Thursday, July 29, 2010

Random Thursday: 1910 Boston Red Sox and why ERA sucks

Some Thursdays just be random. Today is one of those days. So The Common Man used Baseball’s random function to jump from the 1995 American League standings to 100 years ago, and the pitching stats from the 1910 Boston Red Sox, a terrific example of how baseball has changed and why some of our record books may need to be rewritten.

The Sox, like most teams in 1910, relied heavily on its pitching staff, which featured a young Eddie Cicotte and an even younger Smoky Joe Wood. The team finished fourth in the AL with an 82-71 record, and third with a 2.45 ERA.

That 2.45 ERA is deceptive, however. We’re so used to seeing the incredible pitching lines, and low batting lines, of the early 20th century that we’ve taken to calling it the Deadball Era. And while scoring was certainly down in the first two decades of the 20th century, the low ERAs of the Joe Woods, Walter Johnsons, and Ed Walshes are exceptionally deceptive, because they tended to allow almost half as many unearned runs as earned runs. Indeed, while Walsh has a MLB record 1.82 career ERA, his RA/9 is a much more normal (but still excellent) 2.65. Wood had a sparkling 1.91 ERA in 1912, when he won 34 games, but his RA/9 was 2.72. Christy Mathewson had a 1.28 ERA in 1905, he actually allowed 2.26 runs per game. If we account for all the runs he allowed, Matty’s sparkling 2.13 ERA balloons to 3.04 R/9. Between 1901 and 1914, the Major Leagues averaged 2.34 unearned runs per each game played, 28.6% of all runs scored at the time.

And while some of those runs are surely actually the fault of the fielders behind the pitchers, rather than the pitcher himself, in many, if not all of the cases, the pitcher had a hand in his own demise. He allowed baserunners on in the first place, or allowed a hit to drive them in. The Common Man’s feelings on unearned runs (and earned runs, for that matter) are best described by Joe Posnanski, who wrote about unearned runs being a “shaky concept” back in March, “Pitchers don’t prevent runs by themselves. But, for more than 100 years, we have lived in a statistical world where they do, where pitchers are entirely responsible for runs allowed and shutouts and hits per innings pitched … and they cannot be held responsible if some dumb fielder botches the ball behind them.” Likewise, the dumb fielders who botch those balls aren’t solely responsible for them either.

So when we marvel at the work done by Alexander, Young, and Plank and their brilliant ERAs, we’re actually not seeing a realistic picture of the pitcher. And because the play by play data from that era is unattainable, we never will. To judge a pitcher today by this same standard does that pitcher a tremendous injustice.

The fault lies not in our metric, but in ourselves. Indeed, ERA was flawed when Henry Chadwick devised the statistic in the 19th century, but we have adopted it wholeheartedly and without inquiry. We have internalized its meaning without confirmation of its ability to tell us what we want to know. How good was Christy Mathewson? Sadly, ERA does not get you there. FIP gets closer, and certainly a more robust examination of Big Six’s stats demonstrate that he was, indeed, a truly special pitcher.

So what do we do now? As TCM pointed out, we’ve internalized ERA. It’s not going away any time soon. Stats like FIP and WAR are being perfected, and may eventually make their way into the public consciousness. But in the meantime, ERA will continue to be king. In light of this, TCM wonders whether it’s time to rewrite the record book a little. It’s true that the number of errors per team, and the percentage of unearned runs allowed has declined consistently over time. Consider the following graph:

(click to embiggen, TCM's raw data is available here)

The decline has mostly been a gradual process, and has gotten to the point where it’s been down under 10% since 1990, including an all time low of 7.53% of unearned runs allowed in 2009. But there are two watershed moments that drastically dropped the percentage of unearned runs scored. The first occurred in the 19th century, when the use of gloves was popularized. There is also a very significant drop in the number of unearned runs beginning in 1921, as webbing in the gloves seems to have become better and more popular. These changes significantly cut down the number of errors in each season, and therefore the percentage of unearned runs scored.

