Improving our understanding of why some adolescents engage in deviant behaviours, such as delinquency, aggression and substance use, while others do not, is crucial for the development of successful prevention and intervention programs. Several factors have been associated with the development of deviant adolescent behaviour, including genetic risk factors, individual factors, and environmental or contextual factors (Arthur, Hawkins, Pollard, Catalano, & Baglioni, Reference Arthur, Hawkins, Pollard, Catalano and Baglioni2002; Herrenkohl et al., Reference Herrenkohl, Maguin, Hill, Hawkins, Abbott and Catalano2000; Jessor, Reference Jessor1992). In recent years, moral disengagement (MD) has received considerable interest due to its ability to predict delinquent behaviour and substance use (Bandura, Barbaranelli, Caprara, & Pastorelli, Reference Bandura, Barbaranelli, Caprara and Pastorelli1996a; Newton, Barrett, Swaffield, & Teesson, Reference Newton, Barrett, Swaffield and Teesson2014; Newton & Bussey, Reference Newton and Bussey2006; Newton, Havard, & Teesson, Reference Newton, Havard and Teesson2012; Pelton, Gound, Forehand, & Brody, Reference Pelton, Gound, Forehand and Brody2004).

Moral disengagement is defined as the tendency to disengage from moral self-control and responsibility that ordinarily regulate behaviour (Bandura, Barbaranelli, Caprara, & Pastorelli, Reference Bandura, Barbaranelli, Caprara and Pastorelli2001; Pelton et al., Reference Pelton, Gound, Forehand and Brody2004). During the process of socialisation, people adopt moral and social standards that serve as guides and deterrents for their actions (Bandura et al., Reference Bandura, Barbaranelli, Caprara and Pastorelli1996a). People apply these standards to everyday life, and it is through the anticipation of self-sanctions, such as guilt and blame, for not adhering to these standards that conduct is kept in line with their personal standards. However, there are many psychological processes that prevent these self-sanctions from being activated or operated (Bandura et al., Reference Bandura, Barbaranelli, Caprara and Pastorelli1996a). Social cognitive theory specifies eight mechanisms clustered into four major points in the self-regulatory control of moral behaviour through which MD can occur. Self-sanctions can be disengaged from immoral conduct by: (1) reconstructing the conduct, through moral justifications, euphemistic labelling or advantageous comparisons; (2) obscuring personal causal agency, through displacing or diffusing responsibility; (3) misrepresenting or disregarding the injurious consequences of one's actions, by disregarding or distorting responsibility; or (4) blaming and devaluating recipients of behaviour, by vilifying the recipients of maltreatment through dehumanising or attributing blame to them (Bandura et al., Reference Bandura, Barbaranelli, Caprara and Pastorelli1996a).

A growing body of evidence has demonstrated associations between MD and delinquency (Pelton et al., Reference Pelton, Gound, Forehand and Brody2004; Shulman, Cauffman, Piquero, & Fagan, Reference Shulman, Cauffman, Piquero and Fagan2011), aggression (Bandura et al., Reference Bandura, Barbaranelli, Caprara and Pastorelli1996a; Barchia & Bussey, Reference Barchia and Bussey2011; Gini, Pozzoli, & Hymel, Reference Gini, Pozzoli and Hymel2014; Obermann, Reference Obermann2011; Paciello, Fida, Tramontano, Lupinetti, & Caprara, Reference Paciello, Fida, Tramontano, Lupinetti and Caprara2008; Pornari & Wood, Reference Pornari and Wood2010), and alcohol and other drug use, including underage drinking among schoolchildren and adolescents (Barnes, Welte, Hoffman, & Dintcheff, Reference Barnes, Welte, Hoffman and Dintcheff1999; Newton et al., Reference Newton, Barrett, Swaffield and Teesson2014; Newton et al., Reference Newton, Havard and Teesson2012; Passini, Reference Passini2012). Bandura and colleagues (Bandura et al., Reference Bandura, Barbaranelli, Caprara and Pastorelli1996a) were the first to develop a self-report Moral Disengagement Scale (MDS) to evaluate the disinhibitory effects of the four components of MD in relation to transgressive behaviour (Pelton et al., Reference Pelton, Gound, Forehand and Brody2004). The MDS has been adapted for use among schoolchildren and adolescents (Paciello et al., Reference Paciello, Fida, Tramontano, Lupinetti and Caprara2008), and variants have been used in many different contexts and with a range of behaviours, including sport (Lucidi et al., Reference Lucidi, Zelli, Mallia, Grano, Russo and Violani2008) and underage drinking (Quinn & Bussey, Reference Quinn and Bussey2015). Studies that have examined the factor structure of measures of MD have identified a number of different structures (Boardley & Kavussanu, Reference Boardley and Kavussanu2007). Many studies in this area have proposed a unidimensional solution (Bandura et al., Reference Bandura, Barbaranelli, Caprara and Pastorelli1996a; Paciello et al., Reference Paciello, Fida, Tramontano, Lupinetti and Caprara2008); however, a four-factor solution, which corresponds with the four sets of MD mechanisms identified in Bandura's conceptual model, has been proposed in other studies examining MD (McAlister, Bandura, & Owen, Reference McAlister, Bandura and Owen2006; Osofsky, Bandura, & Zimbardo, Reference Osofsky, Bandura and Zimbardo2005; Pozzoli, Gini, & Vieno, Reference Pozzoli, Gini and Vieno2012). In addition, an eight-dimension solution representing the eight mechanisms of MD proposed by Bandura (Reference Bandura, Kurtines and Gewirtz1991) could also logically be considered (Boardley & Kavussanu, Reference Boardley and Kavussanu2007).

