By definition, statistical discrimination occurs when a decision-maker use observable characteristics of individual to infer about another attributes that are relevant but harder to observe. Therefore, agents’ ideas about a specific group will bias their perception and/or determine the way they will treat an individual member of that group. This paper provides a good overview of the theory and clarifies that the decision-makers aim to maximize their utility or profit and face, in the most part of the cases, imperfect information issues about some relevant characteristics such as individuals’ productivity, qualifications, propensity to engage in criminal activity, etc. The rational response to this is to use group statistics as proxies of those unobserved characteristics. Observable characteristics often invoked are easily recognizable physical traits, which are used in the society to broadly categorize demographic groups by race, ethnicity or gender. Also, sometimes this categorization may be based in endogenously chosen characteristics such as club membership or language.
One of the best examples of statistical discrimination happens in the labour market. Race, gender, or any group affiliation is used in the recruitment process and in the determination of wages because it is assumed that workers’ productivity exogenously depends on the group identities. This study using data on American employees between 1977 and 2010 shows empirical evidence of statistical discrimination on the basis of race and gender. This paper published in 2004 did a field experiment randomly assigning resumes to African-American or White-sounding names and simulating a job-search exercise. Results show that, controlling for everything else, White names received 50 percent more callbacks for interviews. These are vexing and shocking numbers.
Theoretical approaches, like Coate and Loury (1993) show that firms may rely on the race of the worker as a useful tool to assess about worker’s skill. This arises a big issue related with investment incentives: if firms believe that workers from a certain racial group are less likely to have the right skills to the job, and in consequence of that impose more conditions in order to assign these workers to their jobs, it will end up lowering these workers’ investment incentives. In the medium run, discriminated minorities will invest less in education, becoming less skilled and therefore meeting firms’ initial pessimistic belief. This work goes deeper in this approach showing peer effects’ importance. The authors prove that peer group effect increases the cost of investment for the group with high investment and decreases the cost of investment for the other group. This is expected to lead to an equalization in the fractions of people that invest. It is shown that under some conditions, a strong integration culminates with groups acquiring the same level of human capital. Therefore, ultimately, this would result in benefits not only for the employees but also to the market itself – which could later translate into benefits for the society.
The labour market is just an example of where we can find statistical discrimination. Data shows that it is a big source of inequality. And specially in this case, when talking about the labour market, it is easy to conclude that this inequality of circumstances and opportunities will lead to an inequality of income and therefore will tend to favour a hypothetical poverty condition for groups who are victims of statistical discrimination.
An obvious solution to this problem is antidiscrimination legislation. In USA, federal law started prohibiting any kind of durable discrimination. Evidence suggests it had a big impact in reducing these kind of behaviours. However, it is still far away from solving the whole problem.
I would say that the solution has to be deeper and longer lasting and is related to a paradigm shift. It is crucial that a collective effort be made to change the widespread perception that statistical discrimination is innocent and harmless, just resulting from a rational process.
Some writers appear to take the position that statistical discrimination is perfectly acceptable because of what was said, trying to excuse discrimination based on the argument that the agent “only” wants to reach an efficient decision and has no intent to harm others during the process. But it seems to me that a sense of morality also has to play a role in here. There are certain groups of people who have been undermined for decades based on too generalist and some obsolete assumptions. And one should not shield himself in concepts of efficiency to justify it.