Moral Identity Complexity: Situated Morality Within and Across Work and Social Roles

Advancing Conventional Approaches to Moral Identity

The Content of Moral Identity

No operationalization can cover all possible dimensions of identity attributes that individuals may hold. Self-complexity research (Linville, 1987; Woolfolk et al., 1995, 2004) thus seeks to identify a core, but not exhaustive, set of attributes to serve as representative indicators to estimate individuals’ levels of complexity. We used four criteria in our selection. Our first criterion was to select dimensions with substantial applicability to a broad and general sample. We sought dimensions from mainstream ethical thinking that are generally held by modern society (e.g., we did not include Stoicism and its associated attributes like chastity). If the attributes used are not fairly commonly held or understood by the population of interest, then it could artificially inflate indices of participants’ identity complexity. This is because individuals may rate themselves low on a certain attribute merely because it is not applicable to (or understood) within the population. As a second criterion, we ensured each dimension is grounded in strong theoretical underpinnings and primary bodies of ethical thought as described below.

Third, it was important that each dimension could be appropriately operationalized using self-identity attributes. For example, while utilitarianism (the consequentialist belief that the moral acceptability of any act depends on the acceptability of its consequences) meets our first and second criteria, it fails the third. While certain individuals may use utilitarianism to consider their own or others’ actions and morality (the external world), it is not inherently something guided toward self-construal (the internal world). While one could potentially conceive of some self-defining attributes conceptually related to utilitarianism (e.g., pragmatic or outcome focused), those would be vague for our purpose and would be difficult to defend as being inherently “moral” attributes. Fourth and finally, we sought dimensions that are highly applicable to organizational contexts and that management research has linked to important outcomes.

Our four criteria led us to select four representative moral dimensions for this first test of moral identity complexity. The first is Justice, which is central in Aristotelian virtue ethics (Curzer, 2012), and its conceptualization is extended in a large canon of subsequent inquiry such as Rawls’s (1971) theory of justice. Due to the scarce resources, competition, and hierarchical power inherent in organizations, it is important that members treat each other and develop systems that are just and fair. Justice and fairness have thus received high levels of attention in management research and have been shown to be an important factor positively influencing individuals’ emotions, attitudes, thoughts, and behaviors at work (Colquitt, Conlon, Wesson, Porter, & Ng, 2001; Greenberg, 1990).

Benevolence is the second dimension, which also has a rich history in ethical thought, tracing through, among others, Aristotle’s pathos, Hume’s (1739/1973) benevolence, and Gilligan’s (1982) ethics of care. Given the demands and stressors faced in organizational life, forms of benevolence (e.g., compassion, care, empathy) have been of significant interest to the management field (e.g., Atkins & Parker, 2012) and are core to theories such as trustworthiness (Mayer, Davis, & Schoorman, 1995) and emotional intelligence (Boyatzis, Stubbs, & Taylor, 2002).

The third dimension is Obligation. This paradigm is grounded in Immanuel Kant’s deontology, focused on moral duties (Beller, 2008; Brink, 2006). Obligation is critical in organizations, wherein members are expected to subordinate self-interest to serve organizational goals and objectives, follow norms and codes of ethics (Folger, Ganegoda, Rice, Taylor, & Wo, 2013; Scanlon, 1998), and maintain accountability for one’s actions (Frink & Klimoski, 2004). Duty is also known to increase ethical behavior and reduce deviant behavior at work (Hannah, Jennings, Bluhm, Peng, & Schaubroeck, 2014). Such dutiful acts cannot be fully explained through the lenses of Benevolence or Justice. In fact, fulfilling obligations can serve to cause harm or injustice, and avoiding harm may prevent fulfilling one’s duties. A business leader, for example, may fulfill fiduciary duties to provide profits to shareholders at the sacrifice of employee pay and benefits. Thus, how individuals construe themselves relative to obligations is distinct and important in organizations.

Finally, while there are deeper arguments as to what constitutes our fourth dimension, Integrity (e.g., Becker, 1998), we focus on its simple form—that of being honest, trustworthy, and truthful. Integrity is also grounded in Aristotelian virtue ethics (Curzer, 2012) and the belief that human flourishing and moral functioning are driven, in part, by individuals’ desire to seek and speak the truth and to be trustworthy in word and deed (Hunter, 2000; Solomon, 1992). Indeed, along with benevolence as described previously and competence/ability, integrity is the third component of trustworthiness, which enables effective social relationships in organizations (Mayer et al., 1995). The etiology of integrity relates to having soundness and completeness, to align thought and word with deed. From the approach of virtue ethics, integrity thus has its own morality. It is “right” to be truthful, and trustworthy is what one should seek to “be,” regardless of whether being so assists or harms others or supports or counters one’s duties.

Brief coverage of these four moral dimensions highlights their important differences, which suggests that individuals can construe themselves relative to each dimension (to varied extents). Consistent with the self-complexity literature, such construals are not static or global psychological facets like traits, but they are dynamic structures that vary across time and roles based on unique social learning and developmental experiences (Markus & Kitayama, 1991; Woolfolk et al., 1995, 2004). Thus, individuals come to subscribe to certain moral dimensions more or less strongly across various societies, communities, organizations, professions, and roles (Hunter, 2000; Leavitt et al., 2012; Reicher & Levine, 1994). We therefore suggest that these four theorized dimensions of moral identity are also empirically discriminant.

The Structure of Moral Identity

We propose that moral identity is not only dimensional within role (based on the extent to which individuals endorse moral attributes reflecting each dimension of Benevolence, Justice, Obligation, and Integrity) as described previously, but that it is also differentiated between roles. Empirical research on self-complexity has shown that self-identity is a highly differentiated structure comprised of various sub-identities (Linville, 1987; Woolfolk et al., 1995, 2004). These sub-identities are thought to develop through social learning as individuals operate in various roles (e.g., manager, accountant, parent, daughter, etc.) over time and begin to define themselves relative to each role (Harter, 1999; Markus & Wurf, 1987). While self-identity typically has a requisite level of unity across roles, it also typically reflects significant differentiation across those roles (Kihlstrom, Beer, & Klein, 2003; Markus & Wurf, 1987). Despite these findings, research has proceeded to conceptualize moral identity as a unitary construct, assuming individuals have a moral identity, as described previously, versus moral content organized across sub-identities. This assumption has not been tested. We propose an alternate view—that one might possess an attribute (e.g., compassionate) in both her manager and mother roles, for example, but the attribute can be more self-defining in one role compared to the other role.

