saber stuff from old forums
Posted: Tue Nov 21, 2006 4:25 pmI copied this from the old. I thought of posting it here since the old forums are gone. The text comes from my phone's limited capacities, sorry if the formatting is not right. Also either because of TSN or my phone I'm going to have to split the original in two posts.
------------Pt 1.
peteAll-Star
Posted February 08, 2003 11:55 AM
since we have a healthy chat going about probability, I was wondering if anyone ever looked into sabermetrics??
I have just started looking into it and find it fascinating but I was hoping for some input for a good reference (bill james?)
JaserDAll-Star
Posted February 08, 2003 12:05 PM
I'm just starting to learn more and more about sabermetrics. As a huge Red Sox fan, having Bill James in the front office has motivated me to get in learning gear.
As an example of the power of saber, here is a post I am copying from a poster (Eric Van who deserves huge credit for his saber insights) on a Red Sox board I belong to regarding the maximazation of the Sox lineup next year.
Real interesting stuff (thanks Eric):
Let's see if we can figure out from numbers and game strategy what the best batting order is. We'll look at 10 players -- Mueller, not Shea, and both Millar and Ortiz.
Who needs to be knocked in?
The stat here is (H - HR + BB + HB - CS - GDP) / PA. If you don't see why you want to deduct HR (or GDP), we'll talk.
3 yrs RHP 2002
Giambi .347
Mueller .313 .357 in '99
Ramirez .326 .358
Damon .329 .320
Garciaparra .320 .290
Walker .308 .309 Fenway should boost it
Millar .302 .305
Ortiz .308 .282
Nixon .305 .285
Varitek .293 .284
AL ave. .274
I regard Mueller and Giambi's 3-year splits vs. RHP as misleading, so I've left those columns blank.
This is actually a subjective ranking. What's very clear is that there are two or three guys who will be on base to be knocked in a ton: Lil' G, Mueller (if he's on his game), and Manny (if he continues like last year) . Conversely, there are three guys who you don't want to bother putting before the RBI hitters: 'Tek, Nixon, and either Millar or Ortiz. Just below the top group is Damon and a regular Manny, below them Nomar (assuming a partial comeback) and Walker (because I expect his OBP to rise from playing in Fenway).
Who can knock 'em in?
This is new stuff. My Contextual Runs metric includes an extremely accurate projection for the percentage of runners waiting on base that get knocked in. It's basically 10% of them plus a SA-like percentage, except the weights are different (1B is .48, 2B is .83 and 3B or HR is 1.06) and the denominator is outs made, not AB or PA.
So I've calculated that percentage for everybody. This time I didn't count CS among the outs but did include GDP. What this tells you is the percentage of waiting runners that would get knocked in by a team composed of nine of the players. (I thought a ton whether that would give the results I was looking for and concluded that it would; again, I'm open for discussion.)
3 yrs RHP 2002
Ramirez .314 .345
Garciaparra .306 .278
Millar .272 .272
Damon .256 .240
Walker .247 .245
Nixon .248 .226
Ortiz .236 .243
Giambi .236
Al ave. .217
Varitek .211 .206
Mueller .208
This is surprising, but on reflection it makes sense. You can actually see that the much aligned BA is actually meaningful; it actually gives you a better idea of a player's RBI capabilities than SA. Guys like Dwight Evans with relatively low BA's but high OBP's thanks to tons of walks, and lots of power, should be hitting just in front of the very best hitters, not just behind them.
Jeremy Giambi wouldn't be a bad #5 hitter, but he would not drive in a ton of runs because he'd be walking so often. He would essentially be passing the RBI buck, much of the time, to whomever was next. That wouldn't be bad, but it would be a misuse of his skills. His biggest potential value to the team is as a guy for Manny and Nomar to knock in; his power is then gravy. And batting him higher in the order gets him up more often, never a bad idea for a hitter this good.
Putting it together so far.
Most of these hitters are actually neutral in terms of where they rank in these lists. They are equally suited to being a tablesetter or an RBI guy; some are just better at both than others. Manny, Damon, Walker, Nixon or Ortiz, Varitek -- in that order (although of course any kind of breakout year for the last three could vault them past Walker).
There are four hitters who are not neutral. Bill Mueller's value is totally as a tablesetter (I didn't even bother to calculate an RBI value for his last good full year of '99). Surprisingly, Jeremy Giambi is also much better suited to this role. Nomar and, even more so, Kevin Millar, have considerably more value as RBI guys (let's do the lineup with Millar first and then re-do it with Ortiz instead).
This suggests a first draft of:
Mueller
Giambi
Manny
Nomar
Millar
Damon
Walker
Nixon
Varitek
There are several problems with this. First, Damon is used to hitting leadoff. Moving a guy to a position where he is uncomfortable, causing a drop in production, is not worth a slight gain in lineup efficiency. And Damon is 4th on the table-setting list, way above average, so he's certainly not going to be wasted in that spot. So let's move him into leadoff, drop Mueller to 9th where he is still setting the table but is totally out of the spotlight, and move the last three guys up one. That's actually a heck of a good lineup (and Walker has hit 6th a lot). But I want to look at one more factor.
Who benefits from a runner on 1B?
OPS difference, last 4 years, with a guy on 1B vs. 1B unoccupied. Some of this is going to be random chance but much of it will be the benefit of the 1B hole when a runner is held.
Ramirez .111 .145 last 3
Walker .101 .172 last 3
Damon .076
Garciaparra .067
Nixon .066
Mueller .033
Millar .030
Varitek .024
Ortiz -.081
Giambi -.100
To use this data, we of course want to look at:
Who is on first.
(H - 2B - 3B - HR + BB + HBP - SB - CS) / PA.
3 yrs RHP 2002
Ramirez .319 .326
Mueller .279 .322 in '99
Giambi .318
Walker .256 .250
Varitek .256 .245
Garciaparra .247 .219
Millar .236 .247
Ortiz .243 .219
Nixon .237 .225
Damon .203 .219
You can see this is worth paying attention to. Bill Mueller is on 1B almost 50% more of the time than Johnny Damon, and Todd Walker gains 100 or 170 or something points of OPS when first base is occupied. You do not want him hitting second behind Damon.
How are we doing so far with this? Jeremy Giambi actually has a big negative split with a runner on 1B. That may be a fluke, but he certainly has shown no sign of benefitting from the 1B hole. Johnny Damon is on 1B less often than anyone on the team. Damon, Giambi is a perfect combination. Manny adores a runner on 1B and Lil' G lives there. Nomar likes a runner on 1B a lot and Manny lives there. Millar benefits just a bit and Nomar is there kind of middling (for this team).
Todd Walker hitting behind Nomar and Millar is not good. Millar is on 1B relatively infrequently, Nomar not much more.
