Monday, April 1, 2013

Pythagorean Wins for Managers

Pythagorean Wins is a very real concept developed by Bill James to attempt to measure how well a team did using the runs scored. Since it compares very well with actual results, it’s used as a marker to measure “luck” or other supposedly intangible elements of the game. After much consideration, I will contend it isn’t luck as much as something else far more basic.

The Pythagorean Wins formula is:
Runs2/(Runs2 + Runs Allowed2)
Using the 2012 Chicago Cubs as an example, we see that they scored 613 runs and gave up 799. Using the formula, we would expect their Win-Loss record to be:
6132/(6132 + 7992)                 375,769/(375,769 + 638,401)                .371 * 162 games ≈ 60-102
The Cubs finished 61-101, so somehow the team “exceeded” its expectations, if such a term can be used on a 100-loss team. These numbers fluctuate from year to year, which is why it is common to hear Pythagorean Wins referred to as the “luck factor.” If Pythagorean Wins can be applied to a team (and it appears to work well, and not just in baseball), can we then apply it to managers and see how much better or worse they did than expected?

This chart shows the top 40 managers whose careers began after 1950 (Hall of Famers highlighted in yellow), with the following column explanations:
R=Runs RA=Runs Against PW% the Pythagorean Win percent EW extra wins over predicted. For fun, the last columns refer to how the manager did in 1-run games, with the last column being the percentage of games that the team was in first.

The best of the lot appears to be Bruce Bochy, at about 30 wins over his expected win total. However, to attain that total, he had to manage almost 2900 games, making it appear that he (or his teams or a combination of both) were able to outperform in 1 game out of 100. To see how this looks for an individual manager, lets view Tony LaRussa, the man who can recite the alphabet backwards with the best of them:

Up and down and up and down and up and down…There’s no pattern to whether the teams outperformed expectations or not, which is the main reason why no manager has Pythagorean Win totals that are very high, and possibly the main reason why Pythagorean Wins aren’t shown as any kind of manager statistic. It took me awhile, and people smarter than me have probably known this for years, but the seeds to this lie in some simple facts. This table shows the winning percent of teams by runs scored from 1950-2012:
















From 1950-2012, the average runs scored is 4.4, and yet teams still manage to win games when they score as few as 1 run—not many, but it’s still 1 in 10. In the end, that’s what Pythagorean Wins ends up measuring—how well a team does when they DON’T score a lot of runs. Should the manager be credited with this? Sure, he deserves some recognition, but no one credits Jim Leyland with being a genius for running Justin Verlander out every fifth day—he’d be an idiot not to. But no matter the era, a low-run or high-run era, how a team does when they score below average runs is what determines whether they’re on the plus or minus side of the Pythagorean Wins ledger.


The Cost of an Error

Not all errors are created equal. What happens when a shortstop boots a ball but the runner doesn’t subsequently score? What happens when a third baseman juuuuuuuuuuuuuust misses reaching that ball? Combing through play-by-play data can tell us if individual errors result in runs, but there’s a simpler, if less-than-perfect method to do it. We’ll begin with defining how many runs translates into a win, which has been well-established by Bill James and others. Simply divide the runs scored in a given year by the wins. For 2012, this would mean 21,017 runs divided by the 2430 wins, giving a value of approximately 8.65 runs equaling a win. This chart shows the historical value from 1901-2012:


The historical value has always been considered to be near 10 runs per win, with a historical value of 8.85. The higher the value, the greater the level of offense, and the lower, the more that pitching is dominant. This works well when viewing data in terms in these terms (all years approximate):
1901-1919                            The Dead Ball Era
1920-1945                            The Lively Ball Era (the dip at the end is due to World War II)
1946-1960                            The No Label Era
1961-1993                            The Expansion Era (with a Mini Dead Ball Era from 1962-1968)
1994-2007                            The “Enhanced Offense” Era
2008-                                     The “Back To Normal” Era

2012 crossed underneath that 112-year average line, and when that occurs, other facets of the game, like base running and defense come to the front. With regard to defense, if we can establish the number of runs equals a win, we can also see how many errors equal a run in a similar fashion, which is simply unearned runs/errors. Before I show that, this next chart shows the percent of runs that were unearned from 1901-2012:
















This is one piece of evidence that defense DOES continue to improve, either through better technique, better positioning, or more likely a combination of that and other factors. Despite the rapid improvement from 1901-1920, from 1920 on the number of unearned runs has decreased by a factor of 2, which is tremendous. This next chart shows the ratio of unearned runs to errors:

















Most of the fluctuation is due to the scale—in this 100+ time span, the value of an error lies between .55 and .65 runs, or in other words, every two errors will lead to a run. Of course, this isn’t game-specific, but it helps see what errors can cost a team.

Base Running Errors



The chart that follows shows the number of base running (I can find no definitive source that tells me whether the term is correctly written as “base running,” “base-running” or “baserunning”—I’ve chosen what I’ve chosen) miscues, which can be grouped by:
1.       Being thrown out trying to advance
2.       Any fielder’s choice
3.       Force outs are NOT included—a runner on 1st isn’t responsible that the hitter drilled one right at the shortstop
4.       Base stealing and pickoffs are also NOT included—this table only shows how many times a team had miscues on the base paths.

It’s surprising how few outliers (definitely the Angels on the high side and perhaps the A’s and Giants on the low side) there are. The White Sox sit firmly in the middle—there’s no doubt that in this period (or at least in 2010-2011) that the Sox were a bit more aggressive on the base paths in terms of stealing attempts and bunts, but as far as basic base running goes, they were pretty run-of-the mill. The take-home message is that a base running miscue will occur about once every three games—the question is whether that is a large number or small one.
 

What Happens When the First Batter of an Inning Gets on Base?

Just as the swallows return to Capistrano every March 19th, there are comments that can be counted on in a sports broadcast:
1.  In football, when a team has a 2nd-and-1 situation, the color guy will state "This is a great passing situation," completely ignoring that since 2000, teams runs the ball 70% of the time in this situation. (click here to see for yourself for 2000-2012 seasons)
2.  In basketball, you'll hear "You can't coach height"--yes you can, or else every 7-footer would have outlandish success in the NBA. They don't.
3. When the leadoff hitter in an inning gets on base, you'll hear "It's tough on a pitcher when that leadoff hitter gets on." With play-by-play data, that notion is easily measured, and this chart shows how that has worked out in the 2010-2012 seasons:








To explain, in 2010 there were 43,441 inning leadoff plate appearances (the number varies due to whether home teams bat in the bottom of the 9th or if a game went into extra innings). Of those, the leadoff hitter hit a single 6,814 times and eventually scored 2,556 times, or 37.5% of the times the leadoff hitter got on base with a single. As would be expected, hitting a double significantly increases those chances to 58.5%, hitting a triple to 83.3%. Walks and hit-by-pitch are similar to singles (as would be expected), and in total, the announcer is pretty darn close to being correct—when 40+% of inning leadoff hitters that get on base eventually score, it IS difficult on pitchers.

I was personally surprised by the triple number—this means that around 15-20% of leadoff triples DO NOT SCORE, which to me is an astounding number. This means that teams AT WORST had two sacrifice opportunities and weren’t able to capitalize. I looked at individual teams to see if any teams jumped out, and nothing really did—Arizona (19 of 29, 65.5%) and Tampa Bay (22 of 22, 100%) were the extremes, but every other team was pretty much in that 80-85% range.