Building Out the Files

March 31, 2020
                               Building Out the Files

 

            In yesterday’s work I outlined a spreadsheet to explain what I am trying to do, tracking the total "Runs Prevented" by each of the 11 factors that we can measure.   Today I am going to add three more columns to that, and speculate a little bit about where we seem to be headed.  Today we are looking at Home Runs, Wild Pitches, and Hit Batsmen:

            The team preventing the most runs ever by not allowing Home Runs was the 2015 Pittsburgh Pirates, who allowed only 110 Home Runs, 106 Park Adjusted, which helped them to finish 98-64.   If they had been 4 standard deviations below the norm they would have allowed 200 home runs, so they beat that by 94 home runs, or about 132 runs.

            The team preventing the fewest runs by not allowing Home Runs was the 1913 St. Louis Cardinals, 51-99.  They allowed 57 home runs, park-adjusted 66, while the zero standard would have been 68.   They were only about 3 runs away from having zero ability in this area.   They led the league in Home Runs Allowed by 40% (57-40) while playing in a park with a Home Run Factor of 67.  The Home Run factor may be misleading, because in that era many Home Runs were inside-the-park Home Runs. 

            The team preventing the most runs by Wild Pitch Avoidance was the 2018 New York Mets, who threw 26 Wild Pitches against a zero-competence standard of 112, saving themselves about 14 runs.

            On the other end of that scale was the 1958 Dodgers, who threw 70 Wild Pitches, which is about 1 run WORSE than the zero-competence standard.    They threw almost twice as many Wild Pitches as any other team in the league. 

            The team which prevented the most runs by Hit Batsmen Avoidance was the 1907 Chicago White Sox, who hit only 22 batters with pitches, while every other team in the majors hit at least 38 that season.   The zero-competence standard would have been 105 Plunkers, so the White Sox saved themselves about 25, 26 runs by not hitting people with pitches.   The 2004 Atlanta Braves did almost as well.  They hit only 27 batters with pitches, in a season in which the zero-competence standard would have been 109. 

            On the other end of the scale is Ty Cobb’s 1922 Tigers, who hit 84 batters with pitches, in a season in which no other team in the American League hit more than 46.  They were actually worse than the zero-competence standard, the only team that was. 

 

 

            OK, let’s do a little bit of assembly work.   We’ll start with the 1951 New York Giants, since that’s a very famous team, and suits our other porpoises.   They won the National League, of course, with a 98-59 record. This is our summary of their defensive performance so far:

 

Team:  1951 New York Giants  (98-59)

 

 

Runs Prevented By:

 

Strikeouts

76

Control

111

Home Run Avoidance

155

Hit Batsmen Avoidance

14

Wild Pitch Avoidance

6

Balk Avoidance

 

Fielding Range (DER)

 

Fielding Consistency (F Pct)

 

Double Plays

 

Stolen Base Control

 

Passed Ball Avoidance

 

 

 

Sum of the Above

362

Actual Runs Prevented:

761

Error/Discrepancy:

 

 

            And, to take a team of my childhood, the 1964 Kansas City A’s; they finished 57-105:

Team:  1964 Kansas City A's (57-105)

 

 

Runs Prevented By:

 

Strikeouts

105

Control

66

Home Run Avoidance

88

Hit Batsmen Avoidance

7

Wild Pitch Avoidance

7

Balk Avoidance

 

Fielding Range (DER)

 

Fielding Consistency (F Pct)

 

Double Plays

 

Stolen Base Control

 

Passed Ball Avoidance

 

 

 

Sum of the Above

274

Actual Runs Prevented:

581

Error/Discrepancy:

 

 

            Overall, all 2,550 teams in the study, there are supposed to be 1,783,676 Runs Prevented which need to be accounted for, and 832,149 which have so far been accounted for.  That’s 47%.   The 1964 Kansas City A’s were at 47%; the 1951 New York Giants were at 47.6%.     That’s why I chose them for illustration, although there are several hundred teams around 47%. 