Looking down the order, Varitek is on 1B a bit more than Nomar and Mueller is there all the time. Why not drop Walker down to 9th? This has the added very large advantage of separating Walker and Nixon, the two guys in the lineup who invite opposing lefty relievers (more on that below).
So now we have:
Damon
Giambi
Manny
Nomar
Millar
Nixon
Varitek
Mueller
Walker
which is of course the lineup I've been suggesting for a few weeks. Now you know why! (In case you can't tell, in this study I actually did re-address the issue from the beginning, with an open mind.)
How is this lineup for dealing with opposing lefty relievers? It's great. Nixon's the first of the lefties, which is very desirable. If they bring in a lefty, you can pinch-hit with Gant (or whomever). Varitek's platoon split is slight, Mueller likes lefties better. Burying Walker in the middle of this sequence is key. You can pinch-hit Mirabelli or Merloni, and guarenteed the opposing manager will leave the lefty in. Because if they bring in the righty now, you send Ortiz up, and then the righty faces Damon and Giambi instead of the lefty. (Yeah, I know that Damon has a reverse and Giambi a neutral platoon split, but those splits are against all lefties; against tough lefty relievers they would have at least a mild split the normal way. And nothing we can do with the lineup is going to prevent those guys from seeing some lefties.)
In contrast, if you bat Walker second you will always be burning two players if you pinch hit, because the opposing manager immediately brings in his righty to face 'Belli (or Merloni), Manny, Nomar, forcing you to hit Ortiz for 'Belli. I think there are better uses for 'Belli than to be announced as a PH.
Let's not think about it.
What's the best lineup with Ortiz instead of Millar? I don't think there's one anywhere near as perfect as this. Millar is the perfect #5 hitter for this team (at least until Nixon or 'Tek has a breakout season). I would seriously entertain hitting Walker 5th, though (although you might want to wait a month to see if he takes to Fenway as well as we hope). A radical solution would be to hit Damon 5th. But any of these changes would force other adjustments. I don't want to spend that mental energy until I know I need to.
peteAll-Star
Posted February 08, 2003 02:57 PM
JaserD,
Awesome post, both for the fact that its a good discussion of the sabermetrics approach, and to the fact that it's about the Sox (which are dear to my bleeding heart as well)...lol
now I just want to digest it for awhile...
I've been trying to use some part of the sabermetric system in my analysis of the ratings disk information, but, to this point this is all I have (given the not so complete amount of data on the disk)...
I got this from
strato saber discussion
-------------------
Below you'll find batters' SOM Card Avg, Oba, Slg and OPS with a couple of other things thrown in. This is what the players would hit if all rolls came off the hitter's card ONLY. I figured BP-HR and BP-S using a neutral park (1-9).
OPS = On Base + Slugging -- A quick and pretty accurate measure of offensive performance
Rate = (.3*OPSvLH) + (.7*OPSvRH) -- Because approximately 30% of all plate appearance come vs. LH pitching, this is a pretty good estimation of each players overall value.
Bval = Rate * Plate Appearances
--------------------
it's not rocket science but it may be enough to help weed out some players, I wouldn't mind including the equation from Erics post above, but the stats disk does not have the complete breakdown needed
[This message was edited by Pete on February 08, 2003 at 03:07 PM.]
Mike_JAll-Star
Posted February 08, 2003 03:27 PM
Talk to luckyman, because, as far as I've seen, he's the sabermetric king on this board. I have a grounding in the basics, and I use it with the ratings disk to do evaluations. I use a 2-step process, first calculating runs created for players based on 216 PA's (I use what's on their card for 1/2, then an estimate I can live with of the average pitcher). I then use that to get a per 27 out number (number of runs scored if that player was every batter in a game), which I think is a better measure of offensive efficiency. Both concepts, though, are straight from Bill James's mouth.
I can give (relatively) simple formulas for this stuff, if you'd like.
peteAll-Star
Posted February 08, 2003 03:30 PM
so to me, what I included, is measuring a players ability to get on base and the ability to advance people on base effectively...
I would also recommend the following...
intro to sabermetrics
from this page I have also found a few formula's that I may attempt to incorporate, for example...
(HITS + WALKS) (TOTAL BASES)RUNS = ---------------------------- AT-BATS + WALKS
my biggest problem is do I want to try to collect data from other locations to crunch, or to stick with the ratings disk...
I would prefer to go more the route of the linear comparison because it seems to be a bit more accurate than bill james original version
JaserDAll-Star
Posted February 08, 2003 03:47 PM
Pete, I haven't seen the ratings disk (just ordered it for the first time for 2002), but the problem I foresee in trying to analyze alot of "real player data" from other places is that (it's my understanding) the Strat cards already take alot of this into consideration but then adjust for ball parks.
The data you gather in other places most likely will not have ball park adjustments (at least the common stat sites don't) so it may skew your player analysis in taking the "real"data and then making similar assumptions for Strat cards. I could be wrong on this, but it's my percpetion.
This would also hold true for the example I posted from Eric Van in looking at "real" statistical data for true OBP (i.e. -HR), Contextual Runs, frequency of getting to 1B only and OPS delta with a runner on 1B only and trying to correlate this info into a Strat lineup strategy. While it's probably close, I doubt it's as accurate as the real data (again for BP adjustments on the cards)
Mike_JAll-Star
Posted February 08, 2003 04:17 PM
The great thing about incorporating sabermetric formulas with the ratings disk is that all the hard, solid numbers you need are right there in Xcel, ready to be plugged into calculations. The problem is that certain things are going to be hard, or impossible, to measure. And linear weights are difficult because they involve manual counting- the disk tells you total number of hits, walks, HR's, TB's, OB chances, out of 108, but doesn't tell you anything about doubles, triples, etc.
So, using the info on the disk, you can run a very simple runs created (per side) by taking OB chances* TB chances/ 108. Its important NOT to translate OB and TB's into percentages, because that's not what the RC formula needs- it needs amounts, because its a cumulative measure, not an average. If you want to incorporate BP effects, its pretty simple. Choose your own Park Averages (as a percentage- 1-10 would be .5) based on circumstance (i.e, home park average, or league average, or combo).
You get (OB chances +(Park Average * BPsingles)+ (Park Average* BPHR's)) * (TB chances + (Average*BPsingles)+ (4*(Average*BPHR's)))/ 108.
This gives you a runs created total for the 108 chances on a hitter's card.
To translate that to more of an efficiency measurement, convert it to a RC per 27 outs.
This formula is. (Runs created/(108- (OB chances+ ( Park average * BPsingles)+(average*BPHR's))*27.
You can work other factors into these formulas, but they are based more on guesswork and situational analysis. Stuff like GBA's, baserunning speed, steal ratings, ** singles, clutch, and defense, are harder to figure into these formulas, but at least these formulas give you a solid foundation. The RC formula was accepted as valid because of its predictive abilities- plug in the OB, TB, and PA numbers for entire teams, and you get very close to the actual number of runs they scored.
