Melson-Silimon, Harris, Shoenfelt, Miller, and Carter (Reference Melson-Silimon, Harris, Shoenfelt, Miller and Carter2019) revisit some important issues concerning the proper and improper usage of clinical and subclinical personality scales in organization research and practice (see also Guenole, Reference Guenole2014). However, the conclusions and recommendations made in their focal article depend on critical assumptions about both the nature of personality as well as the appropriate ways to assess and analyze personality. Specifically, an understanding that underlies many of their recommendations is that normal and abnormal personality traits fall on a common five-factor structure that is largely sufficient for organizing the space of personality traits, and that normal and clinical measures of personality therefore measure essentially the same things, differing mostly in terms of extremity. This leads them to argue that since selecting on the basis of non-job-relevant mental disorders has been deemed illegal in US case law, in selection contexts “practitioners should target personality facets that have low correlations with traditional PDs [personality disorders]” (Melson-Silimon et al., Reference Melson-Silimon, Harris, Shoenfelt, Miller and Carter2019, p. 129) and should particularly avoid measures that assess subclinical variants of PDs. We think this advice is misguided for a number of reasons, particularly regarding the usage of subclinical measures. We organize our arguments around three points below.
#1. Are normal-range and clinical traits really “the same thing”? Not really
The core of the authors’ argument is based on an understanding of personality that suggests that all traits, clinical or otherwise, can be organized under something close to the Big Five or Five Factor Model (e.g., Krueger & Eaton, Reference Krueger and Eaton2010; Trull, Reference Trull2012; Widiger & Presnall, Reference Widiger and Presnall2013). It is incorrect to represent this “lumper” perspective as representing the consensus of personality researchers for a number of reasons. First, as Melson-Silimon et al. (Reference Melson-Silimon, Harris, Shoenfelt, Miller and Carter2019) note, the relationship between normal range measures and measures of PDs is far from unity. The correlations the authors report from Saulsman and Page’s (Reference Saulsman and Page2004, table 8) meta-analysis peak at r = .55. While this is certainly non-trivial, it is quite far from two constructs being “the same thing,” which indicated by a correlation of 1.Footnote
1
A counterpoint to the lumper perspective can be found within personality psychology among the numerous researchers working on functionalist and process approaches (e.g., Borsboom, Kievit, Cervone, & Hood, Reference Borsboom, Kievit, Cervone, Hood, Valsiner, Molenaar, Lyra and Chaudhary2009; Cramer et al., Reference Cramer, Van der Sluis, Noordhof, Wichers, Geschwind, Aggen and Borsboom2012; Denissen, Wood, & Penke, Reference Denissen, Wood and Penke2012; Fleeson & Jayawickreme, Reference Fleeson and Jayawickreme2015). Such models explain trait structures and covariation by understanding phenotypic traits such as the Big Five as composites or combinations of multiple psychological antecedent factors that include motives, abilities, and perceptual tendencies (Harms, Spain, & Wood, Reference Harms, Spain and Wood2014; McClelland, Reference McClelland1951; Wood, Gardner, & Harms, Reference Wood, Gardner and Harms2015). This approach is not inconsistent with traditional clinical approaches where personality disorders are treated as syndromes or clusters of distinct, though often comorbid, traits rather than a single unidimensional trait. For example, psychopathy is thought to consist of “a constellation of personality traits” (Lilienfeld, Reference Lilienfeld1994, p. 18) such as boldness, lack of empathy, and impulsivity (Patrick, Fowles, & Krueger, Reference Patrick, Fowles and Krueger2009). Reflecting the multidimensional nature of these personality disorders is important, as some of the underlying functional antecedents may be more likely to drive negative behavior than others and some may even be associated with positive outcomes (e.g., Grijalva et al., Reference Grijalva, Harms, Newman, Gaddis and Fraley2015; Landay, Harms, & Credé, Reference Landay, Harms and Credé2019; Neo, Sellbom, Smith, &Lilienfeld, Reference Neo, Sellbom, Smith and Lilienfeld2018).
