Best lefty off all time? My vote is Koufax!

[Snowman]

Originally Posted by tschock

To put it another way: If a statistician's model is good at analyzing the past, then it should be reasonably good for predicting the future. Otherwise your model needs adjusting to consider other factors. That didn't seem to play out very well when 'the best team in baseball' this year didn't even get close to winning the World Series (as one example).

At the end of the regular season, every statistical model worth its salt would have said that the Dodgers were the best team in baseball this season. They also would have given the Dodgers a mere 25% chance of winning the world series despite being the best team because there is a tremendous amount of short term luck involved in baseball. This doesn't happen in football, basketball, hockey, or soccer. The best teams in those sports win the championship far more often.

[Snowman]

Let y = 2x + 3

If x = 5, then y = 13

BobC - "Well that's just like, your opinion, man."

[tschock]

Originally Posted by Snowman

At the end of the regular season, every statistical model worth its salt would have said that the Dodgers were the best team in baseball this season. They also would have given the Dodgers a mere 25% chance of winning the world series despite being the best team because there is a tremendous amount of short term luck involved in baseball. This doesn't happen in football, basketball, hockey, or soccer. The best teams in those sports win the championship far more often.

Then I would counter that your statistical model needs tweaking because it's not accounting for all the variables. The Dodgers weren't "The Best Team in Baseball" as they didn't win the championship. Though they may have had the best group of individuals in baseball playing for the same team. There's a difference.

[Snowman]

Originally Posted by tschock

Then I would counter that your statistical model needs tweaking because it's not accounting for all the variables. The Dodgers weren't "The Best Team in Baseball" as they didn't win the championship. Though they may have had the best group of individuals in baseball playing for the same team. There's a difference.

Every regression model has coefficients that estimate the value of impact of each variable in the model along with confidence intervals for those estimates and an error term. That error term encapsulates both things you haven't yet accounted for but otherwise could and random chance. While it is always true that there are probably ways to account for more of the variance in the data by finding new variables to control for, and thereby reducing your error term, it is a mistake to assume that all variance is explainable if only you had built a better model with more explanatory variables. There is always an element of randomness in sports, but baseball in particular is exceptionally subject to randomness, far moreso than other mainstream sports. A slight shift in the breeze could be the difference between a grand slam and a 2nd strike, a "bad hop" is the difference between an out and a hit, the angle of the ball leaning the bat depends on whether it made contact on the seams or the leather, etc. If you set up a 5 gallon bucket at random locations on the infield and told players to hit the bucket while major league pitchers were launching 100 mph fastballs, curveballs, and change-ups at them, even the best hitters in the league would be lucky to hit the bucket 2 times out of 10. They simply don't have that level of control. It's not possible. At best, they could hit the ball within something like 10 feet of the bucket, and that's if the bucket is on the ideal side of the field for them. If they have to push to the opposite field to hit the bucket, it's even harder. A player can control directional accuracy by trying to push or pull the ball or to swing up on it or down, but once they make contact, they have little to no control over whether that ball is hit right at the shortstop or just out of his reach. They just make contact and roll the dice most of the time. The worst teams in the league can beat the best teams on any given night. A bad team often takes 3 games out of 10 against a good team. This doesn't happen in the NFL. The worst team in the league almost never beats the best teams. And the team that wins the super bowl is far more likely to have also been the best team in the regular season in football as well. Same with basketball and hockey. But baseball is different. A 7 game series is simply insufficient for determining which team is better in baseball. Luck plays too big of a role. This is evidenced by the outcomes of past seasons. Any random team can, and often does, win the world series. But that doesn't mean they are the strongest team.

[tschock]

Originally Posted by Snowman

Every regression model has coefficients that estimate the value of impact of each variable in the model along with confidence intervals for those estimates and an error term. That error term encapsulates both things you haven't yet accounted for but otherwise could and random chance. While it is always true that there are probably ways to account for more of the variance in the data by finding new variables to control for, and thereby reducing your error term, it is a mistake to assume that all variance is explainable if only you had built a better model with more explanatory variables.

