Fan vs. PECOTA Projections

[Lucky Jim]

This I agree with but Uribe is without a doubt a surer thing.

BTW, i think Payton goes in the Bedard or BRob deals, so this a moot point...Just using Payton as an example.

He's a surer thing - but only if you place value on his (likely) marginal improvement offensively this year (which I discount a bit). But in order to get there you need to be seriously optimistic.

AND ignore that the value returned for Payton (or his replacement) might very well yield more wins/runs, etc. if applied elsewhere.

I don't give up ANYTHING for Uribe...that could be applied elsewhere for more long-term value.

You'd have to determine that the "surer" component of the probability is substantial enough to outweigh the both the salary difference and the opportunity cost of sending Payton to Chicago for Uribe (the opportunity cost being the return value of the trades that could be made instead but weren't due to trading for Uribe.)

[Lucky Jim]

Can't we have one discussion on defense (or any topic for that matter) without you two fighting? I've seen married couples who fight less than you two. Heck, my two kids fight less. I don't know why you two don't put each other on ignore.

I agree completely, they should try to get someone better than LH. Unfortunately, as yet they haven't done so. I also have to wonder if it is as easy to find said replacement as some might think - and don't even tell me Uribe is that guy. Having said that, I still think AM will indeed find another SS to compete against LH. But if he doesn't bring in an obvious replacement, then LH should be the favorite to win the job going into ST as the roster is currently constructed. If he does win the job, then I hope Trembley is creative in getting other players time at the position (even if on a limited basis).

There's no doubt in my mind that LH is good enough defensively to be at least close to a league average SS overall next season. If he proves me wrong the team might win 65 games as opposed to 63 games. Not a big deal at this point.

As for the whole Offense/Defense thing. A run saved is worth more than a run earned. Want evidence? Using Pythag, and a league where the average team scores five runs per game. Team A is league average defensively, but is +1 run per game better than average offensively - 94 wins. Team B is league average offensively, but +1 run defensively (in other words, gives up 4 runs per game) - 97 wins. Do the same exercise but make the team +2 offensively and average defensively - 105 wins. How about +2 defensively, and average offensively? 116 wins. The trick is finding the right mix of players to make it work.

I'd just like to say - and I know it's irrelevant - that it's posts like these that make OH what it is (and such a relief from bickering/speculation about trades and insider information.) Thanks.

*Helps that I agree, of course.

[RShack]

Can't we have one discussion on defense (or any topic for that matter) without you two fighting? I've seen married couples who fight less than you two. Heck, my two kids fight less. I don't know why you two don't put each other on ignore.

I just wish we could discuss *anything* in one measly thread without it devolving into Trade Mart like all the rest of them.

As for the whole Offense/Defense thing. A run saved is worth more than a run earned. Want evidence? Using Pythag, and a league where the average team scores five runs per game. Team A is league average defensively, but is +1 run per game better than average offensively - 94 wins. Team B is league average offensively, but +1 run defensively (in other words, gives up 4 runs per game) - 97 wins. Do the same exercise but make the team +2 offensively and average defensively - 105 wins. How about +2 defensively, and average offensively? 116 wins. The trick is finding the right mix of players to make it work.

So it sounds like league-average offense with great pitching and defense is a promising idea. I have no clue about the relative difficulty of increasing the O vs. P&D performance further once you get up to average.

As for the question you asked, it disappeared before I answered it, so I guess you changed your mind.

[RShack]

Yes, I decided I'm not ready to discuss that just yet.

But you could answer a couple of questions for me.

What correlation would you need to see between a RZR to +/- system to an existing +/-?

What else would you like to see?

The problem is that it doesn't have to be super-high if you're doing it to get a good picture about a group of many guys. But since what (I think) you want is something that's useful for looking at individual players, that raises the bar a ton. The rule-of-thumb I remember is that, whatever the correlation is, square it to get an idea about how confident you can be. Whatever you asked before was there for just a sec, but when I started to reply it was gone, but I think I saw .80-something. So, .80 squared would be .64 (or a "D"), .85 would be .73 (or a low-ish "C") and .89 would be .79 (or a very high "C"). If you want a low "A", then you need to be up in the mid-.90's. You see the problem. (When engineers are doing anything safety-critical, they gotta be way, way up in the very-high .90's or they can't proceed.)

Now, the good(?) news is that, when it comes to baseball, it seems that people don't realize this (or else they hope nobody else does ;-) I don't really know, but judging from what I remember from the numbers people mentioned in that old thread about reliability of predictions (based on 3-yr ML data), it seems that this is a domain where the standards don't seem very demanding. I'm not dissing the baseball-stat guys, it's just that the problem is hard. This means that "good progress" is one thing, but having something good enough to actually think about making individual decisions on is something way, way different. For example, when I was reading THT about that formula that they said was .7 (and which gave me very crappy results compared to the real numbers), their comment about .7 was, "Not perfect, but pretty good." Well, for what we're talking about, it's not even in the same county with "pretty good". It's very much of a long-distance call from .7 to "pretty good". For looking at individual guys and coming up with the right individual scores, it's actually very bad. But it seems "pretty good" to them, because they were starting with something that was .5, and .7 is a huge improvement over that. (The square of them shows that going from .5 to .7 doubles the goodness of it, from .25 to .49. But .49 is still a very-flunking grade.) So, I can see how the guy was happy about the improvement. But the guy also said "In fact, I don't suggest that you apply the math to individual players, because the equation will differ considerably from position to position." I agree with him that you shouldn't use it for individual players, but I'm not so sure the main reason for that is position-to-position differences. If it was just that, he'd tweak it by position and give different constants to use in his formula for SS vs. 3B. So, I don't think that's the real reason. I think the real reason is because .7 is just way too low to use for individual scores for individual people.

