Return to Baseball Leftovers

Player Evaluation Formula

Moderator: Baseball Moderators

Postby shortsavage » Wed Dec 06, 2006 3:21 pm

Tavish wrote:
shortsavage wrote:Have you guys heard of derived formulas?

I mean, it's not like I came up with them randomly...had a pretty good idea of what I wanted the formula to do.


What is it you want the formula to do?


I want an offensive player value formula.

Really, this is just a method that I'm using to do my fantasy rankings this year...kind of an experiment. I struggled through calculus, so if I ever wanted to sharpen this formula enough to become something that I'd try to get published in a scientific journal, it would take me a lot of time.

Prospects are my thing...I'm just playing around with some ranking techniques that I would have had no idea how to pull off three years ago. And I bet that a majority of fantasy publications come up with their rankings through much less scientific methods than the one that I'm attempting here.

If anyone goes more complex than this with their rankings, I'd love to hear more about what you do (not trying to be cheeky, I really would like to know).
Mike Pelfrey > Matt Garza
shortsavage
General Manager
General Manager

User avatar
Cafe WriterWeb Supporter
Posts: 2931
(Past Year: -15)
Joined: 25 Aug 2003
Home Cafe: Baseball
Location: UC Davis

Postby shortsavage » Wed Dec 06, 2006 4:46 pm

I'm anxoius for some replies here...thanks in advance for your time and thought.
Mike Pelfrey > Matt Garza
shortsavage
General Manager
General Manager

User avatar
Cafe WriterWeb Supporter
Posts: 2931
(Past Year: -15)
Joined: 25 Aug 2003
Home Cafe: Baseball
Location: UC Davis

Postby Koby Schellenger » Thu Dec 07, 2006 2:45 am

If you're doing this to purely rank players you're on the right track but if you're trying to find fantasy value you're missing a step. OPS doesn't correlate well with runs, rbi, sb, hr or avg. There is some correlation between walk and strikeout ratios to average but again, you're missing a major step.

The coefficients should be found through a derivation. This can be done by simply finding the relative scarcity of each fantasy stat and figuring out how OPS, BB/K etc impacts each stat. You can then use the weighted value of each stat with the correlation between your input variables to find an accurate coefficient.

Lastly, you might want to try different inputs. You are presently counting strikeouts against a player twice (in BB/K and contact ratio) and you are counting walks in favor of the batter (BB/K) and then counting walks against the batter by using ABs. If you substituted PA for AB you would be accurately weighting playing time and you would not unfairly penalize players for drawing walks.

I would begin by running a regression of the 5x5's against your inputs to determine non-arbitrary coefficients though and work off that (weighting for stat-scarcity if you'd like)
** http://www.moundtalk.com ** MLB June Mock Draft ongoing.
Koby Schellenger
College Coach
College Coach

User avatar

Posts: 274
Joined: 4 Nov 2006
Home Cafe: Baseball
Location: Madison, WI

Postby shortsavage » Thu Dec 07, 2006 7:53 pm

Good stuff Koby but that sounds like a lot of work....scientific journal type.

I'm moving across the country and starting a work experience, so I'm just looking for something I can get going on between when I graduate and when I start my job. And I don't really think in math...just push my self to try to.

Does anyone have any examples of what they've done?
Mike Pelfrey > Matt Garza
shortsavage
General Manager
General Manager

User avatar
Cafe WriterWeb Supporter
Posts: 2931
(Past Year: -15)
Joined: 25 Aug 2003
Home Cafe: Baseball
Location: UC Davis

Postby RugbyD » Thu Dec 07, 2006 9:16 pm

for roto, just add up the standard deviations across all categories being used. the sample i use is what i consider potentially draftable players for each position. for projections, i'll prorate historicals for the last 3yrs to 162 (adjusting/eliminating anomolies), use the 50/33/17 weighting, and pro-rate down to projected GP/IP. for positions that serve or lack a specific purpose i'll add a coefficient to the particlular category to appropriately emphasize or de-emphasize it. The rarity of SBs in C and 1B overstates the value if you use the full standard deviation. Inverse with SVs and closers.
RugbyD
Hall of Fame Hero
Hall of Fame Hero

