Sackett et al. (Reference Sackett, Zhang and Berry2023) provide a useful more practice-oriented discussion of Sackett et al. (Reference Sackett, Zhang, Berry and Lievens2022) report that reexamined meta-analytic corrections for a wide variety of selection tools, across common content and process domains. We expand on their discussion of implications regarding the new validity estimates for the classic validity–diversity tradeoff by arguing that the benefits of diversity are still underestimated when assessing this tradeoff. To be fair, this issue is not limited to Sackett et al.’s efforts but rather represents a shortcoming of the field at large. Regardless, these limitations mean that if diversity benefits were better understood by the field and properly accounted for in tradeoff estimates, even greater reductions in the usefulness of predictors with high group mean differences would likely be observed. We make three key points. First, we argue that the benefits of group diversity are not included in selection decisions, leading to underestimations of diversity benefits. Second, we elaborate on the central role of interdependence as a condition that maximizes the importance of diversity. Finally, we connect these issues to the long-term implications of assessment decisions containing adverse impact.
Underestimation of diversity benefits
To understand the problem with estimating the diversity–validity tradeoff, we need to understand how selection, performance, and diversity work together. To organize this discussion, a multilevel theoretical model depicting the relationships of interest, inspired by Sackett et al. (Reference Sackett, Putka, McCloy and Kozlowski2012), can be found in Figure 1. At its base, the point of employee selection is to find individuals who are good fits for open positions. Organizations enact processes to select applicants who are the best fit for established job requirements through identifying individual differences that predict important outcomes on the job, namely job performance (Heneman et al., Reference Heneman, Judge and Kammeyer-Mueller2019). This represents the essence of Linkage 1. Practitioners use assessments of applicant trait levels to bet whether applicants with higher trait levels will perform better on the job than applicants with lower trait levels. A further assumption captured in Linkage 2 is that selecting individuals with greater potential performance at the individual level will propagate upward to improve the overall performance of groups and organizations through a process of human capital resource emergence (e.g., Ployhart & Moliterno, Reference Ployhart and Moliterno2011). In other words, the collection of individuals with desirable skills and abilities should not only support the performance of those individuals but emerge into higher level group resources and outcomes.
Figure 1. Multilevel Depiction of Effects of Diversity on Performance.
When we select individuals into an organization, their combined characteristics give rise to group-level characteristics, which we can observe and quantify to summarize the outcome of Linkage 3. This includes group diversity, which represents the “the distribution of differences among members of a unit with respect to a common attribute, X, such as tenure, ethnicity, conscientiousness, task attitude, or pay” (Harrison & Klein, Reference Harrison and Klein2007, p. 1200), but includes all surface and deep level characteristics. Variety diversity refers to differences in kind or category and, from an information processing perspective, has positive effects on group performance by allowing groups greater access to informational resources to improve problem solving. Following this, and supporting Linkage 4, evidence suggests diversity can support group-level performance, especially in complex environments, although the importance of different types of diversity may change over time and the effects are complex (Martins & Sohn, Reference Martins and Sohn2022). Thus, there exists a case where identified job-relevant traits should support individual performance, which in turn supports group performance, and a case wherein differences in individual traits produce diverse groups that should also improve group performance.
The problem arises with how we currently approach diversity in selection. Historically, mean differences across racial groups in predictors with the highest validities have resulted in a “validity–diversity tradeoff,” which positions organizations as choosing between maximizing validity or workforce diversity (Pyburn Jr. et al., Reference Pyburn, Ployhart and Kravitz2008). This effect has also been labeled the performance–diversity tradeoff (Sackett, Reference Sackett, Camara and Kimmel2005), emphasizing that increased diversity will instead, on average, lower individual performance. Thus, the assumed relationship captured by Linkage 5 is negative. This creates a situation wherein increases in diversity through selection should help group performance, but hurt predicted individual performance, which should (paradoxically) in turn negatively affect group performance.
The current gold-standard approach to balancing validity and diversity to meet organizational needs lies in the Pareto-optimal solution. Introduced to the organizational literature by De Corte et al. (Reference De Corte, Lievens and Sackett2007), Pareto optimality was borrowed from operations research where decision makers need to maximize outcomes under competing constraints. In the case of validity and diversity, traditionally competing objectives are to maximize hiring quality while minimizing adverse impact. Given a set of predictors with known validities and group differences, a series of combined regression weights can be calculated along a Pareto front where each objective (i.e., maximizing validity or minimizing adverse impact) is considered equally good and changes to improve validity or adverse impact would come at the cost of the other. Research suggests the Pareto-optimal approach indeed outperforms other approaches in achieving a balance between diversity and validity (De Corte et al., Reference De Corte, Sackett and Lievens2020). However, we argue that it will tend to underestimate the value of diversity for performance because it does not account for the potential positive effects of diversity on group performance. In essence, the Pareto-optimal approach addresses Linkages 1, 3, and 5. To do so, it relies on assumptions about the population-level distributions of variables and the applicant pool, and that selection occurs in a top-down manner. Of necessity, the process relies on group-level estimates of differences and validity to make individual-level selection decisions. The hope is then that the best individual selection emerges into higher level group and organizational performance. However, because Linkage 4 is ignored and Linkage 4 is theorized to be positive, the value of diversity for performance will tend to be underestimated in selecting the optimal weights for use in selection.
