Development and validation of the Moral Outcomes of Relationship Aggression Scale: A measure of moral distress following intimate partner violence perpetration

Abstract

Research with military veterans has established that distress may arise in response to perpetrating violent behaviors that violate individuals' moral beliefs. To date, no studies have similarly examined morally-related cognitive and emotional responses specifically among intimate partner violence (IPV) perpetrators. However, research on moral cognitions and emotions in response to IPV perpetration may inform understanding of the behavior and potential mechanisms for intervention. In the current series of four studies, we used classical test theory to develop a measure of moral distress following IPV perpetration that focuses on thoughts about the actions (assimilated cognitions), thoughts about the self due to one's actions (accommodated cognitions), and emotions experienced due to one's actions (moral emotions). Items were developed and tested among two samples of undergraduate students, and psychometric properties of the final measure were confirmed among two community samples. The final measure consists of three subscales consisting of five items each. Results demonstrate support for internal consistency and test–retest reliability, convergent, discriminant, and incremental validity, and factor structure. This measure can be used in future research designed to examine how individuals respond to their IPV perpetration, and to study the implications this may have for long-term outcomes and behavioral change.

1 ∣. INTRODUCTION

Intimate partner violence (IPV) is a prevalent public health issue in the United States. Each year, at least 22% of partnered individuals engage in IPV; rates among women are at least as high as those for men, and typically both partners engage in violence (Langhinrichsen-Rohling et al., 2012). Male-to-female IPV generally results in more severe consequences (Lawrence et al., 2012) and is more disapproved of than female-to-male IPV, though approval rates for both forms of IPV are low (Spencer et al., 2021). Incidents of IPV often occur repeatedly (Testa et al., 2003). Studies suggest that IPV peaks during early adulthood and declines over time, but stays relatively stable among individuals who remain with the same partner (Shortt et al., 2012). Perpetration may also increase with the occurrence of stressful life circumstances (e.g., economic stress, parenting stress; Capaldi et al., 2012). Johnson's (2008) research on typologies of IPV suggests that “situational couple violence” is most common and characterized by the aggression that occurs due to reactive escalation during the conflict, whereas “intimate terrorism” is defined by systematic efforts to control one's partner. Although situational couple violence is typically moderate in severity, it can have serious negative consequences (Whitaker et al., 2007).

A greater understanding of the experiences of IPV perpetrators will allow us to better understand how the behavior functions and inform ways to intervene. Therapeutic interventions often attempt to help clients conduct a functional analysis of their behavior, which involves identifying antecedents, the maladaptive behavior that occurs, and the consequences of the behavior (Haynes & O'Brien, 1990). With regard to antecedents of IPV perpetration, trait-level characteristics that are hypothesized to function as situational antecedents are well-established (Capaldi et al., 2012), and effectively serve to identify individuals who may be at risk for perpetration. For example, meta-analyses suggest that two common trait level characteristics strongly associated with IPV perpetration are shame and anger (e.g., Lawrence & Taft, 2013; Norlander & Eckhardt, 2005), and there is some evidence that anger may be a salient emotional state immediately before incidents of aggression (Marshall et al., 2018). However, less research exists in regard to the internal consequences of IPV perpetration, which could inform a complete functional analysis of the behavior and reduce the risk for further perpetration.

Application of behavioral theories of aggression suggests that IPV may occur repeatedly due to negative reinforcement, such that aggressive behavior reduces distressing emotions by ending the conflict (Bushman et al., 2001). Yet for many perpetrators, especially those whose aggression occurs reactively rather than instrumentally, there may instead be a punishing effect that occurs via increased negative cognitions and emotions pertaining to oneself. This would largely be a socially appropriate response, yet certain cognitions and emotions may be more or less adaptive. For example, some research on moral emotions suggests that experiences of guilt (i.e., internally focused emotion) may promote behavior change, whereas shame (i.e., externally focused emotion) may pose risk for additional aggression (e.g., Tangney et al., 2011). Thus, understanding of consequences for perpetrators could shed light upon how IPV is intrinsically reinforced or punished, which may represent more readily modifiable targets for intervention.

In other areas of violence research, primarily conducted among military veterans, researchers have begun to examine the concept of moral injury as a potentially distressing response to perpetrated violence. Although definitions of moral injury have varied, this concept generally refers to distress resulting from having perpetrated or witnessed behaviors that transgress one's moral beliefs (Griffin et al., 2019). The social-cognitive model of moral injury (Dennis et al., 2017) proposes that, after veterans perpetrate behaviors during military service that violate their moral beliefs, they may experience heightened moral emotions (i.e., guilt). Those who are unable to adequately reconcile their perceptions about the actions with their understanding of themselves may experience heightened psychological distress, which may then lead to changes in self-concept. These changes may be either intrapersonal (e.g., “I am flawed”) or interpersonal (e.g., “Others cannot be trusted”). In turn, either type of belief may contribute to maladaptive behavior, such as hostility and aggression toward the self or toward others.

Existing evidence supports the social-cognitive model of moral injury (Dennis et al., 2017). Specifically, military exposure to events that violate one's moral beliefs is associated with mental health sequelae (e.g., Currier et al., 2015). One study of service members found that despite perpetration-based morally transgressive events being less common, they are strongly associated with unwanted memories of the events as well as moral cognitions and emotions (e.g., guilt; Litz et al., 2018). Furthermore, veterans’ wartime involvement in potential moral transgressions is associated with their severity of physical aggression toward others (e.g., Dennis et al., 2017). Although less well established, some research documents moral injury following transgressive behavior among nonmilitary populations, and moral injury among perpetrators of interpersonal violence has been identified as an area for future research (Griffin et al., 2019).

IPV differs from war-related violence in that it does not occur within an institutional context characterized by violence. IPV perpetration has generally low levels of societal approval (Spencer et al., 2021), and may represent a moral violation of the commitment of love and affection that is typical of intimate relationships. Thus, moral distress may be an important, but historically overlooked, reaction to IPV perpetration. Existing research on IPV perpetration has documented a range of potential mechanisms associated with the behavior that may suggest a similar social-cognitive model of moral distress among IPV perpetrators. With respect to the first part of the aforementioned model, unfortunately, little research exists to inform whether IPV perpetrators tend to acknowledge their behavior as wrong or as violations of moral beliefs. Some research has explored retrospective attributions that IPV perpetrators make for the causes of their behavior, finding that common attributions for physical aggression include situational factors (e.g., attention-seeking) and distressing emotions (e.g., anger; Neal & Edwards, 2017). Relatedly, both men and women perpetrators engage in more partner blame than self-blame and minimize the severity of their behaviors (Henning et al., 2005). Although perpetrators may be motivated to make situational rather than self-blame attributions to avoid repercussions, moral injury research suggests that individuals may also attribute their transgressive behaviors to the broader context to reduce distressing emotions (Litz et al., 2009). Thus, IPV perpetrators may perceive their behavior as wrong but attribute their behavior to the context to feel better.