Linear weights theories, to me, always seem to be adjusted so that the numbers come out right- a single is worth this much, a double this much, because it makes the formula more accurate, not neccessarily because it measures any "true" value.
luckymanAll-Star
Posted February 08, 2003 04:38 PM
Sabermetrics is a very very interesting (and effective) to analyse the contribution of any baseball player, and I've adopted some of its basics into my playing (thanks for the compliment, Mike_J )
In my next post, I'll tackle Pete's problem, of how to adjust sabermetrics logic with the ratings you have on your strat disk, but first let me say in this post one or two things.
Basically, the logic behind sabermetrics is, as James argued, that "a hitter should be evaluated by his ability to create runs for his team". So what we are looking for is a statistic we will call "Runs created" (RC). The site Pete proposes is a good introduction to the subject, although two clarifications must be done.
First, a lot of different formulas exist. Despite the differences in formulas, for general purpose, their quality to predict the quality of an offensive performance is similar. For those of you interested to look at some of these:
www.baseballstuff.com/fraser/stats2000/stats.html
www.stephent.com/jays/erp.html
Second, the linear weight formula that is proposed in the intro paper has two mistakes. First, it doesn't include outs. Second, since the formula has been introduced, some people showed that the original formula was biased. To correct it, you must give a negative value to outs. In short, the logic is that when you have an out, your chance to score runs is reduced (as opposed to a foul ball, for example, an event where your chance to score a run is not reduced). So the linear formula people now use has changed for something like (notice the minus aside outs):
RC = .47*1B + .78*2b + 1.09*3b + 1.40*HR + .33*(BB+HBP) - .3*(AB-H).
If you want to adjust it for stolen bases and caught stealing, it becomes:
RC = .47*1B + .78*2b + 1.09*3b + 1.40*HR + .33*(BB+HBP) - .3*(AB-H) + .31*SB - .60*CS.
penngrayAll-Star
Posted February 08, 2003 04:54 PM
I wish the ratings disk broke out the 2b,3b etc.
peteAll-Star
Posted February 08, 2003 05:25 PM
Thanks luckyman and Mike_J for the input...
lucky, I had found the formula you stated previously and that is what had led me to thinking that I would have to get stats elsewhere(2b's, 3b's, etc) to complete the formula...
so for now I will baby step, I may stick with the formula I listed above, even though it is flawed, I want to start with some base of reference and work from there...
the formula being
(HITS + WALKS) (TOTAL BASES)RUNS = ---------------------------- AT-BATS + WALKS
I was also thinking that if I ran 2002 player stats through the formulas it would help to spot the cream rising to the top, it may not neccessarily pinpoint the best strat-o performers but at least give a good estimation on how they "should" perform
luckymanAll-Star
Posted February 08, 2003 06:01 PM
Mike_J addresses the problem quite right:
quote:
And linear weights are difficult because they involve manual counting- the disk tells you total number of hits, walks, HR's, TB's, OB chances, out of 108, but doesn't tell you anything about doubles, triples, etc.
What you have, on the strat disk, is on-base, total base and hits. By the way, notice that, since you have hits, you can calculate how much walks and hbp each card has by doing:
on-base - hits = (W+HBP)
Thus, I suggest you that you create an additional column in Excel Strat sheet with the upper formula (if you don't know how, I could explain.)
As Mike_J says, you could do ONB*TB/108, but the problem is, if you go that way, you won't be able to adjust for speed, defense, and the like. To incorporate the latter stats, you really have to keep the logic of linear weight. But you don't have the stats for singles,double, and triples, so what can you do?
Here is the answer. First, look at the weights I came up with in the last formula of my last post. The weights are:
(W+HBP) ---> 0.33
1B --------> 0.47
2B --------> 0.78
3B --------> 1.09
HR --------> 1.40
Outs-------> minus 0.3
Since this is a linear model, I can add or multiply by a constant, and the order among the players will be preserved (in other words, if Jeter is better than Tejada in the original weighted formula, he will still be better in the transformed formula). So a add 0.3 and multiply 2.8 each value, I will end up with the following weights:
(W+HBP) ---> 1.76
1B --------> 2.16
2B --------> 3.02
3B --------> 3.90
HR --------> 4.77
Outs-------> 0
Now, suppose we ADD for each card chances of on-base and chances of total bases, as listed on your strat disk. You know that, for chances of total bases, you have one point for a single, two points for a double, etc. So what this value (on-base chances plus total base chances) is really doing is to give the following weights:
(W+HBP) ---> 1
1B --------> 2
2B --------> 3
3B --------> 4
HR --------> 5
Outs-------> 0
If you compare the last two lists of weights, you see that, except for walks, the two lists are very similar.
Let's fix the weight for walks. In the former list of the two lists we are comparing, walks had a value of 1.76, and singles a value of 2.16. So the value for walks was the equivalent of 80% the value of singles. So if we give the value of 2 for singles, we should have a weight of 1.6 for walks.
So here is the twist I suggest: to have a value of "Runs created" that approximates very well the linear formula from the stats you find on the STRAT disk, you just have to create another column in Excel Strat sheet with the following formula:
=ONB + TB + 0.6*(W+HBP).
(0.6 instead of 1.6 because ONB incorporates 1 point for (W+HBP))
This will give a linear weight of Runs created, with the underlying following weights:
(W+HBP) ---> 1.6
1B --------> 2
2B --------> 3
3B --------> 4
HR --------> 5
Outs-------> 0
Since we have a linear weight RC, we will be able to adjust this RC for:
- ball park adjustment
- stolen bases
- defense
- gbA
- clutch
- star (forcing holding)
I won't get in the details of each but here the final formula I end up with. For each side (vs lhp and vs rhp), compute:
1) RC=(W+HBP)*0.6 + ONB + TB + BPadjustment + "weak" adjustment + clutch/4 + gbA/5;
where BP adjustment is adjustment for Ball Parks homerun. The formula for this is
BP adjustment = k/4 * BP
-BP is the number of Ball park homeruns on a given card
-k is the stadium homerun ratio (for Coors, it would be 19, for a neutral park, it would be 10).
AND where "weak adjustment" is :
-7 for a player being weak
-5 for a player NOT having ball park singles on his card
-12 for a player having both characteristics
0 for normal players with ball park singles
2) running adjustment= (stealing + speed/2)-6
where stealing is
+6 for a playing having a star and a 19 number for stealing
+5 for a playing having a star and a 18 number for stealing
+4 for a playing having a star and a 17 number for stealing
+3 for a playing having a star and a 16 number for stealing
+2 for a playing having a star and a 15 number for stealing or less
and where speed is speed rating (ex.for Jeter, 17)
3) offensive RC = 0.3*(RC vs lhp) + 0.7*(RC vs rhp) + running adjustment
4) overall RC: offensive RC - defensive RC
I come back to defensive RC in another post.
