Speer, Dutta, Chen, and Trussell’s (Reference Speer, Dutta, Chen and Trussell2019) article, which systematically incorporates insights from turnover theory research along with practical considerations, offers extremely timely guidance for attrition modelers. Frameworks like this serve as a handy resource for practitioners who have access to large-scale organizational data with limited substantive knowledge in the relevant psychological literature; as such, it exemplifies the true spirit of the Society for Industrial and Organizational Psychology’s (SIOP) scientist–practitioner model. Speer et al.’s major assertion is that data-driven modeling approaches would benefit from theoretical understanding of various psychological predictors of turnover as documented in the academic literature. In this commentary, I take this assertion one step further by suggesting that the information/knowledge flow can, and should, go the other way around as well.
Just as with any other data-driven approach to describing and predicting important organizational phenomena, attrition modeling has significant potential for advancing theory of turnover and related phenomena, and thus warrants the same level of attention and seriousness from those in academia. Theoretical advancement can occur in three distinctive ways (Woo, O’Boyle, & Spector, Reference Woo, O’Boyle and Spector2017): (1) to test and confirm a theory via a priori hypotheses (hypothetico-deductive approach); (2) to refine, extend, and/or combine existing ones based on unexpected findings at hand (abductive approach); or (3) to detect a new phenomenon that calls for theoretical interpretation (inductive approach). Below I cite a few examples of past turnover research that has taken a data-driven (inductive and/or abductive) approach toward gaining novel theoretical insights on a relatively understudied topic and discuss how attrition modeling with larger data sets would facilitate such goals. Next, I identify some areas within turnover research that need more theoretical development or a brand new theory altogether and discuss how taking a large-scale inductive modeling approach would be advantageous in those specific contexts.
Examples of data-driven turnover research: How attrition modeling might help
Example 1: Stayers and job seekers
Woo and Allen’s (Reference Woo and Allen2014) study consisted of two independent online samples of U.S. workers (Ns = 582 and 536) who were surveyed on a variety of turnover-related constructs and identified four major profiles of job seekers and stayers in the organization: embedded stayers, detached stayers, dissatisfied seekers, and script-driven seekers. These profiles need to be further replicated and validated using larger data sets spanning across multiple time points. One major advantage of attrition modeling lies in the ability to explore the dynamic nature of these configurations over time; for example, are these profiles of stayers and job seekers stable over time? Which profiles are more volatile than others? Are these profiles meaningfully associated with when, why, and how one leaves the organization? Empirical insights gained from such explorations would be informative for further enriching the theory of stayer–leaver profiles with a consideration of “time” and “trends” (Mitchell, Burch, & Lee, Reference Mitchell, Burch and Lee2014), which is sorely missing from the literature.
Example 2: Resignation styles
Klotz and Bolino (Reference Klotz and Bolino2016) used a grounded theory approach to inductively derive a taxonomy of resignation styles based on 53 people’s responses to (open-ended and close-ended) questions about how they experienced the resignation process. Seven distinct resignation styles were identified (i.e., by the book, perfunctory, grateful goodbye, in the loop, bridge burning, impulsive quitting, and avoidant) and differentially linked to antecedents (e.g., social exchange relationships) and outcomes (e.g., supervisor’s emotional reactions). In order to further strengthen the generalizability and applicability of such a taxonomy, attrition modeling with larger data sets could be applied: Organizations may collect written responses (textual data) to the same set of open-ended and close-ended questions used by Klotz and Bolino (Reference Klotz and Bolino2016), which will then be analyzed using text mining techniques to derive overarching themes as well as using machine-learning algorithms to find a classification scheme that maximizes prediction of outcomes.
Example 3: Prequitting behaviors
Gardner, Van Iddekinge, and Hom (Reference Gardner, Van Iddekinge and Hom2018) introduced the notion of using prequitting behaviors, which are observable behaviors exhibited by employees prior to quitting, and developed an initial measure that showed promise for forecasting future turnover beyond other established turnover predictors (e.g., demographics, performance). Attrition modeling (using a larger data set) would greatly enhance the researcher’s ability to derive a more nuanced theoretical structure of prequitting behaviors, possibly uncovering new or unexpected relationships between certain types of behaviors that are currently not theorized in the turnover literature.
Current gaps in turnover research that attrition modeling could address
Gap 1: Contextualizing existing theories
As discussed in a recent review of the turnover literature (Lee, Hom, Eberly, & Mitchell, Reference Lee, Hom, Eberly and Mitchell2017), most (if not all) existing theories of turnover do not effectively take into account contextual factors such as national and industry-specific culture and economic (and job market) conditions of local community. The attrition modeling approach, together with technological advances in the general area of data management, storage, and sharing, introduces a tremendous potential for new developments in multilevel theories of turnover based on massive data that span across cultural/national and organizational boundaries.