The social-cognitive model of moral injury (Dennis et al., 2017) focuses on guilt as a key moral emotion that may arise related to perceiving one's behavior as transgressing morals. For IPV perpetration, the role of guilt has not been well-documented, but there are consistent associations with shame (e.g., Lawrence & Taft, 2013). Shame is a moral emotion closely related to guilt but differing in terms of 1) greater emphasis on how others perceive one's actions, and 2) greater emotional intensity (Tangney et al., 1996). Within cross-sectional studies, associations have been inferred to mean that shame predicts perpetration, but these associations may alternatively be interpreted such that shame is a consequence of perpetration and/or that they operate bidirectionally. Shame may be the emotion related to IPV perpetration with the clearest relevance to moral distress. Additional research documents associations of anger and hostility with IPV perpetration, which may be either self- or other-focused (Norlander & Eckhardt, 2005). Thus, the distressing emotions associated with IPV perpetration may arise as a consequence of believing one's behaviors are wrong and may heighten the risk for subsequent IPV.

When individuals feel heightened distressing emotions in response to believing their behavior to be wrong, they may have difficulty reconciling this with how they understand themselves (e.g., “I am a good person”). When this occurs, Litz et al. (2009) suggest that they may resolve the conflict by either attributing their behavior to the context in which it occurred or by updating their understanding of the self to account for the transgressive behavior. With respect to IPV perpetration, updating how one understands the self can be either adaptive or maladaptive. When adaptive, IPV perpetrators may recognize themselves as someone who has done transgressive things but determine themselves as capable of changing their behavior and taking responsibility for their actions. These individuals may also be most receptive to change, given that motivation and readiness to change are important to IPV interventions (Eckhardt et al., 2013). When maladaptive, IPV perpetrators may make global, stable attributions about the self-being innately flawed or irredeemable, as seen for military-related moral injury (e.g., Litz et al., 2009). This may also impair social relationships, such that individuals experiencing moral distress are thought to have difficulty trusting others due to potential judgment (Vargas et al., 2013) and experience both self- and interpersonally-directed hostility (Owens et al., 2008). Relatedly, research suggests that IPV perpetrators may perceive their partners as threatening (e.g., angry or hostile), even when they display positive affect (Marshall & Holtzworth-Munroe, 2010). This may suggest maladaptive beliefs about the self among IPV perpetrators that create internal distress and perpetuate risk for subsequent IPV.

To our knowledge, no studies have directly documented the moral cognitions and emotions that IPV perpetrators hold about themselves due to their behaviors. Advancing this area of research is limited by a lack of adequate measurement tools. Some researchers have examined beliefs that individuals hold about IPV perpetration generally (e.g., Spencer et al., 2021), but these have not been anchored specifically to the thoughts and feelings that individuals hold in relation to their own behaviors. The majority of measures that do exist to assess moral injury have been developed and validated among military populations and contain military-specific content (e.g., 'I feel betrayed by fellow service members who I once trusted'; Nash et al., 2013) that would not translate well to civilian populations or specifically to IPV perpetration.

A recent measure was developed to assess moral distress arising from nonmilitary transgressions (Steinmetz et al., 2019), but this measure bears limitations for assessing the complexities of IPV. First, the measure was not developed specifically for violent transgressions and the psychometric studies included data anchored to a wide variety of perpetrated behaviors (e.g., abortion, infidelity). The rates at which individuals in the validation samples responded to the items specifically with respect to IPV perpetration were too low to draw psychometric conclusions specific to IPV perpetration. The items also conflate distinct cognitive and emotional components of moral distress specified by prior models (e.g., Dennis et al., 2017).

Given that no measures exist to adequately assess moral distress with respect to IPV specifically, herein we report on the systematic development of a measure of this construct among perpetrators of IPV, the Moral Outcomes of Relationship Aggression Scale (MORALS). Because there is some evidence that physical IPV is appraised more negatively than psychological IPV (e.g., Hammock et al., 2017), we expected that individuals are likely better able to identify perpetration of physical violence as morally right or wrong. Thus, we mostly focused on evaluating the measure specifically among individuals who have perpetrated physical IPV. Although perpetration-related moral distress and victimization-related moral distress (i.e., “betrayal-related moral injury”; Griffin et al., 2019) may both occur within the context of IPV, we focused on perpetration because existing trauma treatments are well-established and able to address moral cognitions and emotions for victims of traumatic experiences such as IPV (e.g., Resick et al., 2016), whereas interventions for IPV perpetrators are often ineffective (Eckhardt et al., 2013). Thus, the identification of moral injury as a potential mechanism of change may have more substantial implications for IPV perpetration interventions.

1.1 ∣. Defining the construct

Our conceptualization of moral distress related to IPV perpetration drew from the literature pertaining to IPV and related cognitions and emotions, moral injury, and posttraumatic stress disorder (PTSD). We defined IPV perpetration-related moral distress to be negative cognitions about the self and associated emotions that arise due to appraising the behavior to have violated moral values. Consistent with existing models of cognitive and emotional responses to moral transgressions (i.e., Litz et al., 2009; Tracy & Robins, 2006), our conceptualization encompasses three components: 1) assimilated cognitions, or beliefs that the actions were wrong, 2) accommodated cognitions, or negative beliefs about the self that arise because of these actions, and 3) distressing moral emotions that one feels in response to their actions.

1.2 ∣. Development and validation approach

The MORALS was developed and validated using a rational approach to test construction and classical test theory analyses (e.g., Clark & Watson, 1995; Nunnally & Bernstein, 1994) across four distinct studies. In Study 1, we developed and conducted a pilot test of items in a sample of undergraduate students. In Study 2, we administered revised items to a separate sample of undergraduate students and assessed reliability and validity. Undergraduate students were chosen for the initial two studies to increase feasibility and sample size while acknowledging that IPV perpetration is highly prevalent in young, college student samples (Marshall et al., 2011). In Study 3, we administered the items on two occasions to a sample of individuals via Mechanical Turk (MTurk) and examined the psychometric properties of the final measure and subscales. In Study 4, we administered the MORALS to low-income community couples to further examine its psychometric properties, while extending it to psychological aggression. Each study included online administration of the MORALS, specified that participation was voluntary, received Institutional Review Board approval from The Pennsylvania State University, and deidentified participants to promote their comfort when disclosing sensitive information about their IPV perpetration. See Supporting Information: Figure S1, for measure development procedures.

2 ∣. STUDY 1: MEASURE DEVELOPMENT AND INITIAL EVALUATION

2.1 ∣. Purpose

To develop an initial item set and revise based on item and scale characteristics.

2.2 ∣. Participants and procedures

Within a sample of 423 undergraduate students participating in a larger study for course credit, the measure was administered to 37 participants who endorsed having perpetrated IPV. See Table 1 for demographic characteristics.

TABLE 1.

