FAQ and Forum on Advanced Stats

[Eminor3rd]

I think they've done quite a bit of this and it's where they derive the statistics from. So many of these variables show such little significance towards the runs scored with such high variability that it's impossible to say that one is more important than the other, but it essentially follows basic guidelines and sounds logic.

 

1) A strikeout is the best guarantee for an out.

2) A walk is never an out.

3) If the batter hits it, you prefer it to be on the ground because it's almost never a home run and the batter will usually not advance beyond 1B.

4) A flyball is preferable to a line drive, but those can be dangerous.

5) You do not ever want to give up line drives.

 

Your ideal pitcher is one who has good command, gets a fair amount of strike outs, and keeps the ball on the ground.

 

I'd argue that the best pitcher - starter or reliever - of the modern era is Mariano Rivera. His numbers:

 

8.22 K/9

2.01 BB/9

52.5% GB%

 

Those can be supplemented, and you can be great without 1 of them, but I generally think you need at least two of them to be a great pitcher.

 

You left out most shocking number: 0.50 HR/9

[Eminor3rd]

I'd argue that the best pitcher - starter or reliever - of the modern era is Mariano Rivera. His numbers:

 

8.22 K/9

2.01 BB/9

52.5% GB%

 

Also, guess who this is:

 

9.26 K/9

2.13 BB/9

45.7% GB

[witesoxfan]

Also, guess who this is:

 

9.26 K/9

2.13 BB/9

45.7% GB

 

I saw a few of them, but I initially just looked at the 2000s. My initial guess is Joe Nathan.

 

EDIT: Nope, way too many flyballs for that ground ball, but it wasn't a bad guess.

Edited February 26, 2014 by witesoxfan

[witesoxfan]

Also, guess who this is:

 

9.26 K/9

2.13 BB/9

45.7% GB

 

I can't find numbers that match that exactly, but it's incredibly similar to Sale's career numbers. I'm honestly not sure Sox fans appreciate just how talented Chris Sale is. If he stays healthy - and I can't knock on enough wood - he should end up with multiple Cy Youngs.

[witesoxfan]

For others, if you want to mess around, this is all of the numbers, sorted by K/9 with BB/9 and GB% right next to them.

 

http://www.fangraphs.com/leaders.aspx?pos=...&sort=9%2cd

 

And one of my favorite pitchers ever is #2 on that list. I'm hoping Billy Wagner gets elected, but I think people forget how truly dominant he really was.

[cabiness42]

Also, guess who this is:

 

9.26 K/9

2.13 BB/9

45.7% GB

 

Somebody other than Ervin Santana?

[Eminor3rd]

I can't find numbers that match that exactly, but it's incredibly similar to Sale's career numbers. I'm honestly not sure Sox fans appreciate just how talented Chris Sale is. If he stays healthy - and I can't knock on enough wood - he should end up with multiple Cy Youngs.

 

Correct! Those are Chris Sale's numbers as a starter, his RP numbers removed. That's what makes that line incredible -- it's WAY easier to pitch well one inning at a time.

[Eminor3rd]

Somebody other than Ervin Santana?

 

This is even MORE correct than wite's answer of the exact correct pitcher.

[witesoxfan]

Correct! Those are Chris Sale's numbers as a starter, his RP numbers removed. That's what makes that line incredible -- it's WAY easier to pitch well one inning at a time.

 

Well, then it's worth pointing out just his 2013 too

 

9.49 K/9

1.93 BB/9

46.6% GB%

 

That is a K/BB of almost 5. The only problem thus far has been that his flyballs have a tendency to leave the yard a bit more often than others.

[ptatc]

Well, then it's worth pointing out just his 2013 too

 

9.49 K/9

1.93 BB/9

46.6% GB%

 

That is a K/BB of almost 5. The only problem thus far has been that his flyballs have a tendency to leave the yard a bit more often than others.

Fangraphs is a great sortable resource. I think I found all the data I need to start. Thanks for showing it. I'll run a few models and post the results.

[witesoxfan]

Fangraphs is a great sortable resource. I think I found all the data I need to start. Thanks for showing it. I'll run a few models and post the results.

