VII.  Returning to Effective Runs Allowed

We are dealing here with the issue of whether some pitchers may have an ability to pitch especially well when they have an opportunity to win the game.  We were working before with John Burkett, Mike Morgan and Whitey Ford, and let’s throw Jim Clancy and Tom Seaver into the mix as well—Seaver, for obvious reasons, and Clancy, because he is the third man in the Burkett/Morgan comparison:

Working with one run of offensive support, Burkett’s teams were 3-30, Clancy’s were 3-37, Morgan’s were 3-39, Ford’s were 10-28, and Seaver’s were 12-56:

Except that that’s not exactly the method that I used in some cases.   That’s exactly the method I used for Clancy, Seaver and Morgan above, but not for Ford and Burkett.

We get into problems with small groups of data.  Suppose that a pitcher went 0-12 when working with one run, or 0-1, or 0-39.  What is his runs allowed rate?   By the method above, you can’t get an answer.   You wind up searching for the square root of infinity, and, of course, only the Dalai Lama and Al Gore know what is the square root of infinity.

To prevent this from happening, I added “placekeeping data” to all groups of games involving less than 40 decisions.   At one run, since the winning percentage of teams with one run was .101, I added one (placekeeping) win and nine losses—treating John Burkett as if he was 4-39, rather than 3-30, and Whitey Ford as if he was 11-37, rather than 10-28.    These additions change the chart above to the following:

I added these “placekeeping numbers” at all levels, to all pitchers who had less than 40 decisions at the level, always adding one “placekeeping win”, but varying the number of losses as follows:

These are the values for the five pitchers at the levels 2 runs and 3 runs:

Whitey Ford is doing really well here.   Some people tend to think of Whitey Ford as a pitcher who won a lot of games because the Bronx Bombers scored seven runs a game for him.  As the charts above show, this is anything but true.   Ford won a lot of games 2-1, 3-1 and 3-2.   This chart summarizes the three charts above:

In the other method, Seaver was number one because I was measuring the total distance away from an average pitcher, which gave Seaver an advantage over Ford because he had pitched 75 to 80% more games within our data.   But on a per-game basis, Ford is more than holding his own.  (The “Effective Runs Allowed” above are not actually integers, and may not add up exactly as you might expect for that reason.)

At the four-run level Seaver bests Ford and the three musketeers go Burkett-Morgan-Clancy, but the cumulative order stays the same:

Ford was the only one of these pitchers—and one of few pitchers in the study—who was able to deliver a winning record for his team in games with four runs or less.   At the five-run level, Whitey Ford’s teams were a fairly astonishing 36-3.   Everybody was able to win over half the time with five runs, but Morgan’s teams were only 29-23:

In the first of this series of articles I mentioned that Burkett was 18-4 when working with six runs, whereas Morgan was 16-4, which I represented as an advantage for Burkett.   But actually, when we look at the team record, rather than the individual pitcher’s won-lost record, Morgan beats Burkett at six runs.   And Whitey Ford once again beats everybody:

At the seven-run level Jim Clancy’s teams were 20-2, which is even better than Seaver’s teams.   But Ford’s Yankees went 27-2:

We come, finally, to our last category, which is “eight runs or more”.  There’s not much point in tracking it beyond eight runs because the winning percentages are getting so close to 1.000, but there is a difference between the “eight-run” category and the others, which is that up to now we have known exactly how many runs the team was working with on offense.   We will assume that the offense, in those games in which it scores eight runs or more, averages nine runs. 

Jim Clancy with eight runs or more was 40-1:

OK, I’ve got some housekeeping details to take care of here, but first let me say:  I think that we have effectively demonstrated at this point—unless I am missing something--that Mike Morgan was not as effective a pitcher, in terms of delivering a win, as were Clancy and Burkett.   In terms of ERA and runs allowed per nine innings, he was just the same.   In terms of his effectiveness at delivering a win, given a certain number of runs to work with, he was not the same.   We have done that, and we have created a framework for measuring the difference—measuring the costs to Mike Morgan’s teams of his failure to respond to the situations.    We’ll pick up on those things in a moment.

The two housekeeping issues that we need to deal with are

1)  What do we do with shutouts? and

2)  Why are the “Effective Runs Allowed Rates” so low?

To this point we have not dealt with games in which the pitcher’s team was shut out, because, if your team doesn’t score any runs, it doesn’t make any difference how well you pitch, you can’t win.    What can we do about these games?

There seem to be four options, which are:

1)  To assume that every pitcher had the same level of effectiveness in these games—let’s say 4.50 runs per game or something like that,

2)  To assume that every pitcher had the same level of effectiveness in these games that he had overall,

3)  To ignore them entirely, and

4)  To ignore them entirely, except that we display them as losses.

