On Intersectionalities and Employer Concerns in the Context of Becker’s Model of Discrimination ()
1. Introduction
Becker’s (1957, 1971) theory of discrimination in labor markets set the stage for much of the economics literature on wage discrimination over the last half century. Becker’s model focused on deliberate attempts by employers to treat equally productive people from different groups differently. In Becker’s model, each employer’s taste for discrimination, or the employer’s willingness to pay for treating members of different groups differently, was represented by a discriminatory coefficient denoting the premium that the employer was willing to pay for discriminating.
Becker started out with two critical simplifying assumptions: 1) all the workers in a given labor market are equally productive, and 2) the labor market is competitive. In retrospect, these assumptions seem less designed to describe what really happens in labor markets, than to show how competitive and efficient markets would eventually eliminate wastefulness, including costly discriminatory wage differentials. In this sense, Becker’s original model might be seen as more of an apologia for free markets than a true description of discriminatory labor market practices. Even so, Becker’s analysis provided a reference framework for answering important questions about the nature of group-based wage differentials. Much as our models of perfectly competitive markets provide a reference context for our understanding of imperfect competition, so Becker’s approach gave us a reference backdrop against which we can juxtapose some of the intractable complexities of real labor markets. The very simplicity of Becker’s approach, along with some of its counterintuitive conclusions, made it an appealing reference framework for considering the problem and the nature of discrimination. It is in this context that Becker’s approach has influenced how generations of economists have approached the problem of labor market discrimination.
It goes without saying that if all the workers in a given labor market are equally productive, then all observed wage differentials must be discriminatory. It is in this context that Becker interpreted his discriminatory coefficient as representing the percentage wage penalty or bonus, relative to a reference wage, that would make an employer indifferent to hiring from separate groups. Based on this reasoning, Becker concluded that the discriminatory coefficient would increase labor costs for discriminating employers. Becker’s logic was simple: In the context of a basic model of competitive and efficient labor markets, we would expect labor compensation to be based on the value that workers bring to their employers, i.e., we would expect wages to equal the value of labor’s marginal product. Thus, in competitive markets, differences in the compensation of workers must be based either on differences in their productivity, or else on whatever job-specific compensatory differentials are necessary to attract workers to different occupations with dissimilar non-pecuniary costs or benefits. This conclusion was subsequently echoed by Arrow (1973) who, following Becker, elaborated further on the general equilibrium conditions under which competitive pressures would eventually reduce discrimination.
Wage differentials that are based on a “taste for discrimination” rather than on differences in productivity, or indirect labor costs, should then not be sustainable. Discriminating employers would face higher labor costs, which would put them at a competitive disadvantage vis-à-vis their non-discriminating counterparts. Thus, eventually, discriminating firms would tend to be eliminated from product markets. And, when the discriminating firms disappear, labor would be reallocated to the remaining non-discriminating firms. Based on this reasoning, we would expect discriminatory wage differentials, or differentials that are based on non-productivity related characteristics, to be temporary and transitional. Of course, a corollary argument would be that discriminatory differentials might persist for some time based on market failures associated with imperfect competition, imperfect adjustments due to transaction costs, or if discriminatory employers are willing to accept less than normal profit.
While Becker explained wage differentials strictly in terms of the employers’ taste for discrimination, a number of later authors also examined discriminatory coefficients based on statistical discrimination, or else on differential costs associated with hiring from different groups. Not surprisingly, different assumptions about the nature of the discriminatory coefficient yield profoundly different long-term policy implications for how best to deal with the problem of discrimination.
As explained by Jonsson (2001), dissimilar reasons for group-based wage differentials may be observationally equivalent. For example, as explained by Lang (1986), intergroup communication and cultural differences can impose costs on employers. In addition, it may be less risky for an employer to hire from groups that the employer is familiar with. Thus, asymmetric information, where employers do not have good information about candidates from a minority group, can lead to statistical discrimination based on risk aversion even when employers expect the average productivities of different groups to be the same.
