From the Economist:
A TRULY informed diner would choose a restaurant based on the quality of the menu and the chef’s experience. The discerning investor would decide which company to back after studying the business plan and meeting the founders. In reality, people often copy the choices of others. Diners pick the crowded restaurant over the empty one. Investors go with the company that already has multiple backers.
Such bandwagon effects are not necessarily irrational. Often, the buyer knows less about a product than the seller; the collective wisdom of the crowd can correct for such “asymmetric information”. It can also be a way of coping with a surplus of choice: rather than study 100 models of music player, why not assume the market has already figured out the duds?
The existence of bandwagon behaviour can be hard to prove. A product or an asset usually becomes popular (or unpopular) in the first place because it is genuinely superior (or inferior). But some have tried to isolate the self-fulfilling effects of popularity. One 2004 study* by Alan Sorensen, now of the University of Wisconsin, examined accidental omissions from the New York Times bestseller list. By comparing the sales of books that did make the list and unlisted books that should have, the author could isolate the effect of inclusion—a modest boost to first-time authors’ sales. In a 2008 study by Matthew Salganik of Princeton University and Duncan Watts, now at Microsoft Research, participants tricked into believing a song was more popular than it actually was were more likely to download it.
Scholars are now asking whether herd behaviour also prevails in labour markets. Although the number of long-term unemployed in America is coming down (see left-hand chart above), it is still near an all-time high. Such workers may simply be losing out to candidates with more qualifications or experience for the jobs that come open. A more ominous possibility is that long-term unemployment is at least partly self-fulfilling: employers may be reluctant to hire someone others have already passed over.
To find out, Kory Kroft of the University of Toronto, Fabian Lange of McGill University and Matthew Notowidigdo of the University of Chicago devised an experiment in which they applied for 3,000 clerical, administrative, sales and customer-service jobs advertised online by submitting 12,000 fictitious cvs. The submissions were designed so that applicants with similar backgrounds, education and experience went for the same job. The only difference was how long the applicant had been jobless, a period that ranged from no time at all to as much as 36 months.
Creating cover identities worthy of the CIA, the researchers assigned the fictitious applicants local phone numbers bought for the experiment, as well as e-mail addresses. They found that the odds of an applicant being called back by an employer declined steadily as the duration of unemployment rose, from 7.4% after one month without work down to 4-5% at the eight-month mark, where the call-back rate stabilised (see right-hand chart above).
These results, the authors say, cannot be because employers found some qualitative flaw in the longer-term unemployed that was hidden from outsiders, since the applicants were similar in other respects. Another explanation for long-term unemployment—that people make less effort to find work as their time out of the labour force lengthens—is also not applicable here.
A third possibility is that employers equate lengthening unemployment with atrophying skills and thus falling productivity. But this should be true whether the economy is booming or in recession. The decline in call-back rates was much more pronounced in cities with tight labour markets; call-back rates changed relatively little when higher unemployment prevailed locally. From this, the authors infer that employers are more likely to overlook a long period of unemployment if overall economic conditions are stacked against candidates.
These results strongly suggest that long-term unemployment is at least partly self-fulfilling. Like patrons who avoid restaurants purely because they are empty, employers were reluctant to hire someone other employers didn’t want.
Although such bandwagon behaviour may be rational, it can also be deeply harmful. Two decades ago Abhijit Banerjee, now at the Massachusetts Institute of Technology, devised a model of “rational herding” in which market participants base their decision on a combination of their own information and the actions of others. Over successive rounds of transactions, participants responded less to their own information and more to the herd.
That can lead to poor outcomes. Imagine a newly unemployed worker who narrowly misses out on the first job he applies for. That initial failure reduces his odds of landing the second job he applies for, and so on, until he ends up as one of the long-term unemployed. The growth of cv-screening software may exacerbate the trend by reducing the chances that someone in human resources will pull the details of a long-term unemployed worker out of the pile of applications. One near miss can increase the odds of protracted failure.
Popularity is not destiny, fortunately. Mr Salganik and Mr Watts could turn an unpopular song into a hit by manipulating its perceived popularity, but could not turn a hit into a loser. Even after being falsely labelled the least popular song, the most popular song rose sharply in the rankings once enough people had listened to it. Similarly, Messrs Kroft, Lange and Notowidigdo found that even someone who has been unemployed for several years has a 4% chance of a call-back. But their study suggests that taking some work is better than none for the recently unemployed. Once the bandwagon starts to roll, it is hard to stop.