I think the best option is (4)—to ignore them entirely, except that we display them as losses in the totals, so that we are displaying accurate totals of wins and losses for the starter’s games.

The other question is, why are the Effective Runs Allowed Rates so low?   Mike Morgan, after all, allowed 4.65 runs per nine innings in real life.   We are saying here

1)  That he was ineffective at pitching well when he had a chance to win, and yet

2)  His “effective” runs allowed rate was 3.99.

What up with that?

Everybody is low.   The reason for this is that the distribution curve of runs scored in a game (or runs allowed in a game) is asymmetrical.   If teams average 4.00 runs in a game, they will never score less than zero, but they will sometimes score more than eight.   That means that there have be more games with less than four than games with more than four, in order to re-balance the system at four.

The Pythagorean formula assumes that, when a team averages four runs a game, this is an average.   In our study it is not an average; it is a constant.   The “four” runs average for the four-run group is 4, 4, 4, 4, 4.   But there will be more “opposing” games under 4.00 than over four, which means that there will be more wins than losses.   If you take two teams which both average four runs a game head to head, but one team always scores exactly four but the other scores a varied number averaging four, the team that always scores four will win over half the time.  The same is true at every level of offense, including one run.   If a team always scored one run in every game and allowed an average of one run per game but in a varied pattern, they would win more than half their games.  

This skews our calculations toward a lower-than-real-life effective runs allowed rate, and we’ll need to adjust for that.  It’s actually kind of interesting how it happens.   You remember this chart, which I presented in the second of this series of articles?

Based on that chart, we can figure what the “effective runs allowed rate” being calculated at each level of offensive effectiveness is:

Again, when teams score 8 runs or more, we assume that they have scored an average of nine.

Anyway, the average runs allowed rate in this study was about 3.50 runs per game, whereas it should have been about 4.40.   It’s 21.4% low.   We can correct for this, then, by multiplying the Effective Runs Allowed Rates calculated before by 14, and dividing by 11.  

OK, picking up the “summary” chart before, but incorporating those two changes, this would be the data for the five gentlemen that we have been following here:

Notes:

1) Tom Seaver’s Effective Runs Allowed Rate here is actually higher than his real-life runs allowed rate (3.15 runs per nine innings).   Seaver ranked first in our earlier competition because Seaver pitched more games with a low ERA than any other modern-era pitcher—thus, he is going to do very well in any kind of a runs-allowed rate competition among modern pitchers.

2)  Whitey Ford’s Effective Runs Allowed Rate is the lowest of any pitcher in our study with a reasonable number of starts.  In the last of this series of articles, posted tomorrow, I will show the data for all 663 pitchers with 100 or more starts in the Retrosheet data.   Ford is number one on the list; Bryan Rekar is number 663.

3)  Jim Clancy surged at the last minute in our data, pushing ahead of John Burkett.  People assume that Mike Morgan pitched for terrible teams, although he actually didn’t, on balance.   Jim Clancy pitched for worse teams than Morgan did, but Clancy had a better won-lost record.    If Clancy had matched the won-lost percentage of his team in every season, his career record would have been 154-161 (.489).

I’m not entirely happy with Clancy’s evaluation.   Clancy’s teams went 40-1 when they scored eight runs or more, which creates a very low Effective Runs Allowed Rate for him in those games, but if his team’s had gone, say, 38-3, then we would have a very different calculation for him in those games, driving his overall final figure up by 15 points to 4.55.   The system allows a disproportionate impact of a very few games in that case, which it should not do; I should have devised some way to prevent that from happening.   Charts sometimes act irrationally near their boundaries; you probably all know this.  

VIII.  Final Thoughts

In saying that Mike Morgan did not pitch well when he had chances to win ball games, in saying that Whitey Ford and Bob Gibson did, I am not offering a moral judgment.   We have not established, and I am not saying, that these performance deviations are beyond what could have occurred by chance.

But I am saying that these performance anomalies are “real” in this sense:  that wins and losses resulted from them, and that therefore we can and should appropriately consider this in evaluating these players.

Mike Morgan, I think, is shown as being 36 runs below average as a pitcher for his career.   The real number, I would argue, is closer to minus 150.   Morgan allowed 4.65 runs per nine innings—but he won ballgames with the consistency of a pitcher allowing 5.08 runs.  That’s a difference, over the course of his career, of 132 runs.

That completes the written part of this analysis.  In tomorrow’s installment we’ll list the data for all pitchers with 100 or more career starts in the Retrosheet data.