Gap 2: Internal job transfer and relocation
Echoing Speer et al.’s point about the importance of research (and the lack thereof) on “employee churn” (i.e., internal job transfer), I suggest that findings from attrition modeling can and should feed back into the theoretical development around this topic. Currently available theories of turnover do not offer clear explanations for (a) why certain employees choose to change and/or relocate jobs within firms as opposed to changing firms altogether; (b) whether and how intra-organizational job mobility affects subsequent performance and well-being outcomes; and (c) how different embeddedness-related constructs (e.g., on-the-job vs. community; focal employee vs. family’s embeddedness) may be affected by internal job transfer or relocation (Lee et al., Reference Lee, Hom, Eberly and Mitchell2017). Attrition modeling serves as a powerful tool to detect and discover patterns in the data based on these variables, which will then be used for constructing a systematic theoretical understanding that is unique to the internal (vs. external) job mobility.
Gap 3: Clarifying the interconnection of withdrawal behaviors
Scholars have proposed a few theories about how specific types of job and work withdrawal behaviors such as absenteeism, lateness, and turnover might relate to one another (Rosse & Miller, Reference Rosse, Miller, Goodman and Atkin1984). One possible theory is that these behaviors occur independently of one another (the “independence forms” model), whereas others argue that engaging in one type of withdrawal behavior (e.g., taking a day off) compensates the person’s need to avoid a negative work situation by engaging in other types of withdrawal behaviors (e.g., taking longer breaks while at work, leaving the job for good; the “compensatory forms” model). Another theory suggest that engaging in milder forms of work withdrawal (e.g., showing up late) leads to more severe withdrawal behaviors (e.g., not showing up for work, leaving for good) down the road (the “progression” model). As clearly articulated by Harrison and Newman (Reference Harrison, Newman, Weiner, Schmitt and Highhouse2013), the literature currently suffers from a lack of direct empirical support for any of the existing models. This is an area where attrition modeling using human resource information systems data containing daily attendance records spanning a large number of years would lead to particularly important theoretical insights: Such big-data approaches can uncover the unique and nuanced patterns of interconnections among specific withdrawal behaviors over time that potentially vary by individual and contextual factors such as personality, organizational tenure, education, and performance, among many other variables.
Closing remarks
Although many scholars in the field of industrial and organizational psychology are starting to appreciate the scientific value of data-driven approaches, there remains a critical need for cultivating greater openness toward explicitly inductive and/or abductive research endeavors that meaningfully contribute to theory development. Such intellectual openness among academics is a crucial ingredient for successfully actualizing the SIOP’s scientist–practitioner model in this big-data era.
Speer, Dutta, Chen, and Trussell’s (Reference Speer, Dutta, Chen and Trussell2019) article, which systematically incorporates insights from turnover theory research along with practical considerations, offers extremely timely guidance for attrition modelers. Frameworks like this serve as a handy resource for practitioners who have access to large-scale organizational data with limited substantive knowledge in the relevant psychological literature; as such, it exemplifies the true spirit of the Society for Industrial and Organizational Psychology’s (SIOP) scientist–practitioner model. Speer et al.’s major assertion is that data-driven modeling approaches would benefit from theoretical understanding of various psychological predictors of turnover as documented in the academic literature. In this commentary, I take this assertion one step further by suggesting that the information/knowledge flow can, and should, go the other way around as well.
Just as with any other data-driven approach to describing and predicting important organizational phenomena, attrition modeling has significant potential for advancing theory of turnover and related phenomena, and thus warrants the same level of attention and seriousness from those in academia. Theoretical advancement can occur in three distinctive ways (Woo, O’Boyle, & Spector, Reference Woo, O’Boyle and Spector2017): (1) to test and confirm a theory via a priori hypotheses (hypothetico-deductive approach); (2) to refine, extend, and/or combine existing ones based on unexpected findings at hand (abductive approach); or (3) to detect a new phenomenon that calls for theoretical interpretation (inductive approach). Below I cite a few examples of past turnover research that has taken a data-driven (inductive and/or abductive) approach toward gaining novel theoretical insights on a relatively understudied topic and discuss how attrition modeling with larger data sets would facilitate such goals. Next, I identify some areas within turnover research that need more theoretical development or a brand new theory altogether and discuss how taking a large-scale inductive modeling approach would be advantageous in those specific contexts.
Examples of data-driven turnover research: How attrition modeling might help
Example 1: Stayers and job seekers
Woo and Allen’s (Reference Woo and Allen2014) study consisted of two independent online samples of U.S. workers (Ns = 582 and 536) who were surveyed on a variety of turnover-related constructs and identified four major profiles of job seekers and stayers in the organization: embedded stayers, detached stayers, dissatisfied seekers, and script-driven seekers. These profiles need to be further replicated and validated using larger data sets spanning across multiple time points. One major advantage of attrition modeling lies in the ability to explore the dynamic nature of these configurations over time; for example, are these profiles of stayers and job seekers stable over time? Which profiles are more volatile than others? Are these profiles meaningfully associated with when, why, and how one leaves the organization? Empirical insights gained from such explorations would be informative for further enriching the theory of stayer–leaver profiles with a consideration of “time” and “trends” (Mitchell, Burch, & Lee, Reference Mitchell, Burch and Lee2014), which is sorely missing from the literature.
