PDA

View Full Version : Ups and Downs of Randomness, Observation, Stats and Baseball



GullyFoyle
06-01-2006, 01:32 PM
Few people understand the concept of randomness and percentages in statistics.

For example:

Some people complain that iTunes is not random on shuffle, it plays certain groups more often, even though Apple has stated multiple times that it is indeed random.

If you play online games like World of Warcraft you always get somebody in a group that says a roll is not random because the same number shows up twice or three times in a row.

What most people don't realize is that random does not mean that you wont be able to find a pattern in it. It also does not mean that all the given outcomes are equally distributed. This was demonstrated on a 60 minutes show a couple of years ago. They had a statistics professor come on the show and talk about his basic stats class. In the class he asked all the students to either

A) flip a coin 50 times and mark down the results (heads or tails) each time

or

B) make it up and don't actually flip the coin

He could always tell the difference between the two groups right away because the people who actually flipped would end up with stretches of 4-6 heads or tails in a row while the people making it up wouldn't. People in general don't realize that long stretches of the same outcome is very likely in 50 flips. What people expect from randomness is a nice even distribution of heads and tails (or songs or hits).

I flipped a coin 50 times and this is what I got:

HHHTTTHHHHTHHTTTTHTHHHHTHHTHTHTTTHTHTTTTTTHHHTTHTT

24 heads, 26 tails and a series of 6 tails in a row and multiple series of 4 in a row


Now lets take this principle to baseball. Lets say I have a .333 avg hitter. He gets a hit 1 out of every 3 times. Now lets take a 12 sides die (because there are no three sided die and yes I'm an old D+D geek). If I roll 1-4, its a hit and 5-12 it is not and I'll roll it 50 times.

Here is what I got (H=hit, O=out)

OOOHHHOOHOHHHOOHOOOOOOOHOOOOOOHOOHOOHOOHOOHOOHHOHO

17 hits in 50 rolls (slightly better than average), but also a streak of 14 rolls with only one hit!

Now imagine what people would have said about this hitter during that 1 in 14 at bat span. You would have seen all sorts of threads taking apart his stance, swing, who is hitting in front and behind, he needs a day off, change positions, swing more, swing less, etc.

Guess what... a .333 will go through periods of not getting a lot of hits and this is just in 50 at bats.

This is why you can't take too much from observation and small sample sizes. The mind wants to find patterns in everything it experiences. It is not well equipped to see randomness.

People expect that if something is random then it means that it is well distributed and this is just not the case. So as our favorite team gets hot and cold remember that even the best hitters will go long times without hits and the best teams can go long stretches without winning (a .500 team is the same as the coin toss above).

(BTW, The purpose of the 60 minutes show was to demonstrate how the IRS catches cheats because people are bad at making up random numbers on their tax returns.)

Hope people found this interesting... back to what I should be doing... :)

savafan
06-01-2006, 01:38 PM
One night when I was over at Ravenlord's house visiting he and his brother, along with another friend of ours, I was playing with a pair of dice. I rolled and got 7. After two more rolls, again both times I rolled 7. I think we went up to some ludicrous number like 19-20 consecutive rolls and each time I rolled I ended up with 7.

GullyFoyle
06-01-2006, 01:40 PM
One night when I was over at Ravenlord's house visiting he and his brother, along with another friend of ours, I was playing with a pair of dice. I rolled and got 7. After two more rolls, again both times I rolled 7. I think we went up to some ludicrous number like 19-20 consecutive rolls and each time I rolled I ended up with 7.

Would have been a good time to hit the Craps table!

(of course there are six combinations on two dice that add to seven so not that unheard of)

dabvu2498
06-01-2006, 01:41 PM
I hit blackjack 4 times in a row playing 100 a hand one night. I will never gamble again.

I will never lie again, either.

SeeinRed
06-01-2006, 01:57 PM
The most common problem people have with statistics is that they look at a series of events and say that they go against the norm, i.e. a series of 10 flips of a coin. Say you get 8 Heads and 2 tails. Some people would say it is an oddity. In reality its because they want to relate a series of events that are independent of each other. Each time you flip a coin, there is a 50 percent chance of getting heads, and a 50 percent chance of getting tails. The next time you flip the coin, the percentages stay the same reguardless of the outcome of the last flip, or series of flips. Thats why streaks can happen. But, in the same reguard, the larger the sample size, the closer to 50-50 your data. Now this is a nice cut and dry example with no outside factors like there are in baseball (pitchers faced, situation, weather, and so on). In baseball there are events that can be related to the success or failure to get a hit in each individual at bat. When a human becomes a factor, statistics are not as reliable, which I guess you could say that a person could not possibly flip a coin the same way every time with a coin. But with both of these cases, large sample sizes are the only way to make those errors less of a factor.

To sum it up, and pretty much probably repeat what GullyFoyle said, statistics are only used to predict the outcome of the overall outcome in the long term. If you use them to predict a singular event, the chances of being right are severly decreased.

gonelong
06-01-2006, 01:58 PM
Few people understand the concept of randomness and percentages in statistics.
...
Now imagine what people would have said about this hitter during that 1 in 14 at bat span.
...


Very. Nice. Thread.

GL

SeeinRed
06-01-2006, 02:00 PM
Very. Nice. Thread.

GL

I second that. :beerme: Very, very good points, and a clear message.:thumbup:

pedro
06-01-2006, 02:03 PM
Thanks for posting this.

dabvu2498
06-01-2006, 02:13 PM
Let's face it. Most people (myself included at times) find in depth and abstract statistics a complete bore.

It's much easier to sit back, swill your beer, and say "Frickin LaRue, 1 for 14."

Most real baseball people (and even some fans) are much more patient and understaning when it comes to short-term vs. long-term trends.

I can't remember where I read this, but there was a study done and the results showed that luck was responsible for 4 runs per game while talent was responsible for 1 run per game. Over a 162 game season, luck evens out and talent shines through. The same can be said for individual talent, I would guess.

People just aren't that patient (me included.).

Cyclone792
06-01-2006, 02:20 PM
Very. Nice. Thread.

GL

^^^ What GL said.

Excellent post, Gully.

pedro
06-01-2006, 02:23 PM
I can't remember where I read this, but there was a study done and the results showed that luck was responsible for 4 runs per game while talent was responsible for 1 run per game. Over a 162 game season, luck evens out and talent shines through. The same can be said for individual talent, I would guess.



along the same lines pretty much every team is going to win 60 games and lose 60 games. It's what you do with the other 40 that counts.

GullyFoyle
06-01-2006, 02:25 PM
Let's face it. Most people (myself included at times) find in depth and abstract statistics a complete bore.

It's much easier to sit back, swill your beer, and say "Frickin LaRue, 1 for 14."
....

People just aren't that patient (me included.).

Very true, and I'm as impatient as the next person.

It seems a lot of the most heated arguments on the board comes from people not realizing their impatience or their own subjectivity...

But that never happens anywhere else either /sarcasm :P

RedsManRick
06-01-2006, 02:33 PM
Great post. This is why "playing the percentages" bugs me in so far as how it's typically used. Let's pretend our manager is named Lony RaTussa, and we have player A and player B.

Player A
Season line: .280/.350/.470
Career line vs. Pitcher Z: 5-24, 1 2B, , 6Ks

Player B
Season line: .250/.310./.390
Career line vs Pitcher Z: 4-9, 1 HR, 5 RBI

It seems like every game Mr. RaTussa coaches we get treated to some crap about how player B really hits Pitcher Z well and that's why he's getting the start.

Baseball gives managers too much opportunity to tweak and "play the percentages". As such, they are almost unamimously over-reacting to small sample size and failing to let things play out.

Of course, all this is horribly complicated by the fact that players are not static entities with fixed outcome percentages (like dice). Sometimes that .333 hitter really IS playing like crap and his 1-14 streak is indicitive not of a random streak, but of a change in his "true" ability. What separates the good managers from the bad ones is the ability filter out the randomness.

While Joe Torre has been given some of the best talent, I commend him for being willing to let things work themselves out, rather than tinker needlessly.

dabvu2498
06-01-2006, 02:51 PM
Baseball gives managers too much opportunity to tweak and "play the percentages". As such, they are almost unamimously over-reacting to small sample size and failing to let things play out.

Of course, all this is horribly complicated by the fact that players are not static entities with fixed outcome percentages (like dice). Sometimes that .333 hitter really IS playing like crap and his 1-14 streak is indicitive not of a random streak, but of a change in his "true" ability. What separates the good managers from the bad ones is the ability filter out the randomness.


Good work! You'll notice that "playing the percentages" goes down in crunch time in late August and September and of course the playoffs. Then you typcially saddle the horses you rode in on.

I wish Manager RaTussa could get us there.

Ltlabner
06-01-2006, 04:25 PM
A different thread on static vs ever changing line ups touched on this idea. A players average for success (whatever the measurement) is built over the course of many games that sometimes streaches over many, many seasons. Those stats (let's say OBP against lefties) have been created over the course of many streaks, many slumps, many lucky situations, many unlucky situations and some just plain freak occurances.

While all those situations average out to give us that OBP percentage, it's only a good indicatator of what a player is likely to do over another simular period of time (ie. if he's hit .312BA for the past 3 years, it's likly he'll hit .312BA in the next 3 years, only assuming that all things remain equal).

It's not a good predictor of what that player will do in a specific matchup on a given night. There are just too many variables to consider. His odds for success over the long haul of hitting against lefties will be .312 but his odds of getting a hit against a specific pitcher in a specifc at bat is not 30%.

This is why I'd rather coachs assign roles, put players in a psedu-static line up (I know that there will always be minor tweeks) and let the averages play out over the course of a season instead of trying to manage the averages of each given situation which is next to impossible.

GullyFoyle
06-01-2006, 04:53 PM
Of course, all this is horribly complicated by the fact that players are not static entities with fixed outcome percentages (like dice). Sometimes that .333 hitter really IS playing like crap and his 1-14 streak is indicitive not of a random streak, but of a change in his "true" ability. What separates the good managers from the bad ones is the ability filter out the randomness.



