An Interview with Bill James

[Honus Wagner Rules]

Here is a great interview with Bill James from December 2004. It is in three parts. Enjoy. :D

http://baseballanalysts.com/archives/2005/02/breakfast_with.php
http://baseballanalysts.com/archives/2005/03/breakfast_with_1.php
http://baseballanalysts.com/archives/2005/03/breakfast_with_2.php

[west coast orange and black]

many thanx, wags.
i copied and will read after the holidays.

[digglahhh]

BJ: Stats are backward looking by nature. That is one of the limitations of them. One might be able to step from the stats to an assessment of the skills in a more pure form. It's debatable whether there is an ability or a skill involved in hitting with runners in scoring position so at that point, you might cut that off. As long as you are simply dealing with what the stats mean, they are always backward looking. The danger is that because stats are backward looking, if you're not careful, you could be the last person to see something.

There may be a pitcher who adds a pitch and the scout may see immediately that, wow, that pitch looks good and it's going to make him into a totally different pitcher. But, if you are just looking at the stats, you won't see that until two years later when the value of it has gone...so there are some situations in which you need to be aware of that.



This argument sounds vaguely familiar. ;)

[SABR Matt]

James is too linear and analogue...he's not a real statistician.

Statistics are only backward looking if you use direct explanatory stats rather than make a real attempt to understand probability theory and how it applies to FORWARD looking analysis.

[Honus Wagner Rules]

James is too linear and analogue...he's not a real statistician.

Statistics are only backward looking if you use direct explanatory stats rather than make a real attempt to understand probability theory and how it applies to FORWARD looking analysis.
James has never claimed he was a "real" statistician, whatever that means. He uses "stats" as mearly tools to write about baseball history and its inner workings. My favorite James' books have always been the more historical books.

The first half of the Historial Abstracts
The Politics of Glory
The Manager's Book

[SABR Matt]

I know James has never claimed to be a career statistician...my point was that he shouldn't be looked at as an expert in the field of statistics and how they can be used.

[misterdirt]

I know James has never claimed to be a career statistician...my point was that he shouldn't be looked at as an expert in the field of statistics and how they can be used.


Although I haven't always agreed with the answers he came up with, I have always admired Bill James for asking the right questions. That's a large part of being a successful analyst. And on top of that he is a great writer.

[SABR Matt]

Again...agreed. He asks the right questions and he's one of the most engaging writers I've ever had the plpeasure of reading in the field of baseball analysis...but that doesn't mean he should be taken seriously when he claims that statistics are by definition backward looking.

[Honus Wagner Rules]

Again...agreed. He asks the right questions and he's one of the most engaging writers I've ever had the plpeasure of reading in the field of baseball analysis...but that doesn't mean he should be taken seriously when he claims that statistics are by definition backward looking.

Show some respect Matt. :o The man is one of giants of sabermetrics, the Isaac Newton of his field. Since you are only 24 years old I think James might know a "little bit" more than you about sabermetrics since he's been doing it since before you were born. Disagreeing with him is fine and welcomed of course, but your "he doesn't know what he is talking about" mentality is a little over the top.

[SABR Matt]

Show some respect Matt. :o The man is one of giants of sabermetrics, the Isaac Newton of his field. Since you are only 24 years old I think James might know a "little bit" more than you about sabermetrics since he's been doing it since before you were born. Disagreeing with him is fine and welcomed of course, but your "he doesn't know what he is talking about" mentality is a little over the top.

I didn't say he didn't know what he was talking about. Don't put words in my mouth. I don't think he's done any research into the field of predictive sabermetrics beyond the favorite toy, though...and he himself would tell you that.

I have. I'm still way far away from producing anything workable...but my early models worked pretty impressively for just scratching the surface. And no...I'm not saying I know more about everything than him...he's got a lot more experience...duh. But James has a tendency to write off what he hasn't researched in detail...he's doing it again here.

I *STRONGLY* believe predictive sabermetrics can and will get much MUCH better than they currently are...and that at the forefront will be developing an understanding of FORWARD-LOOKING statistics.

[Honus Wagner Rules]

I didn't say he didn't know what he was talking about. Don't put words in my mouth. I don't think he's done any research into the field of predictive sabermetrics beyond the favorite toy, though...and he himself would tell you that.

I have. I'm still way far away from producing anything workable...but my early models worked pretty impressively for just scratching the surface. And no...I'm not saying I know more about everything than him...he's got a lot more experience...duh. But James has a tendency to write off what he hasn't researched in detail...he's doing it again here.

I *STRONGLY* believe predictive sabermetrics can and will get much MUCH better than they currently are...and that at the forefront will be developing an understanding of FORWARD-LOOKING statistics.
Fair enough. I look forward to reading your PCA stuff in the future.

[dabum1]

what does it mean that the stats are "backward looking" and "forward looking" ?

[SABR Matt]

James was saying that when you look at the kinds of stats he's familiar with...the common stats associated with baseball (even ordinary sabermetrics like Win Shares)...they describe what has happened...even explain (if properly analyzed) why it happened...but that's all...no view toward the future.