On the other hand, the more information that employers have about each individual, the less likely we are to see them practice statistical discrimination of any kind. Lang and Spitzer (2020) give an informative overview of some of the complications involved. The bottom line is that it is not easy to distinguish the effects of statistical discrimination, based on productivity projections, from the effects of willfully malicious discrimination.
Most importantly, if observed wage differentials actually represent genuine cost or productivity differences, then employers who adjust wages based on that should end up with a competitive advantage rather than a disadvantage. In other words, under those circumstances, seemingly discriminating firms would then end up more profitable than the seemingly non-discriminating firms. It was in this context that Sowell (1994: p. 114) rather blithely suggested that “group membership may in fact be used as a proxy for economically meaningful variables rather than reflecting either mistaken prejudices or even subjective affinities and animosities.” The bottom line is that compensation differentials that are based on real differences in productivity, or in employment costs, are likely to last as long as such differences persist.
2. Hiring Externalities, Psychological Concerns, and Pre Labor-Market Discrimination
The make-up of Becker’s hypothetical discriminatory coefficient has obvious policy implications. Specifically, if things were as simple as he suggested, and if the compensation differentials that we observe are purely discriminatory, and if markets are really competitive, then there would be no real reason to do anything about discrimination since the problem should take care of itself. The corollary to this idea would be that, if and when we observe significant and lasting wage differentials across separate groups, these differentials must either be based on something other than desire to discriminate, or else the labor markets in question are inefficient and imperfectly competitive. But things are even more complicated than these considerations suggest. In some cases, observed real productivity differences may indeed be caused by discrimination per se.
For one thing, the evidence suggests that potential workers from minority groups may suffer more from pre-labor market rather than direct-labor-market discrimination (see Ang, 2021). This may prevent them from obtaining the skills needed to compete for available jobs. To put this in context, a recent NCES (2024) report, that looked at the reading comprehension of 9-year-olds, made clear that Black students in the US remain far behind their counterparts. Thus, 47% of white students were able to read proficiently in 2022, which represented a massive drop from 84% in the half century since 1971. On the other hand, the Black students remained at 14% in both years, 2022 and in 1971. This represents a considerable obstacle, since students with reading difficulties tend to have problems finishing high school or obtaining any further academic or vocational education. This result may further be exacerbated by group-based expectations of current and future discrimination. After all, if you believe that you will be denied opportunities in the future, this will affect your current effort to prepare for that future. Indeed, if you cannot have them, then the grapes must be sour (see Elster, 1983).
Beyond these considerations, how well a worker fits in a particular workplace matters. Thus, we sometimes see externalities in hiring, where the productivity of existing workers is affected by new hires. The resulting hiring externality can be either positive or negative. After all, a collegial workplace where people feel that they are well treated, and where they like and enjoy the company of their co-workers, generally has a positive effect on productivity (see Staw et al., 1994); whereas an unfriendly or hostile work environment has the opposite effect. Indeed, as explained by Sutton (2007), the presence of a single unpleasant disruptor can decrease both the productivity and the retention of existing employees.
A related argument would be that workers who do not feel valued are also less likely to put forth extra effort. As Akerlof (1982) argued, sometimes there is an element of gift exchange in employer-worker relations. Stakhanovite motivations notwithstanding, low pay is typically also perceived as lack of appreciation. Thus, expectations of low productivity may become self-fulfilling: low productivity expectations yield low wages which in turn yield less effort. Thus, efficiency wage arguments (see Yellen, 1984), where effort is tied to compensation, may have profound implications for the persistence of discriminatory wage differentials.
These effects can be tied to the intersectional characteristics of different workers in the sense that workers from different backgrounds, with different identities, might feel most comfortable in different surroundings. People generally feel most at home in an environment that values and accepts their identities in terms of their attitudes, ideals, ethnicities, religion, gender, education, background, and sexual preference, etc. Moreover, how comfortable you are in your specific workplace tends to affect how you interpret different situations there. Thus, workers who feel defensive might also be more sensitive to perceptions of slights in the workplace. Just as a comment made by a stranger may be interpreted in a different way than the same comment made by a close friend would be, so a casual comment that might be perfectly fine in one environment could be seen as a mean-spirited microaggression in a different one.