Participant demographic characteristics across studies

Variable	Study 1 (N = 37)	Study 2 (N = 79)	Study 3 (N = 221)	Study 4 (N = 96)
Age M (SD)	18.97 (1.30)	19.26 (1.33)	31.21 (5.13)	34.40 (7.45)
Education M (SD)	12.78 (1.25)	12.84 (1.14)	14.67 (3.09)	13.47 (2.17)
Annual household income M (SD)	1.79 (2.45)	2.88 (5.90)	55.40 (32.66)	33.62 (28.56)
Gender N (%)
Male	3 (8.1)	15 (19.0)	98 (44.4)	37 (39.4)
Female	34 (91.9)	63 (79.7)	123 (55.7)	57 (60.6)
Neither male nor female	0 (0)	1 (1.3)	0 (0)	0 (0)
Race/ethnicity N (%)
American–Indian/Alaskan Native	0 (0)	1 (1.3)	10 (4.5)	2 (2.1)
Asian	3 (8.1)	6 (7.6)	23 (10.4)	2 (2.1)
Native Hawaiian/Pacific Islander	0 (0)	0 (0)	1 (0.5)	0 (0)
Black/African–American	1 (2.7)	8 (10.1)	26 (11.8)	25 (26.6)
Middle Eastern/North African	0 (0)	0 (0)	2 (0.9)	0 (0)
White	33 (89.2)	60 (75.9)	163 (73.8)	59 (62.8)
Hispanic or Latino	2 (5.4)	6 (7.6)	16 (7.2)	11 (11.7)
Unknown	0 (0)	0 (0)	1 (0.5)	–
More than one race	0 (0)	4 (5.1)	–	6 (6.4)
Sexual identity N (%)
Lesbian/gay	1 (2.7)	1 (1.3)	13 (5.9)	3 (3.2)
Bisexual	1 (2.7)	4 (5.1)	46 (20.8)	4 (4.3)
Queer	0 (0)	1 (1.3)	2 (0.9)	–
Heterosexual	33 (89.2)	69 (87.53)	160 (72.4)	84 (89.4)
Don't identify/did not disclose	–	4 (5.1)	–	3 (3.2)

Note: Income in thousands of dollars. Studies 1–3: Outliers removed on education and income. Studies 1, 2, and 4: Race is mutually exclusive, not ethnicity. Study 3: race and ethnicity are not mutually exclusive.

2.3 ∣. Measures

2.3.1 ∣. Moral distress

Initial items for the MORALS were developed to address our construct definition. First, naïve undergraduate research assistants were instructed to complete the initial measure as if they were study participants and provided feedback regarding item clarity, ease of responding, and time to complete the measure. Items were refined, then reviewed by six content experts who made recommendations for the addition and removal of items, clarifications of item wording, and modifications to the response format. Additional measure revisions were then made, resulting in 49 items from three subscales that aligned with our construct definition: assimilated cognitions (20 items), accommodated cognitions (16 items), and distressing emotions (13 items), each rated on a 6-point Likert scale (1 = strongly disagree; 6 = strongly agree).

2.3.2 ∣. Intimate partner violence

The 39-item Revised Conflict Tactics Scales (CTS2; Straus et al., 1996) assess perpetrated and experienced psychological, physical, and sexual aggression. We focused on the 12-item physical aggression perpetration scale (e.g., “I grabbed my partner”); those who endorsed at least one such behavior were administered the MORALS. The CTS2 demonstrates internal consistency and test–retest reliability (Straus et al., 1996; Vega & O'Leary, 2007) and convergent and discriminant validity (e.g., Straus et al., 1996). Because the original CTS2 response scale limits the upper end of behavior frequency estimates, we modified the response format to allow entry of the number of times each behavior occurred.

2.4 ∣. Analyses

We examined the MORALS’ item and scale level frequency distributions, skewness, and kurtosis to determine whether there was the adequate scale and within-item variability, with desired absolute skewness and kurtosis values less than 2 and 7, respectively. We also examined Cronbach's α, with desired values above .70.

2.5 ∣. Results

Eight items demonstrated high positive skew (one assimilated, five accommodated, and two distressing emotions). No items were highly kurtotic. There was greater than 25% of respondents who endorsed the lowest response option (“strongly disagree”) for all items on the accommodation and emotion subscales, and who endorsed the highest possible response option (“strongly agree”) for eight items on the assimilated subscale. Item means for the total measure ranged from 1.36 to 4.24 (on a 1–6 scale), with the mean item average being 2.38. The assimilated subscale ranged from 1.51 to 4.24, with a mean item average of 3.13. The accommodated subscale ranged from 1.36 to 2.53, with a mean item average of 1.65. Finally, the distressing emotions subscale ranged from 1.56 to 3.31 with a mean item average of 2.19.

As shown in Table 2, skew and kurtosis were acceptable for the total measure and the subscales, but were slightly high for the accommodated subscale. There were floor or ceiling effects (i.e., >15% endorsing the lowest or highest scores; Terwee et al., 2007) for the accommodated and distressing emotions subscales, but not for the total measure or assimilated subscale. Reliability was strong (α's = .94–.97). Results informed item revisions, with a focus on eliciting fewer extreme responses. No items were dropped at this point.

TABLE 2.

Total MORALS and subscale distributions across studies

	α	M (SD)	α Men/women	M Men	M Women	# Items	Minimum	Maximum	Skew (SE)	% Minimum	% Maximum
Study 1
Total	.97	114.67 (48.84)	–	–	–	49	49	294	0.28 (0.39)	11	0
Assimilated	.97	61.11 (27.01)	–	–	–	20	20	120	−0.25 (0.39)	14	0
Accommodated	.97	23.69 (13.90)	–	–	–	16	16	96	1.84 (0.39)	36	0
Emotions	.94	29.77 (15.19)	–	–	–	13	13	78	0.62 (0.39)	19	0
Study 2
Total	.94	52.04 (19.60)	.97/.94	53.40	51.92	15	15	90	0.13 (0.27)	3	5
Assimilated	.90	19.95 (7.39)	.97/.92	21.00	20.00	5	5	30	−0.41 (0.27)	5	11
Accommodated	.88	15.91 (7.13)	.94/.85	15.60	16.00	5	5	30	0.30 (0.27)	10	5
Emotions	.93	16.18 (7.71)	.90/.91	16.80	15.00	5	5	30	0.21 (0.27)	9	9
Study 3
Total	.91	61.91 (13.85)	.87/.92	62.49	61.44	15	15	90	−0.52 (0.16)	1	2
Assimilated	.83	21.63 (5.33)	.74/.87	21.66	21.60	5	5	30	−0.72 (0.16)	2	6
Accommodated	.81	19.60 (5.50)	.73/.84	20.30	19.04	5	5	30	−0.58 (0.16)	2	3
Emotions	.88	20.68 (5.63)	.85/.90	20.53	20.80	5	5	30	−0.66 (0.16)	3	5
Study 4
Total	.92	47.33 (16.16)	.92/.91	49.59	45.57	15	15	90	−0.04 (0.25)	2	0
Assimilated	.89	14.40 (6.55)	.91/.87	15.73	13.54	5	5	30	0.36 (0.25)	12	1
Accommodated	.78	15.37 (5.87)	.81/.77	15.92	14.86	5	5	30	0.13 (0.25)	8	2
Emotions	.89	17.58 (6.64)	.89/.90	17.95	17.24	5	5	30	−0.13 (0.25)	5	6

Abbreviation: MORALS, Moral Outcomes of Relationship Aggression scale.

