Recognizing that Speer, Dutta, Chen, and Trussell’s (Reference Speer, Dutta, Chen and Trussell2019) focal article was not intended to provide readers with an all-encompassing list of components that should be included in attrition models, there is a significant factor that appears to have been overlooked in their guide to attrition modeling: workforce diversity. The omission of diversity from much of the discussion surrounding input variables is of critical importance, as the turnover process does not unfold uniformly for individuals of various demographic traits. This assertion is substantiated by recent workforce data showing that Black Americans in the civilian labor force experience voluntary turnover at a rate that is more than 50% higher than that of White Americans in the labor force (U.S. Bureau of Labor Statistics, 2018a). Speer et al. (Reference Speer, Dutta, Chen and Trussell2019) briefly acknowledge the nuanced effects of demographic traits on voluntary turnover in their discussion regarding the relationship between employee age and different types of turnover, but their article largely neglects the role of diversity in voluntary attrition. The following commentary supplements their work by providing readers with insights as to where issues of diversity should be considered in attrition modeling and why such considerations matter.
Diversity is an essential component to accurately forecasting attrition
Speer et al. (Reference Speer, Dutta, Chen and Trussell2019) describe a variety of contextual variables that should be included in attrition models, but this section is notably silent on the topic of diversity. For example, although the article contends that commuting distance could be incorporated into attrition models, there is no discussion as to how the differential commuting trends of men and women should be considered. From a practical standpoint, this is problematic because women have a voluntary attrition rate that is approximately 7% higher than that of males in the U.S. workforce (U.S. Bureau of Labor Statistics, 2018b), but the mean commute time of women is more than 10% lower than that of their male counterparts, with approximately 58% of women in the U.S. workforce commuting under 25 minutes (United States Census Bureau, 2017). The tension in these statistics suggests that commute time may be unrelated to attrition. The more likely interpretation, however, is that the effect of commute time on voluntary attrition is not universal, and thus, this variable may have to be considered within the context of employee traits that have the potential to influence the relationship between commute time and likelihood of attrition.
Employee diversity should also be considered when including factors of withdrawal and embeddedness into attrition models. Speer et al. (Reference Speer, Dutta, Chen and Trussell2019) suggest that absences, increases in absences, and midweek absences are all indicators of turnover intentions. Although this may be true for the majority of the workforce, one must consider whether it is reasonable to flag a single parent as a flight risk simply because this employee has had a string of out-of-character absences. Such absences could merely be an indicator that the employee does not have a strong support network that can assist when a child is sick. In such a case, alternative work arrangements may be more appropriate than preparation for attrition. The article also proposes that embeddedness indicators such as social connections and internal employment references could play an important role in attrition models. Managers should consider, however, how the incorporation of these factors into attrition models will relate to occupational minorities. If organizations are recruiting outside of their traditional networks in order to reach more diverse populations, this should result in increased onboarding of employees with limited relationships within the organization. Consequently, employees who were recruited through efforts targeted at increasing organizational diversity may be flagged as flight risks by attrition models that rely on relationship-based job embeddedness indicators to predict voluntary turnover. These examples illustrate the potential for attrition models to inaccurately forecast voluntary turnover when said models do not account for workforce diversity.
Effective turnover interventions should consider diversity
Another section in which the article is notably silent on the topic of diversity pertains to the inclusion of attitudes in turnover models. This section discusses using constructs such as job satisfaction, organizational commitment, and engagement as predictors for turnover. Although the article suggests analyzing attitudinal variables at the group level to address issues around confidentiality, such aggregation may preclude organizations from realizing the full benefits of rich attitudinal datasets. Though an extensive body of literature has established relationships between a multitude of attitudinal variables and turnover intentions, there is also evidence that demographic characteristics can play an important role in the formation of attitudes.
For example, McKay et al. (Reference McKay, Avery, Tonidandel, Morris, Hernandez and Hebl2007) found that although the negative relationship between organizational commitment and turnover intentions was similar for White and Black managers, the negative indirect effect of diversity climate perceptions on turnover intentions through organizational commitment was considerably larger for Black managers than it was for White managers. In this case, although aggregating data for attrition modeling would still allow organizations to predict and prepare for turnover based on measures of organizational commitment, this aggregation of data would preclude them from identifying racial or ethnic groups that may be experiencing systematically lower levels of organizational commitment than their White peers due to factors such as diversity climate perceptions. Consequently, the aggregation of attitudinal data for attrition modeling could prevent organizational leaders from identifying interventions that could positively influence diversity climate perceptions, thus indirectly reducing turnover intentions for Black managers.