The focus of the present study is on modified version of the MDS, which includes items specific to Australian children's antisocial and criminal conduct (NSW Attorney General's Department, 2003; Wagland & Bussey, Reference Wagland and Bussey2005; Wagland & Bussey, Reference Wagland and Bussey2015). These items relate to the most commonly publicised transgressions called before the New South Wales (NSW) Children's Courts, and represent aspects of common delinquent behaviour in Australia that are not currently assessed in Bandura's original MDS (Bandura et al., Reference Bandura, Barbaranelli, Caprara and Pastorelli1996a). Although this modified version of the MDS for Australia is a potentially useful tool, particularly among samples where delinquent behaviour is an issue of concern, it is important to examine the psychometric properties of the scale and validate the addition of the criminal and antisocial items against other measures of delinquency. Establishing the psychometric properties of a scale involves assessing its reliability and validity. Construct validity refers to the extent to which a measure is related to another variable with which it should theoretically be associated (Cronbach & Meehl, Reference Cronbach and Meehl1955). Previous research has shown that high scores on MD are significantly associated with increased delinquent behaviour (Pelton et al., Reference Pelton, Gound, Forehand and Brody2004; Shulman et al., Reference Shulman, Cauffman, Piquero and Fagan2011). Given that the adapted MDS purports to measure delinquency, it is expected that the MDS would be positively correlated with other measures of delinquency (DeVellis, Reference DeVellis2003). The present study sought to establish the construct validity of the adapted MDS using the Australian Self-Report Delinquency Scale (Mak, Reference Mak1993), a culturally relevant measure of delinquent behaviour among Australian adolescents. The overall aim of the present study was to explore the factor structure, reliability (internal consistency), and validity of a version of the MDS adapted for the Australian context.

Method

Participants

A total of 452 participants were recruited from five Sydney independent (private) schools. Letters outlining the aims of study were sent to principals at a random selection of Sydney independent schools. Once principals approved the study, assessment times were arranged and information and consent forms were distributed to students’ parents or guardians. Only those students who received parental consent were eligible to participate. All aspects of the study were approved by the administering research institution's Human Research Ethics Committee.

Thirty-nine cases were excluded from the confirmatory factor analysis procedures due to missing data on individual items.

Assessment and Measures

Participants completed the self-report survey using paper and pencil in a classroom setting supervised by a member of the research team.

Demographics

Participants’ gender, age and country of birth were measured.

Moral disengagement (MD)

Moral disengagement was measured using a 39-item scale that assessed proneness to disengage from different forms of detrimental conduct in diverse contexts and interpersonal relationships. Bandura's (Reference Bandura1995) original MDS consisted of 32 items, 31 of which were used in the current study. One item was omitted due to ethical concerns related to ‘getting high’. Consistent with previous research on MD in Australia, an additional eight items relating to the most commonly publicised transgressions called before the NSW Children's Courts in 2003, including arson, theft, property offence, and assault, were included (see Table 1; NSW Attorney General's Department, 2003; Wagland & Bussey, Reference Wagland and Bussey2005). Items in the revised scale covered the four major points in the self-regulatory system identified by Bandura and colleagues (Reference Bandura, Barbaranelli, Caprara and Pastorelli1996) where internal moral control can be disengaged from detrimental conduct. These included: reconstructing the conduct, obscuring personal causal agency, misrepresenting or disregarding the injurious consequences of one's action, and vilifying the recipients of maltreatment by blaming and devaluating them. Participants rated each item on a 4-point Likert scale, indicating their degree of acceptance of moral exonerations for such conduct from 0 (strongly disagree) to 3 (strongly agree).