As a boundary to our theorizing, we focus on levels of self-complexity found in normal, psychologically healthy individuals. Excessive self-complexity can be pathological, reflected in maladies as severe as schizophrenia. Psychologically healthy individuals have adequate self-concept unity, defined as the maintenance of adequate coherence across sub-identities (Block, 1961; Campbell, Assanand, & Di Paula, 2003). Psychologically healthy individuals would, for example, describe themselves as possessing some level of honesty across roles (e.g., as a salesperson and as a parent) yet still have meaningful variance in honesty across those roles.

Self-complexity thus entails differentiation counterbalanced with adequate integration (Woolfolk et al., 1995, 2004). Understanding this is critical for interpreting the following hypotheses. Specifically, given the expectation of substantial self-concept unity in healthy individuals, we do not expect full compartmentalization, in which all four moral dimensions vary across every possible pair of role identities that an individual may hold. Instead, we expect to find reasonable unity yet meaningful variance in moral dimensions across role identities.

Effects of Moral Identity Complexity on Moral Self-Regulation

The generally positive effects of moral identity on ethical thought and behavior have been established (for review, see Jennings et al., 2015). Moral identity is theorized to drive these outcomes based largely on self-consistency motives (Blasi, 1980). Individuals are driven to align their ways of thinking and regulating behavior with their identity to be true to themselves in action (Harter, 1999). This is because individuals seek to verify their identity through acting in ways that validate that they are the person they believe themselves to be (Higgins, 1996; Stahlberg, Peterson, & Dauenheimer, 1999). Moral identity thus serves a critical self-regulatory, executive control function that drives moral action (Erikson, 1964). Consistent with this logic, moral identity has been positively related to ethical attitudes and behavioral intentions (e.g., Aquino, McFerran, & Laven, 2011; Lu & Chang, 2011) and ethical behavior (e.g., Reynolds, 2008; Stets & Carter, 2012) and negatively related to deviance (Mayer et al., 2012; Reynolds & Ceranic, 2007).

A self-complexity approach may enable a more refined appreciation for how moral identity wields its influence on ethical outcomes, enhancing our understanding of situated ethics. If the structure of moral identity is complex as proposed in Hypotheses 1a and 1b, whereby individuals define themselves more or less to various attributes across roles, then that structuring has significant implications for cross-situation variance in ethical intentions and behaviors based on which role (with its constituent moral attributes) is most activated at a given time.

Moral Identity Complexity and Cross-Role Variance

Given its vastness, individuals lack the capacity to access their self-identity in its entirety at any one time, and it would be inefficient to do so as only certain aspects of the self are relevant in any one context (Kihlstrom et al., 2003). Instead, individuals create complex identities over time tailored to various roles (Linville, 1987; Woolfolk et al., 1995, 2004). Thus, only relevant portions of the self are activated in any situation while others are suppressed. Markus and Wurf (1987) term this activated portion the working self.

Given its powerful self-regulatory functions, it follows that the activated part of identity provides the most proximal and significant influence on self-regulation (Mischel & Shoda, 1998). The idiosyncratic nature with which individuals structure moral identity attributes across various roles (e.g., as co-worker, son, etc.) would thus create differences in moral self-regulation based on which role is activated. To our knowledge, the structure of moral identity across roles has not been assessed. There is evidence, however, that individuals’ preferences, values, and ethical choices and behaviors change when different sub-identities are primed (LeBoeuf et al., 2010; Reicher & Levine, 1994). Most relevant to our research, Leavitt et al. (2012) found that when different professional identities are primed in an individual (e.g., engineer vs. manager), the individual responded with different ethical judgments and behaviors. This prior research, however, has not evidenced what is driving those cross-role differences. Based on the primacy of the working self on self-regulation (Markus & Wurf, 1987; Mischel & Shoda, 1998), we propose that when a role is primed, the attributes of moral identity related to that role activate and wield the most proximal influence on ethical intentions and behavior associated with that role.

Here, we reinforce a key theoretical boundary described previously concerning the self-concept unity found in normal, psychologically healthy individuals (Block, 1961; Campbell et al., 2003). This is critical for interpreting the following hypotheses. Given that moral content may be shared in varying levels across roles, we do not always expect a focal role identity (e.g., co-worker) to be exclusive in its positive association with ethical outcomes (e.g., deviance) in the same role. If an individual construes herself as “kind” as both a co-worker and a daughter, for example, then we would expect the shared variance in the moral identities of both roles to have some level of association with acts of kindness in ether role. Due to the influence of the activated working self on self-regulation (Markus & Wurf, 1987), however, we do expect that the focal role’s moral identity will exert disproportional influence on (un)ethical intentions and behaviors in that role. This would be evidenced empirically if the focal role identity (e.g., co-worker) fully or disproportionally accounts for the variance in a given ethical intention or behavior in that role relative to the moral identity of other roles (e.g., son/daughter). In other words, accounting for the effects of other role identities will not significantly increase the variance explained.

Moral Identity Complexity Instrument Development

We followed procedures typical of research on self-complexity (e.g., Woolfolk et al., 2004) in developing the moral identity instrument. First, we selected attributes to serve as representative indicators for the dimensions of interest: Benevolence, Justice, Obligation, and Integrity. We included Aquino and Reed’s (2002) nine traits (i.e., caring, compassionate, fair, friendly, generous, helpful, hardworking, honest, kind) for initial consideration (some were later pruned), and we generated 27 additional potential attributes based on review of the behavioral ethics literature and consultation with behavioral ethics scholars. We next sought to reduce the attributes to create a more parsimonious measure. To do this, six scholars outside the research team specializing in organizational behavior and/or behavioral ethics assessed content validity by classifying each of the 36 attributes into the four dimensions in a card-sorting exercise. We took the 28 attributes that these scholars reliably sorted into the intended dimension as the initial list (see Table S1 in the online supplement) to test further empirically.

Following Woolfolk et al. (2004), we placed all 28 attributes on each of four separate sheets, one sheet for each of four roles: co-worker, follower, son/daughter, and friend. We selected these representative roles for this first test of moral identity complexity as they establish a balance of work and social roles that were general enough to be held in common by all participants in the diverse samples. We used separate attribute sheets for each role, and we conducted distraction and priming tasks (described in the following) between the administrations of each sheet so that participants better deliberated on the unique attributes associated with each role separate from other roles.

We used various primes to activate semantic and episodic/autobiographical self-identity structures in participants prior to their rating the attribute list for each role (see description of priming in the online supplement). We asked participants to “please take a moment and think about yourself when in the role of a co-worker interacting with others who are peers,” then to “envision a typical day or a recent day interacting with your co-workers and accomplishing work tasks. Think about how you have felt when working as a co-worker and what types of things you may have talked about or thought about.” Next, we presented participants with two priming tasks. First, we asked participants to “please write 4 words that describe you as a co-worker interacting with others who are peers.” Second, we asked participants to “list the hair color and the eye color of two of the co-workers with whom you work.” After these priming tasks, following the method of Woolfolk et al. (1995, 2004), participants rated from 0 (not at all like me) to 5 (very much like me) the extent to which each of the 28 attributes described them as a co-worker.

