Are Major League Umpires Prejudiced?

Umpire Tony Randazzo and pitcher Ray King

That's the question supposedly answered by a new study by Daniel Hamermesh, a economics professor at the University of Texas. He claims that umpires are inclined to give pitchers of the same race an extra strike now and then -- but you have to really pay attention to notice, because it only happens 1% of th time. From Time.com:

In the new study, Hamermesh's team analyzed the calls on 2.1 million pitches thrown in the Major League between the 2004 and 2006 seasons. Controlling for all other outside factors, such as the pitcher's tendency to throw strikes, the umpires' tendency to call strikes and the batter's ability to attract balls, researchers found evidence of same-race bias - and the data revealed that the bias benefits mostly white pitchers. Not surprising, since 71% of MLB pitchers and 87% of umpires are white.

[...] Interestingly enough, Hamermesh's research found that the race of the batter didn't seem to matter - the correlation was only between the pitcher and the home-plate ump. Rich Levin, an MLB spokesman, refused to comment on the research findings.

I'm confused. Assuming this bias is true (which I'm not actually ready to do), why are umpires predisposed to reward pitchers? If a white umpire is calling a game with a white pitcher on the mound, how come a white hitter gets screwed?

There's also this interesting fact:

Though his research confirms that bias exists, Hamermesh says it can be easily reduced or eliminated. When a game's attendance is particularly high, when the call is made on a full count or when ballparks use QuesTec, an electronic system that evaluates the accuracy of umpires' calls after the game, the biased behavior disappeared, according to the study. "The umpires hate those [QuesTec] systems," Hamermesh says. "When you're going to be watched and have to pay more attention, you don't subconsciously favor people like yourself. When discrimination has a price, you don't observe it as much." Right now, the QuesTec system is used in 11 of MLB's 30 ballparks, mostly in the American League.

So the total study included 2.1 million pitches, but the bias only exists in games played in 19 parks, only on days the park isn't full and only early in the count ... and even then, it still only appears 1% of the time. After all of those qualifiers, is the sample size of pitches thrown still relevant anymore? (Update: I was probably wrong on that last point -- see explanation below.) Or is that 1% difference just statistical noise, completely meaningless until a researcher with a specific agenda attempts to define it?

I honestly don't know. Those are the questions other researchers should be asking. Maybe after this study goes through a peer review (which, as David Pinto of Baseball Musings points out, hasn't happened yet) I might give it some credence.

On a side note, Hamermesh says he was inspired to conduct the study after hearing about the study claiming there was racial bias among NBA refs. Unfortunately, what neither Hamermesh or Katie Rooney, the author of this Time article, never mentioned was that the NBA study was publicly refuted by both the NBA and the players. One of the biggest holes in that study was that the researchers didn't use data of specific calls made by specific referees, instead comparing the aggregate number of calls in a game with the racial makeup of the crew of referees assigned to the game.

It's irresponsible to mention a study without also disclosing that the results were at the very least disputed, if not wholly incorrect. But, then again, an article such as this is the print equivalent of talk radio -- purposely controversial and short on facts or differing perspectives. The goal is to simply get attention, not inform. To that end, it's obviously been successful.

Update: I received an email from Andrew G., a reader with far more experience in statistical studies than I, who explained that I probably made incorrect assumptions in discrediting the study:

With regards to the umpire being less biased on pitches made in parks with Questec, in full stadiums, and on full counts, Hamermesh wouldn't have been able to simply throw those pitches out of his sample. (How would he have known those pitches weren't biased?) Instead, he likely ran the study on all 2.1 million pitches (and I'm guessing this was a regression analysis) and found that the coefficients for pitches thrown on full-counts, in Questec parks, and in full stadiums leveled the bias to essentially zero (the size and sign of the coefficients would depend entirely on what his dependent variable was. Having not actually read the study, I won't venture a guess).

In other words, the sample remained all 2.1 million pitches, and even with those non-biased pitches, he found a bias 1% of the time. If you threw all of those pitches out of the sample (which is paramount to self-selecting your own statistics), the bias would probably have been much higher.

Even more proof that I made the right decision to be an English major instead of a statistician. That said, I'd still like to hear what other qualified researchers have to say about the study before I accept any of the assumptions it claims to make.

Another update: Is the sample size of umpires too small for the results to be significant?