The Bullpen Has Blown 7 Games for Trachsel

[Fan4Life]

For me, stats are only as good as the consistency level they represent. I'm not sure there is a way to evaluate that other than looking at the stats in a more micro level. As a hitter, going 12-40 can be broken down into a 1-20 slump followed by an 11-20 streak... or a 4-14, 4-13, 4-13, etc etc.... their stats will look the same in the aggregate but they are much different hitters if those types of numbers are typical of each player. Frobby had a post about Hernandez earlier remarking on how terrible he was last year in July and August I think but that in September last year, he had an OPS over 1.1. That hot September last year, and perhaps again this year, would likely bring his numbers up to a reasonable level so that when looking at his overall 2007 season, his stats won't reflect how terrible he was when we needed him. The question I would have is: Is this type of hitter (2-15 and then 10-25) more common than the more consistent 4-12, 4-13 guy?

[Old#5fan]

No. I'm tired of playing these games and splitting hairs. If you don't think past performance gives you a clue into future results, that's your prerogative.

I would say "don't give a clue" is similar to a "crapshoot" would you not agree? As such, I am on that side of this argument. In fact, I bet if you took a poll of who believes that past statistics (especially minor league) will be of signicifant use to predict future major league performance that most people (over 50%) don't believe it. (Not sure about this board, but in real life).

[Sports Guy]

I would say "don't give a clue" is similar to a "crapshoot" would you not agree? As such, I am on that side of this argument. In fact, I bet if you took a poll of who believes that past statistics (especially minor league) will be of signicifant use to predict future major league performance that most people (over 50%) don't believe it. (Not sure about this board, but in real life).

Well, there would be a lot of wrong people out there Eli.

[DrungoHazewood]

I would say "don't give a clue" is similar to a "crapshoot" would you not agree? As such, I am on that side of this argument. In fact, I bet if you took a poll of who believes that past statistics (especially minor league) will be of signicifant use to predict future major league performance that most people (over 50%) don't believe it. (Not sure about this board, but in real life).

No.

My belief is that past performance is strongly related to future performance.

Your belief is that it's not, especially when you're talking about minor league performance.

If your supposition is true most people would be wrong.

[RShack]

No. I'm tired of playing these games and splitting hairs. If you don't think past performance gives you a clue into future results, that's your prerogative.

Come on, Drungo, that's the kind of all-or-nothing logic that Eli used: near-certain reliability or complete-and-utter crap shoot, nothing in between. Of course stats that model *some* aspects of past performance give you a clue. They can give you many clues. Sometimes they can provide clues that you wouldn't/couldn't see without them. But it's also true that there's a big diff between something that's a clue and something that's a truly reliable predictor. It's also true that stats about past baseball performance do not reflect everything that's relevant, they only reflect the subset of attributes for which data is readily available and easily acquired. This is not splitting hairs, these things are just facts.

[ScottieBaseball]

Stats are an effective tool. Stats are an effective tool. Stats are an effective tool. Stats are an effective tool. Stats are an effective tool. Stats are an effective tool...

Stats are an effective tool and a primary one at that when coupled with personal observation and the variables that can't be accounted for/captured via sabremetrics (i.e. adjusted stats). Neither stats nor "the good set o' peepers in my head" are the be-all, end-all in predicting the future. If that's what you're looking for and you're not A) God or B) holding the keys to a time machine, then you're not going to have any luck.

[DrungoHazewood]

Come on, Drungo, that's the kind of all-or-nothing logic that Eli used: near-certain reliability or complete-and-utter crap shoot, nothing in between. Of course stats that model *some* aspects of past performance give you a clue. They can give you many clues. Sometimes they can provide clues that you wouldn't/couldn't see without them. But it's also true that there's a big diff between something that's a clue and something that's a truly reliable predictor. It's also true that stats about past baseball performance do not reflect everything that's relevant, they only reflect the subset of attributes for which data is readily available and easily acquired. This is not splitting hairs, these things are just facts.