We then presented participants with “filler questions” as a distraction task prior to rating attributes for the next role to limit rater effects. Between roles, participants listed their favorite food, movies, sports, vacations, and so on. Participants then completed the same exercise for the other three roles (follower, son/daughter, and friend), similarly priming each role prior to rating the attributes and then conducting a distraction task before rating the next role. We controlled for order effects by counterbalancing the order in which we presented the roles. One-fourth of participants rated the co-worker role first, one-fourth rated the follower role first, and so on. Participants then rated the additional measures unique to each sample for hypothesis testing described in the following.

Sample 1 Participants and Measures

We recruited participants at a university in the United States via a campus-wide e-mail to all university staff (e.g., administrators, career services employees, athletic department staff, etc.), coupled with in-class notices to business school students. Using snowball sampling, we invited respondents to bring others interested in participating. We paid each participant $20 cash, and we collected all data using paper surveys. To promote disclosure given the sensitive questions (e.g., deviant behavior), surveys were anonymous and without identifiers. Two of the 124 participants had numerous missing values and were deleted, producing a final sample of 122 (56 males and 66 females). Participants included university staff members, working professionals from outside the university, and business undergraduate and graduate students, with an average of 6.7 (SD = 9.3) years of work experience. The percentage of participants in each age range were 18 to 23 (59%), 24 to 28 (17%), 29 to 33 (6%), 34 to 38 (4%), and 39 or over (14%). Highest degrees included high school (38%), bachelor’s (37%), master’s (21%), beyond master’s (3%), and other (1%).

Descriptive statistics and scale reliabilities for Sample 1 are in Table 1, Panel A. For moral identity, participants rated from 0 (not at all like me) to 5 (very much like me) the extent to which each of the 28 attributes described them across each of the four counterbalanced roles. For testing H2, we then focused on how role moral identity may influence deviance across two distinct roles (son/daughter and co-worker). We assessed past deviance in each role using eight items from the measure by Spector et al. (2006). We adapted item wording slightly by referring to the target role (son/daughter, co-worker). We presented the eight items on separate sheets for each role after they responded to the various MI measures and engaged in a distraction task. After reminding participants that their reports were anonymous, participants rated how frequently they exhibited each deviant behavior from 1 (never) to 7 (very frequently). An example item was “Blamed my co-workers for an error that I made.” We used self-reports of deviance given that meta-analyses have shown that self-ratings of deviant behaviors are highly correlated with others’ ratings and that ratings by others do not explain additional variance beyond self-reports (Berry, Carpenter, & Barratt, 2012; Berry, Ones, & Sackett, 2007). Further, as individuals usually try to hide their deviance from others, scholars suggest that anonymous self-reports of deviant behavior may provide more accurate assessments (Aquino & Douglas, 2003; Spector & Fox, 2002). This is, in part, because deviance is a deprecating rating and thus is less prone to things like self-enhancement motives that often bias self-ratings of more “positive” constructs. We included control variables of age, gender, and education level in all analyses.

OPEN IN VIEWER

Table 1

Scale Descriptive Statistics, Cronbach’s Alpha, and Correlations

Panel A. Sample 1

Scale	Number of Items	Mean	SD	1	2	3	4	5	6	7
1. Son/daughter moral identity	12	4.27	.53	(.87)
2. Co-worker moral identity	12	4.37	.40	.61**	(.81)
3. Son/daughter deviant behavior	8	2.19	.68	−.52**	−.29**	(.71)
4. Co-worker deviant behavior	8	1.70	.66	−.08	−.38**	.27**	(.76)
5. Age	1	1.97	1.44	.20*	.13	−.17	.11	N/A
6. Education	1	2.90	.88	.05	−.07	.02	.15	.49**	N/A
7. Gender	1	N/A	N/A	−.14	.04	−.03	−.15	.02	−.05	N/A

OPEN IN VIEWER

Panel B. Sample 2

Scale	Number of Items	Mean	SD	1	2	3	4	5	6	7	8	9
1. Son/daughter moral identity	12	4.34	.45	(.89)
2. Co-worker moral identity	12	4.27	.38	.59**	(.86)
3. Son/daughter ethical intentions	4	4.51	.47	.35**	.16	N/A
4. Co-worker ethical intentions	4	4.08	.53	.28**	.38**	.51**	N/A
5. Son/daughter deviant behavior	8	1.82	.60	−.48**	−.29**	−.36**	−.33**	(.69)
6. Co-worker deviant behavior	8	1.86	.74	−.30**	−.41**	−.20*	−.40**	.47**	(.84)
7. Age	1	3.16	1.07	.14	.21*	.11	.23**	−.13	−.02	N/A
8. Education	1	3.22	.52	.00	−.17	.02	−.16	.06	.09	−.04	N/A
9. Gender	1	N/A	N/A	−.04	.17	−.05	.09	.08	−.22*	.05	−.19*	N/A

OPEN IN VIEWER

Panel C. Sample 5

Scale	Number of Items	Mean	SD	1	2	3	4	5	6
1. Son/daughter moral identity	12	4.25	.47	(.83)
2. Employee moral identity	12	4.26	.37	.57**	(.75)
3. Son/daughter scenario	1	3.63	.97	.16*	.14	N/A
4. Employee scenario	1	4.42	.69	.10	.13	.12	N/A
5. Age	1	31.34	5.68	.10	.11	−.02	−.08	N/A
6. Gender	1	N/A	N/A	.04	.13	−.04	.00	−.19*	N/A

Note: In Panels A and B, sample sizes range from 121 to 122 in Sample 1 and 124 to 127 in Sample 2 because of missing values. Cronbach’s alphas are shown in parentheses on the diagonal. We aggregated the four ethical intentions items in Sample 2 to form an index score. In Panel C, sample sizes range from 159 to 164 because of missing values. Cronbach’s alphas are shown in parentheses on the diagonal. We aggregated the three ethical intentions items in the Sample 5 experimental scenarios to form an index score. Because we only used Samples 3 and 4 for confirmatory factor analysis for the structure of the moral identity measure, we do not include correlation matrices as there are no other variables used in those samples.

*

p < .05 (two-tailed test).

**

p < .01 (two-tailed test).

Sample 2 Participants and Measures

A year after we collected Sample 1, we recruited a different set of employees via a campus e-mail notice to the staff at the same university as in Sample 1, again using snowball sampling. We also recruited working professional MBA students via in-class notices. We gave $20 cash to each participant. All surveys were paper-based, anonymous, and without identifiers. The sample included 127 participants (78 males and 49 females). Participants had an average of 10.6 (SD = 7.7) years of work experience. The percentage of participants in each age range was 18 to 23 (1%), 24 to 28 (32%), 29 to 33 (34%), 34 to 38 (17%), and 39 or over (16%). Highest degrees earned included high school (3%), bachelor’s (72%), master’s (23%), and beyond master’s (2%).