            Let me point out to you a couple of things that we can do with this data, IFF we can make the system work.   The 1951 New York Giants have a very high number of runs prevented by Home Runs, a fact which is hidden in the raw data since they gave up a pretty large number of home runs, but it was a Home Run park; the Giants always gave up homers because it was short down the lines at the Polo Grounds.   The Giants’ manager was Leo Durocher, who had great success with his first two teams and significant success with the Cubs of the 1960s, getting them not into first place, but close.  

            Suppose that you line up all of Durocher’s teams, and look at where they were in each Run Prevention area before Durocher took over the team, and how they changed when Durocher took over the team.   Compare Durocher to Casey Stengel, let’s say, or Joe McCarthy.  In what areas of run prevention were one manager’s teams better than the other’s?   In what areas were they worse?  What tradeoffs were they making?   It is possible that you might gain some understanding of how one manager was different than another.  It is possible that, at some point in the distant future, 30 years from now, there might be a general understanding of a manager’s role in shaping his team which is different than the understanding that we now have, because we have a method that measures the success of the manager’s teams in a series of different areas. 

            The 1964 Kansas City A’s—this is from memory, so don’t quote me—but the 1964 Kansas City A’s gave up 220 home runs, which I believe was a major league record at that time, and remained a major league record for quite a while after that.  Because of that, I expected the A’s to be near the zero-competence standard for home runs allowed.

            But they’re actually not all that close to it.   The A’s that spring had moved their fences way in.  They had traded that winter for two sluggers, Rocky Colavito and Jim Gentile, and Charley Finley had visions of battering opponents into submission with the A’s new power.   Their home runs allowed, park-adjusted, are not all that bad; we park-adjust the 220 Home Runs down to 172, and they’re still 63 homers away from the zero-competence standard, the misery line.

            They’re not actually HORRIBLE in that area, but they’re not good, either.   If we say they gave up 172 "actual" home runs, that’s 241 runs, and we see that they "prevented" 88 runs by Home Run prevention.   That’s a record, for home run prevention, of 88-241.   Their success rate, in that area, was 27% (.267. . ..88/329).   Suppose, then, that you figured each team’s success rate in each area.  Do you see where I am going with this?   It seems like you would gain some real understanding of why certain teams succeeded, and why they failed.    

 

 

            But the 1951 Jints and the 1964 A’s are "good" examples; those are examples of the system apparently working more or less the way it is supposed to work.   I mean, 47% seems too low; I’ve already accounted for strikeouts, walks, and home runs allowed; I should probably be up around 70, 75%, but that’s not a real problem, since I’m just guessing at the values.  I can straighten that out later in the process.  

 

            However, frankly, all of the news here is not good news; the good news is not actually winning the battle.   I appear to have some real problems.   I’ll give you the two worst exemplars:  the 1968 Dodgers and 1900 Boston Braves:

 

Team:  1968 Los Angeles Dodgers (76-86)

 

 

Runs Prevented By:

 

Strikeouts

127

Control

111

Home Run Avoidance

193

Hit Batsmen Avoidance

11

Wild Pitch Avoidance

8

Balk Avoidance

 

Fielding Range (DER)

 

Fielding Consistency (F Pct)

 

Double Plays

 

Stolen Base Control

 

Passed Ball Avoidance

 

 

 

Sum of the Above

450

Actual Runs Prevented:

461

Error/Discrepancy:

 

 

 

Team:  1900 Boston Braves (66-72)

 

 

Runs Prevented By:

 

Strikeouts

50

Control

43

Home Run Avoidance

35

Hit Batsmen Avoidance

15

Wild Pitch Avoidance

4

Balk Avoidance

 

Fielding Range (DER)

 

Fielding Consistency (F Pct)

 

Double Plays

 

Stolen Base Control

 

Passed Ball Avoidance

 

 

 

Sum of the Above

146

Actual Runs Prevented:

1150

Error/Discrepancy:

 

 

            The 1968 Dodgers are supposed to have 461 Runs Saved, and we’ve already credited 450 of them.   We’ve already accounted for 98% of their Runs Prevented, and we still have six categories of information to be added to the study.  And we’re probably going to have to increase the values in the categories we have studied so far.   Even if the Dodgers were fantastically awful in all of the areas not yet incorporated into the system, they would still show as saving substantially more runs than we believe that they actually saved—probably 60% more.  I’m going to have a huge error in that team’s estimate of Runs Prevented.   

