Offense vs. Defense article

[Mean Dean]

I don't have anything in particular to say about Jorge Cantu, other than what the system I laid out might say about him. I do want to say that this point...[quote:b057a36edb]Player A has .390 OBP/.450 SLG and Player B has .350 OBP/.495 SLG, the two will have NERP values per 216 plate appearances of about the same. However, Player A will have accumulated almost 9 less outs for his team. This means then, obviously, that his team has 3 more innings in which to score runs, and said player's value should be higher because of his contributing to increasing team offensive opportunities. [/quote:b057a36edb]... gets made every time I discuss anything, but I don't see how it can be right, because if it were, NERP would not predict team scoring correctly, and yet it does. (To put it another way, part of the NERP formula is that the out is assigned a value and you get penalized for making outs.)

That said, I do agree that it's technically more correct to describe the player's contribution as the difference between the team run scoring with him and the team run scoring without him. James has moved to that approach with Runs Created in the last few years. But, the thing is that he can do that because the runs scored by real life teams are known facts. When we're dealing with fictional lineups as in SOM, we don't know how many runs the lineup "really" scored, so I think in that situation, you unfortunately have to fall back to the more abstract approach. (Which I hasten to add really is not sacrificing very much accuracy at all, and again, you can check that by running the formula on teams.)

[maligned]

Do you have any references to data to show how accurate NERP is in predicting runs scored by an offense across different run-scoring environments?

It only makes sense that the .390/.450 hitter becomes more and more valuable in comparison to the .350/.495 hitter as the run-scoring environment (i.e. quality of hitters around him) becomes greater (each out becomes worth a greater value of runs). NERP alone seems not to consider changes in environment. A team with 8 .320/.380 hitters plus a .340/.420 hitter will certainly yield less runs than with 8 .320/.380 hitters plus a .365/.395 hitter. However, this difference becomes greater and greater as the 'team' stats increase.

[Mean Dean]

[quote:f81f83f47e]Do you have any references to data to show how accurate NERP is in predicting runs scored by an offense across different run-scoring environments?[/quote:f81f83f47e]Yeah, there's a study [url=http://www.baseballthinkfactory.org/btf/scholars/furtado/articles/accuracy.htm]here[/url] of every team between 1955 and 1997 that demonstrates NERP's accuracy. (I have no clue why I didn't put this in the article to begin with. But I did plan to revise the article anyway...)

I don't really disagree with most of your points in the abstract. The negative value of an out does depend on the overall offensive context. And a high-OBP player has a little more value when added to a low-OBP/high-SLG team, and the opposite of that is also true. But, I think we're talking about a) very small effects that b) have no apparent way to be fixed. (Unless you want to put all your league's rosters and ballparks into the PC game, set up all the computer managers appropriately, sim a whole bunch of seasons without the player, and then sim a whole bunch of seasons with him.) It makes sense to me to go with estimations that have proven to be good estimations.

[maligned]

I don't think I've been fully clear with my overall point. Let me give an example using NERP. Consider the following two player cards, assuming all BPHR/BPSI/CL results have been factored in and that neither player stole a base or grounded into a double play (for simplicity of example):

Player A:
52 OB
32 Hits
88 AB
49 TB
56 Outs
20 BB+HBP

NERP: 22.8

Player B:
39 OB
24 Hits
93 AB
62 TB
69 Outs
15 BB+HBP

NERP: 22.8

For both of these player cards, the NERP result is 22.8. I can in no way question the accuracy of whether a player (or team) producing these results for 108 plate appearances will average 22.8 runs scored. I never meant to question NERP's accuracy for predicting runs scored by a team. Here is my BIG problem with NERP as a player value predictor for strat: Player A only recorded 56 outs compared to 69 outs for Player B. In the context of a team environment (not a static 108 PAs), Player A (an everyday player) would garner approximately 40 less outs for his team for a season. Thus, a team of average MLB players would score 7 more runs with him than they would with his counterpart, Player B, because they would have all the extra PAs and opportunities to score associated with the 40 extra outs.
In conclusion, even though NERP does in fact "consider" outs because of the AB component and does in fact predict runs scored within the context of completed results, it DOES NOT consider the effects of an individual player extending his team's scoring opportunities on offense or limiting the opponent's scoring opportunities on defense.
Finally, all I'm suggesting for considering rankings is to consider players' values within the context of an average offense. I'm in no way suggesting you must know the players for each team or the batting order for each team to really know a player's value. I'm simply saying you can consider a player's value more accurately by adding his values to those of the average pitcher he'll face and the league average 8 other hitters he will have on his team--then finding the runs scored for a given number of outs. The final difference in results between each player within the same given context will reveal each one's final value (all you have to do is consider NERP/out for the entire team's calculation and multiply by a standard number of outs). As I've said before, your final results will favor OBP more favorably than static NERP alone.