This 1920 break is a good a place to make a clear distinction between modern pitchers and those of the early 20th century. These pitchers are throwing under vastly different conditions, and playing a much crisper and predictable game. If we must have ERA, we must allow it to be meaningful in a historical context, and it seems clear that the ERAs from 1901-1920 are virtually meaningless.

As such, The Common Man proposes that these are your new career ERA leaders for the modern era (among pitchers with more than 2000 innings):
Hoyt Wilhelm2.52
Whitey Ford2.75
Sandy Koufax2.76
Jim Palmer2.86
Andy Messersmith2.86
Tom Seaver2.86
Juan Marichal2.89
Bob Gibson2.91
Dean Chance2.92
Pedro Martinez2.93

And here are the single season leaders:
Bob Gibson19681.12
Dwight Gooden19851.53
Greg Maddux19941.56
Luis Tiant19681.60
Greg Maddux19951.63
Spud Chandler19431.64
Dean Chance19641.65
Carl Hubbell19331.66
Nolan Ryan19811.69
Sandy Koufax19661.73

(brought to you by Baseball's invaluable Play index)


Anonymous said...

Nice work. Good chewing matter. Which led me to the following thought: Conversely, errors would have been a reliable part of any offense. Yet the benefit is a team one and the individual(batter) is punished while providing something useful. Doesn't this same effect call into question batting averages for pre-1920?

How could one unwind the numbers? Total bases? Something like that?


G.Wo said...

This is also not even considering how much of an effect official scorers have on the whole thing as well. How much more likely/unlikely are official scorers (hired by the home team) to rule marginal plays as errors? This data probably exists in some form, I'm just too lazy to look it up.

This was a great read.

Jay W. said...

Very interesting article. One minor quibble: I think Bob Gibson's 1.12 ERA was in 1968, not 1989.

The Common Man said...

Thanks, Jay. TCM fixed it. Don't know what happened there, except that it was late and the mistake was buried in a lot of HTML code.

Anonymous said...

Would be interesting to see the same lists in terms of run average (both earned and unearned), both post-1920 and for all eras. Great article.

Mark said...

I think the 1921 drop also has to do with outlawing the spitball and giving fielders a clean, white, round, non-slimy ball to catch and throw.

Anonymous said...

Embiggen, huh? It's a perfectly cromulent word.

As for the article, great read. There are just so many differences in today's game than there was 100 years ago: how the ball is made, night games, PEDs, the use of spitballs and other now illegal pitches, the cloth in the uniforms, the bats and gloves are different, air travel vs. train and bus travel, longer seasons, less media coverage, more guys named Rabbit and Mordecai.

You can go on and on, but comparing the different eras won't do much other than start some arguements as to which era had the better nicknames.

Anonymous said...

Doesn't ERA+ deal with this very issue? Unearned runs are just one reason why you can't compare the rates from era to era. I don't see why errors pose a particular problem.

Anonymous said...

It really looks like the unearned runs settled out around 1948. What happened then? Is that just post-war?

The Common Man said...

ERA+ does not address this issue at all, actually. ERA normalizes ERA for ballpark and era, but does not address the overall question of runs scored. You may be thinking of FIP and xFIP.

The Common Man said...

To the first anonymous commenter:

You're right that batters in the early part of the 20th centry, but even moreso in the 19th century, helped their teams immensely by getting on base via error. I've written before about batting average being a misleading statistic in the 19th century in particular, and would argue that the best statistic we have to measure hitter performance in the 19th century may actually be runs scored, though that metric is still not adequate.

Unfortunately, there's no way to unwind those numbers unless you've got a time machine in your garage. Reached on an error does not count in total bases and errors were so commonplace during these eras that there's no reason any popular accounts should have taken note of them. Personally, I think it may just be time to lop some of these numbers off, and just consider them separately.