A different thread on static vs ever changing line ups touched on this idea. A players average for success (whatever the measurement) is built over the course of many games that sometimes streaches over many, many seasons. Those stats (let's say OBP against lefties) have been created over the course of many streaks, many slumps, many lucky situations, many unlucky situations and some just plain freak occurances.

Very good points that I was trying to get my thoughts around. I find the relationship between "true ability" and chance occurrence to be interesting.


My biggest "small sample" pet peeve is during the playoffs when the TV announcers show players' stats for that particular World Series as being 1 for 4 or some other nonsense instead of showing what he did all season long! Drives me up a wall.

Ltlabner
06-01-2006, 05:25 PM
My biggest "small sample" pet peeve is during the playoffs when the TV announcers show players' stats for that particular World Series as being 1 for 4 or some other nonsense instead of showing what he did all season long! Drives me up a wall.

I totally agree GullyFoyle. Along the same lines...When they give a players stats for the serries against a particular team and try to draw some clever analysis from it durring the last game of that serries. First it's a small sample that means nothing. Second, it's a totally new pitcher. Third, it's a new day and all the variables have changed: the weather, if he has a cold, if he got drunk the night before, the umpire, the mood of the crowd, if he's got personal problems, if he stubbed his toe in the clubhouse before the game, if he's got a personal grudge against this new pitcher, etc etc.

I'm getting more intersted in the statistical side of baseball and would like to expand my knoweldge in this area, but I think TV annoucers are especially guilty of trying to be too clever for their own good and use numbers they don't understand, and often times use them incorrectly.

Newman4
06-01-2006, 08:22 PM
The most common problem people have with statistics is that they look at a series of events and say that they go against the norm, i.e. a series of 10 flips of a coin. Say you get 8 Heads and 2 tails. Some people would say it is an oddity. In reality its because they want to relate a series of events that are independent of each other. Each time you flip a coin, there is a 50 percent chance of getting heads, and a 50 percent chance of getting tails. The next time you flip the coin, the percentages stay the same reguardless of the outcome of the last flip, or series of flips. Thats why streaks can happen. But, in the same reguard, the larger the sample size, the closer to 50-50 your data. Now this is a nice cut and dry example with no outside factors like there are in baseball (pitchers faced, situation, weather, and so on). In baseball there are events that can be related to the success or failure to get a hit in each individual at bat. When a human becomes a factor, statistics are not as reliable, which I guess you could say that a person could not possibly flip a coin the same way every time with a coin. But with both of these cases, large sample sizes are the only way to make those errors less of a factor.

To sum it up, and pretty much probably repeat what GullyFoyle said, statistics are only used to predict the outcome of the overall outcome in the long term. If you use them to predict a singular event, the chances of being right are severly decreased.

Very well stated. Who was your math teacher again? lol :D

Newman4
06-01-2006, 08:24 PM
One night when I was over at Ravenlord's house visiting he and his brother, along with another friend of ours, I was playing with a pair of dice. I rolled and got 7. After two more rolls, again both times I rolled 7. I think we went up to some ludicrous number like 19-20 consecutive rolls and each time I rolled I ended up with 7.

By the way, the theoretical probability of rolling a '7' on two dice 20 consecutive dice is 1 in 3.65615844 × 10 to the 15th power. Just saying :)

SeeinRed
06-02-2006, 12:26 AM
Very well stated. Who was your math teacher again? lol :D

You think my statistical knowlege is good, you should see my geometric skills. :D

Nugget
06-02-2006, 12:37 AM
Whilst all that is said is true there are outside factors that statistics do not cater for. For instance, players are human and for some reason a player may perform differently under different conditions. Whilst everyone would like to believe (and mathematically it is) that a postseason game is no different to a game in the regular season for some players it isn't. Whether it be the pressure, the exuberance, the exposure, there are outside factors. Ideally, if you are looking at it you are setting the parameters. Accordingly, where managers play the percentages they are essentially looking at a different set of parameters. For instance, the lefty v right parameter.

Whilst you can look at overall statistics they will be truly random like tossing a coin or rolling the dice. But if you are looking at specific parameters then there may be differences which means that baseball statistics aren't always going to be particularly random.

savafan
06-02-2006, 12:56 AM
By the way, the theoretical probability of rolling a '7' on two dice 20 consecutive dice is 1 in 3.65615844 × 10 to the 15th power. Just saying :)

I don't even know what that means, but it is interesting. :)

GullyFoyle
06-02-2006, 01:05 AM
Whilst everyone would like to believe (and mathematically it is) that a postseason game is no different to a game in the regular season for some players it isn't. Whether it be the pressure, the exuberance, the exposure, there are outside factors. Ideally, if you are looking at it you are setting the parameters. Accordingly, where managers play the percentages they are essentially looking at a different set of parameters. For instance, the lefty v right parameter.

But extremely small sample sizes (like those from a short series) can be indicative of almost any outcome... from bad to great... because even a hypothetical .500 hitter would go 6 or 7 at bats without a hit from time to time. So if someone comes up in the second game of the World Series and he has gone 0 for 4 in the previous game, this tells me nothing about his hitting. Where if I'm told that he batted .300 for the season this gives me an idea of the quality of batter. The same thing is true if a batter has only seen a pitcher a couple of times.

Nugget
06-02-2006, 01:10 AM
True and I agree with that but I wouldn't say the 0 for 4 is totally useless, just as the season long statistic is not totally useless. What I would say is that a post season record for a player in his first season means less than for someone who has been there a couple of times already.

savafan
06-02-2006, 01:26 AM
But what is his batting average when playing in parks where the plate faces the east, at nighttime against left handed Latino pitchers whose first name starts with the letter "J" and were born in the month of August? That's what I really want to know
:D

SeeinRed
06-02-2006, 01:28 AM
Whilst all that is said is true there are outside factors that statistics do not cater for. For instance, players are human and for some reason a player may perform differently under different conditions. Whilst everyone would like to believe (and mathematically it is) that a postseason game is no different to a game in the regular season for some players it isn't. Whether it be the pressure, the exuberance, the exposure, there are outside factors. Ideally, if you are looking at it you are setting the parameters. Accordingly, where managers play the percentages they are essentially looking at a different set of parameters. For instance, the lefty v right parameter.

Whilst you can look at overall statistics they will be truly random like tossing a coin or rolling the dice. But if you are looking at specific parameters then there may be differences which means that baseball statistics aren't always going to be particularly random.

Wow that is way more philosophical than I can even think. I applaud you. Now, just for argument: The basis of statistics is to try to make an outcome on a set of events predictable. This can only work if every factor is taken into consideration. In baseball there are so many factors, that you can't possibly consider them all. They may be small, but they are there. But in the same reguard, even just using numbers, there is always an error rate in statistics. The object is to make that error rate as little as possible. Really though, taking for example batting average, baseball statistics can be very crude. No factors other than what the outcome is of an offical plate apperance being either a hit, or an out, is taken into consideration. Wind is not considered, the pitcher is not considered, and so on. In reality, batting average can give you a general idea of what MIGHT happen in the future, but not really a prediction of a particular at bat. In other words statistics like batting average tell you what HAS happened not what WILL happen, which I guess some will say can be said about statistics in general, but even more so in this case.

Statistics are neat, and serve a real world purpose, but they are often misunderstood. Statistical analysis will tell you that 1 out of 3 men get prostate cancer. But that doesn't mean that because you have 2 brothers, and neither of them have prostate cancer, that you are at high risk for it. The other factors are not mentioned in that evaluation of the data. For instance, family history may be a factor. Diet may be a factor.

Looking at a lot of statistics may say that Player A has a better chance of producing more runs in the future than Player B, but that is a guess at best, and as they say, that is why they play the game. Now don't get me wrong either, statistics are very usefull in baseball because you can look at statistics and put the percentages in your favor. Thats what the good GMs do. But the human factor cannot be overlooked, which is part of the problem I have with the "Money Ballers" who say that stats are all that matter. Stats are part of it, but there is a whole other side that can't be overlooked.

Sorry, I kinda ran from my original point and got a little carried away, but I had fun. Hope someone will find this usefull.

SeeinRed
06-02-2006, 01:31 AM
But what is his batting average when playing in parks where the plate faces the east, at nighttime against left handed Latino pitchers whose first name starts with the letter "J" and were born in the month of August? That's what I really want to know
:D
:laugh:
All I gotta say is... I just wasted 30 minutes of my time if thats all you really want to know.... and my head hurts.

SteelSD
06-02-2006, 01:36 AM
Looking at a lot of statistics may say that Player A has a better chance of producing more runs in the future than Player B, but that is a guess at best...

No, it's really not.


...and as they say, that is why they play the game. Now don't get me wrong either, statistics are very usefull in baseball because you can look at statistics and put the percentages in your favor. Thats what the good GMs do. But the human factor cannot be overlooked, which is part of the problem I have with the "Money Ballers" who say that stats are all that matter. Stats are part of it, but there is a whole other side that can't be overlooked.

One of the primary veins of "Moneyball" is that the subjective cannot be overlooked but it can be dramatically overvalued. But there isn't a single person in the world who thinks that statistics are "all that matters" in regard to baseball- including Paul DePodesta and Billy Beane.

SeeinRed
06-02-2006, 01:47 AM
No, it's really not.

You're thinking too much. Injury could change that outcome. Maybe player A is peaking and player B hasn't peaked yet. The outcome is all we are looking at right now, not what happens to get that outcome, because the factors that are not considered are unpredictable. Injuries don't only happen to the ones who are "injury prone." As a general rule, the future is always a guess.


One of the primary veins of "Moneyball" is that the subjective cannot be overlooked but it can be dramatically overvalued. But there isn't a single person in the world who thinks that statistics are "all that matters" in regard to baseball- including Paul DePodesta and Billy Beane.

Thats why I put Money Ballers in quotes. Meaning so called Money Ballers who look too much at stats. I realize MoneyBall can and has worked in real world applications. I never said that anybody looks solely at stats, but in the same way you say the subjective can be dramatically overvalued, I say it can also be dramatically undervalued.