When understood from this perspective, it ceases to be theoretically interesting or meaningful that there is significant overlap between normal and abnormal traits. Nor should we be surprised that we can make crude approximations of abnormal traits using subfacets of normal traits, since those subfacets more directly reflect the psychological antecedents of those traits. In fact, the covariation between trait dimensions can be almost completely accounted for by the degree to which their antecedent psychological factors overlap (Wood et al., Reference Wood, Gardner and Harms2015). Similarly, it can be shown that many relationships between conceptually distinct scales can be explained simply by the degree to which they contain similar item content (Arnulf, Larsen, Martinsen, & Bong, 2014; Wood & Harms, Reference Wood and Harms2016). At the same time though, these proxy measures will almost always fall short of capturing the range and breadth and abnormal traits because their item content will almost always be deficient in this regard (see DeSimone, Reference DeSimone2014). This can lead to dangerous and misleading results, particularly when these approximations are informed by questionable data and measures (e.g., Murphy, Reference Murphy2018).
The complexity of abnormal traits is also reflected in the different ways in which they are operationalized. While FFM traits are typically assessed using self-report measures, clinical diagnoses are often made on the basis of clinical judgment in conjunction with self-reported symptomology (First, Gibbon, Spitzer, Williams, & Benjamin, Reference First, Gibbon, Spitzer, Williams and Benjamin1997; Meehl, Reference Meehl1957). Thus, in addition to the aforementioned conceptual divergence between normal and abnormal traits, differences in operationalization (including, but not limited to, the role of clinical judgment) further underscore the distinction.Footnote
2
#2. “Dark-side” and “subclinical” traits are often very effective predictors of job performance
Although Melson-Silimon et al. (Reference Melson-Silimon, Harris, Shoenfelt, Miller and Carter2019) note that the ADA and ADAAA require that disabilities must be shown to substantially limit major life activities, they nonetheless argue strongly against the use of subclinical measures based simply on the potential threat of a lawsuit. We believe that this recommendation is misguided for a couple of reasons. First, subclinical traits, by their definition, do not interfere substantially with major life activities. Instead, it is believed that subclinical traits or “derailers” are most likely to manifest in negative ways during periods of stress or pressure when emotional and cognitive regulation breaks down (Hogan & Hogan, Reference Hogan and Hogan2001; Spain, Harms, & LeBreton, Reference Spain, Harms and LeBreton2014). Crucially, these “moments of weakness” may be especially important for workplace outcomes and relationships. For example, there is evidence that dark-side traits regularly account for incremental variance in job performance and leadership outcomes over Big Five traits, sometimes dramatically (e.g., Harms, Spain, & Hannah, Reference Harms, Spain and Hannah2011a; Scherer, Baysinger, Zolynsky, & LeBreton, Reference Scherer, Baysinger, Zolynsky and LeBreton2013). In fact, these supposedly “dark” traits may even be associated with positive work outcomes at times (e.g., Gaddis & Foster, Reference Gaddis and Foster2013; Grijalva et al., Reference Grijalva, Harms, Newman, Gaddis and Fraley2015; Harms, Spain, & Hannah, Reference Harms, Spain and Hannah2011b). Because such traits do not result in chronic problems, are unlikely to result in any problems that limit major life activities, and have been shown to be useful for predicting both training and performance, we believe that the recommendation to avoid such measures is misguided. Additionally, the ADA indicates that if a characteristic is a proximal contributor of the individual’s ability or propensity to complete essential job functions or the ability to perform them in safe manner, they can be used. In our opinion, the fear of lawsuits is insufficient justification for neglecting our obligation as industrial and organizational psychologists to maximize the validity of our selection procedures. The demonstrated value of these measures in forecasting the person’s effectiveness as an employee serves as a stronger legal defense in their use than the authors acknowledged.
#3. We do not have to follow this advice yet
Finally, the authors’ advice is not scientific but rather practical advice offered to avoid the unpleasantness of finding oneself sued in court. However, the perspective of clinical and normal traits existing on a common continuum is not currently the regnant perspective in the medical community. Rather, it was relegated to an “alternative model” of personality disorders within the DSM-5—i.e., it was essentially “tabled” for the time being, because the scientific and clinical communities in the American Psychiatric Association felt that “there was not sufficient evidence at the time to validate the proposed new personality disorder model and to establish its clinical utility” (Oldham, Reference Oldham2015, p. 234–235). As we have noted, we think there are some good scientific reasons for considerable pause in this perspective. But whatever the case, there is no immediate legal threat to assessing job-relevant abnormal or “dark-side” trait measures or item content.