Descriptive statistics and scale reliabilities for Sample 2 are in Table 1, Panel B. In Sample 2, we used the same deviance and control variables as in Sample 1. Based on preliminary testing and refinement of the moral identity measure in Sample 1 (discussed in the following), we used 19 of the original 28 moral attributes in Sample 2. In addition, following Aquino et al. (2009), we assessed intentions for ethical behavior as an outcome of moral identity. Participants indicated the likelihood that they would exhibit four ethical behaviors within each of the two focal roles (son/daughter and co-worker) in the future ranging from 1 (very unlikely) to 5 (very likely). Example items include “tell my co-workers the truth even though it may bring me embarrassment or negative consequences” and “stand up against anyone who wrongs or commits injustices against my co-workers.” Items were equally relevant to both roles, and they were adjusted to each role by changing the target word (i.e., by using parents vs. co-workers).

Sample 3 Participants and Measures

We recruited an additional set of different employees via a campus e-mail announcement to the university staff at the same university as Sample 1 and 2, again using a snowball sampling approach. We gave $20 cash to each participant. All surveys were anonymous and did not have identifiers. The sample included 218 participants; 2 cases were removed because of missing data, leaving a net sample size of 216. These included 89 males and 126 females (1 participant did not indicate gender). Participants had an average of 16.22 (SD = 11.55) years of work experience. Participants were 38.11 (SD = 11.62) years old on average. Highest degrees earned included bachelor’s (52%), master’s (34%), high school (8%), and beyond master’s (6%).

Based on testing and refining the moral identity measure in Sample 2 (discussed in the following), we reduced the measure further to 12 attributes for Sample 3. We collected moral identity only for their role as an employee for use in CFA validation of the dimensional structure of moral identity. Note that in Samples 1 and 2, we wanted to maximize the number of roles to see if the structure of our new measure of MI would hold across multiple roles. In Sample 3, we were not testing cross-sample differences, and thus we only used one role. We believed that employee would be the most universal focal role (since some of the respondents had managerial responsibility).

Sample 4 Participants and Measures

One hundred and fifty-seven (153 male and 4 female) firefighters from a large fire department in the United States participated in the study. Participants were 39.44 (SD = 9.33) years old on average, had 11.20 (SD = 7.17) years of professional firefighting experience, and had 9.99 (SD = 7.17) years working for their current fire department. Some college credit but no degree earned (35%) was the most frequent education classification, followed by 23% bachelor’s degree, 16% associate’s degree, 14% high school graduate (diploma or equivalent), 10% trade/technical/vocational training, and 2% master’s degree.

We collected all data using paper surveys. We assessed moral identity (using the same 12 attributes employed in Sample 3). We collected moral identity only for their role as a firefighter for use in CFA validation of the dimensional structure of moral identity.

Sample 5 Participants and Measures

We recruited a new sample of working professional MBA students at the same university. Participants were each paid $20 cash for participation. The sample included 170 participants, 6 of whom were deleted because they had incomplete data on moral identity, leaving a final sample of 164 participants (95 males and 69 females). Participants had an average of 9.50 (SD = 6.25) years prior work experience, and they were 31.34 (SD = 5.68) years old on average.

This sample was collected to conduct a situated within-participant experiment, described in the hypotheses testing section in the following. We used the same methods and measures to assess moral identity as described previously for two distinct roles (son/daughter and employee). We counterbalanced the two roles to control for order effects. We collected all data using paper surveys. Using repeated measures, we then primed the two distinct roles in participants, one after the other, and we observed the relative influence of each primed role on their ethical cognition. Descriptive statistics and scale reliabilities for Sample 5 are reported in Table 1, Panel C.

Hypothesis 1: The Content and Structure of Moral Identity

Tests for Common Method Variance

We attempted to limit the potential for common method variance (CMV) first by counterbalancing the in-role measures for MI and second by separating the measures of MI from the outcome measures (deviant behavior and ethical intentions) by placing other survey measures and distraction tasks in between them. Prior to testing H2 and H3, however, we also tested for potential CMV using Harman’s single-factor test (Podsakoff, MacKenzie, & Lee, 2003) with Samples 1 and 2. We put all measures for in-role MI and in-role outcomes (deviant behavior and ethical intentions) into a principal component analysis, and we repeated this analysis across all four roles and in both samples. If CMV is a problem, then items would load on a single factor, and that factor would explain the majority of the total variance. Instead, we found that the data consisted of at least eight components with eigenvalues of more than 1.0, and these components collectively explained 66% or more of the total variance. In addition, the factor with the largest amount of explained variance did not exceed 29% in any of the eight tests (the range was from 12.4% to 28.8%). These results suggest that CMV is not a major concern in this study.

Hypothesis 2: Role Moral Identity and Within-Role Ethical Intentions and Deviance

Hypothesis 2 predicts that MI within a role is disproportionally positively related to ethical behavioral intentions (H2a) and disproportionally negatively related to deviance (H2b) in that role compared to the influence of the MI of other roles. Based on the principle of self-concept unity described previously, we expect that all role MIs in psychologically healthy individuals may have some level of relationship with ethical outcomes. H2a-b thus propose that the focal role MI has disproportional influence on ethical outcomes within that role compared to the MIs of other roles. For this and all subsequent hypothesis testing, we focused on two roles (son/daughter and either co-worker or employee) to test results across distinct roles, one from the respondent’s personal life and one from the respondent’s professional life.

Within-role ethical intentions

Table 2 Panels A-B show the separate tests of the effects of each role MI on ethical behavioral intentions within the respective role (H2a), using Sample 2. Note that we checked for multicollinearity among the independent variables before using regression for our tests. The highest correlation between two independent variables was .61 (see Table 1, Panel A). We also examined the variance inflation factors (VIFs; not shown), and all VIFs were less than 2.0. From this, we conclude that multicollinearity should not be a significant concern.