If you can plot your data points against the straight-line goal of a perfect 1.0 correlation on a graph, look to see how the plotted points are off the line. If they're mostly off the line to the same side, then you can prolly tweak yours by playing with a small constant factor. What the constant value is doesn't matter at all, just so it gets your data points closer to the line. If your points are distributed off the line more-or-less equally above and below, then you might be stuck. You could try tweaking constants for some individual factors but not others, and see what happens. If there's any kind of pattern to how the points are off (either moreso at the top vs. the bottom, or whatever), that would be of interest, but I don't know exactly how in the abstract. Looking at the individual data points (real vs. the one you calc'd for Player-X) will show you how wrong it can be. Since we evidently need +25 for GG-level (according to what tangotiger said), then if you show a +25-guy as +12, or a +10-guy as +25, or a +7 guy as -3, then I don't see how it will help us in the endeavor of tracking individual guys. It could be good for tracking teams-as-a-whole though. But I don't think that's what we're looking at, is it.

When you said "What else would you like to see?", I'm not sure what you meant. Were you simply re-asking the 1st question, or were you asking a second question?

[RShack]

I presume that the reason we can't get the real +/- scores is because he's selling the data to MLB teams. Is that right? If so, then I wonder if he's doing the same damn thing throughout the MiL's. If I was a GM (and we're all glad I'm not), I'd sure want that about everybody. Do you think it's safe to assume that they're doing that and we just don't know? If they are, then the Warehouse has all the answers they need about LH, it wouldn't be a sample-size issue at all. Of course, we don't have any idea what the results would be, but they'd have them. (Not that it's necessary. There's certainly something of track record in Baltimore about ID'ing good defensive guys without the benefit of formal +/-data.)

If nobody here knows whether or not they do +/- data for the MiL's, can somebody here with contacts with the club ask? I don't mean "ask for the data". (There's prolly contractual limits on what they can do with it. And even if there aren't, they wouldn't give us squat if they paid for it.) All I mean is just asking whether or not they have it.

[DrungoHazewood]

As for the whole Offense/Defense thing. A run saved is worth more than a run earned. Want evidence? Using Pythag, and a league where the average team scores five runs per game. Team A is league average defensively, but is +1 run per game better than average offensively - 94 wins. Team B is league average offensively, but +1 run defensively (in other words, gives up 4 runs per game) - 97 wins. Do the same exercise but make the team +2 offensively and average defensively - 105 wins. How about +2 defensively, and average offensively? 116 wins. The trick is finding the right mix of players to make it work.

A run saved is worth more than a run scored, but like you just noted that's a small effect. +1 run per game defensively is 162 runs/season. That's the difference between the best defensive team in the league and the 11th or 13th. That's a huge difference, one that any team is very unlikely to have a choice in - you just can't turn over that much of a roster, and you can't find that many black and white defense/offense players to do it with.

Given the choice between one player who's +30 on defense and -20 on offense, and another who's 0 on defense and +50 on offense the difference in wins due to "a run saved is better than a run scored" is measured in small fractions of a win. Down in the noise.

Yes, as an overall organizational philosophy you'll be a little better off saving runs than trying to score 1000 a year, all else being equal (which it rarely is). But when picking one stopgap shortstop it's largely irrelevant whether you're saving runs or scoring them.

[RShack]

A run saved is worth more than a run scored, but like you just noted that's a small effect. +1 run per game defensively is 162 runs/season. That's the difference between the best defensive team in the league and the 11th or 13th. That's a huge difference, one that any team is very unlikely to have a choice in - you just can't turn over that much of a roster, and you can't find that many black and white defense/offense players to do it with.

Given the choice between one player who's +30 on defense and -20 on offense, and another who's 0 on defense and +50 on offense the difference in wins due to "a run saved is better than a run scored" is measured in small fractions of a win. Down in the noise.

Yes, as an overall organizational philosophy you'll be a little better off saving runs than trying to score 1000 a year, all else being equal (which it rarely is).

OK, do we know how to say whether it's harder to improve in one of those dimensions vs. the other for a several-year window? If you're constructing a team for the next 5 years (or whatever-number-of-years greater than 1 or 2), is it harder to add 100 runs of scoring or a 100 runs of prevention? This is probably 2 questions, really:

• If you're way below average on both O and D, and you're trying to catch up to being OK, is it easier to add 100 runs of O vs. 100 runs of D?
• Same question, but say you're starting at average and trying to get above that level.

I realize that you probably need to add both in either case. But for each one of those two scenarios, is it easier to get better at one vs. the other? Or do we even know? Also, do we know if it's easier to have a farm system reliably grow one vs. the other?

But when picking one stopgap shortstop it's largely irrelevant whether you're saving runs or scoring them.

I completely agree with this statement. (The only people who've ever claimed otherwise are anti-LH-ranters.)