The focal article describes a hypothetical scenario in which the ADA is expanded through the classification of normal-range or dark trait personality tests as medical exams. For the reasons argued above, we believe it would be unwise for the ADA to be expanded in such a way. It seems absurd that differences in self-rated agreeableness or narcissism would be afforded the same consideration as certain medically acknowledged mental or physical disorders. Thankfully, this is not currently the case.
It is worth remembering that ultimately the self-report inventories to assess dark personality tendencies consist of items like “Being rude and unfriendly is just a part of who I am,” and “I sometimes hit people to remind them who’s in charge” (both items on the Personality Inventory for DSM-5 [PID-5] Callousness scale; Krueger et al., Reference Krueger, Eaton, Derringer, Markon, Watson and Skodol2011). We suspect that judges who were informed that applicants endorsing such items were not hired because test validation studies have shown such item endorsements to be highly predictive of things like customer complaints and incidents of workplace violence—that is, undesirable, job-relevant outcomes that the applicant is virtually saying they will be more likely to promote by endorsing these statements—would be sympathetic to the argument that the job-relevance criterion has been satisfied.
In essence, we are in agreement with Melson-Silimon et al. (Reference Melson-Silimon, Harris, Shoenfelt, Miller and Carter2019) in suggesting that practitioners validate their tests as valid indicators of job performance (their recommendation #3). Demonstrating items or measures to be valid predictors of performance should largely inoculate against the need to omit a large class of useful items, particularly for a legal threat that has not yet been established.
Conclusion
Given these concerns, we are not convinced by many of the arguments and recommendations put forth by Melson-Silimon et al. (Reference Melson-Silimon, Harris, Shoenfelt, Miller and Carter2019). Simply put, the nature of the constructs and issues at hand are far more complicated than they are being represented. There is no current need to avoid using highly valid measures of dark traits that do not meet the ADA guidelines for disabilities. Instead, careful usage of personality instruments without aggregating to the syndrome level is likely to be legally defensible when the relation to job-relevant outcomes has been empirically established.
You have
Access
Melson-Silimon, Harris, Shoenfelt, Miller, and Carter (Reference Melson-Silimon, Harris, Shoenfelt, Miller and Carter2019) revisit some important issues concerning the proper and improper usage of clinical and subclinical personality scales in organization research and practice (see also Guenole, Reference Guenole2014). However, the conclusions and recommendations made in their focal article depend on critical assumptions about both the nature of personality as well as the appropriate ways to assess and analyze personality. Specifically, an understanding that underlies many of their recommendations is that normal and abnormal personality traits fall on a common five-factor structure that is largely sufficient for organizing the space of personality traits, and that normal and clinical measures of personality therefore measure essentially the same things, differing mostly in terms of extremity. This leads them to argue that since selecting on the basis of non-job-relevant mental disorders has been deemed illegal in US case law, in selection contexts “practitioners should target personality facets that have low correlations with traditional PDs [personality disorders]” (Melson-Silimon et al., Reference Melson-Silimon, Harris, Shoenfelt, Miller and Carter2019, p. 129) and should particularly avoid measures that assess subclinical variants of PDs. We think this advice is misguided for a number of reasons, particularly regarding the usage of subclinical measures. We organize our arguments around three points below.