Similarly, the extant literature has long established a negative relationship between job satisfaction and turnover intentions (Podsakoff, LePine, & LePine, Reference Podsakoff, LePine and LePine2007). Knowledge regarding factors that influence job satisfaction, however, continues to grow. Although a rich body of work has demonstrated a negative relationship between work–family conflict and employee job satisfaction (Kossek & Ozeki, Reference Kossek and Ozeki1998), recent research suggests that how this relationship manifests differs for men and women (Zhao, Zhang, Kraimer, & Yang, Reference Zhao, Zhang, Kraimer and Yang2019). Accordingly, similar to the example provided above, when attitudinal data are analyzed at the group level such that survey responses of men and women are pooled, job satisfaction may be used to help predict turnover, but the identification of interventions that could positively influence job satisfaction may be limited. Though they do not represent an exhaustive list, these examples illustrate the need for attrition models to incorporate microlevel attitudinal data in order to identify appropriate turnover interventions, particularly within diverse populations.
Level of analysis can influence equitability of turnover interventions
There are two aspects to diversity that Speer et al. (Reference Speer, Dutta, Chen and Trussell2019) do incorporate into the article’s discussion of building attrition models. The first proposes that in-group similarity will result in greater liking, thus reducing turnover intentions. This suggestion is consistent with findings that likelihood of voluntary turnover decreases as the representation of one’s own race increases (Zatzick, Elvira, & Cohen, Reference Zatzick, Elvira and Cohen2003), but it requires the use of individual-level data (something the focal article has cautioned against) and provides employers with limited actionable steps to reduce turnover intentions. The second aspect discussed is an organization’s risk of facing legal action should an attrition model be used to implement an intervention that has negative impacts on members of a protected class. The article suggests that the use of group-level data may reduce risk of legal exposure, but again, it provides limited insight as to the factors that organizational leaders should consider in order to reduce discriminatory outcomes resulting from the leveraging of insights from attrition models.
As outlined above, not only are these aspects of diversity insufficient for consideration of diversity in attrition modeling, but the discussion on reducing exposure to legal action may be misleading in such a way that it provides organizational leaders with a false sense of security. I argue that the aggregation of data may limit insights, therefore placing organizations at risk of unknowingly implementing discriminatory interventions. Within the context of several previously discussed considerations, I will now provide a hypothetical example of how an intervention developed in response to analysis of group-level data could potentially increase the turnover intentions of some employees while also placing the organization at legal risk.
Consider an organization that collects attitudinal data and analyzes these data at the group level as a part of its attrition modeling process. In this example, the organization finds that employees have recently reported experiencing lower than normal levels of organizational commitment. Attrition modeling using historical data indicates that organizational commitment is linked to turnover, thus motivating the organization to dig deeper into this issue. The data show that reported organizational commitment is lowest in groups where the mean employee age falls between 25 and 45. The organization then integrates healthcare enrollment data into the analysis and finds that reported organizational commitment is lowest in groups where at least 70% of employees are enrolled in a family healthcare plan (as opposed to a single or married plan). Collectively, the data indicate that organizational commitment is lowest for employees with children. Senior management hypothesizes that organizational commitment is being influenced by high levels of work–family conflict and decides to develop an intervention.
Further investigation into the issue reveals that the standard workday of 7:30 am to 4:30 pm prevents employees from dropping their children off at school in the morning and picking them up in the afternoon. To address this, the company decides to allow employees to exercise a scheduling option whereas they can work from 8:30 am to 5:30 pm, enabling them to drop their children off at school in the morning. One year after the implementation of this intervention, the company measures organizational commitment again and finds that the mean employee organizational commitment score has increased by 12%. It appears that the intervention has been successful. Because the data were aggregated and analyzed at the group level, however, there are intervention effects that cannot be captured by analyses.
