4374 vs. 3 x 9 x 162

[wheels]

I swear, I read in there somewhere that Chewbacca is from Kashyyyk, but lives on Endor; therefore sabermetrics is worthless when used to locate Coruscant.

No words.

[Old school 1983]

So you are suggesting that the people who use statistics are only looking at things in aggregate and not considering the reality of situations as they occur in the context of an inning, game or season?

If that's your contention, I think you're poorly informed. What is linear weights if not an explicit consideration of the realities of situational baseball? Of course, that still doesn't account for the specific pitcher, batter, baserunner, etc. But it does create a well-informed baseline from which reasonable adjustments can be made.

And most particularly, it gets people out of the world in which "the way I was taught", "how we played the game in my day" and "I knew in my gut" are the bases for decision-making and in which every case is a special one, allowing for dismissal of what the averages would suggest is the right move. That doesn't mean it rejects allowance for factoring in the uniqueness of every circumstance. But it does beg a greater level of rigor in the thinking behind a given decision.

It says, we know that this is what happens on average, over time. These are the basic dynamics at play here. If you think you know better, you'll need to be able explain where the math gets it wrong in this case. We know the "4374" doesn't always provide the right answer, but we do know that it has some key improvements over the historical instincts that have developed around "3 x 9 x 162".

If any of us are stating things unequivocally based on aggregates and averages, like "never ever bunt no matter what" perhaps we're being too blunt. But I don't think that's happening often -- and when it does, it's likely a purposeful oversimplification for the purpose of making a counter-point to an approach which eschews the data entirely.

Excellent points. I think sometimes it comes off as never bunt no matter what, or that strikeouts don't matter even in some situations, OPS is the only meaningful offensive stat, productive outs are always meaningless when it comes to scoring runs, or that you should always play for maximum run output. I hope I wasn't coming off as real jerk in the thread about making contact because I buy into a ton of the sabermetric way of thinking, and I know there are sabermetrics used to explain small situations. I just feel like the saber guys come off in absolute terms a little too much for arguements sake and it creates the wrong impression of infallibility and totality with no room for sway or consideration of individual game situations. Actually thank you so much for clarifying that this is not the case!

[defender]

So you are suggesting that the people who use statistics are only looking at things in aggregate and not considering the reality of situations as they occur in the context of an inning, game or season?

If that's your contention, I think you're poorly informed. What is linear weights if not an explicit consideration of the realities of situational baseball? Of course, that still doesn't account for the specific pitcher, batter, baserunner, etc. But it does create a well-informed baseline from which reasonable adjustments can be made.

And most particularly, it gets people out of the world in which "the way I was taught", "how we played the game in my day" and "I knew in my gut" are the bases for decision-making and in which every case is a special one, allowing for dismissal of what the averages would suggest is the right move. That doesn't mean it rejects allowance for factoring in the uniqueness of every circumstance. But it does beg a greater level of rigor in the thinking behind a given decision.

It says, we know that this is what happens on average, over time. These are the basic dynamics at play here. If you think you know better, you'll need to be able explain where the math gets it wrong in this case. We know the "4374" doesn't always provide the right answer, but we do know that it has some key improvements over the historical instincts that have developed around "3 x 9 x 162".

If any of us are stating things unequivocally based on aggregates and averages, like "never ever bunt no matter what" perhaps we're being too blunt. But I don't think that's happening often -- and when it does, it's likely a purposeful oversimplification for the purpose of making a counter-point to an approach which eschews the data entirely.

Linear weights were created to remove the realities of situational baseball and find a pure value for each event. Unless baseball results are random there is no mathematical reason that linear weights would provide the best in game strategy.

If we calculated different linear weights for innings with 0-1 runs, 2-4 runs and 5 or more runs the relative value of each event would be different for all three. There are so many factors in the formula that do not come into play in any given situation. For the noise Linear weights takes away in evaluating hitting seasons, it adds noise when trying make in game decisions.

To explain it mathematically, when evaluating an entire season, you can (for math purposes) assume baseball results are random, in game strategy has to be based on baseball being a contest of skill. You can't use stats based on total runs scored that are created to be situation independent.

[RedsManRick]

Linear weights were created to remove the realities of situational baseball and find a pure value for each event. Unless baseball results are random there is no mathematical reason that linear weights would provide the best in game strategy.