Those variances can always be determined but not necessarily quantified or built into a model, as you noted. But you seem to dismiss them out of hand. I see at least 2 things that could be problematic with your statistical model for declaring Koufax as the best left handed pitcher of all time. 1 ) You may be using a statistical probability analysis to determine a singularity (ie Koufax being the best). This is hinted at by your constant touting of the success of your other statistical probability models. If that's not the case, then those probability results are irrelevant and don't add any value to your claim for Koufax anyway. 2 ) You constantly ignore other intangible items that, while oft times not quantifiable, are still non-zero. This is demonstrated, in part, by claims such as athletes are better today than they were years ago. While true, you continue to ignore the fact that athletes years ago played and trained under different conditions with what was available at the time. You're implying that had those 'lesser athletes' been born and raised in today's environment, they would not have taken advantage of today's methods and still been throwing rocks at a chalk outline on a barn. Maybe try the reverse? Put Koufax back into the 1930s environment and do the analysis and you'll have different variables to consider. It is possible he wouldn't have made it out of the Dodgers or Cardinals 13 or so minor league teams.

[BobC]

Originally Posted by tschock

Those variances can always be determined but not necessarily quantified or built into a model, as you noted. But you seem to dismiss them out of hand. I see at least 2 things that could be problematic with your statistical model for declaring Koufax as the best left handed pitcher of all time. 1 ) You may be using a statistical probability analysis to determine a singularity (ie Koufax being the best). This is hinted at by your constant touting of the success of your other statistical probability models. If that's not the case, then those probability results are irrelevant and don't add any value to your claim for Koufax anyway. 2 ) You constantly ignore other intangible items that, while oft times not quantifiable, are still non-zero. This is demonstrated, in part, by claims such as athletes are better today than they were years ago. While true, you continue to ignore the fact that athletes years ago played and trained under different conditions with what was available at the time. You're implying that had those 'lesser athletes' been born and raised in today's environment, they would not have taken advantage of today's methods and still been throwing rocks at a chalk outline on a barn. Maybe try the reverse? Put Koufax back into the 1930s environment and do the analysis and you'll have different variables to consider. It is possible he wouldn't have made it out of the Dodgers or Cardinals 13 or so minor league teams.

Taylor, Great points, and part of what I've been saying all along.

[Snowman]

Originally Posted by tschock

I see at least 2 things that could be problematic with your statistical model for declaring Koufax as the best left handed pitcher of all time. 1 ) You may be using a statistical probability analysis to determine a singularity (ie Koufax being the best). This is hinted at by your constant touting of the success of your other statistical probability models. If that's not the case, then those probability results are irrelevant and don't add any value to your claim for Koufax anyway.

You seem to be conflating what others claim I said with what I've actually said. I have never once claimed that Koufax is the GOAT. I just said he's in the conversation. My arguments have been that Spahn shouldn't be in the conversation at all, and that while he was great for his time, I heavily discount Grove's statistics because of the overall skill level of his contemporaries.

The model I described above is not aimed at proving Koufax is the best, or anyone else for that matter. What I described is a tool for measuring the impact that something like a change in mound heights or a widened strike zone has on performance. It can also be used to estimate something like the overall talent level decrease across the league during WW2, and pretty much anything else that you want to understand the impact of.

Then, if you want, you could use the results of that model to build a separate model to more accurately evaluate pitching performances from different eras. It would give you a better metric than WAR.

2 ) You constantly ignore other intangible items that, while oft times not quantifiable, are still non-zero. This is demonstrated, in part, by claims such as athletes are better today than they were years ago. While true, you continue to ignore the fact that athletes years ago played and trained under different conditions with what was available at the time. You're implying that had those 'lesser athletes' been born and raised in today's environment, they would not have taken advantage of today's methods and still been throwing rocks at a chalk outline on a barn. Maybe try the reverse? Put Koufax back into the 1930s environment and do the analysis and you'll have different variables to consider. It is possible he wouldn't have made it out of the Dodgers or Cardinals 13 or so minor league teams.