Nugget
06-02-2006, 01:55 AM
I never said that anybody looks solely at stats, but in the same way you say the subjective can be dramatically overvalued, I say it can also be dramatically undervalued.

I think thats the rub - How much weighting you give to statistics and how much weight you give to observation. That my lud will be a question that only the baseball gods can answer...

SeeinRed
06-02-2006, 02:07 AM
I think thats the rub - How much weighting you give to statistics and how much weight you give to observation. That my lud will be a question that only the baseball gods can answer...

I hear that, but my point is that there is a place in the middle that is the right answer. Where that is, nobody knows, but to say that one side is more important than the other is just crazy. I have been preaching that stats can't be used to DEFINATIVELY predict the future because of the human intangibles, but in the same reguard, just because a guy is a great competitor with the tools to play the game, doesn't mean he will be a great player either. There is no sure way to predict the sucess of a player in the future, but the best way to make an educated GUESS at that is to take statistics, and observations from scouts as a package. Dontrell Willis is a good example. His stats are good, but some people worry about his mechanics. Do you take a long term risk on the guy knowing that he may get injured? It all depends on wether you trust statistics or scouts more. Just so you know, I'd take the risk. I like the stats on this one. Not saying the Reds should go get him though.

Nugget
06-02-2006, 02:31 AM
I agree with it all but Dontrelle. The argument about stats versus observations has been a hot topic even during the tmie of the last GM hiring rounds.

What I don't like about the Dontrelle risk is that most of the rumours look like you would overpay for Dontrelle even if he didn't get injured. Unlike the Oakland trades the Marlins appear to be asking for considerably more plus Willis' desire to stay with the REDS once FA appears would be questionable too.

SeeinRed
06-02-2006, 02:51 AM
I agree with it all but Dontrelle. The argument about stats versus observations has been a hot topic even during the tmie of the last GM hiring rounds.

What I don't like about the Dontrelle risk is that most of the rumours look like you would overpay for Dontrelle even if he didn't get injured. Unlike the Oakland trades the Marlins appear to be asking for considerably more plus Willis' desire to stay with the REDS once FA appears would be questionable too.

I couldn't agree more. But it was just a hypothetical situation in an attempt to relate it to stats vs. the human. I don't want to start a "Reds should/shouldn't pursue Willis" argument. I've seen what those can do.

RFS62
06-02-2006, 09:25 AM
The concept of "all things being equal" in the coinflip exercise is what separates it from meaningful application in the everyday managing of a baseball team.

That's why advance scouting is so important.

The advance scouts tell you things about the "current" conditions that large sample sizes may miss.

If a player is injured or having some kind of problem that takes him out of his "norm", you may chose to approach him in a different way. If the league has found a hole in a player's swing, the approach now changes and it's a different set of circumstances than before.

Players get hot and cold, change approaches and adjust based on changes the other team implements towards him, and even personal problems can factor in to performance at any given time.

Stats are much more meaningful in large sample sizes, I agree. But doesn't this mean that the subjective analysis of smaller periods of time, such as advance scouting, is more useful to a manager in crafting his approach to the opponents?

gonelong
06-02-2006, 10:47 AM
Stats are much more meaningful in large sample sizes, I agree. But doesn't this mean that the subjective analysis of smaller periods of time, such as advance scouting, is more useful to a manager in crafting his approach to the opponents?

Yes and No.

IMO you still have to meld the two together, and much of the subjective is just that, subjective. It really depends on who your advanced scouts are, how well they do their job, and much much management is willing to use and trust their data.

You go to watch the Braves play the Mets this week because you have both coming in for a series. Pitcher A strikes out Batter B. Was it because Pitcher A has his slider working or because Batter B can't lay off the slider?

In the mean time Batter B has taken his off day to work on his swing and notices he isn't laying off the slider like he should be.

You come into the series and he is letting them pass, works the count, and your pitcher gets behind and has to come in to him and ... POW - 3 run homer.

GL

RFS62
06-02-2006, 10:54 AM
Yes and No.

IMO you still have to meld the two together, and much of the subjective is just that, subjective. It really depends on who your advanced scouts are, how well they do their job, and much much management is willing to use and trust their data.

GL



I sometimes feel like the very word "subjective" gets a bad rap around here. There have been so many subjective arguments smacked down with pure statistical analysis, that any subjective argument immediately get's the hairy eyeball from many of us here.

Not all subjective judgment is the same. The expertise and experience of the observer varies greatly and is of major importance in any subjective argument.

gonelong
06-02-2006, 11:15 AM
I sometimes feel like the very word "subjective" gets a bad rap around here. There have been so many subjective arguments smacked down with pure statistical analysis, that any subjective argument immediately get's the hairy eyeball from many of us here.

It is what it is.

The older I get and the more experience I gather the more hairy my eyeball gets to subjective analysis/arguments about ANY subject. I realize people have agendas and biases, even if they are doing their level best to represent things as well as they can.

I have turned an even more <pun alert>calculated</pun alert> eye towards baseball when I am confronted by "baseball men" that have opinions that line up with neither my subjective or objective analysis.

OTOH, I am not blinded by the numbers. I realize a 22 year old kid is not a finished product. I realize a pitcher might need a few years before he really pays off. I realize that a players attitude can affect his teammates, and that a player is not just the sum total of his OPS.

I also see that many fans, players, broadcasters, scouts, GMs, managers, etc. are lousy at subjective analysis. It doesn't mean they all are, by any means.


Not all subjective judgment is the same. The expertise and experience of the observer varies greatly and is of major importance in any subjective argument.

I agree 100%, I think that goes without saying.

Which is what I was trying to say with this:


Was it because Pitcher A has his slider working or because Batter B can't lay off the slider?

and this:



It really depends on who your advanced scouts are, how well they do their job, and how much management is willing to use and trust their data.



GL

GullyFoyle
06-02-2006, 11:19 AM
I sometimes feel like the very word "subjective" gets a bad rap around here. There have been so many subjective arguments smacked down with pure statistical analysis, that any subjective argument immediately get's the hairy eyeball from many of us here.

I think this is because to know someone else's ability to judge the game takes a lot of time and interaction with that person. Everyone and his mother can come in here and say "I saw this and xxx and yyy and I've been doing it forever". But until you know that person and their skill level how much weight does their opinion have.

Statistics in general mean the same regardless of the person saying them. The problem, I think, with stats is that people not use to thinking about them don't understand their implications, what they cover and don't cover and what part of the game they take in and don't take in. Runs Created and VORP are great stats but trying to explain to someone how they are created and what they cover on the field of baseball is difficult.

Anyway, my hope with this thread was to give a little more info about what I had learned about a very basic part of statistics.

gonelong
06-02-2006, 11:24 AM
I think this is because to know someone else's ability to judge the game takes a lot of time and interaction with that person. Everyone and his mother can come in here and say "I saw this and xxx and yyy and I've been doing it forever". But until you know that person and their skill level how much weight does their opinion have.

Statistics in general mean the same regardless of the person saying them. The problem, I think, with stats is that people not use to thinking about them don't understand their implications, what they cover and don't cover and what part of the game they take in and don't take in. Runs Created and VORP are great stats but trying to explain to someone how they are created and what they cover on the field of baseball is difficult.

Anyway, my hope with this thread was to give a little more info about what I had learned about a very basic part of statistics.

Spot on.

Spot on.

You did, and I appreciate it.

GL

GullyFoyle
06-02-2006, 11:27 AM
TANGENT ALERT


True and I agree with that but I wouldn't say the 0 for 4 is totally useless, just as the season long statistic is not totally useless.

My question, to those here who have more knowledge about statistics than I do and have time on their hands, is...

I know there is a way to determine whether or not a given sample size is large enough to provide any quality information. But I think I fell asleep that day in Stats class.

Would anyone care to give a short refresher on what that is? (I think how it works might beyond a message board in complexity). ;)

Johnny Footstool
06-02-2006, 11:46 AM
Great post. This is why "playing the percentages" bugs me in so far as how it's typically used. Let's pretend our manager is named Lony RaTussa, and we have player A and player B.

Player A
Season line: .280/.350/.470
Career line vs. Pitcher Z: 5-24, 1 2B, , 6Ks

Player B
Season line: .250/.310./.390
Career line vs Pitcher Z: 4-9, 1 HR, 5 RBI

It seems like every game Mr. RaTussa coaches we get treated to some crap about how player B really hits Pitcher Z well and that's why he's getting the start.

Baseball gives managers too much opportunity to tweak and "play the percentages". As such, they are almost unamimously over-reacting to small sample size and failing to let things play out.

Of course, all this is horribly complicated by the fact that players are not static entities with fixed outcome percentages (like dice). Sometimes that .333 hitter really IS playing like crap and his 1-14 streak is indicitive not of a random streak, but of a change in his "true" ability. What separates the good managers from the bad ones is the ability filter out the randomness.

While Joe Torre has been given some of the best talent, I commend him for being willing to let things work themselves out, rather than tinker needlessly.

In LaRussa's defense, he doesn't rely exclusively on the small samples. He also incorporates scouting in his analysis. He has a good idea what kind of stuff a pitcher has, and what pitches a batter can handle. He blends those matchups and the stats (small samples that they are).

He's not Montgomery Burns "playing the percentages" and sending Homer to pinch-hit for Darryl Strawberry to avoid a lefty-lefty matchup.

RFS62
06-02-2006, 11:49 AM
I think this is because to know someone else's ability to judge the game takes a lot of time and interaction with that person. Everyone and his mother can come in here and say "I saw this and xxx and yyy and I've been doing it forever". But until you know that person and their skill level how much weight does their opinion have.

Statistics in general mean the same regardless of the person saying them. The problem, I think, with stats is that people not use to thinking about them don't understand their implications, what they cover and don't cover and what part of the game they take in and don't take in. Runs Created and VORP are great stats but trying to explain to someone how they are created and what they cover on the field of baseball is difficult.

Anyway, my hope with this thread was to give a little more info about what I had learned about a very basic part of statistics.




Yeah, I really appreciate your input.

From my point of view, I have no real interest in doing the math myself, much less searching for newer and better metrics.

But I do want to know what the sabrmetrics community comes up with as a consensus. I trust them to hammer it out and tweak the formulas.

