It really started here. But that was a blogpost about this. And it was soon followed by something we perhaps didn't know we needed.
While there are many fine new fielding metrics around using PBP (play-by-play data), to me the really important breakthrough is going to come when we get a reliable metric for fielding in the pre-PBP days. That's going to involve some kind of assumption about a piece of information we'll never have for most of baseball history - how many hits was a fielder responsible for.
I suspect it's quite important to start from the team level when calculating the value of fielding. (See Clay Davenport's article in the 2002 Baseball Prospectus for an example.) You then have a choice - either you can calculate the total of opportunities from the assumption that all balls will fall for hits, and thus the caught balls represent runs saved. Or you can look at fielding in terms of runs allowed, by thinking in terms of the number of mistakes. I suspect for a decent 'all of baseball history' metric, we need to think more along the lines of the latter. We need to establish 'expected runs' baseline against which to measure fielding.
I also came to a key realization after reading all this. Looking at the statistics when I was developing my Defensive Winning Percentage, I was surprised at how narrow the band is of fielding effectiveness. This is, I think, a reflection that there's both a floor and a ceiling to the effectiveness of fielders. Some of the credit is all down to the pitcher. All home runs except those inside-the-park, for example, are the pitcher's fault. Only some of the runs scored by those who walk, however, are the pitcher's fault. So there's some fielding input on that. But a walk and home run equals all the pitcher's fault.
Probably, the Fielding Experts will say I'm not being very original here. But I hadn't thought of it that way before.
Showing posts with label Defensive Winning Percentage. Show all posts
Showing posts with label Defensive Winning Percentage. Show all posts
Thursday, 11 June 2009
Tuesday, 9 June 2009
Draft Day & Unfinished Business
Today is the amateur baseball draft. In the olden days, one didn't know much about it until the next day's papers. Then, mlb.com started offering the audio of the conference call. I loved that. Great soundtrack for a June afternoon. Now it's one big party. Can't we go back to the olden days?
I also owe this blog some links to very interesting work related to defensive statistics that appeared on the Web recently. It's made me look back over my Defensive Winning Percentage method. I'll get round to it, I promise.
I also owe this blog some links to very interesting work related to defensive statistics that appeared on the Web recently. It's made me look back over my Defensive Winning Percentage method. I'll get round to it, I promise.
Friday, 18 July 2008
Defensive Winning Percentage - Rivals
There are a couple of other well-known defensive metrics that attempt to implement a zone on seasons without a Zone Rating. One is TotalZone, developed by Sean Smith (AROM at Baseball Think Factory), and the other is Dan Agonistes' Simple Fielding Runs.
Both of these estimate a Zone Rating using the Retrosheet play-by-play data, which only goes back as far as 1953. That's fine, but with the current rate at which new seasons are added means we have to wait another thirty years or so to get all Major League history covered. I don't think I can wait that long, as I'll be near 80 by the time we get there. And I don't expect to live to such a ripe age.
In the meantime, we need another system, which is where I think my Defensive Winning Percentage wins out. With a few tweaks, I think I can improve it for the pre-1954 period, when it is hampered by not knowing the number of doubles and triples a pitching staff has given up.
Both of these estimate a Zone Rating using the Retrosheet play-by-play data, which only goes back as far as 1953. That's fine, but with the current rate at which new seasons are added means we have to wait another thirty years or so to get all Major League history covered. I don't think I can wait that long, as I'll be near 80 by the time we get there. And I don't expect to live to such a ripe age.
In the meantime, we need another system, which is where I think my Defensive Winning Percentage wins out. With a few tweaks, I think I can improve it for the pre-1954 period, when it is hampered by not knowing the number of doubles and triples a pitching staff has given up.
Wednesday, 16 July 2008
A Defensive Metric 4
So far, I've identified these flaws with my Defensive Winning Percentage:
1) The statistics will, by their nature, include plays made on balls out of a fielder's zone. This is not necessarily a good thing.
2) The Positional Adjustment, being based on traditional Zone Rating, doesn't accommodate flaw (1). Thus, one ends up with estimated Zone Ratings in excess of 1, which is a logical nonsense. What I should do is adjust the positional factor by the distribution of plays on each team, but that opens up a nightmare world of a lot of work. Another option would be to cap the estimated Zone Rating. Or you could leave it, and bear it in mind when looking at some of the data.
3) There's no adjustment for groundball tendencies of pitchers, nor handedness of batters. This is quite conscious, in part because I don't see the point of doing this for a metric trying to measure what has happened, as opposed to talent level. In other words, I think a player whose pitching staff is pitching to his strengths, deserves credit. I may yet incorporate these factors, but at this stage I get the impression that on the whole the measurements even out.