Example 2: Resignation styles
Klotz and Bolino (Reference Klotz and Bolino2016) used a grounded theory approach to inductively derive a taxonomy of resignation styles based on 53 people’s responses to (open-ended and close-ended) questions about how they experienced the resignation process. Seven distinct resignation styles were identified (i.e., by the book, perfunctory, grateful goodbye, in the loop, bridge burning, impulsive quitting, and avoidant) and differentially linked to antecedents (e.g., social exchange relationships) and outcomes (e.g., supervisor’s emotional reactions). In order to further strengthen the generalizability and applicability of such a taxonomy, attrition modeling with larger data sets could be applied: Organizations may collect written responses (textual data) to the same set of open-ended and close-ended questions used by Klotz and Bolino (Reference Klotz and Bolino2016), which will then be analyzed using text mining techniques to derive overarching themes as well as using machine-learning algorithms to find a classification scheme that maximizes prediction of outcomes.
Example 3: Prequitting behaviors
Gardner, Van Iddekinge, and Hom (Reference Gardner, Van Iddekinge and Hom2018) introduced the notion of using prequitting behaviors, which are observable behaviors exhibited by employees prior to quitting, and developed an initial measure that showed promise for forecasting future turnover beyond other established turnover predictors (e.g., demographics, performance). Attrition modeling (using a larger data set) would greatly enhance the researcher’s ability to derive a more nuanced theoretical structure of prequitting behaviors, possibly uncovering new or unexpected relationships between certain types of behaviors that are currently not theorized in the turnover literature.
Current gaps in turnover research that attrition modeling could address
Gap 1: Contextualizing existing theories
As discussed in a recent review of the turnover literature (Lee, Hom, Eberly, & Mitchell, Reference Lee, Hom, Eberly and Mitchell2017), most (if not all) existing theories of turnover do not effectively take into account contextual factors such as national and industry-specific culture and economic (and job market) conditions of local community. The attrition modeling approach, together with technological advances in the general area of data management, storage, and sharing, introduces a tremendous potential for new developments in multilevel theories of turnover based on massive data that span across cultural/national and organizational boundaries.
Gap 2: Internal job transfer and relocation
Echoing Speer et al.’s point about the importance of research (and the lack thereof) on “employee churn” (i.e., internal job transfer), I suggest that findings from attrition modeling can and should feed back into the theoretical development around this topic. Currently available theories of turnover do not offer clear explanations for (a) why certain employees choose to change and/or relocate jobs within firms as opposed to changing firms altogether; (b) whether and how intra-organizational job mobility affects subsequent performance and well-being outcomes; and (c) how different embeddedness-related constructs (e.g., on-the-job vs. community; focal employee vs. family’s embeddedness) may be affected by internal job transfer or relocation (Lee et al., Reference Lee, Hom, Eberly and Mitchell2017). Attrition modeling serves as a powerful tool to detect and discover patterns in the data based on these variables, which will then be used for constructing a systematic theoretical understanding that is unique to the internal (vs. external) job mobility.
Gap 3: Clarifying the interconnection of withdrawal behaviors
Scholars have proposed a few theories about how specific types of job and work withdrawal behaviors such as absenteeism, lateness, and turnover might relate to one another (Rosse & Miller, Reference Rosse, Miller, Goodman and Atkin1984). One possible theory is that these behaviors occur independently of one another (the “independence forms” model), whereas others argue that engaging in one type of withdrawal behavior (e.g., taking a day off) compensates the person’s need to avoid a negative work situation by engaging in other types of withdrawal behaviors (e.g., taking longer breaks while at work, leaving the job for good; the “compensatory forms” model). Another theory suggest that engaging in milder forms of work withdrawal (e.g., showing up late) leads to more severe withdrawal behaviors (e.g., not showing up for work, leaving for good) down the road (the “progression” model). As clearly articulated by Harrison and Newman (Reference Harrison, Newman, Weiner, Schmitt and Highhouse2013), the literature currently suffers from a lack of direct empirical support for any of the existing models. This is an area where attrition modeling using human resource information systems data containing daily attendance records spanning a large number of years would lead to particularly important theoretical insights: Such big-data approaches can uncover the unique and nuanced patterns of interconnections among specific withdrawal behaviors over time that potentially vary by individual and contextual factors such as personality, organizational tenure, education, and performance, among many other variables.
Closing remarks
Although many scholars in the field of industrial and organizational psychology are starting to appreciate the scientific value of data-driven approaches, there remains a critical need for cultivating greater openness toward explicitly inductive and/or abductive research endeavors that meaningfully contribute to theory development. Such intellectual openness among academics is a crucial ingredient for successfully actualizing the SIOP’s scientist–practitioner model in this big-data era.