[Spider 67]

Nice to be able to read you discussion of NERP as a measure of performance.

I would contend that NERP DOES consider that Player A has consumed fewer OUTs than Player B. In Player A's case it is NECESSARY that his teammates use those extra outs to advance runners and make up for his lack of power. Player B consumes more outs, but does more with his plate appearances to advance bases and, thus, need less support from teammates.

For example, his teammates don't need as many outs, if he homered or they only need to score him from third or second rather than first.

[maligned]

You're right. For those isolated 108 plate appearances, both players created the same number of runs and both players received "penalties" for outs. This is not my concern. My concern is this: NERP is only measuring value for these isolated results, but in actuality, Player A has extended his team's scoring opportunities by 40 outs in a season because he records 13 less outs per 216 plate appearances.
He and his team will use these 40 extra outs to score an additional 6 or 7 runs that Player B's team will never have. My contention is that a value ranking should incorporate this added value.
Again, the change in calculation is simple:
NERP for given player PLUS nerp for an average pitcher's 108 chances PLUS nerp for 8 average teammates DIVIDED BY total outs recorded by all EQUALS nerp per out. Finally, multiply this by the number of outs you want to consider.
The result, when you plug each player into your formula of average pitching opponent and average hitting teammates, will equal a number you can use to compare all players much more effectively than simply looking at isolated NERP for a given number of plate appearances (which, again, is limited in its effectiveness because it doesn't consider the fact that plate appearances are increased when outs are recorded less often).

[maligned]

One more way to explain my most recent example on the previous page:

An average team will record about 4,350 outs in a season. This is true for all teams. There is no way to earn or reduce this number of outs. Each inning is limited to three outs--period.
Player A in my example using NERP on the previous page will use 40 less of these outs in a season than Player B. Yes, for any isolated 216 PAs, they will produce the same number of runs. What you're not understanding is that his team and he STILL GET TO USE THOSE 40 OUTS HE NEVER RECORDED in the normal course of play throughout the season. One by one, those outs he didn't record add up to a few plate appearances here and a few plate appearances there that will lead to, on average, about 7 more runs over the season. RECORDING LESS OUTS = MORE PLATE APPEARANCES IN A SEASON = MORE RUNS FOR YOUR TEAM.

[Mean Dean]

Yes, I understand that avoiding outs is good. Don't be condescending. I just think you're double-counting it; the "extending inning effect" was factored into the value of the out, which is part of the formula.

The simple fact that team NERP is equivalent to the sums of all the individual NERP, means that what you're saying can't be right, IMO.

[maligned]

I apologize, Dean. I was more responding to the last post trying to explain the concept a little more clearly. I definitely didn't mean to sound condescending. Much of what I've learned about Strat and the function of an offense has been through discussions of this type. I'm far from thinking I have answers that are better than others. I'm trying to learn through our current discussion as well.
With all that said, I see from your last explanation that we're still not on the same page. I'm not an expert statistician, so I'm struggling to put to words the concept I'm attempting to explain.
Let me try another explanation. There are two ways to consider offensive output:
1. Offensive value that has already been accrued (this is NERP's function).
2. Offensive value as it affects the FUTURE (this is where NERP/out or RC/out come in)

The out "penalties" that are being considered in our NERP formula only penalize within the stiff, static context of the 216 plate appearances we're considering or the stiff, static context of pre-determined plate appearances when we are analyzing PAST results. RC penalizes also (basic formula is (OB/PA)*TB). The PA element penalizes the potential runs created.

Ultimately, what I'm suggesting is this: We are not talking about PAST NERP or PAST RC; we are trying to figure out how a player's card will function in the FUTURE.
When we consider our current NERP card, we must understand that it's one and only purpose is to demonstrate how many runs are created by a PRE-DETERMINED number of plate appearances (and at-bats). It is not designed to figure out how a player will function in the real future because FUTURE plate appearances are unknown.

Instead of saying, as I have been, that a "good OBP earns extra outs," I want to say that "good OBP earns extra Plate Appearances." Outs will theoretically be the same for every team. They can neither be earned or lost. Every team plays the same number of games with the same number of outs.
Again, the out "penalty" in our current NERP formula only functions to reduce the number of runs produced for our strict 216 plate appearances.