4) It really is only what I'd call reasonably reliable as far back as detailed statistics from Retrosheet go, including the all-important doubles and triples given up by a team's pitchers. Without those, I'm more likely to believe conventional wisdom where my system challenges it.
I intend to stick with this unsatisfactory metric for the time being. I think it does a good job at giving us an idea of the relative ability of a team's fielders at a given position for an era before modern play-by-play metrics. That's all I ask of it, and that's all I ask you to ask of it.
1) The statistics will, by their nature, include plays made on balls out of a fielder's zone. This is not necessarily a good thing.
2) The Positional Adjustment, being based on traditional Zone Rating, doesn't accommodate flaw (1). Thus, one ends up with estimated Zone Ratings in excess of 1, which is a logical nonsense. What I should do is adjust the positional factor by the distribution of plays on each team, but that opens up a nightmare world of a lot of work. Another option would be to cap the estimated Zone Rating. Or you could leave it, and bear it in mind when looking at some of the data.
3) There's no adjustment for groundball tendencies of pitchers, nor handedness of batters. This is quite conscious, in part because I don't see the point of doing this for a metric trying to measure what has happened, as opposed to talent level. In other words, I think a player whose pitching staff is pitching to his strengths, deserves credit. I may yet incorporate these factors, but at this stage I get the impression that on the whole the measurements even out.
4) It really is only what I'd call reasonably reliable as far back as detailed statistics from Retrosheet go, including the all-important doubles and triples given up by a team's pitchers. Without those, I'm more likely to believe conventional wisdom where my system challenges it.
I intend to stick with this unsatisfactory metric for the time being. I think it does a good job at giving us an idea of the relative ability of a team's fielders at a given position for an era before modern play-by-play metrics. That's all I ask of it, and that's all I ask you to ask of it.
Tuesday, 15 July 2008
A Defensive Metric 3
Having reached a total for Runs Saved Above Average for Washington Senators shortstops in 1970, we can use this to calculate a Defensive Winning Percentage.
Offensive Winning Percentage is a tried and tested sabermetric stat that uses a player's Runs Created to make an estimate of how many games a lineup of nine players with that Runs Created would win. You can find one set of Offensive Winning Percentages at each player's page on baseball-reference.com Ed Brinkman's page, under the Special Batting heading, shows his 1970 OWP was .394. That means a team of Ed Brinkmans would win about 63 games in a 162-game season. Let's look at whether his glove made up for that.
The first step is to calculate how many runs an average team in that league scores. In 1970, 12 teams scored 8109 runs, so that means the average team scored 675.7
Subtract the number of Runs Saved Above Average. That was 24.7 as we saw in the last post. (If it had been negative, as was the case with the Cleveland Indians' shortstops in 1970, we would add it.)
675.7 - 24.7 = 651
Now, with Offensive Winning Percentage, you'd take the runs created for a team of Ed Brinkmans, and divide it by the sum of Ed Brinkmans plus the league average, all raised to the exponent 0.83. That would represent the runs scored divided by the runs scored plus the runs allowed. We do the same thing, but in this case it is the runs allowed that is the variable. Eg,
Average Team's Runs Scored = 675.7
Team with Washington SS Runs Allowed = 651
675.7^0.83/(675.7^0.83 + 651^0.83) = .508
Therefore, the Washington shortstops, backing up a league average offense, would win about 82 games. If you added 82 and 63, and divided by two, you'd get a reasonable estimate that a team of 1970 Washington Senators' shortstops, batting like Ed Brinkman, would win about 72 games. No way is Ed Brinkman going to carry you to the pennant, but he was a better player than his team, which could only manage 70 wins.
However, when you compare the Senators' shortstops to the rest of the league, you can see that, starting from the mean point of DWP's, .501, Washington, with Minnesota, had the best fielding shortstops in the league. The Twins' Leo Cardenas was probably the proper Gold Glove winner that season, as his fielding percentage was slightly better, .978 to Brinkman's .974 .508 is excellent for DWP, but it isn't historically remarkable. You really need to push your DWP past .510 to get into remarkable territory.
Oh, who won the AL Gold Glove that year? It was Hall of Fame shortstop Luis Aparicio, then playing for the White Sox. Their shortstops managed a DWP of only .498, so Cardenas and Twins fans have a good case for feeling aggrieved.
I'm going to look next at a couple of problems with DWP, which I think is good, but with a couple flaws that always need to be borne in mind when using it.
EDIT: Edited to correct the omission of the Exponents!
Offensive Winning Percentage is a tried and tested sabermetric stat that uses a player's Runs Created to make an estimate of how many games a lineup of nine players with that Runs Created would win. You can find one set of Offensive Winning Percentages at each player's page on baseball-reference.com Ed Brinkman's page, under the Special Batting heading, shows his 1970 OWP was .394. That means a team of Ed Brinkmans would win about 63 games in a 162-game season. Let's look at whether his glove made up for that.