[This message was edited by luckyman on February 08, 2003 at 07:00 PM.]
luckymanAll-Star
Posted February 08, 2003 06:21 PM
Penngray, about
quote:
I wish the ratings disk broke out the 2b,3b etc.
.
That's very easy to do with Excel.
First, to be sure you don't loose any data, select the whole set of data, copy it and paste it in another spread sheet (the width of columns will change). Second, I suggest that you delete all the non-eligible cards. Then, you click in the first cell under CA (where you find defensive ratings), and press the "ranking" button that looks like this:
A (with an arrow beside)
Z
By doing so, you'll have all the catchers grouped together. You can copy (not cut) and paste it in another sheet called "catcher".
Then you just repeat the procedures for 1B, 2B etc.
penngrayAll-Star
Posted February 08, 2003 06:35 PM
lol, I know how to do the player categories.
actually I use auto-filters and then just filter my defensive ratings that "contain" ss, c, cf etc. when I need to look at one position.
I was saying, I wish the spreadsheet had the hit totals for doubles and triples
luckymanAll-Star
Posted February 08, 2003 06:43 PM
oops!!!
penngrayAll-Star
Posted February 08, 2003 06:45 PM
my original post was really vague
luckymanAll-Star
Posted February 08, 2003 06:47 PM
For defensive ratings:
for INFIELDERS
defense RC = 0,7*e +((range-1)*pos)
where e is the e rating
range is the range rating
pos is:
3.6 for 1b
10.8 for 2b
12.6 for ss
5.4 for 3b
For lf and rf, the formula is:
defensive RC = rr + e
where e is the e-rating, and rr is:
0 for range 1
6 for range 2
13 for range 3
27 for range 4
35 for range 5
For cf, it's the same, except you multiply rr by 1.5.
P.S.
For ss and 2b, you may also want to adjust for double-plays (gb A or gb 1). A simple way is to substract 6 from defensive rating for ss-1, 5 for 2b-1, 3 for ss-2 and 2 for 2b-2.
A sophisticated way is the follow the next formula. With Excel, there is a formula called MAX, with gives the highest numbers of the ones you specify. The formula is:
adjustment for gbA: =MAX((40-defensive RC)*0,18; 0)
luckymanAll-Star
Posted February 08, 2003 06:52 PM
So Pete,
forgive me if you got lost in the ocean of formulas I just threw in. But the whole idea is that the first step should really be:
offensive RC = 0.6*(W+HBP) + ONB + TB.
This keeps the logic of linear weight formula. It's simple. And it enables you to do all the adjustments, espiecially for defense
(so if you don't want to get in the details of all adjustments, I suggest you compute offensive RC with the simple formula above, adjust it to BP homerun in a neutral stadium, and substract from this the defensive RC allowed by each hitter. This gives the following, in two steps:
offensive RC = 0.6*(W+HBP) + ONB + TB + 2.5*BP
overall RC = 0.3*(RC vs lhp) + 0.7*(RC vs rhp) - defensive RC
where BP is the number of BP homeruns you find on a card)
Mike_JAll-Star
Posted February 08, 2003 07:19 PM
That's some pretty awesome stuff lucky, but I have a few question. Where are the clutch and GBA divisiors coming from? Using the divisors as a judge, you are saying that GBA situations occur about 20%, and clutch about 25%, of the time? Or am I missing something? Interested in the source of those values.
And for the defensive ratings, I'm fairly lost with the logic. I see how the pos values are based on chance ratios for positions, but I don't understand how that formula adds up to what a player's defense will actually allow.
peteAll-Star
Posted February 08, 2003 08:45 PM
lucky,
thanks for the reply, I appreciate everything you put up, though it may take me a bit to absorb...
if you have any more insights it would be welcome
I am currently building an sql database that will allow me to sort by ratings, I will go back and add a column based upon your formula
luckymanAll-Star
Posted February 08, 2003 09:13 PM
Mike_J,
Don't wonder why you can't understand how I came up with these numbers: for reasons of simplicity, I didn't mention it!! So you're not missing something, it's rather me who's hidding my assumptions.
Clutch
Let's take Edmonds numbers vs rhp:
W+HBP= 22
ONB = 52.3
TB = 50.7
BP = 7
clutch = -12
Basic RC = 0.6*(22) + 52.3 + 50.7 + 2.5*(7)
= 133.7
Now, we know that clutch situations happen around 12% of the time (that's on average, the 5th hitter in a line-up gets more, close to 15%, but let's keep it simple and take 12%)
So this basic RC, 133.7, holds 88%. In clutch situations, Edmonds' card lost 12 singles. So his basic RC in clutch situation is :
Clutch basic RC = 0.6*(22) + 40.3 + 38.7 + 2.5*(7)
=109.7
So Edmonds rating, when adjusted to clutch, is:
RC = (0.88)*133.7 + (0.12)*109.7 = 130.82
But notice that we would have approximately the same result if we had use this formula:
RC = basic RC + clutch/4 = 133.7 + (-12/4) = 130.7
gbA
A double-play is worth about two outs. If we adjust our weights to give 0 to outs, a dp is worth about the opposite of a walk. So, in our system, a double play is worth about -1.6.
But, as you know, gbA doesn't turn out in double plays each time. It only does so when there is someone on first with less than two outs. If we knew the probability that this situation occurs, then we would know the weight for gbA. I can't remember the source of this information, but I think that on average, it's between around 18% (again, it depends on where a player stands in the line-up).
So a rough approximation of the weight we should attribute to gbA is:
gbA = 18% X -1.6 = -0.29.
That being said, there will be situation where there will be gbA with guys on first/less than 2 outs, and the outcome won't be a double-play. One example is when defense is playing in, another example is if you look at the chart, with guys on first and third and the reading is gbA (3b), the play will be at home, etc.
So in strat life, gb(A) will be worth roughly between -0.20 and -0.25. So a good approximation will be around to weight :
-0.25*gb(A), or gb(A)/4
or 0.2*gb(A), or gb(A)/5,
but certainly not gb(A) taken as an unity.
(the reason I chose gb(A)/5 is because in fact, a dp is worth a little bit less than the opposite of a walk, roughly 85%, but I'm sure that gb(A)/4 would work out well too).
Mike_JAll-Star
Posted February 08, 2003 09:40 PM
I think that's a great system, all based on sound logic and mathematical principles. One thing I noticed though, is that you have a + for GBA's- shouldn't that be a minus? And after popping the system through the ratings disk, the only thing that leaps out as "questionable" is the possible overvalue of walks- Durazo stands out. But I think that's an issue with sabermetrics in general, not your specific formulas.
penngrayAll-Star
Posted February 08, 2003 10:01 PM
what do you do with the BP singles? they arent on the rating disk either.
