The Reliability and Factorial Validity of the Swedish Version of the Revised Controlling Behaviors Scale

Method

Results

When testing the proposed five-factor structure of the scale, the significant Bollen–Stine (p = .004) suggested a poor fit of the model. Apart from SRMR, all other standard indices suggested an inadequate fit of the model: RMSEA = 0.09, SRMR = 0.05, and CFI = 0.86. A look at modification indices suggested that sufficient improvement could only be achieved by a substantial modification of the model by correlating several error terms, which was deemed inappropriate due to both statistical and theoretical concerns. Table 1 shows that CR indices indicate a good reliability for all factors (all above .70). Indices of convergent validity indicate good validity for all five factors (AVE greater than .50 and lesser than CR). More critically, all AVE values were markedly lower than MSV and also markedly lower than ASV (except for Isolation). In addition, there were strong correlations (.74-.96) between the factors in the model. A subsequent exploratory factor analysis extracted one factor with 47% explained variance.

OPEN IN VIEWER

Table 1.

Indicators of Internal Consistency and Validity.

Type of Control Estimates	CR	AVE	MSV	ASV	1	2	3	4
1. Emotional	.862	.556	.922	.723	—
2. Economical	.710	.382	.781	.706	.775	—
3. Threatening	.764	.450	.922	.758	.906	.831
4. Intimidating	.799	.448	.922	.836	.960	.884	.960
5. Isolating	.918	.654	.750	.654	.741	.866	.775	0847

Note. CR = composite reliability; AVE = average variance extracted; MSV = maximum shared squared variance; ASV = average shared squared variance.

As 24 items are a redundant number of items for a single construct, we needed to restrict the number of items when testing if controlling behaviors are distinct from psychological aggression. In the following CFA, we used only items in the CBS-R (threatening control, intimidating control, emotional control) showing a high degree of similarity with items on the two Psychological Aggression subscales. As the items on the Isolation and Economical Control subscales showing little similarity with the Aggression scales, these items were deemed redundant. Thus, we tested if this model could be represented by three correlated first-order factors; one factor consisting of 14 items from three of the dimensions in the CBS-R (threatening control, intimidating control, emotional control) and two separate factors consisting of items from the two subscales (Minor and Severe, four items, respectively) of psychological aggression (CTS2). Due to low intercorrelations with other items on the Severe subscale of psychological aggression and an item-total correlation of .27, one item (“Threatened to hit or throw”) was removed from further analyses (resulting in three items on the Severe subscale).

The proposed three-factor structure model suggested an inadequate fit of the model; the Bollen–Stine p = .028, RMSEA = 0.09, SRMR = 0.07, and CFI = 0.83. However, the relative moderate correlations (.44-.48) between the factors in the model suggested adequate discriminant validity. The modification indices suggested that allowing one pair of error terms to correlate, and permitting “Call the other unpleasant names” (from the CBS-R) to load on both the minor psychological aggression minor factor and the CBS-R factor would substantially improve model fit. Even though the CBS-R item “Smash the other one’s property when annoyed” and the CTS2 item “Destroyed something of my partner’s” are related in content, it was not considered appropriate to allow for these errors to correlate as these items load on separate factors which would complicate interpretation of the two subscales, and therefore the former item was removed. Also, adding a path from psychological aggression minor to “Call the other unpleasant names” (included in the CBS-R) would complicate interpretation of the two subscales. In addition, this item had several error terms that load onto the psychological aggression minor and was therefore removed. With these modifications to the model, the fit significantly improved; the Bollen–Stine p value was no longer statistically significant, p = .20, and all other fit indices approached acceptable ranges: RMSEA = 0.07, SRMR = 0.05, and CFI = 0.90.

Table 2 shows that CR indices indicate a good reliability for all factors (all above .70). However, indices of convergent validity indicated some convergence validity concerns; all three factors AVE were less than CR but marginally failed to be greater than .50. Importantly, indices of discriminant validity indicate good validity for all three factors (all AVE markedly higher than MSV and ASV). Taken together, this CFA indicates that the items on the CBS-R are distinguishable from the items on the Psychological Aggression subscales of the CTS2, and that the minor and severe forms of psychological aggression are two clearly distinguishable factors.

OPEN IN VIEWER

Table 2.

Indicators of Internal Consistency and Validity.

Type of Control Estimates	CR	AVE	MSV	ASV	1	2
1. Severe PA	.738	.486	.211	.200	—
2. Minor PA	.777	.472	.211	.192	.459	—
3. CBS	.910	.460	.190	.182	.436	.416

Note. CR = composite reliability; AVE = average variance extracted; MSV = maximum shared squared variance; ASV = average shared squared variance; PA = Psychological Aggression; CBS = Controlling Behaviors Scale.