The point here is that dissimilar workers may fit best into dissimilar milieus. Sometimes, what works in one environment might be unacceptable in a different one. Thus, just as workers tend to sort themselves into those occupations where they are the most productive, they may also sort themselves into those workplaces where they fit the best. This would also suggest that the discriminatory coefficient could be tied to how well employers expect different workers to fit with their workplaces and this would be tied to a variety of different intersectional characteristics and identities beyond membership in a larger racial or ethnic group.
Based on these considerations, it seems likely that the intersectional background of the members of minority groups matters a great deal. And, as discussed above, employers will consider all the pertinent information, credentials, and characteristics of potential employees before making a hiring decision. With that in mind, we need to take a closer look at how Becker’s framework might be applied to intersectional considerations in the labor market. Whereas Becker focused on pure wage discrimination in a market made up of just two groups, his reasoning can be extended to examine wage differentials in the context of a hypothetical labor force that is made up of n different intersectional groups and where the wage differential may represent a variety of different employment cost and productivity differences as well as ‘taste for discrimination.’
The bottom line is that interactions between groups tend to be a lot more elusive and more complicated than the economics literature on discrimination has so far recognized. Here, Crenshaw’s (1989, 1991) work on intersectionality made clear that within larger racial and gender groups, we may have multiple sub-groups whose experiences and outcomes are different from those of the rest. Yet, theoretical economic models on discrimination have mostly focused on disparate treatment of representative individuals from just two, rather than from multiple and overlapping, groups.
3. On Intersectionalities and the Nature of Discriminatory Intent
Becker’s original model suggested that wage discrimination would lead to labor segregation. The idea was that if the labor supply of a discriminated against group was large enough for some of its members to be forced to deal with discriminating employers, then all the non-discriminating employers would hire from that group, but at the lower market wage influenced by the discriminating employers. This is illustrated in Figure 1 below, where we have two groups, a and b. To keep things simple, just assume that half of the potential employers discriminate against group a, and then if the labor supply of group a (Sa) is large enough relative to demand to force some of its members to deal with discriminating employers then this will depress the group’s wages. Under those circumstances we will find lower market wages for group a than group b. And this will be indicated by a positive market discriminatory coefficient,
.
At the equilibrium employment level of Q* for group a, even though most of the employers hiring from the group do not discriminate, the fact that the group is large enough that some of its members have to deal with discriminating employers will depress the market wage for the group. As a result, we will see employment segregation in the sense that all of those hired from group a will be hired by employers with a discrimination coefficient smaller than market discrimination coefficient, while the employers with a higher discrimination coefficient will only hire from group b.
Figure 1. The labor segregation prediction of Becker’s model.
Of course this is not really what happens. Many employers do make a point of hiring from multiple groups, which obviously contradicts the segregation predictions of Becker’s basic model. So, we would not expect all the members of a discriminated against group to be hired by non-discriminating employers at a lower wage than what the workers from the favored groups receive. At a minimum, since diversity signaling may help a firm’s image, we might see a certain amount of symbolic recruitment from a group that is seen as disadvantaged. Signaling fairness and decency may even improve recruitment from other groups as well. Moreover, paying a lower wage to members of minority groups sends a signal few employers would be willing to give.
The problem is that discriminatory intent, or what Becker (1957, 1971) referred to as a “taste for discrimination”, is far too complex to be captured by a single market discriminatory coefficient. It is tied to the nature of groups, group perceptions, and group interactions in diverse ways that are not easily disentangled. And, to the extent that different groups are not completely homogeneous, it is extremely unlikely that all the members of a given group face the same level of discrimination.