3 ∣. STUDY 2: EVALUATION OF RELIABILITY AND VALIDITY

3.1 ∣. Purpose

Study 2 involved the examination of item and scale characteristics for the revised measure. We also examined convergent validity and factor structure.

3.2 ∣. Participants and procedures

Among a sample of 442 undergraduate students participating in exchange for course credit, 79 who endorsed having perpetrated IPV completed the MORALS. See Table 1 for demographics.

3.3 ∣. Measures

3.3.1 ∣. Intimate partner violence

See Study 1 for measure description. Cronbach's α = .67 in the current study.

3.3.2 ∣. Moral distress

See Study 1 for the description of the MORALS. We administered the revised pool of 49 items.

3.3.3 ∣. Shame

The 18-item Other as Shamer Scale (OASS; Allan & Goss, 1994; Goss et al., 1994) demonstrates convergent and factorial validity and was used to assess self-perceptions of how others view the participant (e.g., “I think others are able to see my defects”). Items were rated on a 5-point Likert response scale (0 = never; 4 = almost always). Cronbach's α = .96.

3.3.4 ∣. Guilt

The 10-item state guilt subscale of the Guilt Inventory (GI; Kugler & Jones, 1992) was used to assess recent experiences of guilt (e.g., “I have recently done something that I deeply regret”). This subscale demonstrates convergent and discriminant validity and internal consistency reliability (Jones & Kugler, 1993; Kugler & Jones, 1992). Items were rated on a 5-point Likert response scale (1 = strongly agree; 5 = strongly disagree). Cronbach's α = .82.

3.3.5 ∣. Negative affect

The 10-item negative affect scale (NAS) from the Positive and Negative Affect Scales (PANAS; Watson et al., 1988) demonstrates strong internal consistency, test–retest reliability, and convergent validity. This scale was used to assess the extent of distressing emotions (e.g., “hostile,” “scared”) felt in the past year using a 5-point Likert response format (1 = very slight or not at all; 5 = extremely). Cronbach's α = .92.

3.3.6 ∣. Social desirability

We used a 13-item version of the Marlowe–Crowne measure of social desirability (MCSD; Crowne & Marlowe, 1960), which demonstrates internal consistency and concurrent validity (Reynolds, 1982). Items with a low base rate of occurrence (e.g., “I sometimes feel resentful when I don't get my way”) are rated as 0 = “false” or 1 = “true.” Responses are summed consistent with the original measure (Reynolds, 1982), such that higher scores indicate less of a tendency to present oneself favorably. Cronbach's α = .61.

3.3.7 ∣. Posttraumatic stress disorder

The 20-item Posttraumatic Checklist for DSM-5 (PCL-5; Blevins et al., 2015) demonstrates internal consistency, test–retest reliability, and convergent and discriminant validity. The PCL-5 was used to assess the past month's frequency that participants were bothered by PTSD symptoms (e.g., “Being ‘superalert’ or watchful or on guard”). Items were rated on a 5-point Likert response scale (0 = not at all; 4 = extremely). Cronbach's α = .93.

3.3.8 ∣. Anxiety

The 40-item Multidimensional Anxiety Questionnaire (MAQ; Reynolds, 1999) demonstrates strong internal consistency, test–retest reliability, and convergent validity. The MAQ was used to assess overall anxiety (e.g., “I worried about being anxious in front of others”) and symptom frequency was rated on a 4-point Likert response format (1 = almost never; 4 = almost all of the time). Higher total scores indicate greater anxiety. Cronbach's α = .94.

3.3.9 ∣. Depression

The 20-item Center for Epidemiological Studies Depression Scale (Radloff, 1977) demonstrates internal consistency and test–retest reliability, and construct and factorial validity. Items reflect the past week's depressive symptoms (e.g., “I felt depressed”) and were rated on a 4-point Likert response scale (0 = rarely or none of the time; 3 = most or almost all the time). Higher total scores indicate greater depressive symptoms. Cronbach's α = .93.

3.4 ∣. Analyses

As in Study 1, we first examined item and scale level frequency distributions. We then eliminated highly skewed items and retained a final set of 15 best-performing items (five per subscale), while also ensuring adequate representation of the theoretical construct (see Supporting Information: Table S5 for final items). Readability (Flesch, 1949) was consistent with Grade 6 or higher. Final items were used for subsequent analyses. Convergent validity was assessed via correlations between the MORALS and other measures of emotions (i.e., OASS, GI, and NAS) and clinically-relevant measures (i.e., PCL-5, MAQ, and CESD). We also assessed associations with social desirability. Finally, we conducted confirmatory factor analyses (CFA) to test the proposed three-factor structure (Model 1: assimilated, accommodated, and emotions subscales) compared to an alternative one-factor model (Model 2: total measure) and an alternative two-factor model (Model 3: combined cognition subscales and emotions subscale). Models were interpreted with respect to widely accepted guidelines (e.g., Hu & Bentler, 1999) of comparative fit index (CFI) and Tucker–Lewis index (TLI) greater than 0.90, root mean square error of approximation (RMSEA) and standardized root mean square residual (SRMR) below 0.08, and lower Akaike information criterion and Bayesian information criterion values. Modification indices and a Satorra–Bentler correction were applied, and factors were allowed to covary in all models. CFA analyses were conducted using the Lavaan Package (Rossel, 2012) available in R.

3.5 ∣. Results

3.5.1 ∣. Item and scale characteristics

Items demonstrated acceptable skew and kurtosis. As presented in Table 2, the overall measure and subscales demonstrated acceptable skew and kurtosis, no floor or ceiling effects, and strong internal consistency (α's range from .88 to .94). Internal consistencies and scale means were similar for men and women.

3.5.2 ∣. Convergent validity

As shown in Table 3, the total measure and accommodated and distressing emotions subscales exhibited moderate positive associations with shame, guilt, and negative affect, as expected. Greater socially undesirable responding was associated with the total measure and emotions subscale (but not with cognition subscales). As expected, subscales correlated most strongly with each other. Greater PTSD, anxiety, and depressive symptoms were significantly associated with the total measure and accommodated and distressing emotions subscales. The assimilated subscale was only significantly associated with shame and anxiety.

TABLE 3.