And even OPS and RC/27 have been tweaked and modified since their appearance. Bill James himself questioned his past work on clutch hitting, which I thought was fantastic as it showed that the truth is what he's really after.

The whole thing is in a state of evolution. Scouts rely on stats and scientific measurements constantly, and have since day one. The stats they rely upon are just different, less comprehensive. But the basic idea of using statistics to measure performance is as old as baseball itself.

SteelSD
06-02-2006, 12:00 PM
You're thinking too much. Injury could change that outcome. Maybe player A is peaking and player B hasn't peaked yet. The outcome is all we are looking at right now, not what happens to get that outcome, because the factors that are not considered are unpredictable. Injuries don't only happen to the ones who are "injury prone." As a general rule, the future is always a guess.

Sure injury could alter the future. That being said, we can use analysis involving all available data (both objective and subjective) to identify injury risk. A bunch of folks on this very board predicted Paul Wilson going down. Ditto for Eric Milton's awful 2005 due to an incorporation of objective data (past performance history, knee injury that will never heal).

RFS also noted advanced scouting and that is important. You send an advance scout out to gather information and find that the team's leadoff hitter has a tweaked calf muscle the team hasn't told anyone about. Certainly that can alter short-term gameplanning versus an opponent because the information, while observational or anecdotal in nature, is still valid information.

But the problem with too many conclusions derived from observation is that they simply doesn't hold up to the validity litmus test. One of the reasons that's so is the random nature of the game noted by the thread starter. Catch a player during a random cold streak and it may make one think that said player isn't very good when the opposite is true. Seeing something and knowing whether or not you're seeing truth is the primary difference between good scouts and bad scouts.


Thats why I put Money Ballers in quotes. Meaning so called Money Ballers who look too much at stats. I realize MoneyBall can and has worked in real world applications. I never said that anybody looks solely at stats, but in the same way you say the subjective can be dramatically overvalued, I say it can also be dramatically undervalued.

Actually, you said this:

But the human factor cannot be overlooked, which is part of the problem I have with the "Money Ballers" who say that stats are all that matter.

If you believe that, then you're arguing against a position that doesn't exist. If you have, however, revised your position to allow for the possibility that "Moneyballers" (whoever they are) understand that objective and subjective data need to be incorporated in an analysis, then you're closer to truth.

The key is balancing the two and I'd suggest that the folks you consider to be unbalanced are a heck of a lot more balanced than you think. You might say "too much stats". I might say "not enough information". The validation is in who's right more often. And I don't particularly care how someone gets to "more often right" as long as they get there.

And you're right that statistics cannot be used to predict what is to happen with 100% accuracy. But that's another argument no one has ever made. Instead, we're talking about probability and really nothing but probability. Knowing that, we also know that the most valid information does the best job of projecting what is most likely to happen in the future.

Cyclone792
06-02-2006, 12:33 PM
This is a fascinating thread, and I've been trying to dig up some time this morning to chime in.

Anyhow, last fall the authors of Baseball Prospectus published Mind Game, which was essentially an analysis of how the Boston Red Sox approached every detail with the 2004 season, and especially the 2004 postseason.

I don't have the book at my disposal right now since I'm typing this at work, but one particular topic I remember the BP authors diving into was the extreme high level of advance preparation the Red Sox put together on the Cardinals prior to the 2004 World Series. In fact, the book's research goes so far as to even claim that the totality of Boston's preparation on the Cardinals just blew away the advance preparation St. Louis had on Boston. Essentially, the statistical edge and advance scouting edge Boston had on the Cardinals provided them a major advantage heading into the Series, and may have played a key role in the Red Sox sweep.

What's very interesting about the in-depth details provided in the book is that the Boston Red Sox, a highly sabermetric leaning franchise led by Theo Epstein and housing Bill James in its front office, just went out and floored Walt Jocketty and Tony La Russa in advance statistical and scouting preparation. It wasn't that the Red Sox prepared better solely with the use of statistics - they did do that - but also that their advance scouting reports were remarkably and accurately prepared.

It was the combination of using both objective and subjective measures at the highest level that set the Red Sox apart from the Cardinals, and Boston's goal was to arrive as close to absolute truth as possible using every possible measure available to them, with an excellent combination of statistical and scouting tools.

RFS62
06-02-2006, 12:37 PM
It was the combination of using both objective and subjective measures at the highest level that set the Red Sox apart from the Cardinals, and Boston's goal was to arrive as close to absolute truth as possible using every possible measure available to them, with an excellent combination of statistical and scouting tools.


As well it should be. I heard James on XM months ago talking about how much he had learned from the Red Sox scouts since he's been there.

Balance between the disciplines. That's the optimum baseball mind, IMO.

gonelong
06-02-2006, 12:39 PM
It was the combination of using both objective and subjective measures at the highest level that set the Red Sox apart from the Cardinals, and Boston's goal was to arrive as close to absolute truth as possible using every possible measure available to them, with an excellent combination of statistical and scouting tools.
Euphoria. :beerme:

GL

pedro
06-02-2006, 12:41 PM
Too bad Joe Morgan isn't as concillatory about admitting there might be some things that he might be able to learns from others, made especially so ironic when the lessons learned could really apply so directly to understanding some of the reasons why he himself was such a good player.

SeeinRed
06-02-2006, 01:18 PM
Sure injury could alter the future. That being said, we can use analysis involving all available data (both objective and subjective) to identify injury risk. A bunch of folks on this very board predicted Paul Wilson going down. Ditto for Eric Milton's awful 2005 due to an incorporation of objective data (past performance history, knee injury that will never heal).

Thats is the complete opposite of what I said. The fact is that if you want to go by that, then you are assuming that only players with injury history, or bad mechanics will be injured. Simply not true, and by your own admission I'm sure, not predictable. Those are two cases in which injury was evident in their history. How about Griffey before he came to Cincinnati? Did people predict his injuries too?


But the problem with too many conclusions derived from observation is that they simply doesn't hold up to the validity litmus test. One of the reasons that's so is the random nature of the game noted by the thread starter. Catch a player during a random cold streak and it may make one think that said player isn't very good when the opposite is true. Seeing something and knowing whether or not you're seeing truth is the primary difference between good scouts and bad scouts.

To me, you are arguing the exact opposite of what I am. If you read my other posts on this thread, you see that I say there is a balance that has to be made. You say they are both important, but you're arguing the scouting aspect of it. Scouting is hit or miss, but then again looking at a players career stats in high school and college is very iffy at best. That is where scouting is the most important along with in the latin countries. Minors is more of a mix. Stats are more useful because of the familiarity of the level of play in the minors. In the big leagues, scouting almost becomes a non issue because the players can be watched on tape extensively. Stats are usefull, as long as it is not from a small sample size. But again, predicting the future is ALWAYS just an educated guess at best.



The key is balancing the two and I'd suggest that the folks you consider to be unbalanced are a heck of a lot more balanced than you think. You might say "too much stats". I might say "not enough information". The validation is in who's right more often. And I don't particularly care how someone gets to "more often right" as long as they get there.[QUOTE]

To argue that point, you must first establish who I am talking about. By your own admission you do not know, and for good reason. I have not specifically pointed to anyone, or any group as being what I call "so called Money Ballers." I know that real Moneyballers have a balance of stats and scouting. That is why it has worked for a short time in Oakland. But the funny thing is that even Moneyball doesn't make you win every year. Why? Because Moneyball is not the eternal answer to the question of how do small market teams compete with the large market teams. In reality, there is no real answer, just theories like taking the Moneyball approach.

I don't even know how this conversation got to this point. I was trying to show that statistics in baseball are crude and were created to track a player's career, not predict it. Yeah sometimes you can use them to guess, and sometimes you will be right, but that also means you will be wrong a great deal of times too. There is a lot more unpredictability in baseball than a lot of people would like to think. Statistics are a way to TRY and predict the future with some sort of accuracy, if used correctly, and used in large sample sizes. But with the factors that aren't considered in the basic baseball statistics the problems arise when somebody tries to show that a second year major leaguer will have a hall of fame career because of his home run rate in those two seasons. Those are the people I have a problem with, and yes they do exist.

[QUOTE]And you're right that statistics cannot be used to predict what is to happen with 100% accuracy. But that's another argument no one has ever made. Instead, we're talking about probability and really nothing but probability. Knowing that, we also know that the most valid information does the best job of projecting what is most likely to happen in the future.

Really?


Originally Posted by SeeinRed
Looking at a lot of statistics may say that Player A has a better chance of producing more runs in the future than Player B, but that is a guess at best...

No, it's really not

My argument was that it was a guess. Your's is that it isn't from what I can tell. Even if it is an educated guess, it is still a guess.

Look I don't want to turn this into personal attacks, and I hope you haven't taken it that way. I think that there is just an honest misunderstanding between us. I'm starting to think we might be arguing the same point. Reguardless, no matter how much we argue about it, we will both feel the same about it. I respect your opinion, and it could be better than mine, I don't really know. Its about time to let this thread get back to its original topic and leave my opinion out of it.

dabvu2498
06-02-2006, 01:19 PM
As well it should be. I heard James on XM months ago talking about how much he had learned from the Red Sox scouts since he's been there.

Balance between the disciplines. That's the optimum baseball mind, IMO.
Include talent: recognizing, acquiring, and developing and you've got it.

SteelSD
06-02-2006, 06:01 PM
Thats is the complete opposite of what I said. The fact is that if you want to go by that, then you are assuming that only players with injury history, or bad mechanics will be injured. Simply not true, and by your own admission I'm sure, not predictable. Those are two cases in which injury was evident in their history. How about Griffey before he came to Cincinnati? Did people predict his injuries too?

Nope. That's not at all what I'm "assuming" because I assume nothing. I play in the realm of probability. Of course some event can come into play to alter what is likely the most probable future, but we can use history and information to help us minimize the risk of those kind of things happening.