Next, a multigroup model was specified in which fit for both genders was examined simultaneously. In this model, the factor structure was specified identically across groups, and all parameters were freely estimated across genders. This is a method of formally establishing configural invariance (i.e., equivalence in factor structure across the genders). The fit indices of this configural model showed that this model had an adequate fit to the data (Bollen–Stine, p = .24, RMSEA = 0.07, SRMR = 0.07, and CFI = 0.82), suggesting that the three-factor structure is similar across genders, with the same items characterizing each factor.

Discussion

In this study, we looked more systematically at the validity of the CBS-R compared with previous studies. The analyses permitted tentative answers concerning the generalizability, psychometric properties, and practical utility of the CBS-R. Our results provided support for the internal consistency of the CBS-R. The CR indices indicated good reliability for all five factors, and Cronbach’s alphas were adequate and comparable with those found in previous studies (e.g., Graham-Kevan & Archer, 2003b; Próspero, 2008).However, despite the indices of convergent validity indicated good validity for all five factors, the model which was expected to load onto five factors failed to reproduce this structural pattern. That is, the CFA results indicated that the items on the scale share underlying commonality and are therefore better represented by one common factor than five clearly distinguishable factors.

Apart from a small-scale exploratory factor analytic study (Emery et al., 2012), the present study provides findings that have not been reported in previous research. The fact that the items from the CBS-R are better represented by one common factor than clearly distinguishable factors suggests that people who are exposed to controlling behaviors experience a variety of controlling acts rather than controlling acts that are predominately restricted to one or two forms of controlling behaviors. This result does not rule out the possibility that in certain populations, more rigid forms of controlling behaviors could be found. Graham-Kevan and Archer’s (2003a) findings from a cluster analysis nonetheless point to the possibility that a division of CBS into high and low control, respectively, could be of value when predicting aggressive behavior (see also Tiwari et al., 2015). Nevertheless, our results suggest that the division of CBS-R into several subscales is still premature, and that controlling behavior might better be represented by one common factor resulting from a variety of controlling acts.

When items from the two Psychological Aggression (Minor and Severe) subscales of the CTS2 were included in a CFA along with CBS-R items, the feasibility of a three-factor solution was examined (i.e., controlling behaviors, severe psychological aggression, and minor psychological aggression). The models without correlated errors did not fit the data sufficiently well. The source of the lack of fit was mainly locatable within the CBS-R, and modification indices suggested that relaxation of constraints related to these items would improve the fit of the three models appreciably. With minor modifications of the model, the analyses indicated that the items on the CBS-R are clearly distinguishable from the items on the Psychological Aggression subscales of the CTS2. This means that controlling behaviors are distinct from psychological aggression.

Moreover, results from the multigroup model indicated that the three-factor solution applies for both males and females. As controlling behavior has been suggested as being the key feature of psychological aggression that occurs in the context of physical partner violence (e.g., Follingstad, 2007; O’Leary, 2001), the results of the present study, even though leaving out the history, motive, dynamic, context, consequences, or meaning for those involved into account (Stark, 2010), thus boost the use of CBS-R when investigating potential predictors for physical partner violence.

Before firm conclusions can be drawn, there are some limitations that need to be addressed. One could argue that 24 items are a redundant number of items to measure a single construct. However, it is important to notice that these items do in fact consist of concrete acts. Thus, by reducing the number of items on this scale, we may fail to capture the true frequency of controlling behaviors. The present study analyzed data from a community sample with a relatively low prevalence of controlling behaviors. It is important to recognize that CBS-R itself may demonstrate different psychometric properties depending on the population in which it is used, but also that the psychometric properties may be dependent on individual differences (e.g., gender identity, socioeconomic status, language). Consequently, future research needs to gather information about the CBS-R among specific populations. Adhering to this argument, the population in this study came from a small northern region in Sweden with a response rate of 35%. Another limiting factor was that one criterion of inclusion was participants’ understanding of the Swedish language. Although the randomization procedure mostly rules out influence from the aspects of socioeconomic status, minority groups might have been excluded due to such inclusion criteria. Finally, although sample stratification was used, the sample was somewhat skewed across different age groups. Thus, the generalizability of the results could be restricted.

Notwithstanding, this study provided evidence that a measure that was exclusively developed to capture various forms of controlling behavior is clearly distinguishable from psychological aggression as measured on an instrument that has been most commonly used to assess partner violence. In light of the Istanbul Convention (Council of Europe Convention, 2011) underscoring the importance of control and psychological aggression, this is an important finding because it may help us select measurement tools for the examination of some of the effects of controlling behaviors on both psychological and physical aggression in intimate relationships, and the independent contribution of controlling behaviors and psychological aggression to physical partner violence. Furthermore, as IPV is associated with decreased health among women, and the relation between psychological aggression and depression, anxiety or posttraumatic stress disorder (PTSD) seems firm (e.g., Campbell, 2002; Clements & Sawhney, 2000), the use of controlling behavior in intimate relationships as a precursor or mediator of both mental and physical health could be investigated.