In a recent blog article, Karla Hoff talks about why economists should care about affirmative action. In a joint paper with Joseph Stiglitz she proposes a framework that links performance and self-confidence: People make biased inferences about their probability to succeed in a task, and perceived probability to succeed is positively linked to performance. Historically discriminated against groups such as women or ethnic minorities may have lower confidence, underestimating their likelihood of success and hence performing worse than the dominant group, which in turn leads to self-fulfilling negative stereotypes about members of those groups. In her article, Hoff suggests that affirmative action may help change this vicious circle of low confidence, low expectations and low performance.
Hoff cites a randomized natural experiment in India, evaluated by Lori Beaman, Raghabendra Chattopadhyay, and Esther Duflo, as evidence for the effect of adjustments in perceptions through affirmative action. In this experiment, affirmative action increased the exposure of villagers to women in high level positions by reserving a certain amount of seats in a local government. This helped villagers and the women themselves to learn something about their abilities, which, as Hoff suggests, facilitated the elimination of initial negative stereotypes and changed the way women perceived themselves and how others perceived them. The result was a lasting positive effect: Even after the quota was removed, women were more likely to be elected into local leadership positions than in districts where a quota had never been introduced.
Of course, mechanisms other than learning about one’s own abilities could play a role for the enhanced performance and change in perceptions of women, for instance, actual learning on the job, or better social connections that enhance women’s ability to implement policies. Beaman et al. try to control for that by keeping objective competences constant: They expose the individuals of their study to fictitious speeches where the only variation is the gender of the speaker. After listening to the speech, subjects were asked to assess the effectiveness of the leader, and men that had been exposed to women in leadership positions assessed female leadership more positively than those who had not been exposed.
However, affirmative action may not always eliminate negative stereotypes, as previous research by Stephen Coate and Glenn C. Loury suggests. They show that even if individuals of different groups are ex-ante identical in terms of ability, initial unfavorable beliefs may become self-confirming.
The following example illustrates how this mechanism may work: A set of firms is looking to employ workers. Employees can be classified into two groups, let’s call them the advantaged and the disadvantaged group. There are two jobs, one of which requires a qualified worker in order to be executed profitably. All workers want to be assigned to this qualified job, as it pays a better wage.
Assume further that employers cannot observe whether or not a worker is qualified for a job, but instead have to infer it from a noisy signal, such as the result of a standardized test. Higher values of the signal are more likely if the worker is qualified, and for a given prior, upon observing a signal, firms assign higher likelihood to the worker being qualified if his signal takes a higher value.
A plausible thing to do for firms is to set a threshold value of the test score, such that if a worker’s test score exceeds that value, they will assign him to the qualified task, and to the unqualified task otherwise.
Now add to the picture negative stereotypes, that is, assume that employers believe that on average individuals from the disadvantaged group are less likely to be qualified than those from the advantaged group. This will result in firms setting a higher threshold for workers from the disadvantaged group. Being confronted with higher hiring standards, the expected return to investing in skills are reduced, which in turn reduces the incentives for individuals from the disadvantaged group to invest in skills. Initial negative stereotypes are confirmed.
Introducing affirmative action in the sense of a target level of representation may actually worsen the initial negative stereotypes. This is because in order to achieve the target level of representation employers may lower hiring standards for minority applicants. Again, incentives for minority applicants to invest in skills are reduced because entry is easier for them and employers’ initial beliefs that minority group workers on average invest less in skills are confirmed.
One of Coate and Loury’s conclusions is that when designing affirmative action policies it is important to create the conditions under which they can actually change negative beliefs about minority groups. Only if beliefs are updated in the “right” direction can affirmative action bring about long lasting improvements rather than just a temporary fix.
In the case of the Indian experiment it was necessary to reserve seats for women for a sufficiently long period of time in order for locals to collect enough evidence of the abilities of female leaders. In other circumstances where entry standards are adjusted in order to increase representation one has to be careful not to stigmatize the favored group, which could reduce their self-confidence and eventually their performance, and reinforce negative stereotypes.