TABLE 1

Factor Structure of a Modified Version of the Moral Disengagement Scale

Note: ^a Items relating to the most commonly publicised transgressions called before the New South Wales (NSW) Children's Courts.

Australian Self-Reported Delinquency Scale (Mak, Reference Mak1990)Footnote ¹

The Australian Self-Reported Delinquency Scale contains 34 items designed to examine delinquent behaviour ranging from minor status offences to more serious crimes. It includes nine subscales: cheating, status offences, fighting, stealing of vehicles and parts, drug use, theft, harming others, driving offences, and acts of vandalism and disturbance (Mak, Reference Mak1993). Participants were required to indicate whether they had been involved in each delinquent activity during the past 12 months (Yes/No). For ethical reasons, four items were omitted from the scale. The elimination of two of these items, which related to the use of ‘barbiturates’ and ‘LSD’, is consistent with previous research that found these items to have minimal effect on the overall delinquency score (Carroll, Durkin, Houghton, & Hattie, Reference Carroll, Durkin, Houghton and Hattie1996). The other two items related to forcing sexual acts on another person and driving a car when drunk. Four ‘lie’ items were interspersed among the delinquency items to detect unusually high levels of social desirability. Previous research has found the scale to have high construct validity and a clear nine-factor structure relating to the nine subscales, with all items loading between .31 to .75 (Mak, Reference Mak1993). In the present study, this scale demonstrated high internal reliability (α = 0.88).

Statistical Analysis

Descriptive analyses and investigations of skewness and kurtosis were performed using SPSS v22. A multistep approach was used to evaluate the factor structure of the adapted Australian version of the MDS. Using the random selection function in SPSS, the sample was divided into an exploratory sample and a cross-validation sample (n = 221and n = 231 respectively). Independent samples t tests and chi-squared tests were used to compare the split samples (exploratory and cross-validation samples) on all variables. First, principal component analysis (PCA) with varimax rotation was conducted using SPSS with the exploratory sample to determine the factor structure of the adapted MDS. Number of components retained was determined with reference to the resulting eigenvalues, Cattell's (Reference Cattell1966) scree plot, and component interpretability. Due to high skewness for some items, the factor solution was verified using the FACTOR package, which allows the application of PCA based on polychoric correlations (as recommended by Baglin, Reference Baglin2014). To determine the optimal factor solution, model fit for multiple factor solutions were compared within a confirmatory factor analysis framework using AMOS. A root mean square error of approximation (RMSEA) value less than .08, chi-square to degrees of freedom ratio (χ²/df) below 2, parsimony-adjusted comparative fit index (PCFI) greater than 0.5, and comparative fit index (CFI) values greater than .90 are considered indices of an acceptable model fit (Byrne, Reference Byrne2010). The expected cross-validation index was also used to compare models (lower values indicate lower expected discrepancy between analysed sample and another from the same population). Scale refinement was informed by modification indices and individual item loadings. Items that loaded more than 0.5 were retained in the factor on which they had the highest loading. Although item loadings of 0.3 to 0.4 are generally considered acceptable (Bryant & Yarnold, Reference Bryant, Yarnold, Grimm and Yarnold2004); the more conservative item loading of 0.5 used in this study is consistent with research that has found the size of item loadings (component saturation) to be the most important influence on pattern stability under a broad range of manipulations (Fava & Velicer, Reference Fava and Velicer1992a, Reference Fava and Velicer1992b; Guagdagnoli & Velicer, Reference Guagdagnoli and Velicer1988; Velicer & Fava, Reference Velicer and Fava1987). Items with high modification indices (>10) were also reviewed.

The final factor solution was then confirmed in the cross-validation sample using AMOS confirmatory factor analysis and with reference to model fit indices as described above. The internal reliability of the MDS revised subscales were assessed using Cronbach's alpha, and construct validity of the scale was examined by calculating Pearson correlations between MDS scores and an existing validated measure of delinquency, the Australian Self-Report Delinquency Scale.

Results

Descriptive Characteristics

Participants ranged in age from 10 to 15 years (mean age of 12.79 years; SD = 1.93) and 50% were female. A total of 179 students were from Year 5 classes in primary schools (mean age = 10.49, SD = .54), and the remaining 273 participants were from Year 9 classes in secondary schools (mean age of 14.29, SD = .47). The majority of participants were of Australian descent (61.2%), with the high minority groups being of British (14.6%) and Chinese (8.5%) descent. Independent samples t tests and chi-squared tests indicated no significant differences between the exploratory and confirmatory sample on any demographic or variables of interest, supporting the similarity of the subsamples (all p values < .05).