#1. Are normal-range and clinical traits really “the same thing”? Not really
The core of the authors’ argument is based on an understanding of personality that suggests that all traits, clinical or otherwise, can be organized under something close to the Big Five or Five Factor Model (e.g., Krueger & Eaton, Reference Krueger and Eaton2010; Trull, Reference Trull2012; Widiger & Presnall, Reference Widiger and Presnall2013). It is incorrect to represent this “lumper” perspective as representing the consensus of personality researchers for a number of reasons. First, as Melson-Silimon et al. (Reference Melson-Silimon, Harris, Shoenfelt, Miller and Carter2019) note, the relationship between normal range measures and measures of PDs is far from unity. The correlations the authors report from Saulsman and Page’s (Reference Saulsman and Page2004, table 8) meta-analysis peak at r = .55. While this is certainly non-trivial, it is quite far from two constructs being “the same thing,” which indicated by a correlation of 1.Footnote 1
A counterpoint to the lumper perspective can be found within personality psychology among the numerous researchers working on functionalist and process approaches (e.g., Borsboom, Kievit, Cervone, & Hood, Reference Borsboom, Kievit, Cervone, Hood, Valsiner, Molenaar, Lyra and Chaudhary2009; Cramer et al., Reference Cramer, Van der Sluis, Noordhof, Wichers, Geschwind, Aggen and Borsboom2012; Denissen, Wood, & Penke, Reference Denissen, Wood and Penke2012; Fleeson & Jayawickreme, Reference Fleeson and Jayawickreme2015). Such models explain trait structures and covariation by understanding phenotypic traits such as the Big Five as composites or combinations of multiple psychological antecedent factors that include motives, abilities, and perceptual tendencies (Harms, Spain, & Wood, Reference Harms, Spain and Wood2014; McClelland, Reference McClelland1951; Wood, Gardner, & Harms, Reference Wood, Gardner and Harms2015). This approach is not inconsistent with traditional clinical approaches where personality disorders are treated as syndromes or clusters of distinct, though often comorbid, traits rather than a single unidimensional trait. For example, psychopathy is thought to consist of “a constellation of personality traits” (Lilienfeld, Reference Lilienfeld1994, p. 18) such as boldness, lack of empathy, and impulsivity (Patrick, Fowles, & Krueger, Reference Patrick, Fowles and Krueger2009). Reflecting the multidimensional nature of these personality disorders is important, as some of the underlying functional antecedents may be more likely to drive negative behavior than others and some may even be associated with positive outcomes (e.g., Grijalva et al., Reference Grijalva, Harms, Newman, Gaddis and Fraley2015; Landay, Harms, & Credé, Reference Landay, Harms and Credé2019; Neo, Sellbom, Smith, &Lilienfeld, Reference Neo, Sellbom, Smith and Lilienfeld2018).
When understood from this perspective, it ceases to be theoretically interesting or meaningful that there is significant overlap between normal and abnormal traits. Nor should we be surprised that we can make crude approximations of abnormal traits using subfacets of normal traits, since those subfacets more directly reflect the psychological antecedents of those traits. In fact, the covariation between trait dimensions can be almost completely accounted for by the degree to which their antecedent psychological factors overlap (Wood et al., Reference Wood, Gardner and Harms2015). Similarly, it can be shown that many relationships between conceptually distinct scales can be explained simply by the degree to which they contain similar item content (Arnulf, Larsen, Martinsen, & Bong, 2014; Wood & Harms, Reference Wood and Harms2016). At the same time though, these proxy measures will almost always fall short of capturing the range and breadth and abnormal traits because their item content will almost always be deficient in this regard (see DeSimone, Reference DeSimone2014). This can lead to dangerous and misleading results, particularly when these approximations are informed by questionable data and measures (e.g., Murphy, Reference Murphy2018).
The complexity of abnormal traits is also reflected in the different ways in which they are operationalized. While FFM traits are typically assessed using self-report measures, clinical diagnoses are often made on the basis of clinical judgment in conjunction with self-reported symptomology (First, Gibbon, Spitzer, Williams, & Benjamin, Reference First, Gibbon, Spitzer, Williams and Benjamin1997; Meehl, Reference Meehl1957). Thus, in addition to the aforementioned conceptual divergence between normal and abnormal traits, differences in operationalization (including, but not limited to, the role of clinical judgment) further underscore the distinction.Footnote 2
#2. “Dark-side” and “subclinical” traits are often very effective predictors of job performance