            The 1900 Braves, on the other, have so far accounted for only 13% of their runs saved.   We’re not going to get there in their case.

            Those are SERIOUS problems for my study.   Those readings indicate that I may have a fundamental error in my concept, quite possibly so fundamental that my effort may fail.  I may have to go back and re-think my outline for the system, and I may not be able to find a way to make it work.   But I’ll keep moving forward, and we’ll see what we’ve got.

            There is SOME tolerance for error; it’s an estimate, after all.   It would be wonderful if the standard error of the estimates, at the end of the process, was 2%.  If the error was 3% or 4%, that clearly would be acceptable; if it was 5%, that would be disappointing.  If it was 10%, that would be intolerable; that would indicate a complete failure of the effort, unless one of my assumptions is wrong somewhere.   The standard deviation of Runs Scored is probably somewhere around 10% of Runs Scored, maybe less, so if you have a 10% error, then you don’t have anything useful; you’ve just got random data. 

            Some of these problems may be simple errors; I may have a data entry error somewhere, or a problem that can be solved by altering one assumption.  I don’t really know how many teams are going to be out of range at the end of the day.   But I’ll just keep plugging away at it.    Thanks for reading. 

 

 
 

COMMENTS (14 Comments, most recent shown first)

bjames
Since every plate appearance represents a potential run, would not total runs prevented be equal to total PA minus runs allowed?

That would be zero DEFENSE. I'm not talking about zero defense; I am talking about zero VALUE defense. Defenders so bad that they have no value to a major league team.
3:55 PM Apr 1st
 
bjames
My own thinking is that you create one zero standard based on standard deviations based on the entire average of all major league history, and then figure runs saved from that.


You can't do that, because the DEVIATION itself does in fact change over time. You might be able to do that in some categories are relatively stable over time, but if you did that for strikeouts and home runs allowed, it would be a disaster. Trust me; it would not work.
3:54 PM Apr 1st
 
Mongo1962
Since every plate appearance represents a potential run, would not total runs prevented be equal to total PA minus runs allowed?

So for a 9-inning game with 40 PA and 4 runs allowed, there would be 40 - 4 = 36 runs prevented.




11:45 AM Apr 1st
 
Brian
My own thinking is that you create one zero standard based on standard deviations based on the entire average of all major league history, and then figure runs saved from that. This is because

1) I think you are trying to create a raw number here. Adjusting for context seems to take away from that.

2)Era and park context might cause some fluctuation, but basically each event is worth the same amount of runs throughout history. Context just changes how much each event happens.

3) On the other side of it, when you arrived at the runs created formula, you didn't use any context adjustments for those. Once you had runs created, you then converted them to wins by adding the adjustments after the fact.

So what I would suggest is to first determine runs saved, then adjust for context after you have done that.

5:33 PM Mar 31st
 
willibphx
Perhaps there is a simple explanation. Are the runs prevented for the two teams correct? Directionally is not runs prevented equal to 2 times average teams runs minus the individual teams runs allowed? If you do this you get 601 runs prevented for the Dodgers and 744 runs prevented for the Braves which brings them more into line with the individual factors. For the Giants it was pretty close at 768 vs 761 in the article, A's were 476 versus 581. Or, God forbid, I really did misunderstand the sanitizer discussion.
4:19 PM Mar 31st
 
Guy123
Maybe the rolling average will fix years like 1900 and 1968 that are outliers within their decade. Another approach would be to calculate the SDs as a percentage of the mean (coefficient of variation) rather than a fixed number, and then use that to calculate yearly SDs. So, for example, if the 1960s SD for HR was say 15% of the average HR total, then the 1968 HR SD would be defined as .15 * 100 = 15.

