12 Ethical issues
12.1 IRBs
Major learning objective(s) for this section:
- Know what an IRB does.
- Know what is included in informed consent.
Institutional Review Boards (IRBs) are organizations designed to protect human subjects in scientific research. IRBs review research proposals and request changes if a research study is suspected to be unethical or illegal.
IRBs typically require that each participant in a research study has enough information to make an informed decision about whether to participate in the research study. This informed consent can involve discussion of potential benefits and potential harms of the research to the participant and an indication that the participant will not be penalized for not participating in the research study. This informed consent can be waived in particular situations, such as audit studies testing for real-world racial discrimination, because [1] the success of audit studies requires that participants not be aware that they are in a research study and [2] the audit studies typically have little chance of harm to the participants (or, perhaps. more accurately, the harm to participants of evaluating a fake resume is arguably relatively small).
12.2 Statistical discrimination
Major learning objective(s) for this section:
- Know what taste-based discrimination and statistical discrimination refer to.
Let’s discuss two types of discrimination. Taste-based discrimination is justifiable only by taste. For example, if the most qualified applicant is a woman, but the employer decides to not hire the woman applicant merely because the employer does not like women and/or prefers men to women, that would be an example of taste-based discrimination.
Statistical discrimination is when unknown information for an individual is estimated based on known or perceived data for the individual’s group. For example, consider a man and a woman who are identical on all observed characteristics, such as age 40, BMI of 20, and a non-smoker. If a company charges a higher price for a life insurance policy for the man than for the woman based only on the fact that women on average live longer than men do controlling for factors such as age, BMI, and smoking status, that would be an example of statistical discrimination. Statistical discrimination might be unfair at the individual level, but statistical discrimination can cause more fairness at the group level and can help produce better outcomes on at least some dimensions…
For example, historically, male drivers have been much more likely than female drivers to die in a motor vehicle crash. Let’s imagine a 16-year-old boy and a 16-year-old girl, none of whom have a driving history. If the auto insurance charges the boy a higher rate than they charge the girl – merely because of how boys and girls on average typically drive – in a sense, at the individual level, that’s unfair to the boys. Instead, we can set the auto insurance rate for 16-year-old boys to be equal to the auto insurance rate for 16-year-old girls, but – even though girls will pay exactly as much as boys for auto insurance – the boys will almost certainly cost the insurance company more in payouts on average than the girls cost the insurance company on average . In a sense, at the group level, treating 16-year-old boys and 16-year-old girls the same way would be unfair to the girls.
Sometimes in real life, imperfect information causes a situation in which the choice is between different types of unfair outcomes.
Sample practice items
Suppose that the CEO of a firm graduated from Illinois State University and gives preference in hiring to applicants who graduated from Illinois State University, even though the CEO has no data that applicants from Illinois State University are any better or worse than other applicants. Which of the following types of discrimination does this better reflect?
- statistical discrimination
- taste-based discrimination
Answer
- taste-based discrimination
Suppose that the CEO of a firm graduated from Illinois State University and gives preference in hiring to applicants who graduated from Illinois State University, because, in data from applicant hires over the past ten years, applicants who had graduated from Illinois State University have performed better on average than other similar applicants have performed. Which of the following types of discrimination does this better reflect?
- statistical discrimination
- taste-based discrimination
Answer
- statistical discrimination
12.3 Kelley’s paradox
Major learning objective(s) for this section:
- Know how information about an individual’s group can help improve predictions about the individual.
Let’s discuss another example in which information about an individual’s group permits better predictions about the individual, at least on average. Kelley’s paradox concerns the fact that two persons who have the same score on a multiple-choice test can differ in their true ability in a way that is predictable on average based on the groups that the persons are drawn from. The key mechanism for Kelley’s paradox in this situation is that guessing a response will on average help test-takers with lower ability more than such guessing will help test-takers with high ability.
Consider a 100-item test in which each item has only two response options and in which each option is equally likely to be correct. If you know 100 items, guessing help you get 0 more items correct on average. If you know 80 items, guessing help you get 10 more items correct on average (because you would guess at the 20 items you don’t know and get about 50% of those items correct). If you know 50 items, guessing help you get 25 more items correct on average. If you know 20 items, guessing help you get 40 more items correct on average. If you know 0 items, guessing help you get 50 more items correct on average.
The same mechanism works at the group level: guessing will on average raise the scores of lower-performing groups more than guessing will raise the scores of higher-performing groups. Therefore, a test-taker from a lower-performing group is more likely to have been helped by guessing than a test-taker from a higher-performing group.
Kelley’s paradox can produce a situation in which it is possible for different measures of fairness to produce different results. If we set the threshold for entry to a gifted math program to be a top 10% threshold score on a multiple-choice math test for all students no matter their group, then that is fair in the sense of treating students equally as individuals and using only a student’s observed performance in our decision. But that type of individual-level fairness might mean that there is group-level unfairness, if guessing helps some groups get into the gifted program more than guessing helps other groups get into the gifted program.
Sample practice items
Suppose that, on a statewide standardized multiple-choice test of science knowledge, the average score was 40 among male students and 50 among female students. However, on this test, female student Amy scored a 70, and male student Bob scored a 70. Each student provided a response to each item on the test. Based on only this information, which of the following is correct?
- It is likely that Amy has a higher science knowledge than Bob has.
- It is certain that Amy has a higher science knowledge than Bob has.
- It is likely that Bob has a higher science knowledge than Amy has.
- It is certain that Bob has a higher science knowledge than Amy has.
- None of the above.
Answer
- It is likely that Amy has a higher science knowledge than Bob has.
Female students had a higher average score than male students did, so male students are one average expected to benefit more from guessing at responses. So, with only the given information, we would expect that Bob benefited more from guessing than Amy did.