It's splitting hairs when I've said numbers aren't the be-all end-all many times over but my words gets twisted around to the point where people are challenging me to predict Corey Patterson's OPS against lefties down to three decimal points, and if I decline that's held up as proof that numbers are almost meaningless.

I've said my piece. I'm done with this thread.

[Old#5fan]

No.

My belief is that past performance is strongly related to future performance.

Your belief is that it's not, especially when you're talking about minor league performance.

If your supposition is true most people would be wrong.

Most people would be wrong where? Here or the man on the street? I am 52 years old and have been an avid Oriole, Colt, and now Ravens fan since I attended my first game at Memorial stadium in 1963. I can assure you that most common fans I know (and most of my friends, and family are sports fans) would not put all the stock into stats that is prevalent on this board.

[Sports Guy]

Most people would be wrong where? Here or the man on the street? I am 52 years old and have been an avid Oriole, Colt, and now Ravens fan since I attended my first game at Memorial stadium in 1963. I can assure you that most common fans I know (and most of my friends, and family are sports fans) would not put all the stock into stats that is prevalent on this board.

Well, you and your friends would be wrong.

[Old#5fan]

Come on, Drungo, that's the kind of all-or-nothing logic that Eli used: near-certain reliability or complete-and-utter crap shoot, nothing in between. Of course stats that model *some* aspects of past performance give you a clue. They can give you many clues. Sometimes they can provide clues that you wouldn't/couldn't see without them. But it's also true that there's a big diff between something that's a clue and something that's a truly reliable predictor. It's also true that stats about past baseball performance do not reflect everything that's relevant, they only reflect the subset of attributes for which data is readily available and easily acquired. This is not splitting hairs, these things are just facts.

Yeah, but apparently these stats guys wish to use the stats after the fact but whenever you try to pin them down to predict something they are pretty much useless or as someone might say unreliable in that context.

[Sports Guy]

Yeah, but apparently these stats guys wish to use the stats after the fact but whenever you try to pin them down to predict something they are pretty much useless or as someone might say unreliable in that context.

You just clearly don't understand the concept though.

You bring up CPat...I showed you that 3 years he has been good versus lefties and 3 years he has not been good against lefties.

Small sample sizes can easily change from year to year. This shouldn't be that hard to figure out.

[Old#5fan]

Well, you and your friends would be wrong.

No, they would not be. Every single time someone is asked here to solely use stats to predict anything all they do is hedge and give possibilities. So in reality just how useful are these stats? They are not much better than guessing when it comes to "predicting anything." As a measurement of what has already happened they have great value but other than that they do not stand on their own as much value at all. A observation by a scout would be of more benefit in most cases.

I will give you an example. Suppose a guy is hitting 300 in the high minors but in reality he is hitting 400 on fast balls and 200 on curves. So if you look at his stats he gets promoted to the majors. Once he gets to the majors scouts have already told other teams the guy has trouble with the curve. Now he gets a steady diet of even better quality (i.e. major league curves) and hits below the mendoza line and gets sent back down. Now how reliable were the stats in this case? I will tell you, they are not even presenting an accurate picture period, or the kid wouldn't even be called up until he learns how to hit a curve ball much better. So relying on a scout is far better than any stats.

[Old#5fan]

You just clearly don't understand the concept though.

You bring up CPat...I showed you that 3 years he has been good versus lefties and 3 years he has been against righties.

Small sample sizes can easily change from year to year. This shouldn't be that hard to figure out.

So where is the value of that in predicting what he will do next year?

[RShack]

Care to quantify these ideas?

What do you mean by unreliable? What r value are you thinking off?

No, I don't care to sift thru 11 textbooks on stats and human performance and create a summary. There's just too much stuff to deal with here. Part of the problem is that it's a complicated business, and people who have access to db's want to believe there's a simple cookbook. There isn't. However, I can give you an example that might shed some light on the main difficulty. This isn't about baseball, but I believe the lessons are relevant to baseball.