The first step is to calculate how many runs an average team in that league scores. In 1970, 12 teams scored 8109 runs, so that means the average team scored 675.7
Subtract the number of Runs Saved Above Average. That was 24.7 as we saw in the last post. (If it had been negative, as was the case with the Cleveland Indians' shortstops in 1970, we would add it.)
675.7 - 24.7 = 651
Now, with Offensive Winning Percentage, you'd take the runs created for a team of Ed Brinkmans, and divide it by the sum of Ed Brinkmans plus the league average, all raised to the exponent 0.83. That would represent the runs scored divided by the runs scored plus the runs allowed. We do the same thing, but in this case it is the runs allowed that is the variable. Eg,
Average Team's Runs Scored = 675.7
Team with Washington SS Runs Allowed = 651
675.7^0.83/(675.7^0.83 + 651^0.83) = .508
Therefore, the Washington shortstops, backing up a league average offense, would win about 82 games. If you added 82 and 63, and divided by two, you'd get a reasonable estimate that a team of 1970 Washington Senators' shortstops, batting like Ed Brinkman, would win about 72 games. No way is Ed Brinkman going to carry you to the pennant, but he was a better player than his team, which could only manage 70 wins.
However, when you compare the Senators' shortstops to the rest of the league, you can see that, starting from the mean point of DWP's, .501, Washington, with Minnesota, had the best fielding shortstops in the league. The Twins' Leo Cardenas was probably the proper Gold Glove winner that season, as his fielding percentage was slightly better, .978 to Brinkman's .974 .508 is excellent for DWP, but it isn't historically remarkable. You really need to push your DWP past .510 to get into remarkable territory.
Oh, who won the AL Gold Glove that year? It was Hall of Fame shortstop Luis Aparicio, then playing for the White Sox. Their shortstops managed a DWP of only .498, so Cardenas and Twins fans have a good case for feeling aggrieved.
I'm going to look next at a couple of problems with DWP, which I think is good, but with a couple flaws that always need to be borne in mind when using it.
EDIT: Edited to correct the omission of the Exponents!
Monday, 14 July 2008
A Defensive Metric 2
I promised an example of how to work out Defensive Winning Percentage, so let's use Eddie Brinkman, Washington Senators shortstop for 1970, his last season before the Detroit Tigers got him in a trade.
The first part of the process is derived from Charlie Saeger's system of Context-Adjusted Defense. This allocates outs and hits to the infield and the outfield.
The first step is to estimate how many groundball and flyball outs the team allowed for the season.
Groundball Outs = Team Assists - Catcher Assists - 1b DPs - OF Assists.
Flyball Outs = Team Putouts - Team Strikeouts - Team Assists.
For 1970 Washington, we get the following
GB outs = 1942 - 80 - 145 - 28 = 1689
FB outs = 4369 - 823 - 1942 = 1604
Then, you need to estimate the number of hits held against infielders, and then those against the outfielders.
IF Hits = Subtract from the total Team Hits Allowed the total of (home runs + doubles + triples). Take 70 percent of the difference. Divide this by the ratio of GB outs to (outs-strikeouts).
OF Hits = Subtract from the total Team Hits Allowed the number of home runs and the total of IF hits. Take 70 percent of this number.
For 1970 Washington, we get:
IF Hits = ((1375-(139-232-47))*0.7)/(1689/3339) = 344
OF Hits = (1375-139-348)*0.7 = 625
Now you are ready to estimate the Zone Rating.
Subtract from the number of GB outs the number of Team pitcher assists. Add the number IF hits. Take 30 percent of this number (this is a positional adjustment, like the one used in Win Shares) and add it to the number of errors made by Senators' shortstops.Take the result and use it to divide the number of shortstop assists.The result is your estimated Zone Rating for all Washington shortstops in 1970.
((1689-220)+344)*0.3 = 543
543 + 23 = 566
593/566 = 1.05
Multiply the estimated Zone Rating by the average number of plays made at the shortstop position, as calculated by Chris Dial in his Dr Strangeglove piece on Baseball Think Factory, and then by the Run Value he has calculated for plays made at each position.This is the Runs Saved by all Senators' shortstops in 1970.
1.05 * 532 *.753 = 420.63
Following the same procedures, we get a total for the American League in 1970 of 394.39 Runs Saved by all AL shortstops. Subtract the league runs saved from the Washington runs saved.This gives you how much better a Washington shortstop was than an average AL shortstop at saving runs.
420.63-394.39 = 24.7
Brinkman started 157 games at shortstop for Washington in 1970, so at worst we can credit a substantial portion of that quality to him.It's worth noting that in 1968, when Brinkman only appeared in 77 games at short, Washington shortstops managed to save 0.7 runs, so it seems likely that Brinkman is worth even more than 24.7.
