Sagarin player ratings

[gregmag]

I haven't seen this discussed anywhere else, so I thought I'd bring it up here. Jeff Sagarin does statistical analysis for USA Today online, including player ratings. He uses a Markov chain analysis to calculate how many runs a team made up entirely of, for example, Brady Clarks would score each game, taking into account all of the "raw" offensive stats (i.e., everything the player does himself -- not runs or RBI). I don't think he uses park adjustments. He breaks the players down into broad groups based on numbers of plate appearances. Here's the link:

 

www.usatoday.com/sports/s.../nlb05.htm

 

He also does a "fantasy analysis," which assigns a dollar value to each player based on runs, times on base, total bases, RBI, and steals discounted by caught stealings. Here's how the Brewers' regular and semiregular batters ranked based on the Markov analysis, with fantasy values in parentheses:

 

PLAYERS WITH 502 OR MORE PA ( 67 players total):

 

23. Jenkins 6.59 ($20.08)

31. Overbay 5.84 ($18.10)

33. Hall 5.80 ($18.28)

35. Clark 5.70 ($18.49)

42. Lee 5.40 ($23.30)

 

PLAYERS WITH 162 TO 501 PA (135 players total):

 

12. Cirillo 6.10 ($1.92)

15. Branyan 6.03 ($3.21)

22. Helms 5.66 ($0.21)

55. Weeks 4.76 ($12.12)

57. Miller 4.70 ($10.12)

75. Hardy 4.39 ($9.66)

122. Moeller 3.30 ($0.00)

 

I'm not going to bother with the "small sample" category, because I'm lazy, but it does confirm that Chris Magruder sucked.

 

These numbers are what they are, just one method. They do confirm to me that the argument for Carlos Lee's value, at least last year, depends a lot on (a) staying in the lineup and (b) being in the batter's box at the right time. They also make a good case for Overbay's and Clark's value, and they underscore the fine year Jenkins had.

 

Hack away.

 

Greg.

 

Edit: Unwanted emoticons

[rluzinski]

He uses a Markov chain analysis to calculate how many runs a team made up entirely of, for example, Brady Clarks would score each game, taking into account all of the "raw" offensive stats

 

But having a player interact with himself doesn't actually show his true value to a team. For instance, Bonds can't hit himself in, so using a basic runs created for only his stats overstates his contribution to runs scored. That's why Bill James adjusted his RC equation for an individual to have the player interact with an average team.

 

My understanding is that the Markov chain gives you the same results as if you just ran a simulation enough times for it to represent an average performance for the given constraints. It's essentually what any run predictor equation tries to do.

[brettac1]

BP does something on the same line as this, with MLV and MLVr. Those stats are quite useful at times.

[JeffSackmann]

My understanding is that the Markov chain gives you the same results as if you just ran a simulation enough times for it to represent an average performance for the given constraints..

 

Yeah. A markov chain is basically a mathematical expression of the average result of a simulation based on certain probabilities. Sounds really complicated, and it is to do the math rather than running a sim, but it's not.

 

---

www.BrewCrewBall.com

[endaround]

The markov chain is a sim. Basically its a fancy way of looking at a two state variable (there maybe 3 states here for walks, but likely teh two states are on base then not on base with getting on base further broken down once established.)

[Jimbo]

23. Jenkins 6.59 ($20.08)

31. Overbay 5.84 ($18.10)

33. Hall 5.80 ($18.28)

35. Clark 5.70 ($18.49)

42. Lee 5.40 ($23.30)

 

Jenkins 6.59 > Lee 5.40, but Lee $23.30 > Jenkins $20.08.

 

What gives? Both play (corner) outfield.

[endaround]

Lee's position in the lineup will lead to more RBIs. Add in about 15 SB and you have higher fantasy value.