By Roger Weber

For most data in baseball we like to use measurements based on
runs. Runs are usually a pretty good indicator of wins. So if we can quantify player wins based on their run production or
ability to prevent runs we may be able to create a near perfect form of comparison. But more basic statistics, like pitching
wins, are not based on true statistical production. They are based on rules that define the statistic.

To earn a win a pitcher must pitch at least five innings. His
team must have the lead before another pitcher replaces him. And that lead must be good enough for the win.

In theory this sounds like a decent indication of how a pitcher
can earn a win for his team. But take an individual case- the famous incident with Waite Hoyt,
who, in a game on September 22^{nd}, 1927 pitched eight innings of shutout baseball against the Tigers at Yankee Stadium
and left the game with a 7-0 lead. Since the pennant was already wrapped up, the Yankees brought in Babe Ruth to pitch the
ninth. Ruth gave up seven runs and then in the bottom of the inning hit a home run to give the Yankees an 8-7 win. The winning
pitcher: Ruth.

How about another hypothetical situation?
A pitcher strikes out the first 14 batters he faces (4 2/3 innings). Then he gets hurt and is taken out of the game having
given up no runs but having gone just 4 2/3 innings (not eligible for a win). During that time his team has scored eleven
runs, but after the pitcher leaves scores no more. For the rest of the game the team uses 5 relievers, each of whom gives
up 2 runs. Final score: The pitcher's team wins 11-10. So who gets the win? It's up to the scorekeeper, but the one assurance
is that it can't be the starting pitcher. Didn't he pitch most effectively for the longest period of time? But rules can't
be broken.

Pitching wins, while they usually give
a general idea of a pitcher's performance, are not a valuable tool for looking at a player's performance or ability. In 2005
Roger Clemens pitched very well but his team failed to give him any run support in nine games. In many of those he was credited
with a loss despite having pitched a well above average game. If a pitcher pitches nine innings and gives up even one run,
if his team does not score he is the "losing" pitcher.

Pitching wins and losses do not depend
so much on the pitcher but on the offense, especially the offense of the pitcher's team. Wins depend greatly on run support
and simple random timing. They can be distorted by bad performances by relief pitchers, who, if they blow the lead the starting
pitcher helped to gain, often cost the starting pitcher his "win".

Statistically wins and losses hold
little value. They best correlate with each other and with little else. In 2004 the "r^2" between the two was -.50, meaning
as wins increased, generally losses decreased. This makes sense because a winning pitcher is generally not losing. The number
is not higher because of variance in numbers of games pitched. 7 wins is not a high total but if the pitcher only pitched
eight games he has only one or zero losses, also a low number.

Wins correlate with strikeouts and
inning pitched weakly but noticeably. They also had in 2004 a -.16 correlation coefficient with earned run average, meaning
wins do not very well tell a player's ERA, generally viewed as the best pitching statistic, and ERA does not predict a player's
win total, in part because ERA has an average around 4.3 no matter how many innings it measures. Wins have a .15 "r^2" with
innings pitched but have virtually no noticeable relationship with complete games, shutouts or walks allowed.

** **