You're now ready to calculate the Defensive Winning Percentage.
The first part of the process is derived from Charlie Saeger's system of Context-Adjusted Defense. This allocates outs and hits to the infield and the outfield.
The first step is to estimate how many groundball and flyball outs the team allowed for the season.
Groundball Outs = Team Assists - Catcher Assists - 1b DPs - OF Assists.
Flyball Outs = Team Putouts - Team Strikeouts - Team Assists.
For 1970 Washington, we get the following
GB outs = 1942 - 80 - 145 - 28 = 1689
FB outs = 4369 - 823 - 1942 = 1604
Then, you need to estimate the number of hits held against infielders, and then those against the outfielders.
IF Hits = Subtract from the total Team Hits Allowed the total of (home runs + doubles + triples). Take 70 percent of the difference. Divide this by the ratio of GB outs to (outs-strikeouts).
OF Hits = Subtract from the total Team Hits Allowed the number of home runs and the total of IF hits. Take 70 percent of this number.
For 1970 Washington, we get:
IF Hits = ((1375-(139-232-47))*0.7)/(1689/3339) = 344
OF Hits = (1375-139-348)*0.7 = 625
Now you are ready to estimate the Zone Rating.
Subtract from the number of GB outs the number of Team pitcher assists. Add the number IF hits. Take 30 percent of this number (this is a positional adjustment, like the one used in Win Shares) and add it to the number of errors made by Senators' shortstops.Take the result and use it to divide the number of shortstop assists.The result is your estimated Zone Rating for all Washington shortstops in 1970.
((1689-220)+344)*0.3 = 543
543 + 23 = 566
593/566 = 1.05
Multiply the estimated Zone Rating by the average number of plays made at the shortstop position, as calculated by Chris Dial in his Dr Strangeglove piece on Baseball Think Factory, and then by the Run Value he has calculated for plays made at each position.This is the Runs Saved by all Senators' shortstops in 1970.
1.05 * 532 *.753 = 420.63
Following the same procedures, we get a total for the American League in 1970 of 394.39 Runs Saved by all AL shortstops. Subtract the league runs saved from the Washington runs saved.This gives you how much better a Washington shortstop was than an average AL shortstop at saving runs.
420.63-394.39 = 24.7
Brinkman started 157 games at shortstop for Washington in 1970, so at worst we can credit a substantial portion of that quality to him.It's worth noting that in 1968, when Brinkman only appeared in 77 games at short, Washington shortstops managed to save 0.7 runs, so it seems likely that Brinkman is worth even more than 24.7.
You're now ready to calculate the Defensive Winning Percentage.
Saturday, 12 July 2008
A Defensive Metric
Last night, I was trying to explain, for the first time in public, to a couple of guys I know, my fielding metric for pre-Zone Rating players. I don't think it went very well, so I'll try again here.
You need to go back to Chris Dial's method of converting Zone Rating into Runs Saved, which is the foundation for what I do. In the modern era of Play-By-Play accounts, it's possible to create a Zone Rating to show how many balls hit into a player's fielding zone are turned into outs. However, before 1987 we don't have this information, so we have to figure out a way of estimating it.
Chris's research revealed that there is an average number of fielding chances at a position. For example, at shortstop the average team has 532 chances per season of 1440 innings. This number, I'm told, is apparently relatively static over the seasons. Chris also established a Run Value for each chance. Thus, an average team's shortstops save about 400.6 runs per season.
Since we have total values for all those who fielded at shortstop for Major League teams going way back in the history of Organized Baseball, we can, with a degree of confidence based on the recording of other information, estimate a Zone Rating for each position on a given team in a given season. And from that, we can work out how many runs the whole season's worth of players at a given position had saved.
I'll give a detailed example in the next post.
You need to go back to Chris Dial's method of converting Zone Rating into Runs Saved, which is the foundation for what I do. In the modern era of Play-By-Play accounts, it's possible to create a Zone Rating to show how many balls hit into a player's fielding zone are turned into outs. However, before 1987 we don't have this information, so we have to figure out a way of estimating it.
Chris's research revealed that there is an average number of fielding chances at a position. For example, at shortstop the average team has 532 chances per season of 1440 innings. This number, I'm told, is apparently relatively static over the seasons. Chris also established a Run Value for each chance. Thus, an average team's shortstops save about 400.6 runs per season.
Since we have total values for all those who fielded at shortstop for Major League teams going way back in the history of Organized Baseball, we can, with a degree of confidence based on the recording of other information, estimate a Zone Rating for each position on a given team in a given season. And from that, we can work out how many runs the whole season's worth of players at a given position had saved.
I'll give a detailed example in the next post.
Subscribe to:
Posts (Atom)