User avatar
Cafe Ranker
Posts: 5589
(Past Year: -2)
Joined: 7 Dec 2004
Home Cafe: Baseball
Location: punting small dogs and being surly

Postby bigh0rt » Thu Dec 07, 2006 9:37 pm

shortsavage wrote:I'm anxoius for some replies here...thanks in advance for your time and thought.


Does anyone have any examples of what they've done?


When you get on kicks like this you get super annoying.
bigh0rt
Hall of Fame Hero
Hall of Fame Hero

User avatar
CafeholicCafe WriterGraphics ExpertMock(ing) DrafterEagle EyeInnovative MemberCafe MusketeerWeb SupporterPick 3 Weekly Winner
Posts: 24931
(Past Year: 376)
Joined: 3 Jun 2005
Home Cafe: Baseball
Location: Crowding The Plate

Postby shortsavage » Fri Dec 08, 2006 4:19 am

bigh0rt wrote:
shortsavage wrote:I'm anxoius for some replies here...thanks in advance for your time and thought.


Does anyone have any examples of what they've done?


When you get on kicks like this you get super annoying.


It would be cool to see some examples.

Was this post really called for? I mean seriously, who are you?
Mike Pelfrey > Matt Garza
shortsavage
General Manager
General Manager

User avatar
Cafe WriterWeb Supporter
Posts: 2931
(Past Year: -15)
Joined: 25 Aug 2003
Home Cafe: Baseball
Location: UC Davis

Postby greenandgold » Fri Dec 08, 2006 7:57 pm

I've got a method that I have been working on for the last couple of seasons and I think it is good now. It is position and scoring specific for roto leagues and has a way to rate the value of a SB to a HR to a high average. It is kind of involved so if you have any questions let me know.

The idea is to make an average team, constructed by creating an average player in each position and putting the together. I do this by averaging R, HR, SB, RBI, H, AB over the "top" 20 at each position (top 60 for OF) to get the average player and from here I can calculate the stats of my average team. There is a recursive step to calculate the "top" players at each position, for an initial ranking I just use OPS. This idea is that in a typical league only the top 20 players are used at each position so the rest can be forgotten.

Once I have the average team I calculate how adding a specific player to the team (say Pujols) will change the team stats, so in this case subtract the average 1st baseman stats and add in Pujols stats. This allows me to compute how much Pujols (or any other player) will impact the average team.

Next I decide on my roto targets (right now I am using 1000/260/1000/150/.310). Once I have these I compute how close I get to these stats by having Puljos on the team and a way to rate them. I add all 5 of these things up to get a total rank and sort on this total rank to find the "top" players (this is the recursive step).

For Pujols I have
.0713 /.1535 /.0810 /.0809 /.2750
which can be read as Pujols gives me about 7% of needed runs over the average 1st baseman, 15% of homers over the average 1st baseman, 8% RBI, 8% SB, and about a third of my needed average.

For comparison, I give Carl Crawford
.0280 /-.0481 /-.0101 /.8609 /.1693
so he helps TONS in steals but he actually has lower HR and RBI stats then the average OF.


It's not really a projection formula but instead tells you how much a player is worth in a 5x5 roto league in a way that is weighted for position and scoring - for projection I just do some weighted 3 year average and adjust based on who I think will breakout/slump.

The main problem is that it gives high rankings to people who do really well in one catagory but maybe not so good in others. People like Pierre/Roberts/Roberts/Figgins are high because of their SB totals and players with high average get ranked high too. Likewise, people with low average are ranked really low (Dunn and Glaus). If I raise my target average and target SB then this problem is less but I feel bad doing that because it adds my bias instead of just going with targets that will win a league.