OPEN IN VIEWER

Table 2

Panel A. Influence of Son/Daughter Moral Integrity (MI) on Son/Daughter Future Ethical Behavioral Intentions: Sample 2

	Model 1				Model 2				Model 3
	Beta	t	p	Part	Beta	t	p	Part	Beta	t	p	Part
Age	.12	1.26	.21	.11	.06	.69	.49	.06	.06	.71	.48	.06
Gender	−.03	−.31	.76	−.03	−.06	−.69	.49	−.06	−.01	−.16	.87	−.01
Education	.01	.10	.92	.01	.06	.63	.53	.06	.03	.28	.78	.02
Co-worker MI					.21*	2.27	.03	.20	−.03	−.30	.76	−.03
Son/daughter role MI									.40**	3.78	.00	.32
Adjusted R2	−.01				.02				.12
∆R2	−.01				.03*	.10**

OPEN IN VIEWER

Panel B. Influence of Co-Worker MI on Co-Worker Future Ethical Behavioral Intentions: Sample 2

	Model 1				Model 2				Model 3
	Beta	t	p	Part	Beta	t	p	Part	Beta	t	p	Part
Age	.22**	2.53	.01	.22	.17	1.99	.05	.17	.13	1.58	.12	.13
Gender	.07	.83	.41	.07	.08	.92	.36	.08	.03	.39	.70	.03
Education	−.15	−1.70	.09	−.15	−.14	−1.57	.12	−.13	−.09	−1.08	.28	−.09
Son/daughter role MI					.29**	3.37	.00	.29	.11	1.07	.01	.09
Co-worker MI									.31**	2.87	.29	.24
Adjusted R2	.06				.12				.17
∆R2	.06*				.06**	.05**

Note: Sample sizes range from 124 to 125 due to missing values. The values shown represent standardized beta estimates and adjusted R² values. “Part” is the semi-partial correlation.

*

p < .05.

**

p < .01.

In both panels in Table 2, Model 1 reports only controls. Model 2 adds the nonfocal role MI. Model 3 adds the focal role MI. In both panels, Model 2 results show that the nonfocal role MI is significantly related to ethical intentions within the role. As shown in Model 3, however, once the focal role MI is added, the nonfocal role MI is no longer statistically significant, while the focal role is. In addition, adding the focal role MI substantively increases the amount of variance explained. H2a is thus supported.

As a more robust test of whether the focal role MI contributed significantly and substantively more to the overall variance explained than the nonfocal role MI, we assessed the squared semi-partial correlations (Cohen, Cohen, West, & Aiken, 2002). In both instances, the squared semi-partial correlations indicate that the contribution of the focal role MI toward the overall variance explained is substantively greater than the contribution of the nonfocal roles. From Table 2 Panel A, Model 3, the squared semi-partial correlations were .10 (.32 squared) for the son/daughter MI and .00 (−.03 squared) for the co-worker role. From Table 2 Panel B, Model 3, the squared semi-partial correlations were .06 (.24 squared) for the co-worker MI and .01 (.09 squared) for the son/daughter role MI.

Within-role deviance

Table 3 Panels A-D show the separate tests of the effects of each role MI on deviance within the respective role (H2b). We tested this hypothesis separately using first Sample 1 and then Sample 2. Table 3 Panels A and C report the results from Sample 1, while Table 3 Panels B and D report the results from Sample 2. The panels in this table are formatted and interpreted the same as those in Table 2. Results reported in Table 3, Model 2 (in each of the four panels) show that in three of the four cases, the nonfocal role MI is significantly related to deviance within the respective role. Once the focal role MI is added, however, the nonfocal role MI (Table 3, Model 3) is no longer a significant predictor. In addition, in all four cases, the amount of variance explained increases substantively by adding the focal role MI, and the squared semi-partial correlation for the focal role is substantively larger than that for the nonfocal role. H2b is thus supported.

OPEN IN VIEWER

Table 3

Panel A. Influence of Son/Daughter Moral Integrity (MI) on Son/Daughter Role Deviance: Sample 1

	Model 1				Model 2				Model 3
	Beta	t	p	Part	Beta	t	p	Part	Beta	t	p	Part
Age	−.23*	−2.20	.03	−.20	−.17	−1.69	.09	−.15	−.11	−1.21	.23	−.10
Gender	−.02	−.24	.81	−.02	−.02	−.24	.81	−.02	−.11	−1.36	.18	−.11
Education	.13	1.22	.23	.110	.09	.892	.39	.08	.10	1.12	.27	.09
Co-worker MI					−.26**	−2.88	.01	−.25	.08	.84	.41	.07
Son/daughter role MI									−.57**	−5.59	.00	−.44
Adjusted R2	.02				.07				.27
∆R2	.02				.05**	.20**

OPEN IN VIEWER

Panel B. Influence of Son/Daughter MI on Son/Daughter Role Deviance: Sample 2

	Model 1				Model 2				Model 3
	Beta	t	p	Part	Beta	t	p	Part	Beta	t	p	Part
Age	−.13	−1.49	.14	−.13	−.06	−.66	.51	−.06	−.05	−.64	.52	−.05
Gender	.11	1.19	.24	.11	.15	1.70	.09	.15	.09	1.08	.28	.09
Education	.07	.74	.46	.07	.01	.06	.95	.01	.04	.53	.60	.04
Co-worker MI					−.29**	−3.17	.00	−.27	.00	.00	.99	.00
Son/daughter role MI									−.46**	−4.60	.00	−.37
Adjusted R2	.00				.07				.21
∆R2	.00				.07**	.14**

OPEN IN VIEWER

Panel C. Influence of Co-Worker MI on Co-Worker Role Deviance: Sample 1

	Model 1				Model 2				Model 3
	Beta	t	p	Part	Beta	t	p	Part	Beta	t	p	Part
Age	.05	.49	.63	.045	.08	.76	.45	.07	.11	1.14	.26	.10
Gender	−.14	−1.55	.12	−.14	−.16	−1.73	.09	−.16	−.10	−1.21	.23	−.10
Education	.12	1.12	.27	.10	.11	1.03	.31	.09	.05	.49	.62	.04
Son/daughter role MI					−.13	−1.35	.18	−.14	.18	1.67	.10	.14
Co-worker MI									−.50**	−4.64	.00	−.39
Adjusted R2	.02				.03				.17
∆R2	.02				.01	.15**

OPEN IN VIEWER

Panel D. Influence of Co-Worker MI on Co-Worker Role Deviance: Sample 2

	Model 1				Model 2				Model 3
	Beta	t	p	Part	Beta	t	p	Part	Beta	t	p	Part
Age	−.01	−.06	.95	−.01	.05	.57	.57	.05	.09	1.08	.28	.09
Gender	−.21*	−2.27	.03	−.20	−.22	−2.49	.01	−.21	−.16	−1.91	.06	−.16
Education	.04	.48	.63	.04	.03	.33	.74	.03	−.02	−.20	.85	−.02
Son/daughter role MI					−.31**	−3.64	.00	−.31	−.12	−1.19	.24	−.10
Co-worker MI									−.33**	−3.09	.00	−.25
Adjusted R2	.02				.11				.17
∆R2	.02				.09**	.08**

Note: Sample sizes range from 121 for Sample 1 to 124 to 125 for Sample 2 due to missing values. The values shown represent standardized beta estimates and adjusted R² values. “Part” is the semi-partial correlation.