Convergent validity of MORALS with related measures

Measure	Study 2 (N = 79)				Study 3 (N = 221)				Study 4 (N = 96)
	1	2	3	4	1	2	3	4	1	2	3	4
1. Total MORALS	–				–				–
2. Assimilated cognitions	.81***	–			.79***	–			.79***	–
3. Accommodated cognitions	.91***	.58***	–		.86***	.49***	–		.86***	.46***	–
4. Distressing emotions	.92***	.57***	.84***	–	.87***	.53***	.66***	–	.89***	.51***	.75***	–
5. Guilt	.37***	0.14	.43***	.41***	.49***	.25***	.52***	.36***	–	–	–	–
6. Shame	.45***	.27*	.47***	.44***	.44***	.22**	.52***	.36***	.25*	0.02	.34***	.28**
7. Negative affect	.44***	0.22	.46***	.47***	.42***	.21**	.41***	.44***	–	–	–	–
8. Social desirability	.25*	0.19	0.21	.25*	.20**	0.13	.12*	.21***	–	–	–	–
9. PTSD	.27*	0.16	.24*	.30**	.25**	0.03	.36***	.24**	.31**	0.04	.39***	.36***
10. Anxiety	.53***	.29*	.56***	.56***	0.17	−0.02	.18*	.24**	–	–	–	–
11. Depression	.41***	0.17	.49***	.43***	.19*	−0.04	.27**	.23**	.46***	0.19	.44***	.54***
12. Joviality	–	–	–	–	−0.08	−0.12	0.07	−.14*	–	–	–	–
13. IPV perpetration	–	–	–	–	.17**	0.01	.29***	.12*	.25***	0.13	.26***	.26***

Note: Measures differ for guilt, anxiety, and depression in Studies 2–4 and for IPV in Study 4. Study 3 correlations with depression and anxiety N = 130. Higher social desirability indicates greater socially undesirable responding. Kendall's Tau correlations are used for IPV to account for skew. Abbreviations: IPV, intimate partner violence; MORALS, Moral Outcomes of Relationship Aggression scale; PTSD, posttraumatic stress disorder.

^***p < .001.

^**p < .01.

^*p < .05.

3.5.3 ∣. Confirmatory factor analysis

CFA results (Table 4) suggest that the three-factor model fits the data well (CFI, TLI, and SRMR within recommended values) and is better than the one- and two-factor models. Covariances between the assimilated and accommodated factors (B = 0.63, SE = 0.08, Z = 7.82, p < .001) and the assimilated and emotion factors (B = 0.65, SE = 0.08, Z = 8.44, p < .001) were within the desired range. Covariance between the accommodated and emotions factors was higher than desired (B = 0.91, SE = 0.03, Z = 28.19, p < .001). See Supporting Information: Table S1 for factor loadings for the three-factor model.

TABLE 4.

Fit indices and χ² values for confirmatory factor analyses

	AIC	BIC	χ 2	df	RMSEA	90% CI	CFI	TLI	SRMR	χ 2 diff	df diff	p
Study 2
Model 1: Three factor	3846.64	3969.85	137.08***	83	0.09	0.07, 0.11	0.93	0.91	0.07	–	–	–
Model 2: One factor	3927.95	4944.05	195.46***	86	0.13	0.11, 0.15	0.85	0.82	0.11	21.18	3	<.001
Model 3: Two factor	3915.56	4034.03	187.31***	85	0.12	0.10, 0.14	0.86	0.83	0.10	22.78	2	<.001
Study 3
Model 1: Three factor	10233.18	10409.88	171.02***	83	0.08	0.06, 0.10	0.93	0.91	0.05	–	–	–
Model 2: One factor	10371.46	10537.97	278.58***	86	0.12	0.10, 0.13	0.84	0.80	0.08	92.01	3	<.001
Model 3: Two factor	10311.98	10481.89	233.16***	85	0.09	0.08, 0.10	0.88	0.85	0.07	56.15	2	<.001
Study 4
Model 1: Three factor	4400.45	4529.86	116.03*	83	0.07	0.04, 0.09	0.95	0.94	0.06	–	–	–
Model 2: One factor	4477.91	4599.85	190.17***	86	0.12	0.10, 0.14	0.84	0.80	0.10	150.44	3	<.001
Model 3: Two factor	4464.24	4588.67	175.24***	85	0.11	0.09, 0.13	0.87	0.84	0.09	104.17	2	<.001

Abbreviations: AIC, Akaike information criterion; BIC, Bayesian information criterion; CFI, comparative fit index; CI, confidence interval; df, degree of freedom; RMSEA, root mean square error of approximation; SRMR, standardized root mean squared residual; TLI, Tucker–Lewis index.

^***p < .001.

^*p < .05.

4 ∣. STUDY 3: EVALUATION OF RELIABILITY, VALIDITY, AND FACTOR STRUCTURE

4.1 ∣. Purpose

Examination of item and scale characteristics, test–retest reliability, convergent and incremental validity, and factor structure in a different type of sample.

4.2 ∣. Participants and procedures

Participants were recruited through Amazon's MTurk via a four-item prescreen survey, for which they were compensated $0.10. Eligibility criteria were: 1) in the U.S., 2) able to read/write in English, 3) age 18–40, 4) had been in an intimate relationship, and 5) perpetrated physical IPV. Age was limited because violent behavior often decreases substantially after age 40. In total, 395 (22.6%) participants were eligible and invited to participate, of whom 264 (67%) consented and were compensated $2.00 for participation. Some individuals did not complete the MORALS because they chose not to finish the survey or because they had missing IPV data, which was necessary to complete the MORALS, resulting in a final sample of 221 participants. Of these individuals, 130 (77%) completed the measure during a second survey approximately 2 weeks later for an additional $1.50. Two weeks was thought to be appropriate to prevent the recall of prior responses, for mood effects due to answering sensitive moral questions during the first administration to reduce and to minimize the likelihood that significant life events would occur that change responding. The mean completion between surveys was 14.65 days (SD = 1.93). Most participants (>99%) had been in an intimate relationship for at least 1 month of the past year. Demographics were similar for Survey 1 (see Tables 1 and 2).

4.3 ∣. Measures

4.3.1 ∣. Intimate partner violence

See Study 1 for a description of the CTS2 (Straus et al., 1996). We administered the physical aggression subscale during Survey 1. A sum of past year behaviors was calculated. Cronbach's α = .90. One item assessing the number of IPV behaviors perpetrated in the past 2 weeks (approximate time between surveys) was included in Survey 2.

4.3.2 ∣. Moral distress

We administered the final 15-item MORALS (i.e., the same items examined in Study 2). See Supporting Information: Table S5 for the final measure.

4.3.3 ∣. Shame

See Study 2 for a description of the OASS (Goss et al., 1994), which was administered during Survey 1. Cronbach's α = .96 in the current study.

4.3.4 ∣. Affective states

See Study 2 for a description of the negative affect scale of the PANAS (Watson et al., 1988). In Survey 1, we administered a reduced eight-item version (α = .89), as well as the eight-item joviality (e.g., “happy,” “enthusiastic”; α = .91) and the six-item guilt subscale (e.g., “guilty,” “angry at self”; α = .90) from the PANAS-X (Watson & Clark, 1999).

4.3.5 ∣. Social desirability

See Study 2 for a description of the MCSD, which was administered during Survey 1. Cronbach's α = .55, suggesting this measure may not be adequately reliable. When McDonald's Omega was examined, reliability was slightly improved at 0.61.

4.3.6 ∣. PTSD symptoms

See Study 2 for a description of the PCL-5, which was administered during Survey 1. Cronbach's α = .96.