The intervention was very effective for employees who live within close proximity to the organization. Because the organization is geographically located within a predominantly White municipality, the positive effects of the intervention, however, were largely limited to White employees. Employees who commute 30 minutes or more to work were virtually unaffected by the intervention. As 90% of the organization’s Black employees are commuters in this example, most members of this demographic group did not benefit from the intervention. Additionally, the differential effects of the intervention made Black employees feel as though their perspectives were not respected and top leaders were not committed to an inclusive workplace. Consequently, the intervention actually resulted in reduced organizational commitment for Black employees. This reduction, however, was not captured by the data because these employees were distributed evenly throughout the organization, and data were aggregated at the group level.
Moreover, because it could be argued that the intervention was not a business necessity and had a disparate impact on members of a protected class, the intervention could also place the organization at risk of legal action (EEOC, 2019), the exact outcome from which the focal article claims group-level data can help shield organizations. Although, in this example, the use of group-level data limited insights that the organization could glean from analyses, these limitations could be deemed irrelevant in litigation as organizations can be held legally responsible for information that they should have known. Conversely, the use of individual-level data could have assisted the organization with identifying the potential for this turnover prevention intervention to have a disparate impact, thus guiding the organization toward a solution with more equitable outcomes and truly reducing the likelihood of exposure to litigation.
Final thoughts
There is no argument that Speer et al. (Reference Speer, Dutta, Chen and Trussell2019) have provided many valuable insights into factors and methodologies that can be incorporated into attrition modeling, but it is critical that the diversity of the workforce be considered throughout this process. This commentary by no means provides an exhaustive list of areas in which diversity can be incorporated into attrition modeling, but instead, it clearly demonstrates how important this topic is throughout the attrition modeling process, therefore illustrating why factors related to diversity need to be considered in every aspect of attrition modeling. Organizational scholars and practitioners are paying more attention to diversity and inclusion than ever before, but considerations of diversity and inclusion are often isolated from other organizational processes. A truly inclusive organization, however, is successful at incorporating these considerations into all processes in which the organization engages. By considering diversity in attrition modeling, organizations can position themselves to identify groups who may be at greater risk of voluntary attrition and implement interventions to reduce this risk, thus contributing to the development of a more inclusive workplace. Furthermore, these efforts should lead to more accurate attrition modeling and more effective turnover interventions for all employees within the organization.
Recognizing that Speer, Dutta, Chen, and Trussell’s (Reference Speer, Dutta, Chen and Trussell2019) focal article was not intended to provide readers with an all-encompassing list of components that should be included in attrition models, there is a significant factor that appears to have been overlooked in their guide to attrition modeling: workforce diversity. The omission of diversity from much of the discussion surrounding input variables is of critical importance, as the turnover process does not unfold uniformly for individuals of various demographic traits. This assertion is substantiated by recent workforce data showing that Black Americans in the civilian labor force experience voluntary turnover at a rate that is more than 50% higher than that of White Americans in the labor force (U.S. Bureau of Labor Statistics, 2018a). Speer et al. (Reference Speer, Dutta, Chen and Trussell2019) briefly acknowledge the nuanced effects of demographic traits on voluntary turnover in their discussion regarding the relationship between employee age and different types of turnover, but their article largely neglects the role of diversity in voluntary attrition. The following commentary supplements their work by providing readers with insights as to where issues of diversity should be considered in attrition modeling and why such considerations matter.
Diversity is an essential component to accurately forecasting attrition
Speer et al. (Reference Speer, Dutta, Chen and Trussell2019) describe a variety of contextual variables that should be included in attrition models, but this section is notably silent on the topic of diversity. For example, although the article contends that commuting distance could be incorporated into attrition models, there is no discussion as to how the differential commuting trends of men and women should be considered. From a practical standpoint, this is problematic because women have a voluntary attrition rate that is approximately 7% higher than that of males in the U.S. workforce (U.S. Bureau of Labor Statistics, 2018b), but the mean commute time of women is more than 10% lower than that of their male counterparts, with approximately 58% of women in the U.S. workforce commuting under 25 minutes (United States Census Bureau, 2017). The tension in these statistics suggests that commute time may be unrelated to attrition. The more likely interpretation, however, is that the effect of commute time on voluntary attrition is not universal, and thus, this variable may have to be considered within the context of employee traits that have the potential to influence the relationship between commute time and likelihood of attrition.
