Sox sign Gordon Beckham, designate Viciedo for Assignment

[Reddy]

This is my point exactly. By not factoring in things like that, there are biases against certain franchises. If you aren't adjusting player performance for franchise input, you are missing out on key data points.

What data points can you attribute to a player because Herm Schneider is a good trainer? Instead you just see that they avg a 7 win boost from their PECOTA and that's your adjustment.

[Reddy]

Not even close to that. I see PECOTA as a reference, a point of view, not a fact.

I have never once said it's fact. Of course it's a reference. It sets a decent baseline, from which standard deviations exist.

[Kyyle23]

I have never once said it's fact. Of course it's a reference. It sets a decent baseline, from which standard deviations exist.

 

No you didn't say it is a fact, but those of us that don't really fall in line with what they are projecting are out of touch of reality, according to you.

[shysocks]

A similar argument occurred about this Fangraphs post a while back, and essentially I think everyone here is a little bit correct about projections. The post wasn't about PECOTA specifically but about projected team WAR, but I think the same generalities apply.

 

The reason projections are helpful: Of 20 playoff teams the past two seasons, 19 were projected for at least 30 WAR prior to that season. So that is solid evidence that projections are a good assessment of where teams stand.

 

The reason projections are not always helpful: Because when I look at the graph with the faded Orioles logo in its background, I see probably 10 teams who exceeded their projected WAR by 10 or more, and another 7 or so who fell short by that margin. That is a full third of the sample (!!) where the projection whiffed considerably. For whatever reason. Those surprises occur often enough that it would be unwise to praise the projections as gospel.

[Eminor3rd]

No you didn't say it is a fact, but those of us that don't really fall in line with what they are projecting are out of touch of reality, according to you.

 

It's a credible point of data that he's using to support a claim. There's no problem with this. Just because humans can't predict the future yet, doesn't make this information useless.

 

It's not PROOF of anything, because proof of the future doesn't exist. But it is evidence to suggest the likeliness of a particular outcome, or more accurately, range of outcomes.

[witesoxfan]

A similar argument occurred about this Fangraphs post a while back, and essentially I think everyone here is a little bit correct about projections. The post wasn't about PECOTA specifically but about projected team WAR, but I think the same generalities apply.

 

The reason projections are helpful: Of 20 playoff teams the past two seasons, 19 were projected for at least 30 WAR prior to that season. So that is solid evidence that projections are a good assessment of where teams stand.

 

The reason projections are not always helpful: Because when I look at the graph with the faded Orioles logo in its background, I see probably 10 teams who exceeded their projected WAR by 10 or more, and another 7 or so who fell short by that margin. That is a full third of the sample (!!) where the projection whiffed considerably. For whatever reason. Those surprises occur often enough that it would be unwise to praise the projections as gospel.

 

I like this post very much.

[Eminor3rd]

For whatever reason. Those surprises occur often enough that it would be unwise to praise the projections as gospel.

 

But again, no one does. It's a strawman argument (that I know you're not making) that dodges the actual utility of these systems.

[shysocks]

I like this post very much.

Be sure to upvote!

[Reddy]

A similar argument occurred about this Fangraphs post a while back, and essentially I think everyone here is a little bit correct about projections. The post wasn't about PECOTA specifically but about projected team WAR, but I think the same generalities apply.

 

The reason projections are helpful: Of 20 playoff teams the past two seasons, 19 were projected for at least 30 WAR prior to that season. So that is solid evidence that projections are a good assessment of where teams stand.

 

The reason projections are not always helpful: Because when I look at the graph with the faded Orioles logo in its background, I see probably 10 teams who exceeded their projected WAR by 10 or more, and another 7 or so who fell short by that margin. That is a full third of the sample (!!) where the projection whiffed considerably. For whatever reason. Those surprises occur often enough that it would be unwise to praise the projections as gospel.

 

It's because of standard deviations and luck - which can't be projected.