To me, you are arguing the exact opposite of what I am. If you read my other posts on this thread, you see that I say there is a balance that has to be made. You say they are both important, but you're arguing the scouting aspect of it. Scouting is hit or miss, but then again looking at a players career stats in high school and college is very iffy at best. That is where scouting is the most important along with in the latin countries. Minors is more of a mix. Stats are more useful because of the familiarity of the level of play in the minors. In the big leagues, scouting almost becomes a non issue because the players can be watched on tape extensively. Stats are usefull, as long as it is not from a small sample size. But again, predicting the future is ALWAYS just an educated guess at best.[QUOTE]

That's a pretty reasonable take except for the fact that I've never argued that scouting isn't necessary or even crucial to the success of a baseball franchise.

[QUOTE]To argue that point, you must first establish who I am talking about. By your own admission you do not know, and for good reason. I have not specifically pointed to anyone, or any group as being what I call "so called Money Ballers."

Yeah, you actually did establish a "group". According to your post they're folks who "...say that stats are all that matter." The problem is you can't actually find one of those people.


I know that real Moneyballers have a balance of stats and scouting. That is why it has worked for a short time in Oakland. But the funny thing is that even Moneyball doesn't make you win every year. Why? Because Moneyball is not the eternal answer to the question of how do small market teams compete with the large market teams. In reality, there is no real answer, just theories like taking the Moneyball approach.

Yeah, there are real answers. Atlanta, Oakland, Boston, New York, and every other long-term consistent winning team have them.


I don't even know how this conversation got to this point. I was trying to show that statistics in baseball are crude and were created to track a player's career, not predict it.

Unfortunately, that's not a case you'll be able to make because a player's past performance is an exceptionally good indicator of how he's most likely to perform in the future.


Yeah sometimes you can use them to guess, and sometimes you will be right, but that also means you will be wrong a great deal of times too.

Everyone is wrong sometime. Difference is, if you're a skilled analyst who uses relevant valid information you'll be right far more than someone who isn't and doesn't.


There is a lot more unpredictability in baseball than a lot of people would like to think. Statistics are a way to TRY and predict the future with some sort of accuracy, if used correctly, and used in large sample sizes. But with the factors that aren't considered in the basic baseball statistics the problems arise when somebody tries to show that a second year major leaguer will have a hall of fame career because of his home run rate in those two seasons. Those are the people I have a problem with, and yes they do exist.

Then all I ask is that you find me someone who's been using two-year HR rate sample sizes as their sole basis for predicting a HOF career. If you're able to find a person who matches that criteria, then you have an issue with their inability to properly analyze and project. But that's not a data issue. That's a "them" issue. Has nothing to do with the validity of statistical analysis.


Really?

Yes, really.


My argument was that it was a guess. Your's is that it isn't from what I can tell. Even if it is an educated guess, it is still a guess.

With all due respect, I don't throw darts at a board.


Look I don't want to turn this into personal attacks, and I hope you haven't taken it that way. I think that there is just an honest misunderstanding between us. I'm starting to think we might be arguing the same point. Reguardless, no matter how much we argue about it, we will both feel the same about it. I respect your opinion, and it could be better than mine, I don't really know. Its about time to let this thread get back to its original topic and leave my opinion out of it.

Of course I'm not taking anything you're typing as a "personal attack". It's just a discussion. And I think we're close on a number of topics actually. But as is often the case we have two people who disagree as to how beneficial statistical analysis can be when used to project what's most likely to happen in the future. I've used it and seen it used with great success. Maybe you haven't, but teams like Boston, Oakland, and St. Louis (just to name a few) are using high-level valid statistical analytics to unbalance the probability scale in their favor. And those analytics aren't a small aside to scouting.

And just so we're clear, I'm not now nor ever have I been "anti-scouting". I am, however, "anti-bad scouting". But then, I'm "anti-bad data" in any form.

SeeinRed
06-02-2006, 07:31 PM
Nope. That's not at all what I'm "assuming" because I assume nothing. I play in the realm of probability. Of course some event can come into play to alter what is likely the most probable future, but we can use history and information to help us minimize the risk of those kind of things happening.

[QUOTE]To me, you are arguing the exact opposite of what I am. If you read my other posts on this thread, you see that I say there is a balance that has to be made. You say they are both important, but you're arguing the scouting aspect of it. Scouting is hit or miss, but then again looking at a players career stats in high school and college is very iffy at best. That is where scouting is the most important along with in the latin countries. Minors is more of a mix. Stats are more useful because of the familiarity of the level of play in the minors. In the big leagues, scouting almost becomes a non issue because the players can be watched on tape extensively. Stats are usefull, as long as it is not from a small sample size. But again, predicting the future is ALWAYS just an educated guess at best.[QUOTE]

That's a pretty reasonable take except for the fact that I've never argued that scouting isn't necessary or even crucial to the success of a baseball franchise.



Yeah, you actually did establish a "group". According to your post they're folks who "...say that stats are all that matter." The problem is you can't actually find one of those people.



Yeah, there are real answers. Atlanta, Oakland, Boston, New York, and every other long-term consistent winning team have them.



Unfortunately, that's not a case you'll be able to make because a player's past performance is an exceptionally good indicator of how he's most likely to perform in the future.



Everyone is wrong sometime. Difference is, if you're a skilled analyst who uses relevant valid information you'll be right far more than someone who isn't and doesn't.



Then all I ask is that you find me someone who's been using two-year HR rate sample sizes as their sole basis for predicting a HOF career. If you're able to find a person who matches that criteria, then you have an issue with their inability to properly analyze and project. But that's not a data issue. That's a "them" issue. Has nothing to do with the validity of statistical analysis.



Yes, really.



With all due respect, I don't throw darts at a board.



Of course I'm not taking anything you're typing as a "personal attack". It's just a discussion. And I think we're close on a number of topics actually. But as is often the case we have two people who disagree as to how beneficial statistical analysis can be when used to project what's most likely to happen in the future. I've used it and seen it used with great success. Maybe you haven't, but teams like Boston, Oakland, and St. Louis (just to name a few) are using high-level valid statistical analytics to unbalance the probability scale in their favor. And those analytics aren't a small aside to scouting.

And just so we're clear, I'm not now nor ever have I been "anti-scouting". I am, however, "anti-bad scouting". But then, I'm "anti-bad data" in any form.

Look, I don't like to say I have a problem with a post, but in this case I very much have a problem. You are accusing me of stating an opinion not supported by facts, yet you try to disprove my opinion with opinions of your own. Instead of asking me to present proof, maybe you should provide some of your own. I have never attacked a particular team, owner, or GM. So providing proof in that reguard is not my intrest. This is about people outside of baseball. The average person who looks at stats in a way that they were not meant to be used, and say they understand the "Moneyball" philosophy.


But since you want to go with that: New York's success has more to do with the fact that they will pick up a player in his prime and use him until they find a better player. Boston is a large market team also. The Moneyball part of baseball exists because of these small market teams like Oakland, and mid level like St. Louis. But having success for a few years does not prove that Moneyball works. Moneyball has not been around long enough to prove it does or doesn't work. You can't just pull teams like New York and Boston and throw them at me and say look at what they did, they are using Moneyball. Its not the same. Boston uses the SABR method, yeah, but New York uses brute force. I don't see anywhere where you have shown me that the basic statistics, i.e. Batting Average, can tell you with even 75% certainty what will happen in a particular at bat, a particular season, and over a number of seasons. That is the point. Just because you study something before you guess, it doesn't make it anymore than a guess, albeit an educated one. The future is ALWAYS a guess. Show me that it isn't, and I'll concede the argument. If I told you I was going to eat eggs for breakfast, that doesn't mean the with 100 percent certainty I will, even though I have eaten them 90% of the time for breakfast in the last 2 years. Too many factors could change so that outcome is not reached. I could wake up late and not have time for eggs, I could not have eggs in my refrigerator, the electricity could be off, and so on. Therefore, if I tell you I am going to eat eggs for breakfast, that too is just an educated guess.

Now, this is the end of it for me. I am not going to keep arguing opinions. This has gotten rediculous. Any time you attack a person's opinion, it is a personal attack. We are both guilty of that. For that, I apologize.

But, why you are so bent on attacking me, I don't know when there are other posters on here who say the same thing I have been saying. Yeah, I got carried away too, and I admidt that, but I tried to use some class and end this because it is taking away from the original point of this thread, which was a very good one.

Highlifeman21
06-02-2006, 07:37 PM
TANGENT ALERT



My question, to those here who have more knowledge about statistics than I do and have time on their hands, is...

I know there is a way to determine whether or not a given sample size is large enough to provide any quality information. But I think I fell asleep that day in Stats class.

Would anyone care to give a short refresher on what that is? (I think how it works might beyond a message board in complexity). ;)

Ask and you shall receive.

IIRC...

Four questions must be answered to determine the sample size:

1. Best estimate of the population size: You do not need to know the exact size of the population. Simply make your best estimate. An inaccurate population size will not seriously affect the formula computations. If the population is very large, this item may be left blank.

2. Best estimate of the rate in the population (%): Make your best estimate of what the actual percent of the survey characteristic is. This is based on the null hypothesis. For example, if the null hypothesis is "blondes don't have more fun", then what is your best estimate of the percent of blondes that do have more fun? If you simply do not know, then enter 50 (for fifty percent).

3. Maximum acceptable difference (%): This is the maximum percent difference that you are willing to accept between the true population rate and the sample rate. Typically, in social science research, you would be willing to accept a difference of 5 percent. That is, if your survey finds that 25 percent of the sample has a certain characteristic, the actual rate in the population may be between 20 and 30 percent.

4. Desired confidence level (%): How confident must you be that the true population rate falls within the acceptable difference (specified in the previous question)? This is the same as the confidence that you want to have in your findings. If you want 95 percent confidence (typical for social science research), you should enter 95. This means that if you took a hundred samples from the population, five of those samples would have a rate that exceeded the difference you specified in the previous question.

All of this is if you're looking for the smallest sample size for your desired percentage.

If you're going to use standard deviation as the model, then you have a different set of rules.

Three questions must be answered to determine the sample size:

1. Standard deviation of the population: It is rare that a researcher knows the exact standard deviation of the population. Typically, the standard deviation of the population is estimated a) from the results of a previous survey, b) from a pilot study, c) from secondary data, or d) or the judgment of the researcher.