*

p < .05.

**

p < .01.

Hypothesis 3: Cross-Role Influence of Moral Identity Dimensions on Ethical Decisions

Tests of Hypothesis 2 demonstrate the disproportional influence of role MI on in-role ethical outcomes. Taking that a step further, given the dimensionality of MI as demonstrated in testing Hypothesis 1, Hypothesis 3 states that each dimension of moral identity within an active (primed) role will be disproportionally related to dimension-relevant corresponding ethical outcomes compared to the same dimension in other inactive (not primed) roles (H3a) and other moral dimensions within that same role (H3b). We test this by assessing whether high (low) MI integrity in an activated employee role should be associated with decisions reflecting high (low) integrity in that role and the Integrity dimension of other roles (Integrity dimension of the son/daughter role) and other MI dimensions within the same role (e.g., Justice dimension of the employee role) would have significantly less influence on those decisions. To test H3a and H3b, we conducted a within-participant/repeated measures experiment.

Results

We used regression analysis with responses to the two in-role (son/daughter, employee) scenarios as the dependent variable. To compute an index value for ethicality for the scenario responses, we reverse-coded the first two (unethical choice) items, and we summed them with the third (ethical choice) item. Thus, a higher score indicated a more ethical response. Table 4 Panels A and B show the results for the son/daughter role, and Table 4 Panels C and D show the results for the employee role. Table 4, Model 1 (in all panels) includes controls only (note that education was not measured but was naturally controlled for since all respondents in Sample 5 had completed a bachelor’s degree and were actively pursuing an MBA). Table 4, Model 2, Panels A and C add the focal (primed) role MI-Integrity dimension. Table 4, Model 3, Panels B and D add the nonfocal (not primed) role MI-Integrity dimension. Finally, Table 4, Model 4, Panels B and D replace the nonfocal role Integrity dimension with all of the other three focal (primed) role dimensions (Benevolence, Justice, and Obligation). The results provide full support across both roles for H3a and H3b.

OPEN IN VIEWER

Table 4

Panel A. Influence of Son/Daughter Integrity on Son/Daughter Scenario Intentions: Sample 5

	Model 1				Model 2
	Beta	t	p	Part	Beta	t	p	Part
Age	−.03	−.39	.69	−.03	−.05	−.57	.57	−.05
Gender	−.05	−.58	.56	−.05	−.05	−.62	.54	−.05
Son/daughter Integrity					.20**	2.50	.01	.20
Adjusted R2	−.01				.01
∆R2	−.01				.02*

OPEN IN VIEWER

Panel B. Influence of Son/Daughter Integrity on Son/Daughter Scenario Intentions: Sample 5

	Model 3				Model 4
	Beta	t	p	Part	Beta	t	p	Part
Age	−.05	−.59	.56	−.05	−.04	−.51	.60	−.04
Gender	−.05	−63	.53	−.05	−.04	−.54	.59	−.04
Son/daughter Integrity	.20*	2.12	.04	.17	.21*	2.14	.03	.17
Employee Integrity	−.01	−09	.93	−.01
Son/daughter Benevolence					.08	.94	.35	.07
Son/daughter Justice					−.02	−.25	.80	−.02
Son/daughter Obligation					−.05	−.54	.59	−.04
Adjusted R2	.02				.01
∆R2	.03*				.02*

OPEN IN VIEWER

Panel C. Influence of Employee Integrity on Employee Scenario Intentions: Sample 5

	Model 1				Model 2
	Beta	t	p	Part	Beta	t	p	Part
Age	−.08	−.99	.32	−.08	−.11	−1.46	.15	−.12
Gender	−.02	−.18	.86	−.02	−.01	−.01	.99	−.01
Employee Integrity					.26**	3.40	.00	.26
Adjusted R2	−.01				.06
∆R2	−.01				.07**

OPEN IN VIEWER

Panel D. Influence of Employee Integrity on Employee Scenario Intentions: Sample 5

	Model 3				Model 4
	Beta	t	p	Part	Beta	t	p	Part
Age	−.11	−1.45	.15	−.11	−.11	−1.43	.15	−.11
Gender	.00	.02	.99	.00	.01	.10	.92	.01
Employee Integrity	.28**	3.05	.00	.23	.28**	.3.11	.00	.24
Son/daughter Integrity	−.04	−.42	.67	−.03
Employee Benevolence					−.05	−.66	.51	−.05
Employee Justice					−.01	−.13	.90	−.01
Employee Obligation					−.02	−.20	.84	−.02
Adjusted R2	.05				.04
∆R2	.06**				.05**

Note: Sample sizes range from 159 to 163 due to missing values. The values shown represent standardized beta estimates and adjusted R² values. Education is not included as a control as all participants in this sample had bachelor’s degrees. “Part” is the semi-partial correlation.

*

p < .05.

**

p < .01.

First, the results support H3a for the primed son/daughter role. When the MI-Integrity dimension for the nonactivated employee role was added (Table 4, Panel B, Model 3), the son/daughter Integrity dimension remained significant while the employee Integrity dimension was not. Further, the squared semi-partial correlation for the son/daughter Integrity dimension (at .17 squared, or .03) was substantially larger than the employee Integrity dimension. The results also provided support for H3a for the activated employee role. When the Integrity dimension was added from the alternate son/daughter role (from Table 4, Panel D, Model 3), the employee Integrity dimension was significant while the son/daughter Integrity dimension was not. In addition, the squared semi-partial correlation for the employee Integrity dimension (at .23 squared, or .05) was larger than the squared semi-partial correlation for the son/daughter Integrity dimension.

The results for the son/daughter role provide full support for H3b. The son/daughter Integrity dimension was a significant predictor of integrity-related intentions in the son/daughter scenario (Table 4, Panel B, Model 4), and it had a disproportional influence compared to the other son/daughter dimensions. Specifically, when the other three son/daughter MI dimensions were added (Benevolence, Justice, and Obligation), only the Integrity dimension was individually statistically significant (Table 4, Panel B, Model 4). As shown by the semi-partial correlations, none of the other dimensions added substantive variance explained in the ethical choice beyond that provided by the Integrity dimension alone.

The results for the employee role also provide full support for H3b. The employee Integrity dimension was a statistically significant predictor of integrity-related intentions in the employee scenario (Table 4, Panel C, Model 2), and when the other three son/daughter dimensions were added, only Integrity was statistically significant (Table 4, Panel D, Model 4). In addition, the squared semi-partial correlation for Integrity (.24 squared, or .06) was substantively the largest.

