You have mentioned HE WHO SHALL NOT BE MENTIONED!!!
Our doom is nigh!
Printable View
I hope it's pretty clear that OBP is what is most highly correlated to scoring to everyone. It's pretty obvious you have to get on to score.
As far as going for small ball vs a tough pitcher, I think there are a variety of factors to consider. The tougher the pitcher, the farther the linear weight goes from the mean, and the less reliable it becomes, is your pitching likely to match or is matching the oppositions' performance, and the actual in game circumstances. To try to play small ball the entire game with every player would be extremely counterproductive. Don't ask guys who have no business bunting or stealing to do so. Let your big guys air it out. When the situation calls for it though, play for one run. It's a risk or reward payoff type of model. Sometimes it's better to get one, then go for a bunch and get nothing. The likelihood of a big inning with your best batters against a dominate pitcher is less. (Not saying don't go for it) so when you get to the part of the order that has less production potential and you get a runner, it may be time to play small ball because the likelihood of a hit is even smaller.
Now playing smallball a lot of the time requires your lesser players to be less production based players, or feast or famine type players, and more skill based players. I mean able to make contact when contact needs made, able to bunt or hit behind a runner to get that runner over.
That's one of the issues I see with the reds. Most of their lesser players are feast or famine players that maybe have a little higher ops than they should because ops is skewed towards power and we play in great American. The reds already have enough big production guys choo votto bruce Phillips (as far as RBI) that when coupled with the pitching, that they are a playoff level team. That leaves us in the playoff, against top notch pitching, where you are not always going to have maximum outcomes. You're going to need skill guys to help manufacture runs in short term if you big production guys fail. Outside of billy hamilton, the reds really don't have this. I'd be nice to have another high production/high skill guy like votto or choo vs. righties, but if the OPS is in the same range, it might be smart to explore higher skill secondary players.
You sure about that, or is that an assumption? It's all about the relative weighting of the options. If one thing shifts away from the mean like you suggest, then it's likely that about everything will shift accordingly which wouldn't do much to what the model would tell you is the highest percentage play.
It's from a fan graphs article I've linked on an earlier post in this thread. I'm new to the topic, but from my basic understanding, everything will not shift accordingly as far as run expectancy because they are taken as an average of all events not the events occurring with the specific in game participants. Things could go in opposite directions. Great pitcher versus poor hitter, thus skewing you down. Or great pitcher versus great hitters maybe making it go back more towards the mean. I think they are a great baseline to go by, but I also think it is the mangers job to determine the risk/reward of playing for the big inning versus small ball depending on the participants in that particular situation. I'm not saying play small ball versus every dominate type pitcher in every situation, but it should be a larger consideration dependent on the match ups.
If I'm understanding this stuff incorrectly could someone please help me out?
Your understanding of LW as an average across all such situations and best thought of as a baseline is correct.
Given large enough samples, all the uniqueness of particular circumstances comes out in the wash. But in any given circumstance, the strategic decision should be informed by the particulars.
The "trick" is understanding just how much the particular circumstances change the calculus. My biggest complaint is that rather than tackle this challenge, both sides tend to ignore it. The sabermetricians (myself included) don't take the time to state how they thinking the odds shift, erring on the side of the averages. Many "old school" types see that the averages don't account for the particulars of their situation and just throw that information completely out the window.
If you have Billy Hamilton on 1B and Clayton Kershaw on the mound, a successful steal 2B probably helps your chances of scoring more than average and the cost of getting caught is similarly increased.
How does that change the risk profile, the break-even odds? I don't know. But given my lack of precise knowledge, I prefer to err on the side of the most data, the general value of the opportunity.
But above all else, what I'm looking for is a logical case to be built and defended. If you think bunting is the right move, tell me why the averages are wrong in this case. Show your work. I don't mean to denigrate the accrued baseball wisdom guys like Baker have. But if their conclusions are right, I sure would like some insight on where the averages are coming from.
That's how I think as well. But I'd be absolutely shocked if Dusty was using base-out state run probabilities to inform his strategic choices.
His strategic thought process was formed on the back of tradition and a learned sense of what it means to play the game 'the right way'. The degree to which the data support 'the right way' is probably not something he gives much of a flip about.
Like i was saying before he knows, but probably thinks his way is correct. Sometimes I do agree with dusty. What drives me crazy with him at times is not so much his disregard for sabermetrics, but his not putting his players in positions to succeed. For example having Frazier or Mes bunt when those would be guys that'd be more likely to hit for power. I guess more or less play to your players strengths and not ask them to do the complete opposite of it.