Yet Becker focused on the taste for discrimination against a single homogeneous minority group by different employers. He made no attempt to explain or understand the nature of this “taste”. His refusal to explore the nature of discriminatory intent is in line with the general reluctance of economists of his generation to psychologize or try to study the nature of tastes and preferences. Stigler and Becker’s (1977) classic De gustibus non est disputandum article makes this attitude quite clear. And yet, not only does the specific make-up of discriminatory intent matter, it has profound policy implications. Most importantly, focusing solely on discrimination against a single group directs our attention away from the complexity of group interactions.
So, let us consider how Becker’s model could be extended to focus the different intersectional clusters that we might find within larger minority groups. As in the original version of the model, presumably the supply and demand for the labor of each intersectional cluster will determine its relative wage compared to the average wage in the market.
To keep things simple, let us assume that each intersectional cluster is completely defined by the vector of its intersectional characteristics. This vector would include individual attributes, personal history, and intersectional characteristics that are likely to be more relevant for an employer than simply membership in a given larger minority group. This would expand the framework of Becker’s original model and provide a more detailed reference framework for thinking about discrimination, and how it tends to work in practice.
For one thing, to the extent that a particular group is known to have been systematically discriminated against in the past, employers often try to signal that they are not doing so now. That in fact they are now making a systematic effort to hire from previously discriminated against groups. Indeed, most people do care about fairness and decency (see Jonsson, 2011). And most people would like to prevent the injustices of the past from being repeated in the present. Thus, we have seen a systematic push for integration of workplaces, with a focus on diversity and inclusion (see, for example, Brymer & Rocha, 2024; Glastonbury et al. 2021; Orupabo & Mangset, 2022; Tugend, 2018; Woods & Tharakan, 2021, etc.).
With these things in mind the question becomes how are those, set up to represent and signal diversity in hiring, selected? There is little doubt that intersectional concerns do play a role here. Thus, an employer may make a sincere effort to hire from a minority group, but only if the workers from that group also meet certain other intersectional criteria. This might include a variety of different things, such as demeanor, education, dress, attitudes, etc. Sometimes it is hard to gauge these things during a standard interviewing process, so in some cases the employers may attempt a deeper dive to see what the specific candidates are like in a workplace situation. Thus, many employers have a variety of internship and fellowship programs that are specifically geared towards minority groups. While these are no doubt well intentioned, these internships may also serve as attempts at deeper profiling for selecting future employees.
In a comparable way, as explained by Fernandez and Greenberg (2013), employers seem to rely far more on referrals when hiring from racial minority groups than they do when hiring white candidates. This suggests that the referrals represent profiling of minority job candidates. In this context, Bolte et al. (2020) have pointed out that the reliance on referrals is also tied to how easy it is to fire workers. If retaining minority workers is important for the signaling of fairness, this alone might increase the reliance on extra screening of minority workers.
Also, when it comes to intersectional clusters within larger racial and ethnic groups, we should not assume that the productivity related characteristics are the same for all. Thus, while statistical discrimination may continue to take place, most employers will attempt to identify all the relevant intersectional characteristics of each employee to help them gauge how productive that employee is likely to be. At the same time, different employers with different workplaces, different production and different work cultures may well be looking for dissimilar things.
Becker’s original idea was that employers are willing to pay a premium for hiring from their preferred groups. Or, equivalently, they are only willing to hire from groups that they value less, if those workers can be hired at a lower wage. But in some cases what may seem like a discriminatory coefficient is not really about taste for discrimination. Instead, it could represent the premium that an employer would be willing to pay for hiring workers that stand out in terms of fit, employment costs, and productivity. And those people could well come from different racial and ethnic backgrounds.
4. Intersectional Considerations in the Context of Becker’s Framework
Let us first assume that each intersectional cluster within a larger racial or ethnic group can be represented by a vector of individual and intersectional attributes. This vector would contain things related to their education, skills, attitudes and personality, experiences and backgrounds, religion and religiosity, gender and sexuality, etc.