Factor Structure of the Adapted MDS

Exploratory factor analyses conducted in the exploratory sample determined the structure of the adapted MDS. The Barlett's test of sphericity and the Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy indicated the scale was suitable for factor analysis procedures, KMO = .90, Barlett's χ² (741, N = 221) = 4061, p < .001. Principal component analysis (PCA) with varimax rotation revealed eight factors with eigenvalues greater than 1 (12.7, 3.0, 1.9, 1.8, 1.5, 1.2, 1.1, 1.1); however, visual inspection of the scree plot suggested a one- or four-factor solution. A Horn's parallel analysis (Watkins, Reference Watkins2000) compared the factor eigenvalues to the average eigenvalues produced from 100 random data sets. The parallel analysis further supported the four-factor solution, as only these eigenvalues were higher than the corresponding values generated in the parallel analysis. The four factors accounted for 32.5%, 7.7%, 5.0% and 4.8% of the variance, respectively; factor loadings are presented in Table 1. Significant skewness was observed for 11 of the items, and thus the factor solution was verified using PCA based on polychoric correlations (Baglin, Reference Baglin2014), which also supported a four-factor solution.

To confirm the optimal structure, multiple-factor solutions were compared within a confirmatory factor analysis framework. One- and four-factor solutions were compared in view of PCA results and factor structure of the MDS that has been identified previously (Bandura, Reference Bandura and Reich1990; Osofsky et al., Reference Osofsky, Bandura and Zimbardo2005; Paciello et al., Reference Paciello, Fida, Tramontano, Lupinetti and Caprara2008; Pozzoli et al., Reference Pozzoli, Gini and Vieno2012). Comparison of model fit indices for the one-factor and four-factor solution reflected a superior fit to the data on all indices for the four-factor solution (see Table 2). The four-factor solution distinguished items according to the four dimensions proposed by Bandura et al. (Reference Bandura, Barbaranelli, Caprara and Pastorelli1996a). The scale was further refined with reference to the psychometric properties of each item. Eleven items were dropped from the scale as they did not load over 0.5 on any factor; two items were deleted due to extreme skewness; and an additional four items were deleted due to high modification indices (>10), suggesting cross-loading onto another factor. Model fit indices for the revised scale are reported in Table 2 and indicated the refinements improved fit to the data, with the final scale demonstrating adequate-to-good model fit.

TABLE 2

Model Fit Statistics for Determining Optimal Factor Solution

The final scale is a 22-item MDS for use in Australia (the Australian MDS) with four subscales mapping onto the proposed dimensions: (1) reconstructing the conduct (factor 2, five items); (2) obscuring personal causal agency (factor 4, five items); (3) misrepresenting or disregarding the injurious consequences of one's actions (factor 3, five items); and (4) blaming and devaluating recipients of behaviour (factor 1, seven items). The Australian MDS is reported in Table 1.

Confirmatory Factor Analysis

The stability of the four-factor solution was confirmed in the cross-validation sample. All 22 items loaded significantly on the factor to which they had been assigned. Goodness-of-fit indices for the four-factor model and comparison one-factor model are reported in Table 2. The four-factor model demonstrated adequate-to-good fit to the data, and all fit indices indicated superior model fit for the four-factor compared to the one-factor solution.

Factor Correlations and Internal Consistency

The mean factor scores on the Australian MDS were 11.87 (SD = 4.23, range 7–35) for factor 1 (blaming and devaluating), 6.96 (SD = 2.73, range 5–21) for factor 2 (reconstructing the conduct), 12.62 (SD = 4.22, range 5–25) for factor 3 (misrepresenting or disregarding), and 14.91 (SD = 4.08, range 5–24) for factor 4 (obscuring personal agency). Table 3 reports correlations between the four factors of the scale; the factors were all significantly intercorrelated, with correlations ranging from .16 to .61.

TABLE 3

Correlations Between the MDS Subscales and Delinquency Scale

Note: **p < .001, *p < .01.

The internal consistency of the Australian MDS was measured using Cronbach's alpha. The alpha coefficients for factors 1–4 were 0.83, 0.85, 0.82 and 0.67 respectively, indicating good internal reliability for factors 1–3 and adequate reliability for the fourth factor. The total scale score also demonstrated high internal consistency (α = 0.87) and was comparable to the internal consistency of the 39-item MDS (α = 0.93), supporting the utility of the 22-item short version of the scale.

Convergent Validity

To assess convergent validity of the Australian MDS, correlations between each factor and Mak's (Reference Mak1990) Australian Self-Report Delinquency Scale are shown in Table 3. Significant correlations were observed between the delinquency questionnaire and all four factors of the Australian MDS. Correlations were strongest for factors 1–3 (blaming/devaluating, reconstructing, misrepresenting/disregarding), which demonstrated strong correlations with the delinquency questionnaire.