2. Maximum acceptable difference: This is the maximum amount of error that you are willing to accept. That is, it is the maximum difference that the sample mean can deviate from the true population mean before you call the difference significant.

3. Desired confidence level (%): The confidence level is your level of certainty that the sample mean does not differ from the true population mean by more than the maximum acceptable difference. Typically, social science research uses a 95% confidence level.

Hopefully this helps

GullyFoyle
06-02-2006, 08:17 PM
Ask and you shall receive.

Hopefully this helps

Thanks!

Obviously its complicated, but it seems like a basic need is to have some idea of where the information is going to go in order to know how much information is required to get there.

I assume this is fairly easy in most baseball calculations because we know a lot of basic ranges (number of games a year, typical number of bats a season, hitting rates between .100 and .450, etc).

I'm surprised their isn't an established default number at bats used to judge weather any batter/pitcher sample is of appropriate size. Of course it wouldn't be perfect, but there are other numbers (e.g. Mendoza Line .200) that establish basic levels of need in baseball. Something like the "20/50/100" meaning you need at least 20 at bats to get anything better than a guess, 50 to get a decent read and 100 to really judge how someone is doing (of course i just made those numbers up).

GullyFoyle
06-02-2006, 08:42 PM
Not to necessarily keep the argument going between SeeinRed and Steel but some of it just seems to be semantics. I think the discussion comes down to what a guess is or isn't. There is a big difference between a guess and statistical prediction, and there are some predictions that can be more than 75% correct


For an obvious example... you could take my batting average after 100 at bats and then be right more than 75% of the time about my outcomes at the plate against a major league pitcher :) . Now SeeinRed is talking about major league players, but different statistics can be 99% correct (usually not in baseball) or they can be worthless depending on sample sizes, context and the accuracy needed (is there any other sport that goes to three points after the decimal!).

I would also disagree with the term "MoneyBall" used as a synonym for people who use statistics. It is now commonly used as a derogative term (which I think was just used off the cuff in this instance) and this might have started the whole problem. If your going to be accurate "Moneyball" was about finding discrepancies in the pricing of player talents in relation to the economic market. Stats where just a tool used. In theory you could write the same book but with a set of scouts that have a unique way of evaluating talent.

Anyway, I don't mean to stoke the fires. Just thought there was a little cross talking confusion.

Edit: Guess - v. def: suppose something without sufficient information to be sure of being correct... I think there is disagreement over what qualifies as being with or without sufficient information

Newman4
06-02-2006, 10:04 PM
Steel, I believe SeeInRed is correct about no matter how highly correlated certain statistics are to future performance, there is a margin of error. Nothing is a perfect correlation. Don't get me wrong though, I have learned from Redszone that certain statstics such as WHIP are very good indicators of pitching efficiency. But, nothing's 100% reliable.

Statistics and scouting are used differently at different levels as well. When players reach MLB there's quite a bit of numbers to look at and SABR methods are obviously probably more important in judging a player as a whole. However, advance scouting as mentioned, can detect ineffiencies in certain aspects of each player's game. For a rudimentary example, Player A can't hit the curve ball or Player B tends to always throw a first pitch fastball to get ahead. Thus, your mixed approach of SABR and scouting is not only valid but unavoidable. In judging draftees however, particularly high school players, often what numbers are available are flawed. For instance, how many "hits" did we all get in high school that were actually errors or just because of sub par athletes on the other team. Therefore, what a pitcher throws on the gun or how well they do in all star games in AAU, Travel Leagues, etc. weigh very heavy on predicting future success.

GullyFoyle
06-02-2006, 10:13 PM
In judging draftees however, particularly high school players, often what numbers are available are flawed. For instance, how many "hits" did we all get in high school that were actually errors or just because of sub par athletes on the other team. Therefore, what a pitcher throws on the gun or how well they do in all star games in AAU, Travel Leagues, etc. weigh very heavy on predicting future success.

I would only note that scouting is very bad at judging high school prospects too. That age (not fully developed physically or trained well) is always difficult to judge.

GullyFoyle
06-02-2006, 10:21 PM
no matter how highly correlated certain statistics are to future performance, there is a margin of error. Nothing is a perfect correlation. Don't get me wrong though, I have learned from Redszone that certain statstics such as WHIP are very good indicators of pitching efficiency. But, nothing's 100% reliable.

I'm not sure I agree with absoluteness of the sentiment here. Sometimes the margin of error is so small that it might as well be perfect, and a lot of what we consider perfect from day to day is that way.(warning: overly simplistic example ahead) I can prediction that Mount Rushmore will be here tomorrow based on statistics of past monuments being destroyed in various ways for various purposes... and though it is not 100% guaranteed, most people would consider the chance that it wont be to small to act on, so for most purposes the prediction is perfect. Again, it would depend on sample sizes, context and the size error acceptable in the situation.

So saying nothing is perfect is true, but saying the statistics can't correlate enough to be considered perfect, I think, is false.

Newman4
06-02-2006, 10:50 PM
Sometimes the margin of error is so small that it might as well be perfect, and a lot of what we consider perfect from day to day is that way.

Give me a baseball example of a stat that has that small of a margin of error.

SteelSD
06-03-2006, 12:13 AM
Steel, I believe SeeInRed is correct about no matter how highly correlated certain statistics are to future performance, there is a margin of error. Nothing is a perfect correlation. Don't get me wrong though, I have learned from Redszone that certain statstics such as WHIP are very good indicators of pitching efficiency. But, nothing's 100% reliable.

Of course he's correct about that because it's an argument that's never been made. No one, to this date, has ever stated that statistical analysis can produce an error rate of 0%. An error rate of 0% is impossible and, as we both know, arguing that the opposition has a non-existent impossible position is strawman building.

SteelSD
06-03-2006, 12:50 AM
Look, I don't like to say I have a problem with a post, but in this case I very much have a problem. You are accusing me of stating an opinion not supported by facts, yet you try to disprove my opinion with opinions of your own. Instead of asking me to present proof, maybe you should provide some of your own.

At this point, I honestly don't have a clue as to what you're talking about. You've claimed there exists all these folks who say that stats are all that matter. Yet, you can't find a soul who actually believes that.


I have never attacked a particular team, owner, or GM. So providing proof in that reguard is not my intrest. This is about people outside of baseball. The average person who looks at stats in a way that they were not meant to be used, and say they understand the "Moneyball" philosophy.

Then those people are bad. That being said, I've seen no indication that you understand what was written in "Moneyball"- particularly considering you've positioned "Moneyballers" as caring about nothing but statistics.


But since you want to go with that: New York's success has more to do with the fact that they will pick up a player in his prime and use him until they find a better player. Boston is a large market team also. The Moneyball part of baseball exists because of these small market teams like Oakland, and mid level like St. Louis. But having success for a few years does not prove that Moneyball works. Moneyball has not been around long enough to prove it does or doesn't work. You can't just pull teams like New York and Boston and throw them at me and say look at what they did, they are using Moneyball. Its not the same. Boston uses the SABR method, yeah, but New York uses brute force. I don't see anywhere where you have shown me that the basic statistics, i.e. Batting Average, can tell you with even 75% certainty what will happen in a particular at bat, a particular season, and over a number of seasons.

And here I was talking about the incorporation of statistical analysis and you keep bringing it back to "Moneyball". And your claims that I can't give you better than 75% accuracy as to what's most likely to happen are unfounded considering that I've been one of the guys telling folks what was most likely to happen for some time now- and with a great deal of accuracy. I projected the Reds to win 78 games last year based on their projected Run Differential. They won 73 games. Even factoring out the Pythag projection (74 Wins), I was 94% accurate. Factoring in the Run Differntial, I was 95% accurate. I, among others do that a lot. A couple excellent posters noted that the Reds hot start in 2004 couldn't be maintained and they were ridiculed by poster who didn't understand probability. And good gosh, they were right.

And there isn't a creature on the planet who can do that from an AB-by-AB perspective. All they can do is maximize probabilty and matchups (why do you think Managers us effective LHP versus LH Hitters?).

And if you're looking for Batting Average to tell you anything, then you're looking at bad data that doesn't correlate with much of anything. There is a virtual two tons of better data out there.


That is the point. Just because you study something before you guess, it doesn't make it anymore than a guess, albeit an educated one. The future is ALWAYS a guess. Show me that it isn't, and I'll concede the argument. If I told you I was going to eat eggs for breakfast, that doesn't mean the with 100 percent certainty I will, even though I have eaten them 90% of the time for breakfast in the last 2 years. Too many factors could change so that outcome is not reached. I could wake up late and not have time for eggs, I could not have eggs in my refrigerator, the electricity could be off, and so on. Therefore, if I tell you I am going to eat eggs for breakfast, that too is just an educated guess.

Again, the best analysts in the game don't throw darts at a board. Your scenario really has no relevance because it's you choice as to what you eat for breakfast. If you project that you'll eat eggs, but do not then you're choosing to affect the outcome.


Now, this is the end of it for me. I am not going to keep arguing opinions. This has gotten rediculous. Any time you attack a person's opinion, it is a personal attack. We are both guilty of that. For that, I apologize.

No. When an opinion is challenged, that does not equate to a "personal attack". If you could provide relevant supporting evidence then I'd be happy to hear it.


But, why you are so bent on attacking me, I don't know when there are other posters on here who say the same thing I have been saying. Yeah, I got carried away too, and I admidt that, but I tried to use some class and end this because it is taking away from the original point of this thread, which was a very good one.

I haven't "attacked" you once during this discussion. I have, however, challenged your opinion. You have the right to have an opinion. You do not have the right to an unchallenged opinion and I would appreciate it if you would cease your effort to act as if you've been somehow wounded by logical counterpoints to your opinion. You've made a number of unsupportable claims and have positioned a number of strawmen thusfar. All I've asked is that we weed those out before continuing.

SteelSD
06-03-2006, 12:51 AM
Give me a baseball example of a stat that has that small of a margin of error.

OPS, Runs Created, Run Differential. And that's without delving into Linear Weights.