With that in mind, let us set this up the way Becker did. First, assume that workers from any given intersectional cluster i (where
) receive wages of
and that the average wage for all workers from their larger racial or ethnic group is
, where
represents the amount of labor hired from each intersectional cluster i. Let us further assume that we have m different employers. The reservation wage of employer j (where
), for hiring from i, would be
. As we discussed in the previous section of this paper, this reservation wage will depend on how well the particular cluster i, is expected to fit the employer’s workplace.
To keep things simple, let us assume for now that the workers themselves do not discriminate between employers and that they will just seek work wherever it pays them the most. As for the employers, notwithstanding their different evaluations of each group i, each employer still seeks to minimize their labor cost, subject to their expectations about what members of distinct intersectional groups are worth paying. Under these circumstances, the extent to which an intersectional group receives a premium or a deduction relative to standard wages will depend on the demand relative to the number of workers of each group seeking employment.
Each employer j is willing to hire from i as long as
The individual discriminatory coefficient of the employer,
indicates how the employer values members of intersectional cluster i relative to their larger racial or ethnic group. Note that this coefficient,
, would be zero for employers who have no specific preference for or against i, while a negative coefficient,
would indicate favoritism and a positive coefficient,
would indicate discrimination. In any case, regardless of their individual discriminatory coefficient, every employer hiring from the group would pay the cluster specific market wage,
To put this differently, employer j is willing to hire from cluster i, as long as
Overall, whether that means that the wages for cluster i,
end up higher or lower than the average wage for their larger group,
, depends on relative market supply and demand for the members of the cluster.
Consider this a bit further in the context of some simplifying assumptions. First, following Becker, let us assume that we can rank all the employers who would potentially hire from intersectional cluster i, in increasing order of the size of their
. For the sake of simplicity, let us further assume that when we divide the ranked employers into quintiles based on their valuation of group i, and that the first quintile will favor group i, so
, the next three quintiles will be indifferent, so
, and the final quintile would prefer not to hire from the group, so their
. This ranking of the employers then yields a demand for labor from group i, based on the employers’ reservation wage relative to the overall group wage,
This demand can be represented by straight line segments for each quintile as shown in Figure 2 below.
In this diagram, segment EF represents the ranking of employers who favor group i and thus have a negative differential coefficient,
, ranging from −0.5 to 0. Segment FG represents employers who are indifferent, with
. Finally segment GH ranks employers with differential coefficients ranging from zero to infinity, where the last ranked employer would have a discrimination coefficient of
and thus not be willing to hire from group i at any wage.
Figure 2. Demand for labor ranked in order of the employers’ discriminatory coefficients.
In this context, the extent to which intersectional cluster i would receive a wage higher or lower than
, will depend on the size of the cluster relative to the number of employers who might consider hiring from it. If the cluster is small enough, as represented by SLow in Figure 3, then the employers who favor the cluster will pay a premium since the supply curve will intersect the demand in the EF range. If the supply intersects the demand in the FG range, then there will be no observable difference between wages for this cluster and their overall group. Finally, if the cluster is large enough that some of its members will have to deal with employers who disfavor them, represented by segment GH of the demand curve, then the members of the cluster will receive a wage that is lower than the average group wage.
Figure 3. Cluster size relative to the number of job openings determines relative wages.
The point here is that different clusters may well face wages that are quite different from the average wages of their larger racial or ethnic groups. Thus, we may see some clusters that experience significant favoritism while a different cluster from the same overall racial or ethnic group might face discrimination.
One of the critical points in Becker’s reasoning was that if the markets are indeed competitive, then the workers and the employers will all be price (or wage) takers. Thus, if the supply of labor from group i is high, SHigh, as shown in the diagram above, then some of the workers in intersectional cluster i will be forced to deal with discriminating employers. This in turn lowers the wage for all the members of that intersectional group. Even those employers who favor the group would now have an opportunity to hire from the group at a wage that is lower than the average wage,
, that they would have had to pay to hire from other groups. If the supply is high enough that some members of the group are forced to deal with employers who disfavor them, then all its members would receive lower than average wages. Hence, in this basic framework, all the employers will pay this group a lower wage than
.