 

I'm incredibly interested in seeing the R-squared and standard deviation of these results.

[Eminor3rd]

Relevant: http://www.fangraphs.com/blogs/howd-we-do-a-year-ago/

[ptatc]

I'm incredibly interested in seeing the R-squared and standard deviation of these results.

I have no doubt that std dev will be high as the variability is high in many of the factors. I would love to see an r2 of at least 60 but I'm not confident.

[witesoxfan]

I have no doubt that std dev will be high as the variability is high in many of the factors. I would love to see an r2 of at least 60 but I'm not confident.

 

I'd be surprised to even see 40. I ran a very basic regression looking at runs scored for teams over like a 2 year period of time just to see which numbers correlated more heavily with it, but it was very poorly done. My conclusions at the time were that stolen bases and batting average showed nothing conclusive at all about runs scored, while OBP was the most highly correlated at around 37-40%, something like that. Still not a significant figure, which tells you that it's more than just getting on base because so much of baseball is contextually based and luck plays such a large factor, but being that high tells you it's the most important aspect of scoring runs at the very least.

[ptatc]

I'd be surprised to even see 40. I ran a very basic regression looking at runs scored for teams over like a 2 year period of time just to see which numbers correlated more heavily with it, but it was very poorly done. My conclusions at the time were that stolen bases and batting average showed nothing conclusive at all about runs scored, while OBP was the most highly correlated at around 37-40%, something like that. Still not a significant figure, which tells you that it's more than just getting on base because so much of baseball is contextually based and luck plays such a large factor, but being that high tells you it's the most important aspect of scoring runs at the very least.

I ran the numbers this morning and was shocked. I looked at variables predicting runs given up by pitchers over the last two years.

 

The R2 was 47.5 for HR alone, 69.6 for HR,BABIP; 82.2 for HR, BABIP, BB; 86 for HR,BABIP,BB,HR/9 and 90 for HR,BABIP, BB, HR/9 and FIP. All of the models had a sig. of .01 or lower. This was using a stepwise linear regression.

 

Granted it should have gone up when including HR/9 and FIP as one of the internal variables there is HR which is 47.55 of the variability on it's own.

 

So in the two year sample (i only included pitchers with a combined 300 innings in the last two years) there is a high predictability of runs scored against a pitcher with HR, BABIP and BB.

 

I guess FIP is a very good predictor of pitchers performance in giving up runs.

 

I'm going to run a larger sample for a lrger number of years to see if it still holds true over time.

[Eminor3rd]

I ran the numbers this morning and was shocked. I looked at variables predicting runs given up by pitchers over the last two years.

 

The R2 was 47.5 for HR alone, 69.6 for HR,BABIP; 82.2 for HR, BABIP, BB; 86 for HR,BABIP,BB,HR/9 and 90 for HR,BABIP, BB, HR/9 and FIP. All of the models had a sig. of .01 or lower. This was using a stepwise linear regression.

 

Granted it should have gone up when including HR/9 and FIP as one of the internal variables there is HR which is 47.55 of the variability on it's own.

 

So in the two year sample (i only included pitchers with a combined 300 innings in the last two years) there is a high predictability of runs scored against a pitcher with HR, BABIP and BB.

 

I guess FIP is a very good predictor of pitchers performance in giving up runs.

 

I'm going to run a larger sample for a lrger number of years to see if it still holds true over time.

 

If you can isolate factors that account for the variability of HR and BABIP, you may be stumbling upon a revolution.

[ptatc]

If you can isolate factors that account for the variability of HR and BABIP, you may be stumbling upon a revolution.

I'm not sure what you're asking. With this sample HR accounts for 47% of the predictive factor in the runs a pitcher gave up. Add BABIP into the mix and it's 69%. All you need to do is plug their values into the equation and it will show the predicted runs scored.

[southsider2k5]

For others, if you want to mess around, this is all of the numbers, sorted by K/9 with BB/9 and GB% right next to them.

 

http://www.fangraphs.com/leaders.aspx?pos=...&sort=9%2cd

 

And one of my favorite pitchers ever is #2 on that list. I'm hoping Billy Wagner gets elected, but I think people forget how truly dominant he really was.