Supplemental Tests

A logical question one may have concerns the relative predictive validity of the proposed measure vis a vis the five-item internalization scale from Aquino and Reed (2002). We collected the internalization scale in Samples 1 and 2, and we provide exploratory results here. We offer numerous cautions in interpreting these results. First, in Sample 1, the Cronbach’s alpha for the Aquino and Reed (2002) internalization scale was a low .57 (.83 in Sample 2). Second, five of the same moral attributes are referred to in both measures, creating some overlap (the two measures were correlated at .34 in Sample 1 and .36 in Sample 2). Third, because the two approaches conceptualize moral identity quite differently and have divergent operationalizations aligned with those separate conceptualizations, it might be argued that they are not measuring the same defined construct and thus a question of which is a “better” measure may be moot. Due to this third caution, we provide two forms of tests.

The first tests regress deviance and ethical intentions on the internalization scale and the new overall moral identity (depicted as “overall MI mean” in Figure 1) measure separately (i.e., the other measure is not in the model as a predictor), along with age, gender, and education (see Table S5 in the online supplement). Results using Sample 1 show that overall MI mean is related to deviance (r = −.46, p < .01) but the internalization measure is not (r = −.16, p > .05). Results using Sample 2 show that overall MI mean is related to deviance (r = −.48, p < .01), as is internalization (r = −.30, p < .01). Results using Sample 2 also show that overall MI mean is related to ethical intentions (r = .41, p < .01), as is internalization (r = .41, p < .01).

Despite the third caution described previously, in the second exploratory tests, we entered both measures as predictors in the same model along with controls (see Table S6 in the online supplement). Results using Sample 1 show that when both predictors are in the same model, overall MI mean is related to deviance (r = −.45, p < .01) but internalization is not (r = −.02, p > .05). Results using Sample 2 show that overall MI mean is related to deviance (r = −.41, p < .01) while internalization is not (r = −.15, p > .05). Results using Sample 2 also show that both overall MI mean (r = .29, p < .01) and internalization (r = .30, p < .01) are equally related to ethical intentions when included together as predictors.

Theoretical Implications

Prior research on moral identity has made important contributions to behavioral ethics research through enriching our understanding of the self-regulatory functions that identity imposes on moral thought and behavior. Prior research has conceptualized and operationalized moral identity as being a unified structure. We integrate moral identity theory with self-complexity theory (Linville, 1987; Markus & Wurf, 1987), seeking to better understand how individuals structure moral attributes across various sub-identities. We assessed the content and structuring of moral identity to establish its complexity and then assessed the resultant situated ethical responses of participants across five samples, including a within-participants repeated measures experiment.

The application of self-complexity methodologies (Woolfolk et al., 1995, 2004) enabled forming and testing original hypotheses. This includes assessing how the structure of moral identity, based on the content of role-based identities, predicts ethical intentions and behavior differentially across roles. Across six permutations of tests (both Tables 2 and 3), results showed that the moral identity associated with a given role significantly predicted ethical intentions and deviance within that particular role. Further, in all six cases, adding the focal role moral identity as a predictor added significant additional variance explained in the outcomes beyond that explained by another nonfocal role, demonstrating the general disproportional influence of role-based moral identity. While only a first step, this research provides a base of both theory and methodology from which researchers can assess the role of moral identity complexity in situated ethics.

These findings are important to advancing behavioral ethics theory and research. Research in management (Leavitt et al., 2012; LeBoeuf et al., 2010) and social psychology (Reicher & Levine, 1994) has shown that individuals’ ethical thoughts and behaviors change when different identities are primed, and it has conjectured that those identities must differ somehow. Yet, the empirical basis for such cross-role variance has been largely unanswered. Workers in complex organizations often assume multiple roles, and they thus form multiple occupational identities (Johnson, Morgeson, Ilgen, Meyer, & Lloyd, 2006). In a field experiment of project managers who were also professional engineers, for example, Leavitt et al. (2012) showed that when they primed the manager versus professional engineer identity, participants were significantly more likely to commit unethical acts in engineering projects. They theorized (but did not directly test) that the professional engineer identity contained higher levels of ethical standards given the profession’s code of ethics and norms. The approach used here directly assesses the moral identity of participants relative to such roles, allowing a more refined investigation to illuminate the specific identity content and structure driving the situated effects on ethical outcomes.

In these initial studies predicting ethical intentions and deviance, we focused on two distinct roles (employee and son/daughter). We described previously that psychologically healthy individuals have adequate coherence across sub-identities (i.e., self-concept unity, Block, 1961; Campbell et al., 2003), which allows them to maintain adequate consistency across roles. Self-concept unity can be thought of as the inverse of complexity (i.e., the more one is complex, the less unity they have). We theorized that roles that are more similar to one another (e.g., similar role demands, contextual factors, social learning factors) would thus be expected to have greater unity of moral content. Therefore, low unity/high complexity between a given pair of roles (i.e., shared variance) suggests that those roles would explain more unique variance as predictors.

Our intent, however, extended beyond prediction. We foremost sought deeper knowledge of moral identity itself. The self-complexity approach creates a detailed mapping of individuals’ moral identities, and it thus provides a richer formulation of moral identity (see Figure 1). Prior research conceptualized moral identity as being unitary in structure not only across roles but within-role. The current results demonstrate that a four-dimension structure reflecting domains of morality (Benevolence, Justice, Obligation, and Integrity) fits the data well for each role assessed across five samples. Together, these results thus suggest that moral identity is highly complex, being both dimensional within and varying across roles.

Further informing situated ethics—and consistent with our analyses showing that MI is dimensional within roles and that those dimensions generally differ across roles in all five samples (Hypotheses 1 and 2)—we found in testing Hypothesis 3a that when a role is primed, its moral dimensions exert predominant influence on corresponding ethical decisions. Specifically, a counterbalanced within-participant experiment, in which we primed two separate roles in each participant in succession and across two distinct scenarios, showed that the MI-Integrity dimension of the primed role was the predominant predictor of participants’ responses to the integrity-based scenario compared to the MI-Integrity dimension of the nonprimed role.

Hypothesis 3b, focusing on the unique effects of separate dimensions of within-role moral identity in predicting ethical intentions, was also supported. Results from the within-participant experiment showed that when the employee role was primed, the employee MI-Integrity dimension was the predominant predictor of integrity-based ethical choices compared to the other three employee MI dimensions (Justice, Benevolence, Obligation) as hypothesized. The same was true when priming the son/daughter role. We found that son/daughter MI-Integrity significantly predicted the integrity-based decision, and it had an additional unique effect when all four dimensions were included as predictors. The additional dimensions did not offer significant additional variance explained.