Some of this echoes Becker’s original segregation argument. Except, in this case, we are looking at partial segregation by intersectional clusters rather than by larger racial or ethnic groups. If the markets are competitive, then the employers hiring from cluster i would now pay the same wage, wi, which might be higher or lower than
. But, as we discussed in the previous section, we are not assuming that each employer only hires from a single overall racial or ethnic group. After all, in the context of job fit and reputational concerns, many employers make a point of hiring from multiple groups, thus contradicting the original segregation predictions of Becker’s model.
The important point is that employers may evaluate the intersectional characteristics of different members of larger groups in very different ways. We often observe different members of larger ethnic or racial groups, who have different individual backgrounds, face very different experiences in the labor market. For example, a minority student who graduates from a STEM field with a high GPA may receive a significant wage premium, not just compared to their overall larger racial or ethnic group but compared to all other hires. At the same time, another student from the same larger group, who graduates with a low GPA, and who also has some minor criminal record, may have difficulties finding any employment.
5. Conclusion and Suggestions for Further Research
Becker provided a reference framework for examining discrimination in labor markets. Yet, the deceptive simplicity and elegance of his model directed our attention away from real and important problems. In particular, his blithe refusal to examine the nature of discriminatory intent stands out, as the specific make up of his discriminatory coefficient has profound implications for the predictions of his model.
Sometimes discriminatory intent can be provisional and subject to specific circumstances. Moreover, what might be interpreted as taste for discrimination could be about something else. It might be about employer expectations about issues such as worker fit, hiring externalities, or employment cost. In other cases, it could be based on uncertainty and risk aversion or be about statistical discrimination.
The paper has also argued that employers generally want to project fairness and decency in their handling of diversity. This in turn means that most employers are unlikely to focus solely on the larger racial or ethnic group memberships of prospective employees. Instead, the employers will generally consider the individual and intersectional attributes of new hires. In other words, discriminatory concerns are now reflected in the additional screening of candidates from minority groups.
In this context, this paper looked at some of the implications of intersectional clusters within larger racial and ethnic groups and concluded that we may see very different results for different clusters. Thus, sometimes we may see one intersectional cluster from a larger group receive a significant wage premium, while a different intersectional cluster may receive wages that are well below average.
This paper retained Becker’s assumption to the effect that markets were competitive for each cluster. This is of course a simplification, so future research on these issues should consider how things might work in imperfectly competitive markets. For one thing, members of small clusters are likely to have some monopsony power. And, for many situations, employers also have some monopoly power, so sometimes we might be looking at bilateral monopoly situations, especially for highly sought after members of minority groups.
Moreover, the fact that each of us has some unique attributes makes it an oversimplification to presume that all intersectional clusters will be internally completely homogeneous. Employers are likely to consider some uniquely individual characteristics for every person they hire. The same goes for prospective employees, who will generally consider specific job attributes and all their options before accepting any job. Thus, a perfectly competitive framework may be a round peg in a square hole for examining these issues. Indeed, elements of bilateral monopoly may be found in many employment situations. Future research should take a closer look at the implications of this for screening discrimination and intersectional concerns.
The theoretical arguments of this paper call for further empirical investigations of employment screening. While it can be difficult to obtain direct evidence of screening, sometimes indirect evidence can be found. For example, if we have data on internships, and also on permanent hiring from the pool of interns, that may give us some indications. Still, we might have a variety of different reasons why the percentage of hiring from different intersectional intern clusters varies. For one thing, it takes two to tango, and the hiring percentages depend on offers accepted, not just on offers made.
Lastly, the question of how best to deal with intersectional discrimination remains. Before we formulate policies to deal with this problem, we should try to understand it completely. And that calls for a closer look at specific intersectional concerns. Once we know which intersectional clusters are hit the hardest, we may be able to formulate policies to deal with the specific issues that they face.
Acknowledgements
This paper is an extension of a segment excised from a larger paper that was originally presented at the STOREP conference in Milan in June of 2024. The author would like to thank the STOREP participants for helpful comments that led to the writing of this paper.