Yes, humans evolve. Advances in nutrition, kinesiology, sports medicine, and game theory are all part of the evolutionary process. We know more today than we did 100 years ago. We're also 3 inches taller and stronger, on average, today than we were 100 years ago. Should we add a few more inches to Babe Ruth then, and credit him with a few more HRs? Or what about amphetamine usage from before they started testing for it? It was rampant. You can make adjustments for whatever you want to control for. If you want to debate with BobC about how much humans have evolved over the past 100 years, or even whether or not you think it "counts" as evolution to begin with, be my guest. It's just a conversation I'm not interested in having.

[Mark17]

Originally Posted by tschock

Those variances can always be determined but not necessarily quantified or built into a model, as you noted. But you seem to dismiss them out of hand. I see at least 2 things that could be problematic with your statistical model for declaring Koufax as the best left handed pitcher of all time. 1 ) You may be using a statistical probability analysis to determine a singularity (ie Koufax being the best). This is hinted at by your constant touting of the success of your other statistical probability models. If that's not the case, then those probability results are irrelevant and don't add any value to your claim for Koufax anyway. 2 ) You constantly ignore other intangible items that, while oft times not quantifiable, are still non-zero. This is demonstrated, in part, by claims such as athletes are better today than they were years ago. While true, you continue to ignore the fact that athletes years ago played and trained under different conditions with what was available at the time. You're implying that had those 'lesser athletes' been born and raised in today's environment, they would not have taken advantage of today's methods and still been throwing rocks at a chalk outline on a barn. Maybe try the reverse? Put Koufax back into the 1930s environment and do the analysis and you'll have different variables to consider. It is possible he wouldn't have made it out of the Dodgers or Cardinals 13 or so minor league teams.

I'd like to know how Statistician Snowman would factor in cortisone, regarding Koufax. When we talk about performance enhancing drugs, in the literal sense, Koufax is a great example. Without that drug his career might've had about a 2 year peak. Grove came back from injury without such aid.

[Peter_Spaeth]

Originally Posted by Mark17

I'd like to know how Statistician Snowman would factor in cortisone, regarding Koufax. When we talk about performance enhancing drugs, in the literal sense, Koufax is a great example. Without that drug his career might've had about a 2 year peak. Grove came back from injury without such aid.

But on the other hand, Koufax might have been able to extend his career with advances in surgical techniques.

It's all just too speculative when one tries to make direct comparisons.

Another point that has occurred to me, today's pitchers I presume have access to data that literally analyzes every pitch a hitter has ever taken or swung at and I presume there are people who can turn that into useful information. In Koufax's day, they probably had little more than anecdotal information to go on, and in pre-team meetings came up with brilliant strategies like smoke him inside. Counterpoint, I guess, is that batters now have the same information about the pitcfhers.

[BobC]

Originally Posted by Peter_Spaeth

But on the other hand, Koufax might have been able to extend his career with advances in surgical techniques.

It's all just too speculative when one tries to make direct comparisons.

Another point that has occurred to me, today's pitchers I presume have access to data that literally analyzes every pitch a hitter has ever taken or swung at and I presume there are people who can turn that into useful information. In Koufax's day, they probably had little more than anecdotal information to go on, and in pre-team meetings came up with brilliant strategies like smoke him inside. Counterpoint, I guess, is that batters now have the same information about the pitcfhers.

Great points Peter. Though he's got other issues to deal with, it is well-known that Trevor Bauer has gone to unprecedented lengths to scientifically study pitching, aerodynamics, spin rates, and so on using equipment and technology Grove and Spahn never had. And he is also known for his somewhat unique training techniques and exercises. And he was able to parlay all that into at least one Cy Young Award.