Interdependence, complexity, and diversity
The unaccounted-for positive effects of diversity at the group level further builds on Sackett et al.’s (Reference Sackett, Zhang and Berry2023) discussion of the changing importance of GMA as work becomes increasingly team based. A hallmark of team-based workflows is interdependence among team members working together to achieve a common goal. In Steiner’s (Reference Steiner1972) early typology of team tasks, the author pointed out how different tasks were dependent on the lowest, the average, or the maximum level of performance that an individual could provide. Later representations moved away from task and team types, and instead focused on the level of differentiation among team members as a characteristic of how well teams can perform. Newer representations of teamwork capture interdependence from a network perspective and highlight how the contributions that individuals make are dependent to their position within the team (Griffin et al., Reference Griffin, Somaraju, Dishop and DeShon2022). Without facilitation from team member to team member, it is difficult for a given member to complete team projects in isolation, no matter their singular capability. However, the characteristics required for such faciliatory positions can extend beyond simply GMA. Here, personality and even vocational interests may be more important for determining whether an individual has the characteristics required for the job. Moreover, in such positions, an individual’s job performance is better represented through the extent that they can facilitate others’ performance rather than completing their task. That is, knowledge, skills, abilities, and other characteristics for teamwork are important for the functioning of individuals within teams and the facilitation of team effectiveness (Mohammed & McKay, Reference Mohammed, McKay, Farr and Tippins2017), and these may not accurately be reflected in standard validity estimates.
Relatedly, interdependence can be viewed as differing levels of complexity. Bell and Kozlowski (Reference Bell and Kozlowski2002) discussed task complexity on a continuum. On the low end of complexity, tasks are more static and only loosely connected to their contexts and may have minimal temporal spacing or entrainment requirements. In comparison, high complexity tasks are more dynamic, coupled to their contexts, and may have high temporal entrainment and demanding pacing requirements. In highly complex environments, the value of teamwork is likely to increase as it becomes more necessary for individuals to work together effectively to accomplish their shared goals (Joshi & Roh, Reference Joshi and Roh2009). Complex tasks are likely to benefit from diversity more than simple tasks because of the benefits of the variety of backgrounds and expertise that can be brought to solve complex problems.
Taken together, we argue that although GMA may prove more important than noncognitive predictors for some team positions compared to others, many more factors are required to support overall team performance. Teamwork especially requires at the very least some level of diversity among skillsets and perspectives, and these are likely to increase in importance as interdependence and complexity increase. Therefore, it seems likely that the relative importance of GMA in selection will continue to decrease as work becomes increasingly team-oriented and dynamic.
Long-run implications
Finally, we would like to briefly tie this undervaluing of diversity in our present methods to a larger, ongoing discussion regarding the role of organizations and our science in the production of societal inequality. In particular, organizational scholars increasingly recognize that organizational processes affect the creation of wealth disparities and that these differences can compound over time (Amis et al., Reference Amis, Mair and Munir2020; Olenick et al., Reference Olenick, Ryan and Kuljanin2021). Historically, the approach to such disparities as practiced in the United States and largely supported by our field adheres to the equity norms espoused by equal opportunity employment. Unfortunately, despite these ostensibly equitable practices, wealth gaps, such as those between Black and White Americans, continue to grow (Derenoncourt et al., Reference Derenoncourt, Kim, Kuhn and Schularick2022; Somaraju & Olenick, Reference Somaraju and Olenick2022). These gaps are likely to continue to grow in part due to organizational processes as long as the current structures remain, which tend to reproduce past inequities such as the implicit treatment of Whiteness as a positive credential (Ray, Reference Ray2019). We believe correctly accounting for the effects of diversity in our selection processes by taking a more multilevel approach will help make organizations a more positive force for social change.
Acknowledgments
We gratefully acknowledge the National Defense Science and Engineering Graduate Fellowship program (NDSEG; Ajay V. Somaraju) for support that in part contributed to the composition of this article. Any opinions, findings, conclusions, and recommendations expressed are those of the authors and do not necessarily reflect the views of the NDSEG. We have no other known conflicts of interest to disclose.
