Number of Saves vs. Games Over .500

by Jeff Zimmerman on Aug 14, 2009 1:06 AM EDT

Tommy Bennett is helping me with a 2010 baseball draft and we got into a discussion on closers. We are allowed only one reliever in a normal 5x5 league, so this reliever will need to get the as many saves as possible. I said that we need to get a reliever from a team with many chances (e.g. Rivera, Papelbon) while he stated that the team's winning percentage doesn't matter. He tried to Google and couldn't find any articles on it. I decided to go ahead and do a quick look into theory.

I took the games over 500 and the team saves for all teams between 1999 and 2008. Then I plotted the data and generated a regression line along with the r-squared value and here are the results:

This analysis shows that with a r-squared of .46, about 46% of a teams' closers ability to get saves depends on the number of games the team is above .500. Also, I went ahead and figured the standard deviation from the predicted and actual value and ended up with a value of 6.26 saves. Here is a chart of predicted number of saves a team should have depending on their record:

Record	Average number of saves	66% of time # of Saves Between	95% of time # of Saves Between
100-62	48.34	42.1 to 54.6	35.8 to 60.9
81-81	40.32	34.1 to 46.6	27.8 to 52.8
62-100	32.29	26.0 to 38.6	19.8 to 44.8

Using this data, you can see that a closer on a 100 win team has an over 95% chance of having more saves than the average closers on a 62-100 team. There are definite other factors involved in the number of save chances, but if a team doesn't win, there can't be a chance for a save. Hopefully, I am able to convince him now and next the two of us need to decided when we are going to actually draft our closer.

Tweet Comment 16 comments | 0 recs |

Do you like this story?

Comments

I'm confused

Isn’t R (not R Squared) the thing we care about? If I’m not mistaken, R^2 is just used to compare different correlations because it is an absolute value instead of a positive or negative value.

I always though that to derive meaning from an isolated regression, you should use R, which in this case is .69. That would mean that about 70% of a teams ability to win is tied to saves.

Although, I could be completely wrong. I’ve never taken a stats course in my life, so I really don’t have much of an idea what I am talking about.

F*** Billy Beane... actually, I kinda like Holliday

by vivaelpujols on Aug 14, 2009 1:24 AM EDT reply actions

The r-square tells you how much of the change in X is explained by Y. In this case, X is the # of saves and Y is the team’s winning percentage.

Crashburn Alley

by Crashburn Alley on Aug 14, 2009 8:32 AM EDT up reply actions

Did you actually read my comment?

i was wondering what the difference was between R and R^2, besides the obvious. If R is the correlation, than isn’t the only point of squaring it to turn it into an absolute value? I know that Excel squares it automatically, but when saying “x is explained this much % by y” wouldn’t it be better to un-square it?

F*** Billy Beane... actually, I kinda like Holliday

by vivaelpujols on Aug 14, 2009 8:49 AM EDT up reply actions

Difference between R & R^2

R is the one that’s actually pretty useless, it’s mostly only good to tell whether the correlation is positive or negative. R^2 tells how much of the variation in y is due to the change in x. The names (coefficient of correlation for R and coefficient of determination for R^2) can be somewhat misleading as the correlation that it talks about is, as stated previously, positive or negative correlation not strength of correlation.

by mycr0ft on Aug 14, 2009 9:22 AM EDT up reply actions 1 recs

Okay, thanks

F*** Billy Beane... actually, I kinda like Holliday

by vivaelpujols on Aug 14, 2009 9:35 AM EDT up reply actions

Rec'd.

Made sense to someone with minimal statistical education.

@bs_uf15bosox9be The Original Gameday; Learn to use SB Nation

by bdalebs on Aug 14, 2009 12:17 PM EDT up reply actions

Thanks for explanation of r-squared.

I have on my list to go and describe some commonly used advanced stats. In this case I need R-squared and Standard Deviation to tell the “story” in needed.

Also in this case the Standard Deviation passes the eye test in that if you look at say at 0 (81-81 season), most of the points are within 6 saves below and above the line.

Jeff Zimmerman - Protecting the world from RBI's and Wins from my mom's guest house.

by Jeff Zimmerman on Aug 14, 2009 12:33 PM EDT up reply actions

To be absolutely clear...

R should stand for simply the residual, or maybe sum of the residuals. Any individual residual really doesn’t display alot of information, but can be used for various tests for things such as homoskedasticity, or for looking for other possible trends in the data.

Additionally, R^2 isn’t strictly the square of R. It’s a ratio of the sum of the squared residuals over the sum of the squared observations. (sstotal – ssresid) / sstotal, or 1 – (ssresid/sstotal).

This effectively gives you a ratio that defines how close you are to actual. If every Y exactly equaled X, you would have an R^2 of 1. Additionally… R^2 will always increase with additional variables, no matter how unimportant they are.

by Trickman on Aug 14, 2009 1:41 PM EDT up reply actions

Riddle me owned

But wouldn’t it be more fair to do something like expected win percentage, since that’s the information you’re dealing with?

by Tommy Bennett on Aug 14, 2009 2:43 AM EDT reply actions

I agree

The teams that have the most saves may not be the best, but the luckiest (or best in close situations).

F*** Billy Beane... actually, I kinda like Holliday

by vivaelpujols on Aug 14, 2009 2:57 AM EDT up reply actions

True on the expected wins

Jeff Zimmerman - Protecting the world from RBI's and Wins from my mom's guest house.

by Jeff Zimmerman on Aug 14, 2009 9:22 AM EDT up reply actions

Wouldn't it make more sense to plot win% vs. save chances?

Saves should correlate rather well with wins because every save is always a win.

Unless I understand it wrong, isn’t the point of the exercise to see whether or not you need to just find a closer on a good team or if you’re just looking for a great closer on an at least somewhat good team? This takes into account closer quality and save opportunities.

by arzdb on Aug 14, 2009 1:35 PM EDT reply actions

Either way it would be the same results

I figure this form would be more useful for a person to see what a teams predicted record is for the year, say 90-72 and then be able to determine the possible number of saves the team closer may have.

Jeff Zimmerman - Protecting the world from RBI's and Wins from my mom's guest house.

by Jeff Zimmerman on Aug 14, 2009 2:00 PM EDT up reply actions