Employee diversity should also be considered when including factors of withdrawal and embeddedness into attrition models. Speer et al. (Reference Speer, Dutta, Chen and Trussell2019) suggest that absences, increases in absences, and midweek absences are all indicators of turnover intentions. Although this may be true for the majority of the workforce, one must consider whether it is reasonable to flag a single parent as a flight risk simply because this employee has had a string of out-of-character absences. Such absences could merely be an indicator that the employee does not have a strong support network that can assist when a child is sick. In such a case, alternative work arrangements may be more appropriate than preparation for attrition. The article also proposes that embeddedness indicators such as social connections and internal employment references could play an important role in attrition models. Managers should consider, however, how the incorporation of these factors into attrition models will relate to occupational minorities. If organizations are recruiting outside of their traditional networks in order to reach more diverse populations, this should result in increased onboarding of employees with limited relationships within the organization. Consequently, employees who were recruited through efforts targeted at increasing organizational diversity may be flagged as flight risks by attrition models that rely on relationship-based job embeddedness indicators to predict voluntary turnover. These examples illustrate the potential for attrition models to inaccurately forecast voluntary turnover when said models do not account for workforce diversity.
Effective turnover interventions should consider diversity
Another section in which the article is notably silent on the topic of diversity pertains to the inclusion of attitudes in turnover models. This section discusses using constructs such as job satisfaction, organizational commitment, and engagement as predictors for turnover. Although the article suggests analyzing attitudinal variables at the group level to address issues around confidentiality, such aggregation may preclude organizations from realizing the full benefits of rich attitudinal datasets. Though an extensive body of literature has established relationships between a multitude of attitudinal variables and turnover intentions, there is also evidence that demographic characteristics can play an important role in the formation of attitudes.
For example, McKay et al. (Reference McKay, Avery, Tonidandel, Morris, Hernandez and Hebl2007) found that although the negative relationship between organizational commitment and turnover intentions was similar for White and Black managers, the negative indirect effect of diversity climate perceptions on turnover intentions through organizational commitment was considerably larger for Black managers than it was for White managers. In this case, although aggregating data for attrition modeling would still allow organizations to predict and prepare for turnover based on measures of organizational commitment, this aggregation of data would preclude them from identifying racial or ethnic groups that may be experiencing systematically lower levels of organizational commitment than their White peers due to factors such as diversity climate perceptions. Consequently, the aggregation of attitudinal data for attrition modeling could prevent organizational leaders from identifying interventions that could positively influence diversity climate perceptions, thus indirectly reducing turnover intentions for Black managers.
Similarly, the extant literature has long established a negative relationship between job satisfaction and turnover intentions (Podsakoff, LePine, & LePine, Reference Podsakoff, LePine and LePine2007). Knowledge regarding factors that influence job satisfaction, however, continues to grow. Although a rich body of work has demonstrated a negative relationship between work–family conflict and employee job satisfaction (Kossek & Ozeki, Reference Kossek and Ozeki1998), recent research suggests that how this relationship manifests differs for men and women (Zhao, Zhang, Kraimer, & Yang, Reference Zhao, Zhang, Kraimer and Yang2019). Accordingly, similar to the example provided above, when attitudinal data are analyzed at the group level such that survey responses of men and women are pooled, job satisfaction may be used to help predict turnover, but the identification of interventions that could positively influence job satisfaction may be limited. Though they do not represent an exhaustive list, these examples illustrate the need for attrition models to incorporate microlevel attitudinal data in order to identify appropriate turnover interventions, particularly within diverse populations.
Level of analysis can influence equitability of turnover interventions
There are two aspects to diversity that Speer et al. (Reference Speer, Dutta, Chen and Trussell2019) do incorporate into the article’s discussion of building attrition models. The first proposes that in-group similarity will result in greater liking, thus reducing turnover intentions. This suggestion is consistent with findings that likelihood of voluntary turnover decreases as the representation of one’s own race increases (Zatzick, Elvira, & Cohen, Reference Zatzick, Elvira and Cohen2003), but it requires the use of individual-level data (something the focal article has cautioned against) and provides employers with limited actionable steps to reduce turnover intentions. The second aspect discussed is an organization’s risk of facing legal action should an attrition model be used to implement an intervention that has negative impacts on members of a protected class. The article suggests that the use of group-level data may reduce risk of legal exposure, but again, it provides limited insight as to the factors that organizational leaders should consider in order to reduce discriminatory outcomes resulting from the leveraging of insights from attrition models.