------------Pt 1.
peteAll-Star
Posted February 08, 2003 11:55 AM
since we have a healthy chat going about probability, I was wondering if anyone ever looked into sabermetrics??
I have just started looking into it and find it fascinating but I was hoping for some input for a good reference (bill james?)
JaserDAll-Star
Posted February 08, 2003 12:05 PM
I'm just starting to learn more and more about sabermetrics. As a huge Red Sox fan, having Bill James in the front office has motivated me to get in learning gear.
As an example of the power of saber, here is a post I am copying from a poster (Eric Van who deserves huge credit for his saber insights) on a Red Sox board I belong to regarding the maximazation of the Sox lineup next year.
Real interesting stuff (thanks Eric):
Let's see if we can figure out from numbers and game strategy what the best batting order is. We'll look at 10 players -- Mueller, not Shea, and both Millar and Ortiz.
Who needs to be knocked in?
The stat here is (H - HR + BB + HB - CS - GDP) / PA. If you don't see why you want to deduct HR (or GDP), we'll talk.
3 yrs RHP 2002
Giambi .347
Mueller .313 .357 in '99
Ramirez .326 .358
Damon .329 .320
Garciaparra .320 .290
Walker .308 .309 Fenway should boost it
Millar .302 .305
Ortiz .308 .282
Nixon .305 .285
Varitek .293 .284
AL ave. .274
I regard Mueller and Giambi's 3-year splits vs. RHP as misleading, so I've left those columns blank.
This is actually a subjective ranking. What's very clear is that there are two or three guys who will be on base to be knocked in a ton: Lil' G, Mueller (if he's on his game), and Manny (if he continues like last year) . Conversely, there are three guys who you don't want to bother putting before the RBI hitters: 'Tek, Nixon, and either Millar or Ortiz. Just below the top group is Damon and a regular Manny, below them Nomar (assuming a partial comeback) and Walker (because I expect his OBP to rise from playing in Fenway).
Who can knock 'em in?
This is new stuff. My Contextual Runs metric includes an extremely accurate projection for the percentage of runners waiting on base that get knocked in. It's basically 10% of them plus a SA-like percentage, except the weights are different (1B is .48, 2B is .83 and 3B or HR is 1.06) and the denominator is outs made, not AB or PA.
So I've calculated that percentage for everybody. This time I didn't count CS among the outs but did include GDP. What this tells you is the percentage of waiting runners that would get knocked in by a team composed of nine of the players. (I thought a ton whether that would give the results I was looking for and concluded that it would; again, I'm open for discussion.)
3 yrs RHP 2002
Ramirez .314 .345
Garciaparra .306 .278
Millar .272 .272
Damon .256 .240
Walker .247 .245
Nixon .248 .226
Ortiz .236 .243
Giambi .236
Al ave. .217
Varitek .211 .206
Mueller .208
This is surprising, but on reflection it makes sense. You can actually see that the much aligned BA is actually meaningful; it actually gives you a better idea of a player's RBI capabilities than SA. Guys like Dwight Evans with relatively low BA's but high OBP's thanks to tons of walks, and lots of power, should be hitting just in front of the very best hitters, not just behind them.
Jeremy Giambi wouldn't be a bad #5 hitter, but he would not drive in a ton of runs because he'd be walking so often. He would essentially be passing the RBI buck, much of the time, to whomever was next. That wouldn't be bad, but it would be a misuse of his skills. His biggest potential value to the team is as a guy for Manny and Nomar to knock in; his power is then gravy. And batting him higher in the order gets him up more often, never a bad idea for a hitter this good.
Putting it together so far.
Most of these hitters are actually neutral in terms of where they rank in these lists. They are equally suited to being a tablesetter or an RBI guy; some are just better at both than others. Manny, Damon, Walker, Nixon or Ortiz, Varitek -- in that order (although of course any kind of breakout year for the last three could vault them past Walker).
There are four hitters who are not neutral. Bill Mueller's value is totally as a tablesetter (I didn't even bother to calculate an RBI value for his last good full year of '99). Surprisingly, Jeremy Giambi is also much better suited to this role. Nomar and, even more so, Kevin Millar, have considerably more value as RBI guys (let's do the lineup with Millar first and then re-do it with Ortiz instead).
This suggests a first draft of:
Mueller
Giambi
Manny
Nomar
Millar
Damon
Walker
Nixon
Varitek
There are several problems with this. First, Damon is used to hitting leadoff. Moving a guy to a position where he is uncomfortable, causing a drop in production, is not worth a slight gain in lineup efficiency. And Damon is 4th on the table-setting list, way above average, so he's certainly not going to be wasted in that spot. So let's move him into leadoff, drop Mueller to 9th where he is still setting the table but is totally out of the spotlight, and move the last three guys up one. That's actually a heck of a good lineup (and Walker has hit 6th a lot). But I want to look at one more factor.
Who benefits from a runner on 1B?
OPS difference, last 4 years, with a guy on 1B vs. 1B unoccupied. Some of this is going to be random chance but much of it will be the benefit of the 1B hole when a runner is held.
Ramirez .111 .145 last 3
Walker .101 .172 last 3
Damon .076
Garciaparra .067
Nixon .066
Mueller .033
Millar .030
Varitek .024
Ortiz -.081
Giambi -.100
To use this data, we of course want to look at:
Who is on first.
(H - 2B - 3B - HR + BB + HBP - SB - CS) / PA.
3 yrs RHP 2002
Ramirez .319 .326
Mueller .279 .322 in '99
Giambi .318
Walker .256 .250
Varitek .256 .245
Garciaparra .247 .219
Millar .236 .247
Ortiz .243 .219
Nixon .237 .225
Damon .203 .219
You can see this is worth paying attention to. Bill Mueller is on 1B almost 50% more of the time than Johnny Damon, and Todd Walker gains 100 or 170 or something points of OPS when first base is occupied. You do not want him hitting second behind Damon.
How are we doing so far with this? Jeremy Giambi actually has a big negative split with a runner on 1B. That may be a fluke, but he certainly has shown no sign of benefitting from the 1B hole. Johnny Damon is on 1B less often than anyone on the team. Damon, Giambi is a perfect combination. Manny adores a runner on 1B and Lil' G lives there. Nomar likes a runner on 1B a lot and Manny lives there. Millar benefits just a bit and Nomar is there kind of middling (for this team).
Todd Walker hitting behind Nomar and Millar is not good. Millar is on 1B relatively infrequently, Nomar not much more.
Looking down the order, Varitek is on 1B a bit more than Nomar and Mueller is there all the time. Why not drop Walker down to 9th? This has the added very large advantage of separating Walker and Nixon, the two guys in the lineup who invite opposing lefty relievers (more on that below).