GullyFoyle
06-03-2006, 06:09 PM
Give me a baseball example of a stat that has that small of a margin of error.

I can't give you an example of a stat that is always that way, but I defer to Steel in his choices.

But here is my reasoning for thinking its possible.

If I understand you correctly, you are saying that baseball statistics can never reach a degree of precision that we could consider it "perfect" (not 100% but close enough for us to consider it so).

If we take a very extreme case.. someone who doesn't play baseball batting against a major league pitcher... the statistics for how they fair is probably going to be very accurate, accurate enough to determine their rate of failure to 99%.

Can a stat be this accurate in a realistic scenario?... I don't know, but for me, it seems well within the range of possibilty (I'm not a big stats person, though I'm familiar with them and I find the concepts intriguing). Like quantum physics a lot of what statistics can explain can be counter intuitive (Hey, back full circle to the reason for this thread... randomness is often counter intuitive :beerme: )

Anyway, thats why I think Steel is probably right, but your milage might vary. :)

SeeinRed
06-03-2006, 08:22 PM
I can't give you an example of a stat that is always that way, but I defer to Steel in his choices.

But here is my reasoning for thinking its possible.

If I understand you correctly, you are saying that baseball statistics can never reach a degree of precision that we could consider it "perfect" (not 100% but close enough for us to consider it so).

If we take a very extreme case.. someone who doesn't play baseball batting against a major league pitcher... the statistics for how they fair is probably going to be very accurate, accurate enough to determine their rate of failure to 99%.

Can a stat be this accurate in a realistic scenario?... I don't know, but for me, it seems well within the range of possibilty (I'm not a big stats person, though I'm familiar with them and I find the concepts intriguing). Like quantum physics a lot of what statistics can explain can be counter intuitive (Hey, back full circle to the reason for this thread... randomness is often counter intuitive :beerme: )

Anyway, thats why I think Steel is probably right, but your milage might vary. :)


Now, this seems like a rather obvious statement, but players stats change throughout their carrer. To me anyway, it seems inconcievable to project a player's career by using a statistic that changes throughout that same player's career. Then you throw in factors like the park they play in, injuries, and the teams that player faces, it just wouldn't work in a real world situation. Randomness is one thing, but changing values are another. Once again, I say that stats track a player's career, not predict it.

Mathematically speaking, as a general rule, any time there is a margin of error over a tenth of a percent, the outcome is discarded, meaning it is not considered reliable enough in the terms of statistical analysis. That changes from situation to situation, but I am not going out on a limb at all to say that baseball statistics fail in that reguard. How do you get a margin of error to become that small? By considering all possible outcomes, scenarios, and the factors that could change that outcome. Obviously, this is near impossible to do in baseball situations. Maybe I'm missing the point of all of this thread, but it originally showed that randomness can be explained in mathematical terms. (i.e. going hit-less for X amout of consecutive at bats) But this only works with cut and dry experiments like flipping a coin that have small margins of error. However, say you put a weight on one side of the coin. Does that not mean that all of the stats from the original non-weighted coin are irrelavent when trying to predict the outcome of a series of flips of the weighted coin? If you change a factor, you change the outcome, and using the old stats won't help you predict that outcome.

Once again, I argue that just the fact that we are trying to predict the outcome of an event that involves humans makes it that much more difficult. Even the changes in a player's mental approach can change the physical outcome. You eliminate the human error, then yeah, you can make the error margin small. Considering that you can't, your margin of error is going to be large. That is why there are so many failed trades in baseball. There are also trades that are sucessful from a teams standpoint because the player plays to or above expectations. But if your expectations are lower than what that player achieves, then it is still a failed guess. I think that is where a lot of people have a problem with this. They look at a player that out performs expectations and they think it is a sucess. It is in baseball terms, but not in mathematical terms.

I think the rift in this thread is the differing views in stats from a baseball point of view as compared to mathematical point of view. Mathematically speaking, there is no stat in baseball with a small margin of error. Realitively speaking in baseball terms, there are stats with smaller margins of error than others. But remember, not even the most SABR friendly player is a sure thing. Luck is still involved. I guess what I'm trying to show is that it is mathematically to prove events in the real world if a. there is a large sample size, and b. ALL of the factors that change that outcome are considered in the equation. The best example I can think of is the group of college students who used an elaborate system to win big in blackjack at a casino in Las Vegas. Can't recall where the students were from, but they made a team, played blackjack, and won big after they got big enough sample sizes to get an equation to work. That is a real world application to statistics, but the factors are limited in that the cards are random, and human's control of the outcome is limited. Is that possible in baseball? Not in my opinion.

The point is that if you are looking to predict the future of baseball players, stats are not the answer. Neither is scouting. Both only tell you what might happen, not what will happen. You need a crystal ball to say that you can predict the future, and even then it is questionable.

Newman4
06-03-2006, 09:58 PM
OPS, Runs Created, Run Differential. And that's without delving into Linear Weights.
Gully stated that some stats have so little margin error that they can be considered "perfect". You apparently agree. What exactly are the stats you mentioned "perfect" to predict? (realistic example, not if a 4 year old batted against Clemens or something like that)

SteelSD
06-04-2006, 01:53 AM
Gully stated that some stats have so little margin error that they can be considered "perfect". You apparently agree. What exactly are the stats you mentioned "perfect" to predict? (realistic example, not if a 4 year old batted against Clemens or something like that)

Gully is both right and wrong. No metric is a "perfect" predictor of performance.

But that's not what you asked for. You asked for examples of metrics that produced low error rates; which is what I offered you.

And let's be clear- no one should be offering up a metric that can allegedly predict the future with 100% accuracy. I've stated that just a few posts down the thread so I've little idea why you'd think I'd "apparantly agree" to the concept of a "perfect" success rate.

However, the presence of an error rate doesn't invalidate the proper use of metrics when making a projection. That doesn't mean someone will be right all the time, but when utilized properly and within the right context, proper use of performance history can make folks more right more often and that's really what we're looking for (probability gain).

I gave an example in a previous post of a projection I made about the Reds record in 2005. Here's another documented projection from 4/10/06 regarding Tigers CF Curtis Granderson:

<Begin excerpt>

I always hate to project a BA that high for a first full year (.290), but I think you're pretty close. Not sure what PECOTA says, but I'd go .275 BA/.345 OBP/.485 SLG for this year (similar to Grady Sizemore's 2005 numbers). He won't get as many SB as a guy like Mike Cameron (during his prime), so I think 20 SB is a good number. And because he's hitting leadoff and will be benefitting from increased PA, I expect his HR total to be closer to 25 than 20.

In ST, I really had hoped that Jim Leyland would fall in love with Nook Logan- which would have made Granderson available possibly. But Leyland figured out that Granderson was the better option. I have enough confidence in the guy that I took him with my first pick in my keeper league (which is closer to an 8th round selection as we each kept 7 players from last year).

Long-term, I think Granderson will easily raise his Isolated Discipline numbers to 80 points and a .100 IsoD isn't out of the question. If that happens, you've got a consistent .280 BA/.380 OBP/.500 SLG Centerfielder with 30/30 potential.

<End excerpt>

Currently, Granderson is posting numbers of .279 BA/.373 OBP/.473 SLG. That's an OPS of .846 versus my .830 OPS projection. As of right now, I'm better than 98% accurate and I don't see any reason to expect much less than that over the rest of the season. And if Granderson continues on his current path, you know what we're seeing? A .280 BA/.380 OBP/.500 SLG Centerfielder with 30/30 potential.

And how did I come up with my projections? Past performance coupled with a contextual understanding of how that relates to the present and future.

Newman4
06-04-2006, 07:49 AM
And let's be clear- no one should be offering up a metric that can allegedly predict the future with 100% accuracy.

That's what I needed to hear. :)

rdiersin
06-04-2006, 11:40 AM
Give me a baseball example of a stat that has that small of a margin of error.

Well, lets begin with RC and XR. Those stats are estimates of runs scored. Both the correlation coefficient and properties of the error, like the variance, can be determined. For completeness, I'll include the formulas that I'll use here

XR=.5*(H-DB-TR-HR)+.72*DB+1.04*TR+1.44*HR
+.34*(HBP+BB-IBB)+.25*IBB+.18*SB-.32*CS
-.09*(AB-H-K)-.098*K-.37*GIDP+.37*SF+.04*SH

RC=(H + BB - CS + HBP - GIDP)(TB + (.26 * (BB - IBB + HBP)) + (.52 * (SH + SF + SB)) )/(AB + BB + HBP + SH + SF).

So, consider now Runs Created (RC). Using data from retrosheet.org for the 1960-2005 seasons we obtain this histogram of the error (e=RS-RC).

http://www.nd.edu/~rdiersin/rc_hist.jpg


We can see that it is definitely Gaussian in character and furthermore the standard deviation is about 24.7. Relatively, that is a fairly small error, that is about 85% of the estimated runs are withing 25 of the actuial runs scored for a team over the course of a season. Furthermore we can see that the correlation is also good from the following scatter plot.

http://www.nd.edu/~rdiersin/rc_scatter.jpg

Alright now on to XR, just for fun. We will see a similiar standard deviation and Gaussian character.

http://www.nd.edu/~rdiersin/xr_hist.jpg

also we can see the correlation of XR and runs scored for a team in a season.

http://www.nd.edu/~rdiersin/xr_scatter.jpg

So, in summary the error can be examined and its not bad. For more fun lets look at the good old Pythagoream projection for teams from 1960-2005. To do this we'll examine the estimated winning percentage.
W%=RS^1.83/(RS^1.83+RA^1.83)

This is what we will see
http://www.nd.edu/~rdiersin/pyth_hist.jpg

and

http://www.nd.edu/~rdiersin/pyth_scatter.jpg

pedro
06-04-2006, 11:56 AM
Seeinred, not sure if you know anything about Basball Prospectus' Pecota System which is a a compuetr software program a nd database which they use to calculate predictions for players based on past performance, age, and historical data. It pretty interesting stuff. The part of it that speaks to your issue with margins of errors is that they give a variance of possible outcomes (over an entire seasons stats) with percentile chances of that outcome being true. By doing this they are accepting that a large range of outcomes are possible, but the likelihood of each potential outcome is weighted. They've really had a lot of success with it to date.