A forward looking statistic is one that rather than explaining what has already happened...gives you a lot of information about what will probably happen in the near future. Using the laws of probability, and properly analyzing the existing "current" data and how it fits into the great pattern that is the history of baseball (from 1871 to 2005)...you can paint a reliable picture of what is likely to happen in the future...the entire frontier of probability theory is as scarsely researched in sabermetrics...James himself has done some work on log5 probability (a little complicated to explain here)...but only as it applies to single-game match-ups within a season...I'm presently (very slowly) working on extending the idea to a much grander scale, but the overall point here is that the light work done in predictive sabermetrics to date does not represent its potential and I believe James is beginning to get left behind in that regard if he truly believes statistics can only look backward.

[digglahhh]

Matt,

I respect your drive and sincerely hope that you achieve your goal. But what James was really saying was not to ignore things that happen outside the stat sheet, things that happen even off the field, in the offseason, etc.

No stat will tell you if a pitcher has developed a new pitch. No stat will tell you whether a player who exceeds expectations did so as a fluke, or did so because he changed something in his approach or mechanics.

James is not a pure statestician or sabermetrician. That is actually one of the reasons why he is such a good analyst. He understands the numbers, hell he developed a bunch of them, but knows full well that understanding the game is not limited developing metrics. In my opinion, James does a great job of combining several different perspectives in his analysis.

[SABR Matt]

I wouldn't argue against the idea that you have to pay attention to the game on the field and try to come up with explanations when players do significantly better or worse than expectations that go beyond 'it was luck and it won't repeat". Players grow and change with time and things change about their approach...no doubt about it.

I still think James is underestimating the potential for sabermetrics to look forward rather than just backward.

[64Cards]

James was saying that when you look at the kinds of stats he's familiar with...the common stats associated with baseball (even ordinary sabermetrics like Win Shares)...they describe what has happened...even explain (if properly analyzed) why it happened...but that's all...no view toward the future.

A forward looking statistic is one that rather than explaining what has already happened...gives you a lot of information about what will probably happen in the near future. Using the laws of probability, and properly analyzing the existing "current" data and how it fits into the great pattern that is the history of baseball (from 1871 to 2005)...you can paint a reliable picture of what is likely to happen in the future...the entire frontier of probability theory is as scarsely researched in sabermetrics...James himself has done some work on log5 probability (a little complicated to explain here)...but only as it applies to single-game match-ups within a season...I'm presently (very slowly) working on extending the idea to a much grander scale, but the overall point here is that the light work done in predictive sabermetrics to date does not represent its potential and I believe James is beginning to get left behind in that regard if he truly believes statistics can only look backward.
Matt, if you can come up with something that can accurately predict what will happen in future games, please call me, lets move to Nevada, slap down a few bets every day and get rich. But I'm not holding my breath on this...because these are human beings out competing against each other, not just a bunch of statistical figures. And yeah, I can predict with reasonable certainty that Albert Pujols, baring injury, next season, will hit over .300, with 30 plus HR's, and over 100 RBI's and runs scored. Don't have to be Nostradamus to figure that out. As to what Scott Rolen will do, coming off a serious injury to his shoulder, well I'm clueless.

As for James, I've read a few of his books, have enjoyed them for his writing, but to me analyzing a bunch of numbers to death is boring, but then I hated math in school. God, I wish we would of had pocket calculators then.

[SABR Matt]

Not everyone is interested in finding the great patterns that bind things together mathematically. Every sabermetrician knows he's working for a limited audience...didn't need a reminder from you. :D

But seriously, injuries happen...changes in approach happen...players grow and change and conditions in the league change...prediction will never be exact...but it can get a heck of a lot better than it is right now.

[JDD]

I *STRONGLY* believe predictive sabermetrics can and will get much MUCH better than they currently are...and that at the forefront will be developing an understanding of FORWARD-LOOKING statistics. >>><<<

As long as TEAM rosters change, and INDIVIDUAL players indeed get one year older from year to year, forward-looking statistics will remain in the world of crystal balls. There is a huge difference between theoretical and experimental probability. That fact makes this desperate work at best, and a forward-looking whatever is not a statistic, it is a prediction (read:guess).

Forward looking statistics are based on prior data. If someone is 26 or 27, sure we can say that he will remain in his prime for one or two more years because most of his peers have done the same. Father time walks into the room right around 30. But that only goes for players with real talent. Average players drop out of the batting order by then. Those guys only play 100 or so games for about two or three years.

However, those same peak AGES are when the real players produce monster numbers during a time which is often called their "prime". So when an average player is in his prime, he just gets to play. When someone with real talent is in his prime, he sets records and gets headlines.

That is about it for forward-looking metrics. There is not anything new on the way because there cannot be anything new on the way. Everything there is to do is based on trends displayed by players from the past, of similar age, under similar conditions. Anythine else is just a guess using "backwards-looking" data, whatever that means.

The real leap off the cliff is when somone tries to assign value to future results. Meaning, if you think ARod will hit 35 homers in 2007, because a bunch of American League thirdbaseman of recent years with similar ability and age also put up similar numbers relative to his ability and recent yearly results, then that is a good guess. Assigning value to that number is drags league numbers into the metric, and the production of the other eight spots in the batting order. Don't forget the pitching staff. Every run is huge if your offense scores five runs a game, and your pitching staff gives up the same. If their is gap between the offense and defense, then the value of each individual run by a given player will go........ yes.... down.

Consider the value of a run in 1906 versus the value of a run in say 1930 or 1977, and you will see what I mean.