 

Closers have really gotten the shaft in the HOF balloting.

[Eminor3rd]

I'm not sure what you're asking. With this sample HR accounts for 47% of the predictive factor in the runs a pitcher gave up. Add BABIP into the mix and it's 69%. All you need to do is plug their values into the equation and it will show the predicted runs scored.

 

Right, I'm saying that since those are such high factors in runs allowed, then if you can do the same analysis for what accounts for HR and BABIP totals, and find a significant result, you could forecast regression for pitchers. Of course, maybe that's what xFIP does already, not sure.

[witesoxfan]

Actually, maybe we should be paying more attention to SIERA, as it does actually take into accounts balls in play to try and determine talent level.

 

 

[ptatc]

Right, I'm saying that since those are such high factors in runs allowed, then if you can do the same analysis for what accounts for HR and BABIP totals, and find a significant result, you could forecast regression for pitchers. Of course, maybe that's what xFIP does already, not sure.

xFIP doesn't do it. This really doesn't show a regression for pitchers. It will predict how many runs they will give up if HR or BABIP values change.

 

For example if a pitcher gives up an abnormally high number of HR, you can predict the runs for the following year if you change the value. Same with BABIP. If it's high one year, you can predict the decrease in runs if you lower the BABIP to what the average pitcher had that year.

The regression model just uses past variables to determine their weight in the dependent varible. You can use this to predict fututre performance.

 

This is what I use to determine which applicants to our DPT program have a good chance to pass our license exam thus which applicants we accept.

[ptatc]

Actually, maybe we should be paying more attention to SIERA, as it does actually take into accounts balls in play to try and determine talent level.

What is that and where can I find it?

[witesoxfan]

What is that and where can I find it?

 

FanGraphs has it, it's just a little bit lower. Here's a brief write-up on it with more detailed explanations (including the math behind it) down below.

 

http://www.fangraphs.com/library/pitching/siera/3

 

I'll link Sale's page because he's awesome: http://www.fangraphs.com/statss.aspx?playe...&position=P

 

SIERA is listed under the heading batted ball, but it's actually separate from the batted ball statistics. It's listed alongside tERA (which also takes into account batted balls, but does not appear to be anywhere nearly as accurate), xFIP-, and xFIP.

[Eminor3rd]

xFIP doesn't do it. This really doesn't show a regression for pitchers. It will predict how many runs they will give up if HR or BABIP values change.

 

For example if a pitcher gives up an abnormally high number of HR, you can predict the runs for the following year if you change the value. Same with BABIP. If it's high one year, you can predict the decrease in runs if you lower the BABIP to what the average pitcher had that year.

The regression model just uses past variables to determine their weight in the dependent varible. You can use this to predict fututre performance.

 

This is what I use to determine which applicants to our DPT program have a good chance to pass our license exam thus which applicants we accept.

 

I see. Well go invent stuff!

 

You should definitely read The Book and Baseball Between the Numbers to get a sense of what was done early on in DIPS so you don't end up repeating something unnecessarily. But this is an area where a breakthrough could definitely be made.

[ptatc]

FanGraphs has it, it's just a little bit lower. Here's a brief write-up on it with more detailed explanations (including the math behind it) down below.

 

http://www.fangraphs.com/library/pitching/siera/3

 

I'll link Sale's page because he's awesome: http://www.fangraphs.com/statss.aspx?playe...&position=P

 

SIERA is listed under the heading batted ball, but it's actually separate from the batted ball statistics. It's listed alongside tERA (which also takes into account batted balls, but does not appear to be anywhere nearly as accurate), xFIP-, and xFIP.

This is an outstanding stat. It takes into consideration all of the variables at which I was looking. It also does not use somewhat arbitrary multipliers like FIP and xFIP. I don't think as of right now there is much of an improvement on looking at how many runs a pitcher gave up based mostly on what he can influence.

 

I can't find the numerical predictive model on which it's based on however I really like this one and will use it as the best indicator of how a pitcher can influence wins.