So now we have:
Damon
Giambi
Manny
Nomar
Millar
Nixon
Varitek
Mueller
Walker
which is of course the lineup I've been suggesting for a few weeks. Now you know why! (In case you can't tell, in this study I actually did re-address the issue from the beginning, with an open mind.)
How is this lineup for dealing with opposing lefty relievers? It's great. Nixon's the first of the lefties, which is very desirable. If they bring in a lefty, you can pinch-hit with Gant (or whomever). Varitek's platoon split is slight, Mueller likes lefties better. Burying Walker in the middle of this sequence is key. You can pinch-hit Mirabelli or Merloni, and guarenteed the opposing manager will leave the lefty in. Because if they bring in the righty now, you send Ortiz up, and then the righty faces Damon and Giambi instead of the lefty. (Yeah, I know that Damon has a reverse and Giambi a neutral platoon split, but those splits are against all lefties; against tough lefty relievers they would have at least a mild split the normal way. And nothing we can do with the lineup is going to prevent those guys from seeing some lefties.)
In contrast, if you bat Walker second you will always be burning two players if you pinch hit, because the opposing manager immediately brings in his righty to face 'Belli (or Merloni), Manny, Nomar, forcing you to hit Ortiz for 'Belli. I think there are better uses for 'Belli than to be announced as a PH.
Let's not think about it.
What's the best lineup with Ortiz instead of Millar? I don't think there's one anywhere near as perfect as this. Millar is the perfect #5 hitter for this team (at least until Nixon or 'Tek has a breakout season). I would seriously entertain hitting Walker 5th, though (although you might want to wait a month to see if he takes to Fenway as well as we hope). A radical solution would be to hit Damon 5th. But any of these changes would force other adjustments. I don't want to spend that mental energy until I know I need to.
peteAll-Star
Posted February 08, 2003 02:57 PM
JaserD,
Awesome post, both for the fact that its a good discussion of the sabermetrics approach, and to the fact that it's about the Sox (which are dear to my bleeding heart as well)...lol
now I just want to digest it for awhile...
I've been trying to use some part of the sabermetric system in my analysis of the ratings disk information, but, to this point this is all I have (given the not so complete amount of data on the disk)...
I got this from
strato saber discussion
-------------------
Below you'll find batters' SOM Card Avg, Oba, Slg and OPS with a couple of other things thrown in. This is what the players would hit if all rolls came off the hitter's card ONLY. I figured BP-HR and BP-S using a neutral park (1-9).
OPS = On Base + Slugging -- A quick and pretty accurate measure of offensive performance
Rate = (.3*OPSvLH) + (.7*OPSvRH) -- Because approximately 30% of all plate appearance come vs. LH pitching, this is a pretty good estimation of each players overall value.
Bval = Rate * Plate Appearances
--------------------
it's not rocket science but it may be enough to help weed out some players, I wouldn't mind including the equation from Erics post above, but the stats disk does not have the complete breakdown needed
[This message was edited by Pete on February 08, 2003 at 03:07 PM.]
Mike_JAll-Star
Posted February 08, 2003 03:27 PM
Talk to luckyman, because, as far as I've seen, he's the sabermetric king on this board. I have a grounding in the basics, and I use it with the ratings disk to do evaluations. I use a 2-step process, first calculating runs created for players based on 216 PA's (I use what's on their card for 1/2, then an estimate I can live with of the average pitcher). I then use that to get a per 27 out number (number of runs scored if that player was every batter in a game), which I think is a better measure of offensive efficiency. Both concepts, though, are straight from Bill James's mouth.
I can give (relatively) simple formulas for this stuff, if you'd like.
peteAll-Star
Posted February 08, 2003 03:30 PM
so to me, what I included, is measuring a players ability to get on base and the ability to advance people on base effectively...
I would also recommend the following...
intro to sabermetrics
from this page I have also found a few formula's that I may attempt to incorporate, for example...
(HITS + WALKS) (TOTAL BASES)RUNS = ---------------------------- AT-BATS + WALKS
my biggest problem is do I want to try to collect data from other locations to crunch, or to stick with the ratings disk...
I would prefer to go more the route of the linear comparison because it seems to be a bit more accurate than bill james original version
JaserDAll-Star
Posted February 08, 2003 03:47 PM
Pete, I haven't seen the ratings disk (just ordered it for the first time for 2002), but the problem I foresee in trying to analyze alot of "real player data" from other places is that (it's my understanding) the Strat cards already take alot of this into consideration but then adjust for ball parks.
The data you gather in other places most likely will not have ball park adjustments (at least the common stat sites don't) so it may skew your player analysis in taking the "real"data and then making similar assumptions for Strat cards. I could be wrong on this, but it's my percpetion.
This would also hold true for the example I posted from Eric Van in looking at "real" statistical data for true OBP (i.e. -HR), Contextual Runs, frequency of getting to 1B only and OPS delta with a runner on 1B only and trying to correlate this info into a Strat lineup strategy. While it's probably close, I doubt it's as accurate as the real data (again for BP adjustments on the cards)
Mike_JAll-Star
Posted February 08, 2003 04:17 PM
The great thing about incorporating sabermetric formulas with the ratings disk is that all the hard, solid numbers you need are right there in Xcel, ready to be plugged into calculations. The problem is that certain things are going to be hard, or impossible, to measure. And linear weights are difficult because they involve manual counting- the disk tells you total number of hits, walks, HR's, TB's, OB chances, out of 108, but doesn't tell you anything about doubles, triples, etc.
So, using the info on the disk, you can run a very simple runs created (per side) by taking OB chances* TB chances/ 108. Its important NOT to translate OB and TB's into percentages, because that's not what the RC formula needs- it needs amounts, because its a cumulative measure, not an average. If you want to incorporate BP effects, its pretty simple. Choose your own Park Averages (as a percentage- 1-10 would be .5) based on circumstance (i.e, home park average, or league average, or combo).
You get (OB chances +(Park Average * BPsingles)+ (Park Average* BPHR's)) * (TB chances + (Average*BPsingles)+ (4*(Average*BPHR's)))/ 108.
This gives you a runs created total for the 108 chances on a hitter's card.
To translate that to more of an efficiency measurement, convert it to a RC per 27 outs.
This formula is. (Runs created/(108- (OB chances+ ( Park average * BPsingles)+(average*BPHR's))*27.
You can work other factors into these formulas, but they are based more on guesswork and situational analysis. Stuff like GBA's, baserunning speed, steal ratings, ** singles, clutch, and defense, are harder to figure into these formulas, but at least these formulas give you a solid foundation. The RC formula was accepted as valid because of its predictive abilities- plug in the OB, TB, and PA numbers for entire teams, and you get very close to the actual number of runs they scored.