4.3.7 ∣. Anxiety and depression

In Survey 2, we administered the four-item Patient Health Questionnaire (PHQ4; Kroenke et al., 2009), which exhibits convergent and factorial validity. The PHQ4 assesses the past 2-week frequency of two anxiety (α = .68) and two depressive (α = .74) symptoms, rated on a 5-point Likert response scale (0 = not at all; 4 = nearly every day).

4.4 ∣. Analyses

We first examined item and scale characteristics. Convergent validity was examined via associations of the MORALS with shame, guilt, negative affect, joviality, past year IPV perpetration, PTSD, depression, and anxiety. Outliers for IPV behavior frequency were truncated to the next highest value. We also examined associations between the MORALS and social desirability. We tested the incremental validity of the MORALS via two generalized linear models. In Model 1, guilt and shame (Survey 1) were entered as predictors of past 2-week IPV (Survey 2), which was the approximate time frame between surveys. In Model 2, the three MORALS subscales were added as predictors. To address the positive skew of the IPV count variable, we dropped one outlier and specified a negative binomial distribution. Prediction of IPV by the cognition subscales above and beyond guilt and shame would support incremental validity. Due to content overlap with shame and guilt, the distressing emotions subscale was not expected to uniquely predict IPV. We examined test–retest reliability across the two administrations via bivariate and intraclass correlations (ICCs; two-way mixed effects with absolute agreement) and paired samples t-tests. Finally, we examined a CFA using the same approach as in Study 2. We also assessed measurement invariance across men and women via a multigroup CFA by conducting a series of models, each building on the prior model by adding additional constraints to be equal across men and women: 1) a configural model with no constraints, 2) factor loadings constrained, 3) item intercepts constrained, 4) item means constrained, and 5) item residual variances constrained.

4.5 ∣. Results

4.5.1 ∣. Item and scale characteristics

All items demonstrated acceptable skew and kurtosis. As depicted in Table 2, the total measure and each subscale demonstrated acceptable skew and kurtosis, no floor or ceiling effects, and strong internal consistency during both administrations (α's from .80 to .91). Internal consistencies and scale means were similar for both men and women.

4.5.2 ∣. Convergent validity

As displayed in Table 3, we observed small to moderate positive correlations for greater shame, guilt, and negative affect with the total measure and each subscale, consistent with Study 2 and our hypotheses. As expected, joviality was not associated with the total measure or cognition subscales but had a small negative association with distressing emotions. There were small positive associations between greater socially undesirable responses and total moral distress and accommodated and distressing emotions subscales, but the association with assimilated cognitions was nonsignificant. Greater IPV, PTSD, and depression were significantly associated with the total measure and accommodated and distressing emotions subscales. Greater anxiety was significantly associated with accommodated cognitions and distressing emotions. Clinical measures were not significantly associated with assimilated cognitions. Results were similar for men and women (Supporting Information: Table S2).

4.5.3 ∣. Test–retest reliability

As presented in Table 5, the two administrations (approximately 2 weeks apart) of the total measure and each subscale were strongly and positively correlated. ICC's were greater than or equal to 0.70 for the overall measure and accommodated and emotions subscales, but was 0.66 for assimilated cognitions, suggesting moderate test–retest reliability.

TABLE 5.

Test–retest reliability in Study 3

	First administration mean (SD)	Second administration mean (SD)	t	r	ICCagreement (95% CI)	Mean difference (SD)
Total MORALS	61.64 (13.83)	62.68 (14.02)	−0.92	.58***	0.73 (0.62; 0.81)	−1.04 (12.82)
Assimilated Cognitions	21.56 (5.16)	22.14 (5.19)	−1.26	.50***	0.66 (0.52; 0.76)	−0.58 (5.20)
Accommodated Cognitions	19.38 (5.55)	19.52 (5.49)	−0.31	.57***	0.73 (0.61; 0.81)	−0.14 (5.14)
Distressing Emotions	20.69 (5.70)	21.02 (5.59)	−0.68	.53***	0.70 (0.57; 0.79)	−0.32 (5.45)

Note: ICC_agreement = intraclass correlation coefficient for a two-way mixed effects model with absolute agreement.

Abbreviations: CI, confidence interval; MORALS, Moral Outcomes of Relationship Aggression scale.

^***p < .001.

4.5.4 ∣. Incremental validity

On average, 1.33 (SD = 1.89) IPV behaviors were reported between surveys, and 52.3% (N = 68) reported at least one act of violence in the prior 2 weeks. As presented in Table 6, greater shame, but not guilt, significantly predicted greater past 2-week IPV in Model 1. In Model 2, greater assimilated cognitions significantly predicted less IPV, greater accommodated cognitions significantly predicted greater IPV, and greater shame remained a significant predictor of greater IPV. For every one unit increase in shame, there was a 4% increase in IPV incidence, compared to a 16% increase for accommodated cognitions and a 13% decrease for assimilated cognitions, suggesting that MORALS cognitions subscales are stronger predictors. Guilt and the MORALS emotion subscale were nonsignificant. The likelihood ratio test comparing deviance statistics of Models 1 and 2 suggests that including the MORALS subscales better explained IPV [χ²(3) = 38.21, p < .001].

TABLE 6.

Incremental validity via generalized linear regression models in Study 3

	Deviance	B	SE	Exp(B)	Wald χ2
Outcome: Past 2-week IPV
Model 1	131.62
Guilt		−0.02	0.02	0.98	0.63
Shame		0.05	0.01	1.05	21.02***
Model 2	169.83
Guilt		−0.03	0.02	0.97	2.92
Shame		0.04	0.01	1.04	25.05***
Assimilated cognitions		−0.13	0.03	0.87	21.48***
Accommodated cognitions		0.15	0.03	1.16	26.08***
Distressing emotions		−0.02	0.03	0.99	0.30

Note: Predictors were from Study 3 Survey 1 and the outcome was from Study 3 Survey 2 (approximately 2 weeks apart). N = 130. The outcome variable reflects the number of IPV behaviors perpetrated between the first and second surveys. A negative binomial distribution was specified to account for skew in dependent variable.

Abbreviation: IPV, intimate partner violence.

^***p < .001.

4.5.5 ∣. Factor analysis

CFA results (Table 4) suggest that the three-factor model fits the data reasonably well (CFI, TLI, RMSEA, and SRMR within recommended values), and had a better fit than the one- and two-factor models. Covariances of the assimilated factor with the accommodated (B = 0.57, SE = 0.06, Z = 9.62, p < .001) and distressing emotions (B = 0.61, SE = 0.05, Z = 11.41, p < .001) factors were within the desired range, but were slightly high for the accommodated and distressing emotions factors (B = 0.78, SE = 0.04, Z = 19.61, p < .001). Standardized factor loadings range from 0.48 to 0.84 (Supporting Information: Table S1).

4.5.6 ∣. Measurement invariance

As presented in Supporting Information: Table S3, fit indices for the configural models were just below desired levels. However, there were no significant changes in model fit when the loadings, intercepts, means, and residual variances were constrained, suggesting minimal differences in the proposed factor structure for men and women.