 

The '05 White Sox were massively lucky, and got career years out of our entire bullpen - which was the major reason for our hot start out of the gate. That kind of thing happens every once and a while. It could happen for us this year. But the point is that the MOST LIKELY OUTCOME is what these projections are for. Each team doesn't take into account the others, so just because no one in the Central is slated to win over 82 games, doesn't mean no one will. It simply means that for each individual team, the MOST LIKELY OUTCOME, statistically speaking, is that they hit their projected win total. And obviously there are caveats for teams like the White Sox who have a system that can't be projected quite as accurately as the others.

[witesoxfan]

It's because of standard deviations and luck - which can't be projected.

 

The '05 White Sox were massively lucky, and got career years out of our entire bullpen - which was the major reason for our hot start out of the gate. That kind of thing happens every once and a while. It could happen for us this year. But the point is that the MOST LIKELY OUTCOME is what these projections are for. Each team doesn't take into account the others, so just because no one in the Central is slated to win over 82 games, doesn't mean no one will. It simply means that for each individual team, the MOST LIKELY OUTCOME, statistically speaking, is that they hit their projected win total. And obviously there are caveats for teams like the White Sox who have a system that can't be projected quite as accurately as the others.

 

But the "most likely outcome" doesn't mean that it happens a lot, it just means it happens the most often in simulations and based on expectations. The bell curve of possible outcomes has a median of 79 wins, but it's a very large error bar and the odds of hitting exactly 79 wins is incredibly small. In fact, I've seen suggestions that the error bar may be as high as 8 games. If that's the case, then everything within 1 standard deviation leaves the Sox between 71 and 87 wins, which is either one of the worst teams in the league or one of the best.

 

There's all kinds of noise that isn't accounted for in projections because they literally can't account for it. Injuries, roster additions, roster subtractions, players breaking out, players disappointing, luck, and whatever else, they are very inexact. They should be looked at because, as you've noted, they indicate the baseline talent level of the Sox, which is probably an average team (if we used a range of +/- 2 on that 79 wins, it'd 77-81 wins, which is less intimidating), and there are plenty of reasons to believe they can outplay that projection. There are others to believe they will not beat it.

 

I expect about 85 wins. I won't be surprised in the least if they win 90+. I won't be surprised if they win 75, but (not that it's a surprise) something catastrophic will have happened for them to win that little.

[Eminor3rd]

But the "most likely outcome" doesn't mean that it happens a lot, it just means it happens the most often in simulations and based on expectations. The bell curve of possible outcomes has a median of 79 wins, but it's a very large error bar and the odds of hitting exactly 79 wins is incredibly small. In fact, I've seen suggestions that the error bar may be as high as 8 games. If that's the case, then everything within 1 standard deviation leaves the Sox between 71 and 87 wins, which is either one of the worst teams in the league or one of the best.

 

There's all kinds of noise that isn't accounted for in projections because they literally can't account for it. Injuries, roster additions, roster subtractions, players breaking out, players disappointing, luck, and whatever else, they are very inexact. They should be looked at because, as you've noted, they indicate the baseline talent level of the Sox, which is probably an average team (if we used a range of +/- 2 on that 79 wins, it'd 77-81 wins, which is less intimidating), and there are plenty of reasons to believe they can outplay that projection. There are others to believe they will not beat it.

 

I expect about 85 wins. I won't be surprised in the least if they win 90+. I won't be surprised if they win 75, but (not that it's a surprise) something catastrophic will have happened for them to win that little.

 

Understanding this is essential if one is going to make an informed criticism of projections.

 

Also, RE: the bolded -- the errors bars depend upon the confidence level you set. Usually people use 90 or 95%, so you might say "I'm 90% sure that the Sox will win between 71 and 85 games." If you want to be 95% sure, you've got to widen it. If you only want to be 50% sure, you can narrow it. The actual number these projections land on is the mean of whatever confidence interval they set.

Edited January 29, 2015 by Eminor3rd

[ptatc]

This is just a Hawk vs. Sabermetrics situation. Older guys don't appreciate statistics saying their team isn't good as they "feel like" it is.

what some people don't understand is that computer algorithms are not stats in the pure sense of the word and do have biases. They have the biases of the people who created the algorithms.