If we calculated different linear weights for innings with 0-1 runs, 2-4 runs and 5 or more runs the relative value of each event would be different for all three. There are so many factors in the formula that do not come into play in any given situation. For the noise Linear weights takes away in evaluating hitting seasons, it adds noise when trying make in game decisions.

To explain it mathematically, when evaluating an entire season, you can (for math purposes) assume baseball results are random, in game strategy has to be based on baseball being a contest of skill. You can't use stats based on total runs scored that are created to be situation independent.

Again, you're failing to provide the point of comparison. What is the better baseline of strategy you are espousing? It's funny, because you talk about this like there's this crazy amount of nuance going in to these decisions when we constantly observe managers making their decision on an almost rote basis given "the book" of what to do in given circumstances.

As for the rest, it feels like you're smashing together a bunch of buzzwords in to something that has the appearance of an argument without actually meaning much. If a politician was to give a speech against sabermetrics, I think that's what it would sound like.

[vaticanplum]

I swear, I read in there somewhere that Chewbacca is from Kashyyyk, but lives on Endor; therefore sabermetrics is worthless when used to locate Coruscant.

Steeeell!!!!

I feel like this thread is punking me on so many levels.

[AtomicDumpling]

Linear weights were created to remove the realities of situational baseball and find a pure value for each event. Unless baseball results are random there is no mathematical reason that linear weights would provide the best in game strategy.

If we calculated different linear weights for innings with 0-1 runs, 2-4 runs and 5 or more runs the relative value of each event would be different for all three. There are so many factors in the formula that do not come into play in any given situation. For the noise Linear weights takes away in evaluating hitting seasons, it adds noise when trying make in game decisions.

To explain it mathematically, when evaluating an entire season, you can (for math purposes) assume baseball results are random, in game strategy has to be based on baseball being a contest of skill. You can't use stats based on total runs scored that are created to be situation independent.

Linear weights and other sabre stats do not remove the "realities of situational baseball". Quite the contrary, they actually expose the "realities of situational baseball" for what they really are, not what people have assumed or guestimated them to be.

[defender]

Linear weights and other sabre stats do not remove the "realities of situational baseball". Quite the contrary, they actually expose the "realities of situational baseball" for what they really are, not what people have assumed or guestimated them to be.

The realities of situational baseball are: your team scores more runs than the other team while you are pitching and you get a win. Linear weights are the value of your actions independent of when/where they occur.

Unless you assume the events in baseball are random, it is mathematically wrong to use linear weights for in game strategy.

For example linear weights you would choose higher OBP but slower and more KOs than then the traditional faster, bat control but lower OBP 2nd hitter. The failure of linear weights based on total runs, is that the event values are different for each position in the batting order, event values are different for your team and linear weights values are skewed by the rare occurrence of high run innings.

If a manager chooses the traditional 2nd hitter, do not assume it is just because that is how they did it in his day, or he does not want to clog the bases. Linear weights may reveal a different answer, but it might be because ti is the wrong math for the situation.

[Caveat Emperor]

I played lots of RBI baseball on the Nintendo growing up -- that's neither "on a computer" nor "watching games."

Where do I fit in this world view?

[RedsManRick]

The realities of situational baseball are: your team scores more runs than the other team while you are pitching and you get a win. Linear weights are the value of your actions independent of when/where they occur.

We know that, in aggregate, OBP is the PA outcome most strongly correlated with run scoring. Does that mean a walk is better than a productive out in every single circumstance. Not necessarily. But it does give us an evidence-based baseline assumption from which we can operate.

If we choose to ignore OBP because we think baserunning and bunting are actually more important to run scoring, primarily because of their (very real) situational value, we are ignoring reality.

But if you're making a decision that will affect every turn in the lineup solely based on a circumstances which may occur from time-to-time, without considering the full value of that trade-off, then you aren't on sound logical ground. I'm all for consideration of the specific circumstances, the variety of possible situations and outcomes at play. And I'm more than willing to recognize that our aggregate stats miss a lot of information that truly matters in given circumstances, that truly changes the calculus. But every single PA is a situation. And we know what the aggregate dynamics of those situations looks like. And if the aggregate of a manager's preferred approaches is in direct conflict with the broader trends we see, something is wrong. He's probably placing too much weight on some circumstances at the expense of others (see, worrying about the use of a late-game lefty reliever in a tight game instead of how to maximize the value of the plate appearances in the 6 or 7 innings that come first).