Linear weights theories, to me, always seem to be adjusted so that the numbers come out right- a single is worth this much, a double this much, because it makes the formula more accurate, not neccessarily because it measures any "true" value.
luckymanAll-Star
Posted February 08, 2003 04:38 PM
Sabermetrics is a very very interesting (and effective) to analyse the contribution of any baseball player, and I've adopted some of its basics into my playing (thanks for the compliment, Mike_J )
In my next post, I'll tackle Pete's problem, of how to adjust sabermetrics logic with the ratings you have on your strat disk, but first let me say in this post one or two things.
Basically, the logic behind sabermetrics is, as James argued, that "a hitter should be evaluated by his ability to create runs for his team". So what we are looking for is a statistic we will call "Runs created" (RC). The site Pete proposes is a good introduction to the subject, although two clarifications must be done.
First, a lot of different formulas exist. Despite the differences in formulas, for general purpose, their quality to predict the quality of an offensive performance is similar. For those of you interested to look at some of these:
www.baseballstuff.com/fraser/stats2000/stats.html
www.stephent.com/jays/erp.html
Second, the linear weight formula that is proposed in the intro paper has two mistakes. First, it doesn't include outs. Second, since the formula has been introduced, some people showed that the original formula was biased. To correct it, you must give a negative value to outs. In short, the logic is that when you have an out, your chance to score runs is reduced (as opposed to a foul ball, for example, an event where your chance to score a run is not reduced). So the linear formula people now use has changed for something like (notice the minus aside outs):
RC = .47*1B + .78*2b + 1.09*3b + 1.40*HR + .33*(BB+HBP) - .3*(AB-H).
If you want to adjust it for stolen bases and caught stealing, it becomes:
RC = .47*1B + .78*2b + 1.09*3b + 1.40*HR + .33*(BB+HBP) - .3*(AB-H) + .31*SB - .60*CS.
penngrayAll-Star
Posted February 08, 2003 04:54 PM
I wish the ratings disk broke out the 2b,3b etc.
peteAll-Star
Posted February 08, 2003 05:25 PM
Thanks luckyman and Mike_J for the input...
lucky, I had found the formula you stated previously and that is what had led me to thinking that I would have to get stats elsewhere(2b's, 3b's, etc) to complete the formula...
so for now I will baby step, I may stick with the formula I listed above, even though it is flawed, I want to start with some base of reference and work from there...
the formula being
(HITS + WALKS) (TOTAL BASES)RUNS = ---------------------------- AT-BATS + WALKS
I was also thinking that if I ran 2002 player stats through the formulas it would help to spot the cream rising to the top, it may not neccessarily pinpoint the best strat-o performers but at least give a good estimation on how they "should" perform
luckymanAll-Star
Posted February 08, 2003 06:01 PM
Mike_J addresses the problem quite right:
quote:
And linear weights are difficult because they involve manual counting- the disk tells you total number of hits, walks, HR's, TB's, OB chances, out of 108, but doesn't tell you anything about doubles, triples, etc.
What you have, on the strat disk, is on-base, total base and hits. By the way, notice that, since you have hits, you can calculate how much walks and hbp each card has by doing:
on-base - hits = (W+HBP)
Thus, I suggest you that you create an additional column in Excel Strat sheet with the upper formula (if you don't know how, I could explain.)
As Mike_J says, you could do ONB*TB/108, but the problem is, if you go that way, you won't be able to adjust for speed, defense, and the like. To incorporate the latter stats, you really have to keep the logic of linear weight. But you don't have the stats for singles,double, and triples, so what can you do?
Here is the answer. First, look at the weights I came up with in the last formula of my last post. The weights are:
(W+HBP) ---> 0.33
1B --------> 0.47
2B --------> 0.78
3B --------> 1.09
HR --------> 1.40
Outs-------> minus 0.3
Since this is a linear model, I can add or multiply by a constant, and the order among the players will be preserved (in other words, if Jeter is better than Tejada in the original weighted formula, he will still be better in the transformed formula). So a add 0.3 and multiply 2.8 each value, I will end up with the following weights:
(W+HBP) ---> 1.76
1B --------> 2.16
2B --------> 3.02
3B --------> 3.90
HR --------> 4.77
Outs-------> 0
Now, suppose we ADD for each card chances of on-base and chances of total bases, as listed on your strat disk. You know that, for chances of total bases, you have one point for a single, two points for a double, etc. So what this value (on-base chances plus total base chances) is really doing is to give the following weights:
(W+HBP) ---> 1
1B --------> 2
2B --------> 3
3B --------> 4
HR --------> 5
Outs-------> 0
If you compare the last two lists of weights, you see that, except for walks, the two lists are very similar.
Let's fix the weight for walks. In the former list of the two lists we are comparing, walks had a value of 1.76, and singles a value of 2.16. So the value for walks was the equivalent of 80% the value of singles. So if we give the value of 2 for singles, we should have a weight of 1.6 for walks.
So here is the twist I suggest: to have a value of "Runs created" that approximates very well the linear formula from the stats you find on the STRAT disk, you just have to create another column in Excel Strat sheet with the following formula:
=ONB + TB + 0.6*(W+HBP).
(0.6 instead of 1.6 because ONB incorporates 1 point for (W+HBP))
This will give a linear weight of Runs created, with the underlying following weights:
(W+HBP) ---> 1.6
1B --------> 2
2B --------> 3
3B --------> 4
HR --------> 5
Outs-------> 0
Since we have a linear weight RC, we will be able to adjust this RC for:
- ball park adjustment
- stolen bases
- defense
- gbA
- clutch
- star (forcing holding)
I won't get in the details of each but here the final formula I end up with. For each side (vs lhp and vs rhp), compute:
1) RC=(W+HBP)*0.6 + ONB + TB + BPadjustment + "weak" adjustment + clutch/4 + gbA/5;
where BP adjustment is adjustment for Ball Parks homerun. The formula for this is
BP adjustment = k/4 * BP
-BP is the number of Ball park homeruns on a given card
-k is the stadium homerun ratio (for Coors, it would be 19, for a neutral park, it would be 10).
AND where "weak adjustment" is :
-7 for a player being weak
-5 for a player NOT having ball park singles on his card
-12 for a player having both characteristics
0 for normal players with ball park singles
2) running adjustment= (stealing + speed/2)-6
where stealing is
+6 for a playing having a star and a 19 number for stealing
+5 for a playing having a star and a 18 number for stealing
+4 for a playing having a star and a 17 number for stealing
+3 for a playing having a star and a 16 number for stealing
+2 for a playing having a star and a 15 number for stealing or less
and where speed is speed rating (ex.for Jeter, 17)
3) offensive RC = 0.3*(RC vs lhp) + 0.7*(RC vs rhp) + running adjustment
4) overall RC: offensive RC - defensive RC
I come back to defensive RC in another post.