Here are a couple outputs from 2006 stats

R/HR/RBI/SB/AVG/TOT

C
-----
Joe Mauer /.0840 /-.0182 /.0496 /.1024 /.3471 /.5649
Brian McCann /.0051 /.0994 /.0747 /-.0193 /.2037 /.3636
Víctor Martínez /.0714 /.0139 /.0747 /-.0598 /.1432 /.2433
Iván Rodríguez /.0461 /-.0182 /.0077 /.1024 /.0138 /.1518
Paul Lo Duca /.0650 /-.1037 /-.0482 /.0010 /.1394 /.0536
Jorge Posada /.0177 /.0887 /.0747 /.0010 /-.1368 /.0453
Kenji Johjima /.0051 /.0353 /.0272 /.0010 /-.0540 /.0146
Russell Martin /.0177 /-.0502 /-.0035 /.1430 /-.1017 /.0052
Jason Kendall /.0524 /-.1464 /-.0454 /.1633 /-.0194 /.0045
Josh Willingham /.0082 /.1208 /.0216 /-.0193 /-.1469 /-.0155

1B
-----
Albert Pujols /.0714 /.1523 /.0807 /.0811 /.2769 /.6624
Ryan Howard /.0240 /.2485 /.1142 /-.0609 /.1585 /.4844
Lance Berkman /-.0044 /.1096 /.0779 /.0000 /.1626 /.3457
David Ortiz /.0587 /.2058 /.0807 /-.0406 /-.0497 /.2549
Justin Morneau /.0019 /-.0080 /.0611 /.0000 /.2231 /.2781
Travis Hafner /.0114 /.0775 /.0248 /-.0609 /.0958 /.1487
Paul Konerko /.0019 /.0027 /.0137 /-.0406 /.1508 /.1285
Jim Thome /.0366 /.0775 /.0025 /-.0609 /-.0378 /.0179
Lyle Overbay /-.0455 /-.1363 /-.0449 /.0406 /.1449 /-.0412
Nick Johnson /.0114 /-.1256 /-.0868 /.1420 /-.0229 /-.0819
greenandgold
Minor League Mentor
Minor League Mentor

User avatar
Cafe RankerMatchup Meltdown Survivor
Posts: 723
(Past Year: -23)
Joined: 11 Oct 2006
Home Cafe: Baseball
Location: Goin' home with the Armadillo

Postby Niffoc4 » Sat Dec 09, 2006 5:53 am

seems like an interesting project you are working on. Lately I have been using standard deviations added together for ranking in my big league because it uses OPS TB Holds and K/BB.

I think someone mentioned something that I agree with. I don't know that this works too well for fantasy rankings because not all of these are related to fantasy cats. For example, SBs seem to have little to do with any of these.

Like someone else said I think breaking down predictions into individual stats rather than using this for an overall ranking. I know it would be more work... but it really does seem to be more valuable for real life player evaluation than Fantasy. This may be closer to YOUR league's stats though.
One thing this does succeed at doing is giving you rankings that are different from everyone else's. Since certain parts of fantasy are a crapshoot, these rankings will at least get you drafting people in different rounds than other people.
Niffoc4
Major League Manager
Major League Manager


Posts: 1698
(Past Year: 7)
Joined: 26 Oct 2004
Home Cafe: Baseball

Postby wrveres » Sat Dec 09, 2006 5:05 pm

The more I see people create their own formulas, the more I realize that Marcel the Monkey was a genius ... ;-D
wrveres
Hall of Fame Hero
Hall of Fame Hero

User avatar
CafeholicFantasy ExpertCafe WriterEagle EyeInnovative MemberCafe Musketeer
Posts: 31204
(Past Year: -313)
Joined: 2 Mar 2003
Home Cafe: Baseball

PreviousNext

Return to Baseball Leftovers

Who is online

Users browsing this forum: No registered users and 5 guests