Although Melson-Silimon et al. (Reference Melson-Silimon, Harris, Shoenfelt, Miller and Carter2019) note that the ADA and ADAAA require that disabilities must be shown to substantially limit major life activities, they nonetheless argue strongly against the use of subclinical measures based simply on the potential threat of a lawsuit. We believe that this recommendation is misguided for a couple of reasons. First, subclinical traits, by their definition, do not interfere substantially with major life activities. Instead, it is believed that subclinical traits or “derailers” are most likely to manifest in negative ways during periods of stress or pressure when emotional and cognitive regulation breaks down (Hogan & Hogan, Reference Hogan and Hogan2001; Spain, Harms, & LeBreton, Reference Spain, Harms and LeBreton2014). Crucially, these “moments of weakness” may be especially important for workplace outcomes and relationships. For example, there is evidence that dark-side traits regularly account for incremental variance in job performance and leadership outcomes over Big Five traits, sometimes dramatically (e.g., Harms, Spain, & Hannah, Reference Harms, Spain and Hannah2011a; Scherer, Baysinger, Zolynsky, & LeBreton, Reference Scherer, Baysinger, Zolynsky and LeBreton2013). In fact, these supposedly “dark” traits may even be associated with positive work outcomes at times (e.g., Gaddis & Foster, Reference Gaddis and Foster2013; Grijalva et al., Reference Grijalva, Harms, Newman, Gaddis and Fraley2015; Harms, Spain, & Hannah, Reference Harms, Spain and Hannah2011b). Because such traits do not result in chronic problems, are unlikely to result in any problems that limit major life activities, and have been shown to be useful for predicting both training and performance, we believe that the recommendation to avoid such measures is misguided. Additionally, the ADA indicates that if a characteristic is a proximal contributor of the individual’s ability or propensity to complete essential job functions or the ability to perform them in safe manner, they can be used. In our opinion, the fear of lawsuits is insufficient justification for neglecting our obligation as industrial and organizational psychologists to maximize the validity of our selection procedures. The demonstrated value of these measures in forecasting the person’s effectiveness as an employee serves as a stronger legal defense in their use than the authors acknowledged.
#3. We do not have to follow this advice yet
Finally, the authors’ advice is not scientific but rather practical advice offered to avoid the unpleasantness of finding oneself sued in court. However, the perspective of clinical and normal traits existing on a common continuum is not currently the regnant perspective in the medical community. Rather, it was relegated to an “alternative model” of personality disorders within the DSM-5—i.e., it was essentially “tabled” for the time being, because the scientific and clinical communities in the American Psychiatric Association felt that “there was not sufficient evidence at the time to validate the proposed new personality disorder model and to establish its clinical utility” (Oldham, Reference Oldham2015, p. 234–235). As we have noted, we think there are some good scientific reasons for considerable pause in this perspective. But whatever the case, there is no immediate legal threat to assessing job-relevant abnormal or “dark-side” trait measures or item content.
The focal article describes a hypothetical scenario in which the ADA is expanded through the classification of normal-range or dark trait personality tests as medical exams. For the reasons argued above, we believe it would be unwise for the ADA to be expanded in such a way. It seems absurd that differences in self-rated agreeableness or narcissism would be afforded the same consideration as certain medically acknowledged mental or physical disorders. Thankfully, this is not currently the case.
It is worth remembering that ultimately the self-report inventories to assess dark personality tendencies consist of items like “Being rude and unfriendly is just a part of who I am,” and “I sometimes hit people to remind them who’s in charge” (both items on the Personality Inventory for DSM-5 [PID-5] Callousness scale; Krueger et al., Reference Krueger, Eaton, Derringer, Markon, Watson and Skodol2011). We suspect that judges who were informed that applicants endorsing such items were not hired because test validation studies have shown such item endorsements to be highly predictive of things like customer complaints and incidents of workplace violence—that is, undesirable, job-relevant outcomes that the applicant is virtually saying they will be more likely to promote by endorsing these statements—would be sympathetic to the argument that the job-relevance criterion has been satisfied.
In essence, we are in agreement with Melson-Silimon et al. (Reference Melson-Silimon, Harris, Shoenfelt, Miller and Carter2019) in suggesting that practitioners validate their tests as valid indicators of job performance (their recommendation #3). Demonstrating items or measures to be valid predictors of performance should largely inoculate against the need to omit a large class of useful items, particularly for a legal threat that has not yet been established.
Conclusion
Given these concerns, we are not convinced by many of the arguments and recommendations put forth by Melson-Silimon et al. (Reference Melson-Silimon, Harris, Shoenfelt, Miller and Carter2019). Simply put, the nature of the constructs and issues at hand are far more complicated than they are being represented. There is no current need to avoid using highly valid measures of dark traits that do not meet the ADA guidelines for disabilities. Instead, careful usage of personality instruments without aggregating to the syndrome level is likely to be legally defensible when the relation to job-relevant outcomes has been empirically established.