As outlined above, not only are these aspects of diversity insufficient for consideration of diversity in attrition modeling, but the discussion on reducing exposure to legal action may be misleading in such a way that it provides organizational leaders with a false sense of security. I argue that the aggregation of data may limit insights, therefore placing organizations at risk of unknowingly implementing discriminatory interventions. Within the context of several previously discussed considerations, I will now provide a hypothetical example of how an intervention developed in response to analysis of group-level data could potentially increase the turnover intentions of some employees while also placing the organization at legal risk.
Consider an organization that collects attitudinal data and analyzes these data at the group level as a part of its attrition modeling process. In this example, the organization finds that employees have recently reported experiencing lower than normal levels of organizational commitment. Attrition modeling using historical data indicates that organizational commitment is linked to turnover, thus motivating the organization to dig deeper into this issue. The data show that reported organizational commitment is lowest in groups where the mean employee age falls between 25 and 45. The organization then integrates healthcare enrollment data into the analysis and finds that reported organizational commitment is lowest in groups where at least 70% of employees are enrolled in a family healthcare plan (as opposed to a single or married plan). Collectively, the data indicate that organizational commitment is lowest for employees with children. Senior management hypothesizes that organizational commitment is being influenced by high levels of work–family conflict and decides to develop an intervention.
Further investigation into the issue reveals that the standard workday of 7:30 am to 4:30 pm prevents employees from dropping their children off at school in the morning and picking them up in the afternoon. To address this, the company decides to allow employees to exercise a scheduling option whereas they can work from 8:30 am to 5:30 pm, enabling them to drop their children off at school in the morning. One year after the implementation of this intervention, the company measures organizational commitment again and finds that the mean employee organizational commitment score has increased by 12%. It appears that the intervention has been successful. Because the data were aggregated and analyzed at the group level, however, there are intervention effects that cannot be captured by analyses.
The intervention was very effective for employees who live within close proximity to the organization. Because the organization is geographically located within a predominantly White municipality, the positive effects of the intervention, however, were largely limited to White employees. Employees who commute 30 minutes or more to work were virtually unaffected by the intervention. As 90% of the organization’s Black employees are commuters in this example, most members of this demographic group did not benefit from the intervention. Additionally, the differential effects of the intervention made Black employees feel as though their perspectives were not respected and top leaders were not committed to an inclusive workplace. Consequently, the intervention actually resulted in reduced organizational commitment for Black employees. This reduction, however, was not captured by the data because these employees were distributed evenly throughout the organization, and data were aggregated at the group level.
Moreover, because it could be argued that the intervention was not a business necessity and had a disparate impact on members of a protected class, the intervention could also place the organization at risk of legal action (EEOC, 2019), the exact outcome from which the focal article claims group-level data can help shield organizations. Although, in this example, the use of group-level data limited insights that the organization could glean from analyses, these limitations could be deemed irrelevant in litigation as organizations can be held legally responsible for information that they should have known. Conversely, the use of individual-level data could have assisted the organization with identifying the potential for this turnover prevention intervention to have a disparate impact, thus guiding the organization toward a solution with more equitable outcomes and truly reducing the likelihood of exposure to litigation.
Final thoughts
There is no argument that Speer et al. (Reference Speer, Dutta, Chen and Trussell2019) have provided many valuable insights into factors and methodologies that can be incorporated into attrition modeling, but it is critical that the diversity of the workforce be considered throughout this process. This commentary by no means provides an exhaustive list of areas in which diversity can be incorporated into attrition modeling, but instead, it clearly demonstrates how important this topic is throughout the attrition modeling process, therefore illustrating why factors related to diversity need to be considered in every aspect of attrition modeling. Organizational scholars and practitioners are paying more attention to diversity and inclusion than ever before, but considerations of diversity and inclusion are often isolated from other organizational processes. A truly inclusive organization, however, is successful at incorporating these considerations into all processes in which the organization engages. By considering diversity in attrition modeling, organizations can position themselves to identify groups who may be at greater risk of voluntary attrition and implement interventions to reduce this risk, thus contributing to the development of a more inclusive workplace. Furthermore, these efforts should lead to more accurate attrition modeling and more effective turnover interventions for all employees within the organization.