1:19 PM Mar 31st
 
evanecurb
I don't know if this will be helpful or not, but I'm going to throw it out there. Could there be benefit in looking at errors, double plays in Division I college ball or in the low minor leagues to establish what a zero competence level might look like in those categories?
12:53 PM Mar 31st
 
chuck
Regarding the ’68 Dodgers and home runs allowed.
’68 is a season in which there was likely a high number of substandard balls being used, dramatically reducing the number of home runs. This began around August of 1967 and extended through the '68 season. Anyway, the ’68 NL ratio of HR/BFP was .0147, while your figure for the decade norm was .0216. The NL as a whole is already almost 2 standard deviations below the norm.

The Dodgers’ park was not conducive to home runs. They allowed 24 at home, 41 on the road. Their offense hit 25 at home, 42 on the road. The overall home/road ratio is .59. On the road, the Dodgers allowed a HR/BFP rate of .0140, almost the league average. At home their ratio was .0079, and much of that would seem to be explained by the park.

But I think the fact that that season itself is so far off of the decade norm may be what is exaggerating their runs on home runs saved.
12:31 PM Mar 31st
 
SteveN
"suits our other porpoises" Do you think that, maybe, Bill has been inside a tad too long?
11:45 AM Mar 31st
 
willibphx
Sorry if I wasn't clearer, but my earlier comment on Park Effects related to the STdev of HRA. Though it potentially could apply to DER at that phase.
11:40 AM Mar 31st
 
willibphx
A few thoughts,

The 1900 Braves concerns me less than the 68 Dodgers example in accordance with Guy123 comments. In 1900, 85% of PA ended up with a ball in play and for the entire decade 30% of runs were unearned as compared with 65% and 7% respectively in 2019. Hence, run prevention from the defensive components should be much higher.

The 68 Dodgers is much harder to understand. The Stdev for Ks for the 60s looks high based on my calculation which was closer to .15 than .18. Not sure how material that is. The one other thought I had that may effect the Dodgers but potentially all results is whether the park effect should be applied to the calculation of the StDev for each decade. Not sure but Park Effects could be materially contributing to the level of StDev and may smooth results by including.

Finally, are you assuming that the value of each outcome is the same across all eras? At various times I know Pete Palmer adjusted his values by era and you had separate formulas for different eras.

As always thanks for sharing the analysis in process, it has been a great way to spend time thinking about something other than Covid.
11:39 AM Mar 31st
 
Mongo1962
To add to Guy123's post, perhaps consider allowing the "floor" for each category of runs saved to vary according to the year.

I do think that teams save a LOT of runs through ordinary defensive efforts; probably a lot more than twice the average number of runs scored. A very good defensive team might save maybe 110% of the average number of runs saved, while a historically bad team might save 80% of the average number of runs saved. The floors should be set accordingly.
9:55 AM Mar 31st
 
Guy123
Not sure about LAD, but I think the problems with the 1900 Braves are likely caused by three factors:

1) The estimate of 1,150 runs saved seems way too high. Average team playing 142G would give up 740 runs, and Braves gave up 738. Seems like runs saved should be in same ballpark.

2) The 1900 run environment (5.21 R/G) was very different than the decade overall (around 4.0 R/G), so maybe using decade-based SDs is creating problems?

3) DER and Fielding% -- which you haven't covered yet -- were *much* more important in the 1900s. They explain about 65% of team defense in that period (compared to about 30% in today's game). So a big unexplained proportion is not surprising.
9:37 AM Mar 31st
 
Mongo1962
Hello Bill,

It occurs to me that the issue may be due to the specific "floors" for each category of run prevention. It's possible that some categories have a floor that's six or seven SD away from the average, depending on how easy it is for that skill to be mastered by a professional-level player. This would mean that every team would almost automatically get some large fraction of the possible runs saved in that category, simply by playing defense for 9 innings a game.

The result would be that each team's total runs saved would be within some narrower band of runs saved, with a significant number of "automatic" runs saved.
8:56 AM Mar 31st
 
 
©2020 Be Jolly, Inc. All Rights Reserved.|Web site design and development by Americaneagle.com|Terms & Conditions|Privacy Policy