Strat-O-Matic Fielding Page

Thanks for the clarification on the BPH. I now see that in the README file for the spreadsheet; I hadn't understood its significance.

I did see, however that although you recommend NERP,

"Better yet, you should use a linear weight formula such like the NERP that Dean has used here:
http://www.mfooz.com/bblog/wp-content/u ... efense.pdf...",

you don't actually use it,

"Personally, if I were to use Dean's formula,..."

Hey, good catch.

You're right, I don't use the NERP formula for my own ratings, but I use something similar, an updated and modified version of NERP that includes catcher's defense, clutch, speed and uses different weights for almost all events---albeit the weights are very similar to what Dean has used in his NERP.

I'll write a post soon on the strategy section about my ratings (I know I promised this a few months ago, but I recently fixed a problem I had with my ratings...)

As for weighting the value of stolen bases, I wasn't satisfied with the solution of using an approximation or real life data, mostly because I knew that I rarely reproduce with my online teams the numbers of stolen bases that players are having in real life. So I chose an alternative.

I downloaded thirty 80M or 100M online seasons (maybe a bit less...going from memory), and I used some filters to make sure I had data from players with complete seasons (rather than players being used only as a bench player), and I averaged the sb and cs stats. I published those results on this forum, you can get it here:

viewtopic.php?f=5&t=638407&hilit=stealing

So, when I did my ratings earlier this year, I plugged these numbers, adjusted for playing time,in my own ratings (for example, I used 48 sb and 16 cs for Altuve, instead of 56 sb and 9 cs as in real life).

Since then, I have been able to develop my own formula for stolen bases which predicts very nicely the results for SB for 80M strat, and it doesn't rely on real life stats, so you could reproduce it simply with the SOM ratings. This said, my formula is quite complicated and needs more computation by hands than if you were to use Dean's approximation. Anyway, I put it below, perhaps for your amusement:

SB=sent*[[(first-12)/20*9.3+(second-13)/20*7.2]*(first/20-0.13)]-cs+lead_prob

where
sent stands for the chances (on 36 chances) that the runner is sent to second base
first stands for the first steal number
second stands for the second steal number
cs stands for the chances (on 36 chances) that the runner is directly caught stealing
lead_prob is the probability (based on 100%) that the runner will be held

The formula holds only if the first number is greater 12 and the second number only greater to 13, otherwise I use zero, and I round down all results, so the real formula I use in EXCEL to implement this logic is the following:

SB=ROUNDDOWN(MAX((sent*(MAX((first-12)/20*9.3,0)+MAX((second-13)/20*7.2,0))*(first/20-0.13))-cs,0)+lead_prob, 0)

If you take Puig, for example.

In real life, Puig had 11 SB and 7 CS. But based on roughly 10 seasons, Puig in fact had on average 9 SB and 5 CS.

Puig's rating is: *6,7/12 (15-9), hence

sent = 11 (5 chances for the 6 and 6 chances for the 7)
first= 15
second=9
cs=1 (chance of getting caught with 12)
lead=90% (that's based on the assumption that he has a star, which attracts a hold most of the time, but not when pitchers and catchers combine for -4 or worst, which is fairly rare).

SB=sent*(((first-12)/20*9.3)+((second-13)/20*7.2)*(first/20-0.13))-cs+lead_prob
SB=11*(((15-12)/20*9.3) +((9-13)/20*7.2) *(15/20-0.13)) -1 +0.9
SB=11*((1.395 + 0) * 0.62) -1 + 0.9
SB= 9.5 -1 + 0.9
SB= 9.4 which is rounded down to 9 with the Excel formula.

For CS, I take the smallest number between the two:

CS=ROUNDUP(sent*0.85*(MAX((first-12)/20*9.3,0)+MAX((second-13)/20*15,0))*(0.89-first/20)+cs/36*SB),0)

OR
CS= SB+2 (SB is the value for stolen bases I estimated earlier).

For Puig, 6 cs.

Very important: you then have to adjust for playing time, otherwise some players will appear way off the charts. Altuve, for example, needs to be adjusted by assuming that he leads off.

Like I said, the formula is no gimme: SOM ratings don't come with the probability of successfully sending the runner to second when trying to steal (the "sent" variable), so I have to compute it manually.

I didn't test this formula for ATG, and while I'm highly satisfied, some players with more extreme ratings (like 2-12(15-13) or players with heavy cs penalty) were more off, and these players appear much more in ATG. So there is probably some more tweaking to have better formulas.

Final note: it's quite interesting to observe that I didn't need to incorporate onbase in my formulas to predict accurately the stolen bases. Just to make sure, I calculated the correlation between onbase with how far my predicted SB was from reality and it was below r=0.12. Hence, I assume that SOM incorporates how succesful a player will get on-base when creating his sb/cs ratings.

Can't tell for sure, but I don't think this is how salary is constructed in SOM, I'm pretty sure they put a premium for stealing ability that is accompanied by good on-base.

I think this analysis is a bit beyond my current abilities. I will run the numbers for OB+TB, RC, and NERP, see what I get, and then go from there. Thanks again. I look forward to your post in the strategy section.