## Standard Deviation in Rankings

Moderator: Baseball Moderators

Use fantasy points instead of dollar amounts, and you will be on to something.

And factor in the 2002 and 2003 seasons as well, and put more weight on 2004.
--JDD
JDD
Minor League Mentor

Posts: 569
Joined: 28 Dec 2004
Home Cafe: Baseball
Location: Williamsport, PA

Looking at the SD values that I have come up with so far, Closer and Catcher are two of the positions with the least variation.

In the past, my gut has always told me to go with a bargain C or RP because they are a dime a dozen. A significant discovery, my values show 2B as being similarly close together for the 2005 season.

With Alfonzo Soriano's play dropping and Michael Young loosing 2B eligibility, this may be the year to pass on a "top 2B" and settle for a decent one (ex. Chone Figgins).
Mike Pelfrey > Matt Garza
shortsavage
General Manager

Posts: 2946
Joined: 25 Aug 2003
Home Cafe: Baseball
Location: UC Davis

just curious how you guys use StDev to figure batting average (BA) value. my formula is as follows:

((players BA - league total BA)/BA StDev)*(player AB's/league average AB's)

replace league with position specific #'s to figure positional value. is there another way or are you guys using the same formula? even though i seem to be getting good numbers with that formula, what i'm really wondering if you can get a true StDev value for simple .XXX BA numbers or is there another way to figure it that would factor in AB's and result in a 'truer' number?
mbuser
Major League Manager

Posts: 1197
(Past Year: 21)
Joined: 21 Feb 2004

Maybe one of you guys can help me out, but as I noted before, I am not sure SD is a good measure for some things. Specifically, is SD a good measure when the sample size is not a normal bell curve?

Can you fellas tell me if I am wrong here?

If you have 4-6 players of similar value at the high end of the range, then 10-12 with a significant drop afterward, wouldnt the mean by which the SD is derived be completely irrelevant?
Hall of Fame Hero

Posts: 5694
Joined: 7 Jul 2003
Home Cafe: Football

Cornbread Maxwell wrote:Maybe one of you guys can help me out, but as I noted before, I am not sure SD is a good measure for some things. Specifically, is SD a good measure when the sample size is not a normal bell curve?

Can you fellas tell me if I am wrong here?

If you have 4-6 players of similar value at the high end of the range, then 10-12 with a significant drop afterward, wouldnt the mean by which the SD is derived be completely irrelevant?

You'd like to have a sample size of 30+ to begin with. I haven't run any devs yet but it wouldn't surprise me if some came out in a skewed/lognormal distribution because there are no negative raw value numbers. A double hump might pop up as well, but that would just help in solidifying a VORP/tiered draft strategy for that position.
TennCare rocks!!!!
RugbyD
Hall of Fame Hero

Posts: 5591
Joined: 7 Dec 2004
Home Cafe: Baseball
Location: punting small dogs and being surly

RugbyD wrote:
Cornbread Maxwell wrote:Maybe one of you guys can help me out, but as I noted before, I am not sure SD is a good measure for some things. Specifically, is SD a good measure when the sample size is not a normal bell curve?

Can you fellas tell me if I am wrong here?

If you have 4-6 players of similar value at the high end of the range, then 10-12 with a significant drop afterward, wouldnt the mean by which the SD is derived be completely irrelevant?

You'd like to have a sample size of 30+ to begin with. I haven't run any devs yet but it wouldn't surprise me if some came out in a skewed/lognormal distribution because there are no negative raw value numbers. A double hump might pop up as well, but that would just help in solidifying a VORP/tiered draft strategy for that position.

I guess thats my point - if you need to use a sample size of at least 30 then what you are doing is taking all of the starting players in baseball for a certain position - of course this wouldnt be a wise thing to do since in fantasy rarely do you go over the top 15-20 unless you are in a very deep league. Since the sample size is so small, and what you are essentially trying to do is rate the top half of each position, than bell curve calculations (stan. dev.) becomes a bit obsolete. That is why I would endorse a tiered system over using stand dev.
Hall of Fame Hero

Posts: 5694
Joined: 7 Jul 2003
Home Cafe: Football

I just use a system similar to VBD where I will choose a baseline player based on how many I think will be drafted at that position. Then I rate players using their % above or below that baseline.
Maine has a good swing for a pitcher but on anything that moves, he has no chance. And if it's a fastball, it has to be up in the zone. Basically, the pitcher has to hit his bat. - Mike Pelfrey
Amazinz
Mod in Retirement

Posts: 18800
Joined: 16 Mar 2004
Home Cafe: Baseball
Location: in Canada, toughening up figure skaters

Cornbread Maxwell wrote:I guess thats my point - if you need to use a sample size of at least 30 then what you are doing is taking all of the starting players in baseball for a certain position - of course this wouldnt be a wise thing to do since in fantasy rarely do you go over the top 15-20 unless you are in a very deep league. Since the sample size is so small, and what you are essentially trying to do is rate the top half of each position, than bell curve calculations (stan. dev.) becomes a bit obsolete. That is why I would endorse a tiered system over using stand dev.

30+ is ideal, but 15-20 will do. Using std.dev is essentially the same as using tiers because all a tier is is a specific range if std.devs from the mean. The distribution just helps establish the appropriate tiers. The 1B/DH, OF, SP and MR distributions would paint the prettiest pictures of course.
TennCare rocks!!!!
RugbyD
Hall of Fame Hero

Posts: 5591
Joined: 7 Dec 2004
Home Cafe: Baseball
Location: punting small dogs and being surly

RugbyD wrote:30+ is ideal, but 15-20 will do. Using std.dev is essentially the same as using tiers because all a tier is is a specific range if std.devs from the mean. The distribution just helps establish the appropriate tiers. The 1B/DH, OF, SP and MR distributions would paint the prettiest pictures of course.

I understand thats what using SD is trying to do: 1 out from the mean, 2 out from the mean, etc... but the problem is that SD is based on a normal bell curve, and I argue that with a sample size of 15-20 of the best players at a position, there is rarely going to be a normal bell curve. Thats the whole argument. Therefore rather than using SD as a strict formula to determine tiers, I think there is probably a better way that a strict formula base would more than likely hinder.

I also want to add that a tiered approach is only based on SD - but it isnt locked into the strictness of it. Using a tiered approach and using a SD approach are not the same thing.
Hall of Fame Hero

Posts: 5694
Joined: 7 Jul 2003
Home Cafe: Football

Cornbread Maxwell wrote:but the problem is that SD is based on a normal bell curve, and I argue that with a sample size of 15-20 of the best players at a position, there is rarely going to be a normal bell curve.

This is correct. Ideally you would use a T-Distribution.
Maine has a good swing for a pitcher but on anything that moves, he has no chance. And if it's a fastball, it has to be up in the zone. Basically, the pitcher has to hit his bat. - Mike Pelfrey
Amazinz
Mod in Retirement

Posts: 18800
Joined: 16 Mar 2004
Home Cafe: Baseball
Location: in Canada, toughening up figure skaters

PreviousNext