[This message was edited by luckyman on February 08, 2003 at 07:00 PM.]
luckymanAll-Star
Posted February 08, 2003 06:21 PM
Penngray, about
quote:
I wish the ratings disk broke out the 2b,3b etc.
.
That's very easy to do with Excel.
First, to be sure you don't loose any data, select the whole set of data, copy it and paste it in another spread sheet (the width of columns will change). Second, I suggest that you delete all the non-eligible cards. Then, you click in the first cell under CA (where you find defensive ratings), and press the "ranking" button that looks like this:
A (with an arrow beside)
Z
By doing so, you'll have all the catchers grouped together. You can copy (not cut) and paste it in another sheet called "catcher".
Then you just repeat the procedures for 1B, 2B etc.
penngrayAll-Star
Posted February 08, 2003 06:35 PM
lol, I know how to do the player categories.
actually I use auto-filters and then just filter my defensive ratings that "contain" ss, c, cf etc. when I need to look at one position.
I was saying, I wish the spreadsheet had the hit totals for doubles and triples
luckymanAll-Star
Posted February 08, 2003 06:43 PM
oops!!!
penngrayAll-Star
Posted February 08, 2003 06:45 PM
my original post was really vague
luckymanAll-Star
Posted February 08, 2003 06:47 PM
For defensive ratings:
for INFIELDERS
defense RC = 0,7*e +((range-1)*pos)
where e is the e rating
range is the range rating
pos is:
3.6 for 1b
10.8 for 2b
12.6 for ss
5.4 for 3b
For lf and rf, the formula is:
defensive RC = rr + e
where e is the e-rating, and rr is:
0 for range 1
6 for range 2
13 for range 3
27 for range 4
35 for range 5
For cf, it's the same, except you multiply rr by 1.5.
P.S.
For ss and 2b, you may also want to adjust for double-plays (gb A or gb 1). A simple way is to substract 6 from defensive rating for ss-1, 5 for 2b-1, 3 for ss-2 and 2 for 2b-2.
A sophisticated way is the follow the next formula. With Excel, there is a formula called MAX, with gives the highest numbers of the ones you specify. The formula is:
adjustment for gbA: =MAX((40-defensive RC)*0,18; 0)
luckymanAll-Star
Posted February 08, 2003 06:52 PM
So Pete,
forgive me if you got lost in the ocean of formulas I just threw in. But the whole idea is that the first step should really be:
offensive RC = 0.6*(W+HBP) + ONB + TB.
This keeps the logic of linear weight formula. It's simple. And it enables you to do all the adjustments, espiecially for defense
(so if you don't want to get in the details of all adjustments, I suggest you compute offensive RC with the simple formula above, adjust it to BP homerun in a neutral stadium, and substract from this the defensive RC allowed by each hitter. This gives the following, in two steps:
offensive RC = 0.6*(W+HBP) + ONB + TB + 2.5*BP
overall RC = 0.3*(RC vs lhp) + 0.7*(RC vs rhp) - defensive RC
where BP is the number of BP homeruns you find on a card)
Mike_JAll-Star
Posted February 08, 2003 07:19 PM
That's some pretty awesome stuff lucky, but I have a few question. Where are the clutch and GBA divisiors coming from? Using the divisors as a judge, you are saying that GBA situations occur about 20%, and clutch about 25%, of the time? Or am I missing something? Interested in the source of those values.
And for the defensive ratings, I'm fairly lost with the logic. I see how the pos values are based on chance ratios for positions, but I don't understand how that formula adds up to what a player's defense will actually allow.
peteAll-Star
Posted February 08, 2003 08:45 PM
lucky,
thanks for the reply, I appreciate everything you put up, though it may take me a bit to absorb...
if you have any more insights it would be welcome
I am currently building an sql database that will allow me to sort by ratings, I will go back and add a column based upon your formula
luckymanAll-Star
Posted February 08, 2003 09:13 PM
Mike_J,
Don't wonder why you can't understand how I came up with these numbers: for reasons of simplicity, I didn't mention it!! So you're not missing something, it's rather me who's hidding my assumptions.
Clutch
Let's take Edmonds numbers vs rhp:
W+HBP= 22
ONB = 52.3
TB = 50.7
BP = 7
clutch = -12
Basic RC = 0.6*(22) + 52.3 + 50.7 + 2.5*(7)
= 133.7
Now, we know that clutch situations happen around 12% of the time (that's on average, the 5th hitter in a line-up gets more, close to 15%, but let's keep it simple and take 12%)
So this basic RC, 133.7, holds 88%. In clutch situations, Edmonds' card lost 12 singles. So his basic RC in clutch situation is :
Clutch basic RC = 0.6*(22) + 40.3 + 38.7 + 2.5*(7)
=109.7
So Edmonds rating, when adjusted to clutch, is:
RC = (0.88)*133.7 + (0.12)*109.7 = 130.82
But notice that we would have approximately the same result if we had use this formula:
RC = basic RC + clutch/4 = 133.7 + (-12/4) = 130.7
gbA
A double-play is worth about two outs. If we adjust our weights to give 0 to outs, a dp is worth about the opposite of a walk. So, in our system, a double play is worth about -1.6.
But, as you know, gbA doesn't turn out in double plays each time. It only does so when there is someone on first with less than two outs. If we knew the probability that this situation occurs, then we would know the weight for gbA. I can't remember the source of this information, but I think that on average, it's between around 18% (again, it depends on where a player stands in the line-up).
So a rough approximation of the weight we should attribute to gbA is:
gbA = 18% X -1.6 = -0.29.
That being said, there will be situation where there will be gbA with guys on first/less than 2 outs, and the outcome won't be a double-play. One example is when defense is playing in, another example is if you look at the chart, with guys on first and third and the reading is gbA (3b), the play will be at home, etc.
So in strat life, gb(A) will be worth roughly between -0.20 and -0.25. So a good approximation will be around to weight :
-0.25*gb(A), or gb(A)/4
or 0.2*gb(A), or gb(A)/5,
but certainly not gb(A) taken as an unity.
(the reason I chose gb(A)/5 is because in fact, a dp is worth a little bit less than the opposite of a walk, roughly 85%, but I'm sure that gb(A)/4 would work out well too).
Mike_JAll-Star
Posted February 08, 2003 09:40 PM
I think that's a great system, all based on sound logic and mathematical principles. One thing I noticed though, is that you have a + for GBA's- shouldn't that be a minus? And after popping the system through the ratings disk, the only thing that leaps out as "questionable" is the possible overvalue of walks- Durazo stands out. But I think that's an issue with sabermetrics in general, not your specific formulas.
penngrayAll-Star
Posted February 08, 2003 10:01 PM
what do you do with the BP singles? they arent on the rating disk either.