In real, future competition, here is what will happen in approximately 1/3 of a season:

Player A from my NERP example (for his 108 card chances): 56 outs, 52 on-base
Average opponent pitcher/defense (for his 108 card chances): 74 outs, 34 on-base
Average teammate (per 216 PAs): 147 outs, 69 on-base
Cumulative team OB to Out ratio: .489 OB per 1 out (1.489 plate appearances/out)

Player B from my NERP example: 69 outs, 39 on-base
Average opponent pitcher: 74 outs, 34 on-base
Average teammate: 147 outs, 69 on-base
Cumulative team OB to Out ratio: .474 OB per 1 out (1.474 plate appearances/out)

Team outs per 1/3 season: 1450
Player A's team plate appearances: 2159
Player B's team plate appearances: 2137

Now, in the midst of these two seasons, again, I cannot argue the fact that in any given 216 plate appearances for each player, he will contribute the same number of runs. However, again, because Player A has generated 22 extra plate appearances for his team, his team will score about 2.5 runs more than Player B's with the same teammates. Over the course of the season, this will balloon to approximately 7.5 extra runs (generated from 66 extra team plate appearances).

My point in all of it is this: It's not that much more complicated to calculate future results within the context of an average team...why would we not take these 7 extra runs into consideration in our ranking?

As you mentioned Bill James has begun doing this of late. Also, Baseball Prospectus has always done it's famous value calculations from the context of "How many runs would this player's team have if he were a part of an average team?" They explain this clearly in their explanation of Value and Value Over Replacement Player (VORP) calculations. Part of their explanation is that simulating a theoretical team of 1 considered player and 8 average others predicts how a given player will serve to extend games and contribute to the offense in its totality. They use the most updated version of RC/out to do this calculation.

Sorry this was so long-winded again. And again, I apologize that I came across as condescending before. I've learned a lot from your posts and research, so I certainly do not intend to demean you or your work--I just think we can extend and improve the ranking theory you've proposed.

[childsmwc]

Everything maligned has laid out is true. Let me try to reverse engineer this:

A good linear model takes a set of relevant events (singles, hr's, BB's, outs, SB, DP, etc.) and by assigning appropriate values to each event accurately estimates the runs a team will score. Once you have a formula that estimates runs a team scores, you can now apply that formula to each individual on the team, to find their individual contributions. The sum of the individual parts, will total back to the team total.

Now that we have a good RC value that works in this context, we apply it to the expected results from a SOM card over a predetermined number of plate appearances (lets assume 700 on average for a typical season). The result is some positive (and in some rare cases negative) result that represents the number of runs this individual would add to your team over 700 PA's. So we apply this to the entire set and we identify the 9 guys we want.

So to estimate how many runs my team will score I just add up the run value for those 9 players right? Well not exactly because the run value you determined for the individual was over 700 PA's, so your team total now assumes 6,300 PA's. But in a team context it is not PA's that are important but outs, so if that group commits more than 4,374 outs (assume all 9 inning games x 162 game) in those 6.300 PA's, then we have a problem. Our individual RC value that we computed is over stated because we will not get the 6,300 PA's that we have assumed. This overstatement is the cost of outs represented by the NERP maligned was identifying.

On the flip side a team of Ted. Williams might only commit 4,000 outs so adding up 6,300 PA's for that team would understate the runs they will score, because they are going to get more PA's than 6,300 again this represents the NERP of OBP over the average.

NERP bridges the gap between taking the individuals out of their team context. NERP is designed to say the moment I take an average player off of a team and insert T. Williams for 700 PA's, the value of all the remaining Joe's on the team goes up because now they are getting another 10 PA's because T. Williams joined the line up and didn't make as many outs as the guy he replaced.

The linear RC value is still correct, however, Joe's RC value goes up on the same team, because T. Willams creates more events (i.e. PA's) in a team context, because he doesn't consume outs.

The linear RC formula already correctly values the cost/benefit for each event, however, when estimating the value of a player you have to feed in accurate plate appearance data for the end results to be correct. Once you add a player with an OBP above the team average, you have changed this PA dynamic (your input) for everyone else on the team and a new RC value would need to be completed for each.

However, we can short cut this by simply determining that for every average PA, X additional runs (or a fraction of a run as the case really is) are scored. So any hitter that generates fewer outs than the average (I use outs, because DP and CS impact this computation not just OBP) generates additional PA's for his team, the value of which, if it is an average team is X additional runs times additional PA's.

No matter how accurate a linear RC model you have, it is worthless without the correct input. NERP is actually a team input concept. It does not change or challenge the values assigned to any event, it simply says that you may have more or less events than you originally estimated based on the number of outs this individual is estimated to consume and then attempts to value this change in events.

The value of this change in events, should adjust the final value for an individual card.

Bbrool