It is not that the formula for linear weights assumes the events it uses for data are random, it is that linear weights are useful in a statistical sense if sample size is large enough that it as if it is random (no noise).
Statistical analysis depends on a large enough sample size that for math purposes you can assume the data is random. It will tell you who will win an election, not who to vote for.
To address Rick's point about runs correlates most strongly with getting on base: In every situation you need a base runner to score a run. OBP is the sun and all the other factors that go into scoring run are the planets. When you analyze based on total runs scored you are looking at the solar system from the sun.
The structure of baseball is an artifice that makes OBP less important in scoring runs and wins than it would be in a natural state of each team given 4374 outs, or even each team being able to put last innings base runners back on base at the start of the next.
The pull of OBP skews every thing, to the point that total run based stats are not useful for strategy. Managing is not just waiting around for your team to get on base a bunch in a row.
OBP is the talllent, which overall is more important than the managing, managing is using the talent (maybe being nice to it, so it gets on base more). It is a different math problem. There is no reason to believe someone who is a baseball expert does not know base-out state run probabilities. He just might be using better data.
This thread makes me feel like an idiot. I'm just not following defender's critique of saber approaches to baseball. To me, it seems like he's using a bunch of fancy words to say the same thing that most situationalists do -- without acknowledging that "game events" are already factored into the stats that meta-analysis uses to suggest strategy.Quote:
The structure of baseball is an artifice that makes OBP less important in scoring runs and wins than it would be in a natural state of each team given 4374 outs, or even each team being able to put last innings base runners back on base at the start of the next.
Maybe I'm just dumb though. I freely admit that possibility.
Let's move away from linear weights portion of this conversation, because I think you're arguing a straw-man a bit. Quite admittedly, linear weights assigns an average run value to a given event. We know that, over time, players don't tend to be able to systemically control when they produce certain PA outcomes. So while the actual value of a given event can swing wildly based on circumstance, the sum values come out pretty well.
To your point however, the linear weights derived event values aren't a great source for micro-level strategy.
What is a good source for that is base/out state run expectancy models, which simply look at what has happened in reality over time given the state of those two variables:
We can say with a high degree of reliability that, on average, teams score more runs from a 0 out, man on 1B situation (~.85 runs) than a 1 out , man on 2B situation (~.7 runs).
We can say with a high degree of reliability that, on average, teams are more likely to score at least 1 run from a 0 out, man on 1B situation ( ~43%) than a 1 out, man on 2B situation (~41%).
Sure, those are just baselines. Any manager needs to consider the particulars of the situation and make adjustments based on his assessment of likely outcomes. If Ryan Hanigan and the pitcher are coming up, the relative value of getting that guy to 2B probably goes up quite a bit, perhaps flipping the script on what strategies make sense.
My concern is that Dusty either isn't aware of the RE tables say (explicitly, he knows they exist, but I don't know that's really looked at them), simply doesn't believe them, or just isn't willing to address the cognitive dissonance they cause in terms of suggesting some of his preferred strategies are, on average, losing ones.
That is easy. For the manager; man on 1st, no outs is not an option. The AB has to happen.
The manager is choosing the value of runner on second with one out – chance bunt fails vs. value of runner on first with one out + chance of hitter reaching base – double play + chance of out that advances runner.
Total run data in essence ignores the outs. It does not factor that the run needs to score before the 3rd out and that the team needs to score more runs within 27 outs.
Just to clarify (and this is nothing to do with the post above), I am not making a watch the game vs. computer or old school vs. new school, it is purely a math argument.
As for the quote above, the implication is that if the best outcome of a bunt is worse than the current state, not bunting must be better.
I don't think total run data is useful for the true calculation a manager is making. Bunting is rare. Overall only 1% of PA's. The most common time is no outs man on first (374 bunts in 4992 PAs 2013 NL), only 7.5% of the time. 3rd and 4th hitters, only 12 bunts total all year in the NL.
The best hitters don't bunt, and I would bet managers bunt less against the worst pitchers. The 5th hitter has 16 bunts, the 9th hitter has 572. That seems to indicate managers are doing some calculations.
If you throw out the three best hitters on the team, it might turn out the run is more likely to score with one out from second, than first with no outs. Also consider, that the manager may want to make the move to put pressure on the the other team and keep momentum with his team.