How judges, loan officers, and baseball umpires overcompensate for past decisions

The actual title is “Decision-Making under the Gambler’s Fallacy” (pdf) and the authors are daniel Chen, Tobias J. Moskowitz, and Kelly Shue.  Here is one short bit from what is more generally a very interesting paper:

We test our hypothesis in three high-stakes settings: refugee court asylum decisions in the US, a field experiment by Cole et al. (2013) in which experienced loan officers in India review real small-business loan applications in an experimentally controlled environment, and umpire calls of pitches in Major League Baseball games. In each setting, we show that the ordering of cases is likely to be conditionally random. However, decisions are significantly negatively autocorrelated. We estimate that up to 5 percent of decisions are reversed due to the gambler’s fallacy.

To make that more concrete, if a baseball umpire first calls a ball, the next pitch he is more likely to then call a strike.  Of course this may plague your paper refereeing decisions, whether or not you finish your next book, and your dating life.

The original pointer was from Cass Sunstein on Twitter.


There is a signalling interpretation that is different than the gambler's fallacy. This would say that decisions are negatively correlated to increase the chance that third parties see you as objective. In that case, you wouldn't expect to see this effect in private decision making like your dating life (unless isn't not private) or paper refereeing (except maybe if you care about the editor). From the paper:

> We believe that concerns about external perceptions could be an important driver of decisions. However, they are unlikely to drive the results in the context of loan approval, which is an experimental setting where payouts depend only on accuracy.

This doesn't really seem to agree with the fact that negative autocorrelation decreased when there were stronger incentives for accuracy. If people really believe the gambler's fallacy, why would they stop believing it when given higher payouts?

This is a good comment. Though it seems like it should be more important the more public the role. The loan officer data comes from an experiment - maybe it would differ from real loan officers who might work more in private, where no-one could observe the order in which they consider candidates.

The Caldecott Medal (awarded annually for children's book art since 1938) and Newbery Medal ( awarded annually for childrren's book literature since 1922) exhibit the same characteristic. This, in spite of the fact, that the jury for each prize each year is composed of different individuals. The composition of the jury and the process of their deliberation has been, until recently, secret. ( Their names are now known retrospectively). They do seem to read or examine every book any publisher feels has a chance of winning.
For example, Bill Peet's Autobiography (1990 Caldecott Honor Book) which appeared after he had retired from making picture books ( more than 30 from 1961-1986 ) was premeated not for its merit but retrospectively, for having missed all the others.

> if a baseball umpire first calls a ball, the next pitch he is more likely to then call a strike.

Do we really need to point out that there is another human being involved here, and he is also completely aware that the first pitch was called a ball? And he is in fact throwing the pitches that are called?

It's not rocket science, people. Or science.

It has been shown that the pattern holds conditioning on the location of the pitch, for example as discussed here:

"It’s not rocket science, people. Or science."

Or, for that matter, rockets.

Unless Roger Clemens is pitching.

I generally prefer things with rockets.

Bernie Rosellen is a New York State licensed Realtor.

The movie is a very good example for women empowerment
and a straight uppercut message to the men's jaw was delivered on this
movie. In particular, it looks as if new homebuyers will make up a lion's share
of the market as winter turns into spring. The types of mortgage below are the most widespread:.
Please select the type(s) of Alerts you would like to review:.
The slogan of this company is "income for the regular joe" which means that you don't need any website marketing expertise to take advantage of the system.

I'm sure the authors would be interested to know that after last week's Twitter debacle, Fairfax County Schools are closed today, with <1" of snow on the ground.

Presentation ordering is only one factor biasing judges. Part of vetting judicial nominees should be tests of order bias and implicit association tests to look for racial, ethnic, gender, sexual orientation biases etcetera.

Oh never mind. Let us just continue to elect judges in many states so they can reflect median voter biases.

Now I'm sad early in the day.

Re Implicit Association -- as Wikipedia points out, not everyone agrees that the scores mean bias. They may be the result of unversal preferences (e.g. for day over night), salience, ease of mental access to the concept or the default/other organization of language opposition called "marking" in Linguistics (the thing that makes "how old are you" the normal way to ask the question and "how young are you" a bit of a joke).

I would not like to see IA tests used for any non-research purpose.

Or at least, it would be nice to have some actual evidence for the IA score : real world behavior correlation that's being assumed. Otherwise you might as well use a dowsing rod.

So, behavioral economics strikes again: "The harder I try to be fair, the more unfair I get?"

I suppose it's not surprising. If we could intuitively recognize genuine randomness then we'd be far better at casually generating truly random sequences. Since we can't, intuitive attempts to achieve an autocorrelation of zero are sure to fail (in this case, by over-correction).

Or does it? It's hard not to be suspicious of a paper that's so intuitively correct, given the general unreliability of intuition.

Interesting article. No thanks to Sunstein, who appears to be successful in selling books repackaging well-known behavioral economics and psychological phenomena as "breakthroughs".

Asking questions are truly nice thing if you are not
understanding something totally, except this article gives
good understanding yet.

Markov model text are truly nice thing but they are not understanding topic of discussion totally, except this comment link spam yet.

After pregnancy, women tend to weigh an average of 4.
This will help eliminate the unconscious snacking that is so much a part
of being overweight. Like it or not, in today's dating world it
is still largely the guy's job to approach the woman and make first contact.

Comments for this post are closed