot will have very different stats from the F eligible to fill the 6th F slot. Obviously those few D that can also play F would be included in the group where they would provide more comparable value (ie Byfuglien at D).

[OceanMon]

Thank you very much, everybody!