5 ∣. STUDY 4: CONFIRMATION OF RELIABILITY, VALIDITY, AND FACTOR STRUCTURE

5.1 ∣. Purpose

Final confirmation of the MORALS’ psychometric properties, including further examination of discriminant validity, with an extension to psychological aggression.

5.2 ∣. Participants and procedures

Participants include a subsample of 96 individuals from 63 couples who completed an ongoing study of family violence that included online baseline surveys, six semistructured telephone interviews about family violence, and one online follow-up survey that included the MORALS. Participant inclusion criteria were: 1) the primary caregiver or cohabitating partner of a primary caregiver of at least one child age 3–5 years, 2) the child was enrolled in, or eligible for, a Head Start program in the Greater Pittsburgh Metro Region, 3) at least 18 years old, and 4) able to speak English. Participants were recruited from Head Start programs, social media advertisements, and word of mouth. Compensation was $50 per study segment, for a total of $500. Participants were included if they endorsed perpetrating psychological or physical intimate partner aggression (IPA) and had completed the MORALS at the time of data analysis. See Table 1 for participant demographic information.

5.3 ∣. Measures

5.3.1 ∣. Family aggression

The Children, Intimate Relationships, and Conflictual Life Events (CIRCLE) interview is an event history calendar interview to assess incidents of psychological and physical IPA and parent-to-child aggression (PCA). Estimates of IPA and PCA behavior frequencies from the CIRCLE interview demonstrate strong convergent, discriminant, structural, predictive, and incremental validity as well as a relatively high degree of interpartner reporting concordance (Marshall et al., 2017). For this study, participants reported on aggressive incidents that occurred during the prior 4 weeks during each of six interviews to represent a total assessment period of 6 months. The total number of IPA and PCA behaviors were summed separately for each interview and then averaged.

5.3.2 ∣. Moral distress

See Study 2 for a description of the MORALS and Supporting Information: Table S5 for the final measure. Instructions were modified to anchor participants to any perpetrated psychological or physical IPA reported earlier in the study procedures.

5.3.3 ∣. Posttraumatic stress disorder

See Study 2 for a description of the PCL-5. Cronbach's α = .97.

5.3.4 ∣. Depression

See Study 2 for a description of the CESD. Cronbach's α = .95.

5.3.5 ∣. Shame

See Study 2 for a description of the OASS. In this study, modified items assessed perceptions of how their partner views them, rather than others in general. Cronbach's α = .85.

5.4 ∣. Analyses

We examined scale characteristics and convergent validity via associations of the MORALS with PTSD, IPV perpetration, depression, and shame. To examine discriminant validity, we compared whether the MORALS was more strongly associated with IPA perpetration than PCA perpetration (another immoral behavior), with stronger associations for IPA perpetration providing evidence for discriminant validity. To account for the positive skew of the IPA and PCA count variables, models were examined via generalized linear models with a negative binomial distribution. In separate models, each MORALS subscale was examined as the sole predictor of the aggression variable being examined. We re-examined the factor structure and measurement invariance across genders in a manner consistent with Study 3. Although Study 4 draws from dyadic data, a notable portion of participants’ (N = 30, 31%) partners were excluded from the current study because they did not have relevant data available. Of the 33 couples in the current study, only one couple was in a same-gender relationship. Thus, to facilitate interpretability across studies, results presented for Study 4 were not conducted using multilevel modeling to maintain consistency with Studies 1–3. Instead, when relevant we present analyses separately for men and women to remove any dependency in the data.

5.5 ∣. Results

5.5.1 ∣. Item and scale characteristics

As depicted in Table 2, the total measure and each subscale demonstrated acceptable skew, kurtosis, and no floor or ceiling effects. Internal consistencies were good among the total sample during both administrations (α's from .78 to .92). Internal consistencies and scale means were similar for men and women.

5.5.2 ∣. Convergent validity

As displayed in Table 3, we observed moderate positive correlations for shame, PTSD, depression, and IPV perpetration with the total measure and accommodated and distressing emotions subscales. The assimilated subscale was not associated with these measures. With few exceptions, generally similar associations were observed among women and men (Supporting Information: Table S4).

5.5.3 ∣. Discriminant validity

As displayed in Table 7, total moral distress, accommodated cognitions, and distressing moral emotions were significantly and positively associated with IPA perpetration, but assimilated cognitions was not. None of the MORALS scales were significantly associated with PCA perpetration.

TABLE 7.

Study 4 discriminant validity of the MORALS via comparisons of intimate partner aggression (IPA) and parent-to-child aggression (PCA)

	B	SE	Exp(B)	p
Outcome: IPA perpetration
MORALS total	0.02	0.01	1.02	.041
MORALS assimilated cognitions	0.03	0.02	1.03	.250
MORALS accommodated cognitions	0.07	0.02	1.07	.003
MORALS distressing emotions	0.04	0.02	1.04	.070
Outcome: PCA perpetration
MORALS total	0.01	0.01	1.01	.252
MORALS assimilated cognitions	0.03	0.02	1.03	.159
MORALS accommodated cognitions	0.004	0.02	1.004	.838
MORALS distressing emotions	0.02	0.02	1.02	.223

Note: Each of the predictors was examined via separate generalized linear regressions with a negative binomial distribution to account for the positive skew of the aggression count variables. Aggression variables reflect the average number of behaviors across monthly interviews. Abbreviation: MORALS, Moral Outcomes of Relationship ggression scale.

5.5.4 ∣. Factor analysis

CFA results (Table 4) indicate that the three-factor model fits the data well and better than the one- and two-factor models. Covariances of the assimilated factor with the accommodated (B = 0.60, SE = 0.09, Z = 6.78, p < .001) and distressing emotions (B = 0.65, SE = 0.07, Z = 8.84, p < .001) factors were within the desired range, but were high for accommodated and distressing emotions factors (B = 0.87, SE = 0.05, Z = 17.65, p < .001). Standardized factor loadings range from 0.44 to 0.89 (Supporting Information: Table S1).

5.5.5 ∣. Measurement invariance

As presented in Supporting Information: Table S3, fit indices for the configural models were just below desired levels. However, there were no significant changes in model fit when the loadings, intercepts, means, and residual variances were constrained, suggesting minimal differences in the proposed factor structure between men and women.

6 ∣. DISCUSSION

Understanding of morally-related cognitions and emotions that individuals experience in response to their IPV perpetration is limited, despite the potential for this area of study to inform understanding of repeated incidents of violence and targets for intervention. Because there were no adequate measurement tools to assess moral distress with respect to IPV perpetration, we developed and validated a brief measure to assess this construct, the MORALS. The MORALS includes 15 self-report items that assess cognitions about whether one considers their actions to be wrong, cognitions about oneself being changed due to their actions, and distressing moral emotions about one's actions. Overall, the measure is both reliable and valid and would be a useful tool for additional research aiming to understand how individuals respond to their IPV perpetration.