 

PECOTA is a complicated program but one of the biases that stands out to me is that it assumes that older players will regress and younger players will improve. Another is the assumed number of injuries and days off, especially from the pitching staff. I think this is why it predicts the Astros will only be one game worse than the White Sox. On paper they look a lot farther apart than that.

Edited January 29, 2015 by ptatc

[witesoxfan]

Understanding this is essential if one is going to make an informed criticism of projections.

 

Also, RE: the bolded -- the errors bars depend upon the confidence level you set. Usually people use 90 or 95%, so you might say "I'm 90% sure that the Sox will win between 71 and 85 games." If you want to be 95% sure, you've got to widen it. If you only want to be 50% sure, you can narrow it. The actual number these projections land on is the mean of whatever confidence interval they set.

 

A 99% confidence interval for the Sox would be something like "I think they will win between 55 and 103 games this year." Cool story bro.

 

(working through this, some college statistics lessons are coming back to me)

[GGajewski18]

Who knew Gordon Beckham would get 27 pages from SoxTalk. You guys that interested in him?

 

 

 

[LDF]

 

b/c you have good posters writing and not swearing or crazy stuff like that,

[Kyyle23]

Who knew Gordon Beckham would get 27 pages from SoxTalk. You guys that interested in him?

 

 

 

 

As is the case in many white sox threads, we have experienced deviation

[ptatc]

As is the case in many white sox threads, we have experienced deviation

From this board that deviation would be standard.

[ptatc]

Understanding this is essential if one is going to make an informed criticism of projections.

 

Also, RE: the bolded -- the errors bars depend upon the confidence level you set. Usually people use 90 or 95%, so you might say "I'm 90% sure that the Sox will win between 71 and 85 games." If you want to be 95% sure, you've got to widen it. If you only want to be 50% sure, you can narrow it. The actual number these projections land on is the mean of whatever confidence interval they set.

What they really need to do is come up with the Standard Error of Measure based on how far off they've been over the last decade or so. That would give them enough data to predict how far off they are likely to be.

[Dick Allen]

But again, no one does. It's a strawman argument (that I know you're not making) that dodges the actual utility of these systems.

What exactly is the purpose of the BP projections?

[LDF]

What they really need to do is come up with the Standard Error of Measure based on how far off they've been over the last decade or so. That would give them enough data to predict how far off they are likely to be.

ahhh what did you say....

[Reddy]

But the "most likely outcome" doesn't mean that it happens a lot, it just means it happens the most often in simulations and based on expectations. The bell curve of possible outcomes has a median of 79 wins, but it's a very large error bar and the odds of hitting exactly 79 wins is incredibly small. In fact, I've seen suggestions that the error bar may be as high as 8 games. If that's the case, then everything within 1 standard deviation leaves the Sox between 71 and 87 wins, which is either one of the worst teams in the league or one of the best.

 

There's all kinds of noise that isn't accounted for in projections because they literally can't account for it. Injuries, roster additions, roster subtractions, players breaking out, players disappointing, luck, and whatever else, they are very inexact. They should be looked at because, as you've noted, they indicate the baseline talent level of the Sox, which is probably an average team (if we used a range of +/- 2 on that 79 wins, it'd 77-81 wins, which is less intimidating), and there are plenty of reasons to believe they can outplay that projection. There are others to believe they will not beat it.

 

I expect about 85 wins. I won't be surprised in the least if they win 90+. I won't be surprised if they win 75, but (not that it's a surprise) something catastrophic will have happened for them to win that little.

 

at the end of the day we completely agree on the projections of this team, then. I expect around 85 as well, though we differ in that I'll be a *little* surprised if we win 90.

[Jake]

I thought most projection systems use arithmetic mean, which is typically best when searching for expected value.

[Kalapse]

f***ing nerds are ruining all that was once precious in this country.

[Reddy]

f***ing nerds are ruining all that was once precious in this country.

 

 

[Kalapse]

Total nerd move taking shots at Philo Beddoe.