While only the first tests of the complexity of moral identity, these results built across five studies support that (a) individuals formulate differentiated moral content reflecting Benevolence, Justice, Obligation, and Integrity for each of the various “hats” they wear in life; (b) the identity structures associated with each role/hat can be more or less activated at different times; (c) the moral identity content associated with the activated role is the most proximal and significant driver of moral responses in that role; and (d) the moral identity content associated with a specific dimension of MI (i.e., MI-Integrity) is more significant in driving corresponding (integrity-based) ethical choices than the same dimension of MI from other roles and compared to other MI dimensions (i.e., Benevolence, Justice, and Obligation) within that same role.

Considering these results in total suggest that researchers should account for the differentiation of moral identity across and within roles in both their theory and their methods. Doing so can potentially inform theories or models attempting to explain why individuals may enact situated ethics. If the focus of investigation is limited to a single role (e.g., as a leader, as a stockbroker, etc.), then it is important that researchers specify the role of interest and orient respondents to report their moral identity relevant to the focal role versus responding “as a person in general.” Our results suggest that doing so will enhance precision and will not degrade their ability to predict ethical outcomes (at least those assessed here) related to that role. That is, the other role identities did not add explanatory power. When the research question requires investigating multi-role or cross-role behavior, such as in our experiment, however, researchers should include those additional roles relevant to the context/research setting and research question.

Practical Implications

The current results have numerous implications for workplace ethical intentions and behavior. Priming cues in the workplace serve to activate certain identity structures while suppressing others, which can be a largely automatic response (Kihlstrom et al., 2003; Leavitt et al., 2012). Our results show that the activated identity has the most proximal and powerful source of influence on self-regulation (Markus & Wurf, 1987), suggesting that employees’ ethical responses will vary based on which sub-identity(ies) is salient. This might lead to fruitful workplace interventions that create cues and reinforcements that activate desired identities. Might a medical device sales unit, for example, employ primes to activate members’ identities as a health care professional versus salesperson, and does the former lead to more ethical behaviors? Understanding the activations of different moral identities and which cues prime those identities could also help explain why an individual may seem to “bend the truth” when an identity as a company media spokesperson is active, for example, but be less willing to do so if another work (e.g., a safety inspector) or non-work (e.g., a parent) identity is activated.

Our results evidencing the complex content and structure of moral identity also has implications for the formation of occupational and professional identities. These role-based work identities have (or in some cases lack) associated norms and morals the group seeks to inculcate in members (Leavitt et al., 2012; Scanlon, 1998). Our results suggest moral identity is role-based. Therefore, interventions might be best targeted at role- or occupation-based identities, as opposed to general efforts that approach the moral self as being a global entity. This may entail, for example, distinct efforts to emphasize to a pharmaceutical executive the importance of integrity to her role as well as illuminate what integrity “looks like” (expected decisions and behaviors) in practice both when wearing her manager hat and her health care professional hat.

Limitations and Future Research Opportunities

Our nonexperimental samples did not have lagged data, and measures were self-reported. As described previously, we attempted to limit common method concerns by counterbalancing the measures, using distractions tasks between key measures, and using priming tasks before key measures. Further, the series of common method variance tests reported previously suggest common method variance is not an issue. As far as self-reports, moral identity and ethical intentions are inherently self-report measures, and we supported the validity of using self-reports of deviance in the Methods section. Yet, our empirical tests and results need to be interpreted with these potential limitations in mind. Having said that, we believe the experimental study, coupled with extensive replications across multiple samples, should provide confidence in the more complex structure of moral identity that we observed and its effects.

Self-identity is thought to be the vastest set of knowledge structures individuals possess (Kihlstrom et al., 2003). No measure of moral identity, or of any other form of self-identity, can assess its breadth. While we sought to assess a broader expanse of content and structure than previous unitary operationalizations of moral identity, the attributes used here are merely representative. In this first assessment of moral identity complexity, we also included a limited set of two work (follower and co-worker) and two non-work (son/daughter and friend) roles to test the initial hypotheses relating to the structure and content of MI. These roles were chosen in part because they would be common roles held across our diverse samples of participants. However, the methodology used here can be used for any single role (as we did with Samples 3 and 4) or multiple roles, with roles tailored to a research setting or phenomenon of interest. In a study of supervisors in a hospital setting, for example, supervisors could not only be administrators (responsible for efficiencies such as cost control) but also members of the medical profession (physicians or nurses) responsible for effective patient care and for adherence to medical ethics. Those dual roles’ moral identities could be measured and tested for cross-role ethical outcomes.

Future research should consider how many attributes and roles are needed to approximate adequately a participant’s identity content—that is, how many roles and attributes are needed in a self-complexity data matrix until adding additional content begins to offer diminishing returns? Yet, this question is also relative to what one intends to predict. If the ethical outcomes (decisions, behaviors, etc.) of only a certain role(s) and moral dimension(s) are of interest, then obviously a lesser expanse of participants’ moral identities could be measured.

Based on our four criteria presented previously, we focused on moral attributes reflecting aspects of Benevolence, Justice, Obligation, and Integrity. Yet, these dimensions are not intended to be exhaustive. For example, Haidt and colleagues (Graham, Haidt, & Nosek, 2009; Haidt, 2008) proposed that morality is organized around Benevolence (what they label harm/care), Justice (what they label fairness/reciprocity), and in-group/loyalty (which would overlap with Obligation in our framework), as well as authority/respect and purity/sanctity. There are other ethical theories such as Confucianism, spiritualism and divine command theory, existentialism, epicureanism, pragmatism, nihilism, naturalism, and many other ethical perspectives that could be considered in the future relevant to certain samples. Some forms of morality may also be more relevant in different cultures (Haidt, 2008). Thus, our intended contribution is not limited to offering a new measure or set of dimensions and attributes, per se, as much as it is the application of self-complexity theory and methodology to moral identity research broadly. Researchers can extend our approach and methods to other moral domains, dimensions, and attributes.

Further, it is important to distinguish moral identity complexity as operationalized here from those conceptualizations of “moral complexity” that scholars have described as reflecting the richness of moral knowledge structures one possesses, with that richness reflecting either moral self-knowledge or knowledge of a particular domain(s) of morality, such as medical ethics. Those approaches are driven instead by cognitive complexity, and they propose that greater richness enhances moral expertise and functioning (Hannah, Avolio, & May, 2011; Narvaez, 2010). A high score on the attribute honest in a moral identity measure captures only the extent to which individuals define themselves relative to that attribute, not the qualitative richness or expertise with which respondents understand and apply the concept of honesty.

Finally, future research might investigate whether moral identity complexity may inform what leads to moral disengagement or ethical blindness (Moore, Detert, Treviño, Baker, & Mayer, 2012; Tenbrunsel & Messick, 2004) in certain work roles. Such disengagement could reflect that the individual simply lacks adequate moral content in the activated role despite perhaps having higher moral content in other, yet inactive roles. This may explain in part why otherwise “good people” do “bad things” based on whether a morally rich or deficient role is activated.