If a manager chooses the traditional 2nd hitter, do not assume it is just because that is how they did it in his day, or he does not want to clog the bases. Linear weights may reveal a different answer, but it might be because ti is the wrong math for the situation.

Since when is "batting 2nd" a special situation that requires constant situational rethinking? This was precisely my point from earlier. The game just isn't that complicated -- and certainly isn't that predictable on the margins. And to the extent that certain situations do demand adjustment, we often don't see those adjustments being made. When was the last time Dusty dropped Phillips in the lineup vs. a particularly tough RHP? When was the last time Dusty cited a reason for batting somebody like Cozart or Stubbs 2nd that actually made sense given the unique circumstances. Sure, he was considering factors that linear weights couldn't begin to account for. But that's part of the problem. He misses the forest for the trees. You can "overfit" your model.

Even if the manager is using situational logic (e.g. we're going against a tough pitcher > 1 run is likely to be more valuable today > bunting and base stealing is going to be important today) doesn't necessarily mean that his logic is actually producing a sound strategy. He's basing the logic on a rough heuristic of "small ball is good vs. tough pitchers", not a rigorous analysis of the relative marginal contributions of those skills in this specific context relative to the alternative. His situational argument may be based on false assumptions and/or incorrect weighting of the likelihood of future events. It may also be quite smart/prescient. But we shouldn't merely assume that the stock model is wrong simply because one can hypothesize an alternative.

Sure, there are times when a decision is made that may not make sense to us but which is actually very sound -- say using a certain pinch hitter with worse aggregate stats because he's really good against high fastballs from guys that throw from a 3/4 arm slot. But often, It's not as if Dusty is thinking on a whole other level of strategy than us lay people. It's that he's applying stock principals in ways that are silly. He's bringing in the pinch hitter who is the clearly lesser hitter on balance because he's 4 for 10 with a homer against that pitcher in his career.

[wheels]

This thread has the BadFundamentals stamp of approval.

[AtomicDumpling]

The realities of situational baseball are: your team scores more runs than the other team while you are pitching and you get a win. Linear weights are the value of your actions independent of when/where they occur.

Unless you assume the events in baseball are random, it is mathematically wrong to use linear weights for in game strategy.

For example linear weights you would choose higher OBP but slower and more KOs than then the traditional faster, bat control but lower OBP 2nd hitter. The failure of linear weights based on total runs, is that the event values are different for each position in the batting order, event values are different for your team and linear weights values are skewed by the rare occurrence of high run innings.

If a manager chooses the traditional 2nd hitter, do not assume it is just because that is how they did it in his day, or he does not want to clog the bases. Linear weights may reveal a different answer, but it might be because ti is the wrong math for the situation.

I don't think this post holds up to scrutiny. Linear weights doesn't assume that events are random. There is no indication that the batting order changes the run values of events either. Linear weights are certainly not skewed by high-run innings, which in fact are the goal after all.

[Wonderful Monds]

This thread has the BadFundamentals stamp of approval.

I know I'm feeling KRISPy.

[RedsManRick]

Worth considering: RE24. Situation-specific linear weights. Guess what? RBI leaders who don't get on base still don't fare too well.

[Old school 1983]

I haven't dealt much with linear weights prior to this thread, so I did a google search in hopes to learn more. After the search idk how much credence I can put into them in terms of run expectancy. More or less they seemed to be based on average occurances in certain situations with no regard to who is actually participating in that situation in the game. Furthermore, (I provided the link) in a fan graphs article it was stating that against more dominate pitching (see playoffs) linear weights in terms of run expectancy tend to be off and not very helpful. From my brief look into it, it seems they may be used as a good baseline of how to approach a situation, but nowhere near gospel. I see a lot of times dusty being taken to task on his knowledge of the game, and I can say myself that I'm not a huge fan of him as a tactical manager, but it seems as if we question whether dusty keeps up on this stuff or knows it, and the question should be is whether or not dusty thinks it's the correct way to approach the game. Honestly after my reading I think it can be a useful tool to determine baselines, but not something I'd put full faith and credit in 100% of the time.

http://www.fangraphs.com/library/pri...inear-weights/

[PTjvs]

got my first Commodore and haven't seen a game since....just a Matrix-y stream of numbers flowing constantly across a diamond formatted spreadsheet in elysian green....

You get used to it. I don't even see the statistics anymore. All I see is SS, LHP, 1B who is into shopping, designer wear, fancy coffees, and pedicures.

Hey, you uh... want a drink?

jvs