For a bunch of years I taught big CS classes at Georgia Tech. In the first 2 of them, we grew to the point where we had more than 5,000 seats per year. The grading scheme was weekly quizzes and a big final exam, and it was weighted to let people who did well on the final overcome earlier weaknesses in performance. (This is because I didn't want their grade to reflect what they knew in Week 3, I cared what they knew at the end of the course.) In an effort to ID which students might most need extra work, we deployed a big stat effort aimed at early ID of people who would prove to do poorly in the end so we could try to intervene to help them succeed. (Everybody at GT is smart, but an 18-yr-old goofball is still an 18-year-old goofball regardless of his SAT scores.)

We got stat prof's and doctoral students and psychology faculty and everydamn body involved. We had tons of performance data, indexed by topic and difficulty. Nobody in academia had more data about where their many students succeeded and failed than I did. (IMO, most academic record-keeping and grading is a joke.) But despite a wealth of solid performance data, we couldn't do it. There were just too many variables in play that don't show up as nice, neat little numbers. So, after spending God-only-knows how much time and energy banging our heads against a wall, we eventually invented Plan B, which was to build a neural net and feed everything into it, including both detailed performance data and the subjective answers students gave on weekly surveys about how they thought they were doing, what they thought of the class, etc. Then, the neural net chugs away and does computations like crazy, searching for weights and paths through the data that prove to be reliable predictors. After it chewed on data from a year or two's worth of classes, it became able to make accurate predictions. By week 4 of a 15-week semester, the neural net was predicting the final grades each student would wind up with, and it did so with scary accuracy. But (here's the important part) we don't know why. When we looked at the factors it was giving weight to, it made no sense. It tended to discount seemingly solid factors about demonstrated performance, and it tended to give high weight to student responses on seemingly random subjective questions. It made no rational sense of any kind, but it worked like gangbusters.

The point here is that there are many factors in play, and we don't even know what most of them are. Many important factors are things that we don't have data for. The bottom line is that I could look at a kid's grades after 12 weeks of quizzes, and I would have a clue about if he was *likely* to do well on the final. But my clue was not a reliable predictor. We did develop a reliable predictor, but it relied on things that we don't understand.

There is a whole lot of the same stuff going on when you try to predict future human performance in many things, baseball included. If you disagree, then please show me a stats model that predicted in advance things like Finley, Kent, Clemens, Mora, Cuellar, etc., etc., etc.

Bottom line: I am *not* dissing stats. I am simply drawing attention to the fact that we do not have effective cookbook solutions to effectively predicting complex human performance. We do have things that give us clues, but making near-certain predictions based on those clues is a very risky business. Furthermore, it is least risky in precisely the many routine cases when everybody pretty much knows the answer anyway, stats or no stats, and most risky in precisely those "if-fy" cases where we most want/hope/pray that stats can tell us something we don't already know. As for Trax, we know he's gonna fall off the table sometime pretty soon, but we don't know exactly when. The fact that we know this about him is not predominantly due to great insights gained from stats.

[Sports Guy]

No, they would not be. Every single time someone is asked here to solely use stats to predict anything all they do is hedge and give possibilities. So in reality just how useful are these stats? They are not much better than guessing when it comes to "predicting anything." As a measurement of what has already happened they have great value but other than that they do not stand on their own as much value at all. A observation by a scout would be of more benefit in most cases.

I will give you an example. Suppose a guy is hitting 300 in the high minors but in reality he is hitting 400 on fast balls and 200 on curves. So if you look at his stats he gets promoted to the majors. Once he gets to the majors scouts have told other teams the guy has trouble with the curve. Now he gets a steady diet of even better quality (i.e. major league curves) and hits below the mendoza line and gets sent back down. Now how reliable where the stats in this case? I will tell, you they are not even presenting an accurate picture period.

First of all, anyone who looks at BA as a predictor is looking at the wrong thing.

Secondly, what else is that player doing? Are they walking a lot? Are they striking out at an alarming rate?

Pitchers or hitters league?

How old are they?

Are they showing power potential?