I don't think these links are premium content so I believe you should be able to see the articles.

http://www.baseballprospectus.com/article.php?articleid=2515

http://www.baseballprospectus.com/glossary/index.php?context=6&category=true

RedsManRick
06-04-2006, 12:50 PM
See, how much have you read about or looked in to PECOTA. It really seems to address a lot of your concerns. The basic idea is to find the players who are statistically similar to the player you're analyzing and use their careers as a baseline for your player's projection.

It then adjusts for numerous things such as era, injury, ballparks, etc. Then, it provides a series of projections from the very low end (the 10% projection) the "average" (50%) projection, and the high end.

In there in lies the secret with projections. Projections are predictions are very different things. A prediction makes the claim that based on our knowledge, X will happen. The Reds will go 84-78. Projections establish multiple sets of assumptions and seeing some likely outomces given those. Nobody, and no stat(s), can predict the future perfectly. However, we can get an idea about what is likely to happen. Baseball decisions are based around this likelihood. We don't know if ARod is going to hit 30 homers or 50. However, we know that he's VERY likely to hit somewhere betweeen 30 and 50. It's all shades of gray between no clue and perfect prediction.

Newman4
06-04-2006, 01:20 PM
Well, lets begin with RC and XR. Those stats are estimates of runs scored. Both the correlation coefficient and properties of the error, like the variance, can be determined. For completeness, I'll include the formulas that I'll use here

XR=.5*(H-DB-TR-HR)+.72*DB+1.04*TR+1.44*HR
+.34*(HBP+BB-IBB)+.25*IBB+.18*SB-.32*CS
-.09*(AB-H-K)-.098*K-.37*GIDP+.37*SF+.04*SH

RC=(H + BB - CS + HBP - GIDP)(TB + (.26 * (BB - IBB + HBP)) + (.52 * (SH + SF + SB)) )/(AB + BB + HBP + SH + SF).

So, consider now Runs Created (RC). Using data from retrosheet.org for the 1960-2005 seasons we obtain this histogram of the error (e=RS-RC).

http://www.nd.edu/~rdiersin/rc_hist.jpg


We can see that it is definitely Gaussian in character and furthermore the standard deviation is about 24.7. Relatively, that is a fairly small error, that is about 85% of the estimated runs are withing 25 of the actuial runs scored for a team over the course of a season. Furthermore we can see that the correlation is also good from the following scatter plot.

http://www.nd.edu/~rdiersin/rc_scatter.jpg

Alright now on to XR, just for fun. We will see a similiar standard deviation and Gaussian character.

http://www.nd.edu/~rdiersin/xr_hist.jpg

also we can see the correlation of XR and runs scored for a team in a season.

http://www.nd.edu/~rdiersin/xr_scatter.jpg

So, in summary the error can be examined and its not bad. For more fun lets look at the good old Pythagoream projection for teams from 1960-2005. To do this we'll examine the estimated winning percentage.
W%=RS^1.83/(RS^1.83+RA^1.83)

This is what we will see
http://www.nd.edu/~rdiersin/pyth_hist.jpg

and

http://www.nd.edu/~rdiersin/pyth_scatter.jpg

Rdiersin, his statement was the margin of error is so small that it can be considered "perfect" for predicting the future. While your analysis is certainly impressive, from the histogram you can see that all of the mentioned stats are very close to what is actual. However, there are enough examples of instances when the predicted and actual do not match up to not call the relationship "perfect". Which is my point, no method of prediction is going to be "perfect". The formulas illustrated, of course, are very reliable and tweaked over many years of trials. Thus, are as near to "perfect" as one can determine, but when dealing with a sport like baseball, where variables are constantly changing it's impossible to have a predictory stat that is "perfect".

SteelSD
06-04-2006, 02:00 PM
Rdiersin, his statement was the margin of error is so small that it can be considered "perfect" for predicting the future. While your analysis is certainly impressive, from the histogram you can see that all of the mentioned stats are very close to what is actual. However, there are enough examples of instances when the predicted and actual do not match up to not call the relationship "perfect". Which is my point, no method of prediction is going to be "perfect". The formulas illustrated, of course, are very reliable and tweaked over many years of trials. Thus, are as near to "perfect" as one can determine, but when dealing with a sport like baseball, where variables are constantly changing it's impossible to have a predictory stat that is "perfect".

You're right. But then rdiersin isn't arguing for the existence of a "perfect" predictor and we've already established that Gullyfoyle's interpretation of the word "pefect" was a subjective consideration based on context rather than a black and white dictionary definition that involves the concept of a 0% error rate.

And context is a huge thing when we're talking about accuracy. Are performance metrics- as predictors- accurate at an acceptable DNA-level error rate? No. And that's due to randomness and the variables you've noted. However, within the context of baseball, we've got some highly accurate metrics available with very acceptable correlations and/or failure rates when applied to future performance. RedsManRick even noted PECOTA, which is a marvelous system because it breaks down probability to the point of "more likely" and "less likely" for a player.

The fact that we can't find a 100% accurate predictor doesn't invalidate the use of statistical analytics because a 0% error rate is an unreasonable expectation given the context. Ditto for meteorology and actuarial sciences (and high-level sabermetrics does qualify).

Imperfect does not equal in invalid and we measure accuracy in degrees rather than a black and white "right or wrong".

BTW, you can keep stating that nothing is perfect and you'll be right every time you type it. That being said, no one is arguing that anything IS dictionary-definition "perfect". In fact, anyone who'd state that performance analysis is always 100% right is 100% wrong. Ditto for anyone who'd state that because performance analysis is never 100% right it's 100% wrong. The pendulum swings both ways. But the interesting discussion and the real opportunity for knowledge gain is in the gray area between 100% right and 0% right. That's where the real headway is being made.

ochre
06-04-2006, 02:02 PM
Gully is both right and wrong. No metric is a "perfect" predictor of performance.

But that's not what you asked for. You asked for examples of metrics that produced low error rates; which is what I offered you.

And let's be clear- no one should be offering up a metric that can allegedly predict the future with 100% accuracy. I've stated that just a few posts down the thread so I've little idea why you'd think I'd "apparantly agree" to the concept of a "perfect" success rate.

However, the presence of an error rate doesn't invalidate the proper use of metrics when making a projection. That doesn't mean someone will be right all the time, but when utilized properly and within the right context, proper use of performance history can make folks more right more often and that's really what we're looking for (probability gain).

I gave an example in a previous post of a projection I made about the Reds record in 2005. Here's another documented projection from 4/10/06 regarding Tigers CF Curtis Granderson:

<Begin excerpt>

I always hate to project a BA that high for a first full year (.290), but I think you're pretty close. Not sure what PECOTA says, but I'd go .275 BA/.345 OBP/.485 SLG for this year (similar to Grady Sizemore's 2005 numbers). He won't get as many SB as a guy like Mike Cameron (during his prime), so I think 20 SB is a good number. And because he's hitting leadoff and will be benefitting from increased PA, I expect his HR total to be closer to 25 than 20.

In ST, I really had hoped that Jim Leyland would fall in love with Nook Logan- which would have made Granderson available possibly. But Leyland figured out that Granderson was the better option. I have enough confidence in the guy that I took him with my first pick in my keeper league (which is closer to an 8th round selection as we each kept 7 players from last year).

Long-term, I think Granderson will easily raise his Isolated Discipline numbers to 80 points and a .100 IsoD isn't out of the question. If that happens, you've got a consistent .280 BA/.380 OBP/.500 SLG Centerfielder with 30/30 potential.

<End excerpt>

Currently, Granderson is posting numbers of .279 BA/.373 OBP/.473 SLG. That's an OPS of .846 versus my .830 OPS projection. As of right now, I'm better than 98% accurate and I don't see any reason to expect much less than that over the rest of the season. And if Granderson continues on his current path, you know what we're seeing? A .280 BA/.380 OBP/.500 SLG Centerfielder with 30/30 potential.

And how did I come up with my projections? Past performance coupled with a contextual understanding of how that relates to the present and future.
Does this mean I'm allowed to mention Granderson again?

SteelSD
06-04-2006, 02:06 PM
Does this mean I'm allowed to mention Granderson again?

Yeah. Sorry. I should have PM'd you. ;)

You can talk about Jose Lopez too.

Newman4
06-04-2006, 04:25 PM
You're right. But then rdiersin isn't arguing for the existence of a "perfect" predictor and we've already established that Gullyfoyle's interpretation of the word "pefect" was a subjective consideration based on context rather than a black and white dictionary definition that involves the concept of a 0% error rate.

And context is a huge thing when we're talking about accuracy. Are performance metrics- as predictors- accurate at an acceptable DNA-level error rate? No. And that's due to randomness and the variables you've noted. However, within the context of baseball, we've got some highly accurate metrics available with very acceptable correlations and/or failure rates when applied to future performance. RedsManRick even noted PECOTA, which is a marvelous system because it breaks down probability to the point of "more likely" and "less likely" for a player.

The fact that we can't find a 100% accurate predictor doesn't invalidate the use of statistical analytics because a 0% error rate is an unreasonable expectation given the context. Ditto for meteorology and actuarial sciences (and high-level sabermetrics does qualify).

Imperfect does not equal in invalid and we measure accuracy in degrees rather than a black and white "right or wrong".

BTW, you can keep stating that nothing is perfect and you'll be right every time you type it. That being said, no one is arguing that anything IS dictionary-definition "perfect". In fact, anyone who'd state that performance analysis is always 100% right is 100% wrong. Ditto for anyone who'd state that because performance analysis is never 100% right it's 100% wrong. The pendulum swings both ways. But the interesting discussion and the real opportunity for knowledge gain is in the gray area between 100% right and 0% right. That's where the real headway is being made.

I'll agree with you there. rdiersin certainly does well at providing compelling evidence for the use of the RC, XR and Pytha formulas with the graphics provided. Honestly, the correlations for those three items is stronger than most other stats we use in real life for other purposes (i.e. weather forecasting, insurance premiums, etc.) Nice work.