Statistical discrimination refers to a situation where, when selecting between different individuals, a selecting agency uses the average characteristics of groups that these individuals belong to as proxies for the characteristics of the individuals in lieu of direct measurements of these characteristics for the individuals.
Statistical discrimination is to be distinguished from taste-based discrimination.
Statistical discrimination could occur in personal decisions (choice of friends and lovers), employment decisions, admission decisions to educational institutions, political decisions, or any other form of decisions.
Some of the parameters/group identifications used for statistical discrimination are:
|Example of parameter||Examples of what it may be used as a proxy for||Example of situation|
|Educational attainment (how many years of schooling have been completed)||intelligence, curiosity, conscientiousness||It may be statistically the case that people who have complete graduate degrees tend to be more conscientious than high school dropouts. An employer, deciding whether to hire somebody for a job with high conscientiousness requirements, may use a graduate degree as a hiring filter rather than test potential employees for their level of conscientiousness.|
|Age||People in certain age groups may be willing to work longer hours, or may have a better understanding of certain consumer segments, or may have more of the right kind of experience.||A startup hiring workers for extremely intensive jobs may choose to only hire workers within a certain age group rather than testing the willingness of each individual to work the required long hours.|
|Gender||For instance, males may statistically have more physical strength||Gender may be used as a filter for certain jobs that require physical strength even though there may be females who have the requisite physical strength.|
|Racial or ethnic group||people in a particular racial group may be statistically less likely to steal from their employers.||Employers may use racial or ethnic group as a filter for hiring (or seek a higher bar from applicants of particular racial or ethnic groups) rather than conduct tests on employees for their likelihood to steal.|
Motivation behind statistical discrimination
Below are some situations that make statistical discrimination more attractive. The stronger these are, the more likely statistical discrimination is to occur.
- The cost (in time, money or effort) of determining the characteristics for individuals may be too high to justify individual testing.
- For the specified characteristics being tested for and the groups under consideration, the intra-group variation for the characteristic may be sufficiently low to make group averages a reasonable proxy.
- In some cases, erroneous judgments made based on statistical discrimination may be easy to correct.
Statistical discrimination versus other kinds of discrimination
With pure statistical discrimination, the discriminating agency uses a correct (or close to correct) value of the average within the group, and is also aware of intra-group variation and willing to use more accurate estimations of individual characteristics if it is cheap to do so. Taste-based discrimination may combine this statistical discrimination component with others, such as:
- A preference or taste for certain kinds of groups that is not explained by the differences in average characteristics.
- A fairly inaccurate estimate of average characteristics of groups.
- An under-estimation of intra-group variation.
- An unwillingness to use countervailing evidence in individual cases even where such evidence is cheaply and readily available.
Effects of statistical discrimination
Incentives for individuals
If individuals are judged solely on the basis of group characteristics, the following may happen:
- In so far as it is possible to enter and exit groups, individuals have an incentive to enter groups with above average characteristics and exit groups with below average characteristics.
- Individuals have an incentive to express greater concern for the characteristics of other members within their groups because of intra-group externalities. They may use devices such as intra-group shaming or punishment (you're embarassing our entire group) to try to counter other individuals bringing down the group average.
- Individuals have an incentive to be less careful about their own characteristics and will try to free ride on the group's characteristics.
If, however, individuals are judged on a combination of their individual characteristics and group characteristics and can choose which aspects to place emphasis on:
- Individuals within the group who are below the group average have an incentive to play up their group characteristics and play down their individual characteristics.
- Individuals within the group who are above the group average have an incentive to play up their individual characteristics and play down their group characteristics.
- The extent to which individuals have incentives to improve their individual characteristics depends on the extent to which individual characteristics matter, in particular whether such characteristics are easy to measure and can be convincingly demonstrated.
Ethics of statistical discrimination
Ethical questions surrounding statistical discrimination are murky. Some of the typical ethical issues are as follows:
- Suppose there do exist tests to measure individual characteristics, and such tests are moderately but not prohibitively expensive. In other words, the tests aren't cheap enough for the discriminating agency to administer in a way that is clearly profitable to it, but they are a modest expense. Does the agency have an ethical obligation to administer these tests to ensure fairness in hiring?
- Suppose there are no tests to measure some particular individual characteristics. Are certain kinds of statistical discrimination unethical? If so, should employers simply ignore those characteristics when hiring?
There are also more indirect arguments, such as:
- If taste-based discrimination is unethical and if it is not possible in practice to distinguish between statistical and taste-based discrimination, might it be the case that any form of statistical discrimination that could morph into, or be confused for, taste-based discrimination is unethical?
- Does statistical discrimination create negative incentives for people within certain groups, and if so, does that make it unethical?
Parameters in ethical evaluation
Some of the parameters that may be used to evaluate the ethics of statistical discrimination are:
- The degree of benefit or harm to the agent performing statistical discrimination
- The degree of benefit or harm to other individuals subject to or affected by statistical discrimination
- The reliability of the criterion used for statistical discrimination
- The degree of control or agency that affected individuals have over the criteria being used