Importantly, the evaluation of the MORALS met many standard psychometric criteria (e.g., Terwee et al., 2007). First, the measure is internally consistent, with Cronbach's α greater than .78 for the total measure and each subscale in Studies 2 through 4. Test–retest reliability over approximately 2 weeks was also good, with strong positive correlations between the two administrations and ICCs greater than 0.66. No floor or ceiling effects were observed for the final measure or the subscales in Studies 2 through 4. In addition, convergent validity was demonstrated such that greater moral distress was positively associated with related measures of emotions, IPV perpetration, and psychopathology. Discriminant validity was observed via distinct associations of the MORALS with aggression perpetrated toward intimate partners, but not children. There were some instances in which the assimilated cognitions subscale appeared to perform differently from the other subscales. For example, the assimilated subscale was not significantly associated with clinical measures in Studies 3 or 4 (i.e., past year IPV, PTSD, anxiety, and depression) while the other subscales were. This is consistent with moral injury theory (e.g., Dennis et al., 2017), such that greater assimilated cognitions are an important prerequisite, but not a guarantee, for individuals to develop more pervasive and potentially clinically significant distressing accommodated cognitions and moral emotions.

Furthermore, assimilated and accommodated cognitions explained additional variance in future IPV perpetration beyond that explained by guilt and shame, thus demonstrating incremental and predictive validity. Indeed, greater guilt did not predict IPV perpetration, whereas shame did. Greater assimilated cognitions were also associated with less IPV perpetration, whereas greater accommodated cognitions were associated with greater IPV perpetration. This is consistent with the theorized model (e.g., Dennis et al., 2017), such that awareness that the behavior is wrong may prevent future IPV perpetration, whereas maladaptive cognitions pertaining to the self (e.g., self as unworthy), may increase future IPV perpetration.

Despite the MORALS asking about distressing and potentially self-incriminating information, associations with social desirability were small, such that endorsing greater total moral distress or moral emotions was associated with somewhat greater socially undesirable responses among women (but not among men, or for the cognitive subscales). This may suggest that the MORALS is not subject to social desirability bias, which is consistent with research finding that this form of bias minimally impacts IPV reporting (e.g., Visschers et al., 2017). However, the reliability of the social desirability measure was not ideal in Study 2 (α = .61) or Study 3 (α = .55; Omega = 0.61), which is generally consistent with prior research on this measure (Kuder–Richardson-20 = 0.76; Reynolds, 1982) and research indicating that Cronbach's α may be lower for binary response formats than those with multiple response options (e.g., Hilbert et al., 2016). Despite this concern, we observed small positive associations across two studies, in a pattern consistent with past research. Thus, any impact of social desirability is likely minimal.

The MORALS demonstrates several strengths, most notably that it is easy to administer with just 15 items, with the option of administering only the subscales of relevance to the research, and readability (Flesch, 1949) consistent with grade six or higher. In addition, the measure was designed to be consistent with cognitive behavioral therapeutic models. Most notably, the measure captures domains outlined by Cognitive Processing Therapy (Resick et al., 2016), which differentiates assimilated and accommodated cognitions, as well as emotional consequences. Although the measure has yet to be validated among clinically-recruited samples, we consider this an important design for the measure, given the need for effective IPV interventions (Eckhardt et al., 2013) and prior literature suggesting that trauma-informed interventions may be useful for addressing IPV perpetration (Taft et al., 2016). Furthermore, the proposed theory underlying the development of the MORALS was drawn from moral injury research conducted primarily among veterans who perpetrated war-related violence (see Griffin et al., 2019, for review). The current results extend this work by providing evidence that moral distress may occur among civilians and in relation to another form of violence, which may help to advance research pertaining to the role of moral experiences in psychological health.

The current set of studies also advances IPV research that has historically emphasized the role of emotional antecedents for IPV perpetration (e.g., Norlander & Eckhardt, 2005) by highlighting the need to better elucidate cognitive and emotional consequences that may play a role in the recurring nature of IPV for some perpetrators. Although some research suggests that reductions in emotional distress may negatively reinforce and heighten the risk for future aggression (e.g., Bushman et al., 2001), the current research suggests that a more nuanced process may occur. That is, distressing emotions may be immediately relieved following IPV incidents, but this may be followed by increased moral distress that also influences subsequent IPV behavior.

One potential limitation of the current research is the representativeness of the samples across the four studies. The first two studies included undergraduate students, the third included a self-selected sample from a panel of community members and the fourth included low-income community couples who care for young children. Despite this, the measure performed similarly across the studies. In the first two studies, participants were predominantly women, which precluded gender comparisons. However, this was improved in Study 3 (44% men) and Study 4 (39% men), in which our examinations of convergent validity and measurement invariance of the factor structure indicated that the measure performed similarly among men and women. Furthermore, although the majority were in opposite-gender relationships, the sample in Study 3 included nearly 28% of participants who identified as sexual minorities. Although we were unable to examine the performance of the measure for specific racial and ethnic subgroups, the sample in Study 3 was represented by 35% of participants who identified as racial or ethnic minorities and 43% in Study 4. Thus, our analyses suggest that the measure may be expected to perform similarly among more diverse samples, but this will be important for future work to examine.

Future research may be designed to continue the psychometric evaluation of the MORALS. Most notably, additional examinations of predictive validity, factor structure, and performance among samples with greater diversity with respect race, ethnicity, gender, and sexual identities may help support the generalizability of the measure. Of particular importance is examination among couples with diverse gender identities, given that gender differences in how IPV is viewed when men and women are the perpetrators and victims may impact moral evaluations of IPV. This warrants a nuanced, comprehensive, and culturally sensitive examination that was not possible in the current studies. Furthermore, community couples, as examined in the current research, are thought to mostly engage in moderately severe mutual aggression, whereas clinical populations (particularly those in court-mandated IPV treatment) may be more prone to nonmutual aggression characterized by efforts to exert power and control, which may contribute to differing moral responses. Confirmation of validity among treatment-receiving IPV perpetrators would facilitate the establishment of the MORALS's clinical utility in this domain. Measure adaptations may also be an area for future research. We focused on IPV perpetration, with evidence for reliability and validity when anchored to both physical and psychological behaviors. Future validation efforts may also examine how the measure performs when anchored to moral distress that may arise due to experiences of IPV victimization, which would require rewording items to be oriented to the partner's actions. Future research may also examine how moral responses differ for IPV and other immoral behaviors. Finally, future studies may be designed to examine factors that predict IPV perpetration-related moral distress, the degree to which it predicts future perpetration, and how it changes with treatment.

In sum, the MORALS is a brief, psychometrically sound instrument that can be used to assess the moral distress that individuals may experience with respect to their IPV perpetration. The MORALS demonstrates internal consistency and test–retest reliability, as well as convergent, discriminant, incremental, and factorial validity, and it performs similarly for men and women. This measure makes important contributions to understanding moral injury, which has been examined primarily among military veterans to date. This measure will be well-suited for research examining moral psychological processes generally, but also to better understand potential treatment targets for IPV perpetrators.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.