Cluster Analysis in Partner-Violent Men: A Comment to Redondo et al. (2019)

Highlights

• Stability of the anger based-typology of partner-violent men against women victims is questioned

• A general differential response to psychological treatment typology-based was not found

• Absence of untreated control group: Direct attribution of changes to treatment

• But unexpected higher scores in Physical Aggression and Injuries were found in Overcontrolled group (OCA) at Post-treatment

• The capacity to predict criminal justice recidivism by two anger-clusters typology was not found

Summary

This commentary examines Redondo et al. (2019)  study on anger-based subtypes in partner-violent men (PVMs). It challenges the clustering validation process, the absence of a third anger-free cluster that Murphy and colleagues (2007) identified, and the neglect of psychophysiological facets previously considered in the previous study by Ornduff et al. (1995). The interchangeable usage of ‘anger’ and ‘hostility’ is criticized. The commentary disputes the assumed link between intra-cluster complexity and recidivism and calls into question the study's reliance on self-reported measures. It contrasts this approach with Graña et al. (2014) research which encompassed field-relevant variables such as prior arrests, lifetime substance consumption, and serious interpersonal problems. The commentary advocates for a more thorough, multifaceted approach in future research on PVMs and their anger-based classifications.

Resumen

Este comentario examina el estudio de Redondo et al. (2019) sobre los subtipos basados en la ira en hombres violentos contra sus parejas (PVMs, por sus siglas en inglés). Examina el proceso de validación del agrupamiento en clústeres, la ausencia de un tercer clúster sin problemas graves de ira que Murphy y colegas (2007) identificaron, y la ausencia de factores psicofisiológicos considerados en el estudio previo de Ornduff et al. (1995). Se critica el uso intercambiable de ‘ira’ y 'hostilidad'. El comentario discute la supuesta relación entre la complejidad intra clúster y la reincidencia y cuestiona la dependencia del estudio de las medidas autoinformadas. Este enfoque contrasta con investigación previa (Graña et al., 2014) que estudió variables relevantes para el campo como detenciones previas, historial de consumo de sustancias y problemas interpersonales graves. El comentario aboga por un enfoque más completo y multifacético en futuras investigaciones sobre los PVMs y sus clasificaciones basadas en la ira.

Graphical Abstract

Keywords

intimate partner violence against women, partner-violent men, typology, treatment, recidivism

Palabras clave

violencia de pareja contra la mujer, hombres violentos contra la pareja, tipología, tratamiento, reincidencia

Redondo et al. (2019) published a study in the journal Criminal Justice and Behavior. The authors conducted research on 483 court-referred partner-violent men (PVMs) against women. The study had four objectives: (a) to identify an anger-based typology of PVMs, (b) discerning differences in clinically relevant characteristics between the subtypes, (c) evaluating their response to a cognitive-behavioral therapy (CBT), and (d) assessing their differences in recidivism.
     This article critique examined the clustering validation process and the interesting inferences on internal processes that could fill some gaps in the validation of the clusters. The expected typology was identified, consisting of Undercontrolled anger subtype (UCA or Cluster 1; 26.1%) and Overcontrolled anger subtype (OCA or Cluster 2; 73.9%).

Internal Validation and Stability Measure

A Hierarchical and a Two-Step cluster analysis were performed on the State-Trait Anger Expression Inventory-2 (STAXI-2) subscale scores showed by PVMs. In a review study, Clatworthy et al. (2005) assessed the use and reporting of cluster analysis in health psychology publications and noted that the use of two different methods is not the most recommended approach. Another review study evaluating structured and systematically the literature on cluster analysis (Ullmann et al., 2022) noted that repetition of the method has typically been used to validate the techniques—instead of clusters—and added "the more methods that are tried, the more likely it is that one of them produce a satisfactory result by chance.".

    No dendrogram or values comparing within-cluster homogeneity and between-clusters heterogeneity were provided, as recommended in the guidelines of the reviews cited (see above). Redondo et al. (2019) also did not provide detailed procedures, despite naming the methods and criteria used.

  To assess the stability of the identified clusters with the initial sample of 483 PVMs, Redondo et al. repeated a Two-Step Cluster Analysis on a "randomly drawn sample from the same population" (n=258 PVMs; p. 1707). However, the randomization process and relevant characteristics such as the sociodemographic variables and age were not described in this different sample, not allowing the examination of their independence.

   The external validation of the cluster solution was made by comparing the clusters’ scores on sociodemographic variables and the Anger and Hostility subscales from the Aggression Questionnaire (AQ). All data on sociodemographic differences were not reported. Notwithstanding the AQ Anger and Hostility levels between the clusters may not be sufficient for external validation and some overlapping it could be possible.

Psychological changes and Recidivism Between-Subtypes

The UCA subtype, despite starting with higher scores than the other subtype, achieved significant changes after treatment, but still maintained higher scores than the other subtype in Psychological Aggression and in the Use of Dominant and Jealous Tactics, even six months later. Inexplicably, the OCA group surpassed the UCA in Physical Aggression and Injuries post-treatment, a fact not clarified. Additionally, there is a lack of comparison with an untreated control group which would have allowed for a direct attribution of changes to the treatment.

    A Binary Logistic Regression on the levels prior to treatment in some variables and the subtype membership (UCA or OCA) showed that the subtypes were not predictors for recidivism, and the AQ Hostility subscale average scores (for all 483 men before treatment) were predictors. Another Logistic Regression analysis showed that the  average posttreatment scores on the External Anger Control subscale (STAXI-2) for the 385 PVMs who completed the treatment predicted recidivism at 1, 3, and 5 years, although cluster membership had no impact on such recidivism.

Final Considerations

     1. This study reveals insights into PVMs subtypes identified but falls short on empirical support for their validation. The authors stated that these findings corroborated the existence of two anger clusters aligning with prior research findings, but it is necessary to examine such empirical support.

    2. Subotnik dissertation’s findings (1983) were based on an amalgama of different kind of variables, not only anger. Then, Redondo et al. (2019) confirmed the existence of two distinct anger profiles but failed to identify a third cluster without serious anger problems, as found by Murphy et al. (2007, p. 190). Also, the anger-based distinction is not the same as the one between over-and-undercontrolled hostility found by Megargee et al. (1967)—in offenders not family only—and applied to maritally aggressive men by Hershorn and Rosenbaum (1991). Megargee et al. (1967)—in PVMs referred for treatment due to physical marital violence. Since in the study discussed the psychological violence was the most prevalent self-reported type of aggression, despite the contradictory data indicating that 83% were convicted for physical violence. they failed not considering the influence of the different samples.

    They noted that anger is associated with both cognitive and physiological states, in contrast to hostility, which they define as an attitude but they did not examine the role of psychophysiological  aspects of the anger, given this topic has been discussed decades ago by Ornduff et al. (1995) in their commentary on the work of Gottman et al. (1995).

     Surprisingly it seems that the authors understood anger and hostility as interchangeable constructs, despite trying to demonstrate the validity of the anger-based clusters.

    All exposure can be related to the fact that all measures were self-reported and no objective measures from the real context were analyzed in order to obtain supported findings. In this line, external validation of the identified clusters was addressed with other similar measures—AQ Anger and Hostility subscale scores— and other clinically relevant characteristics external to the cluster analyses, but other pertinent variables for the field should be considered to understand the clusters, such as those variables analyzed in previous research (e.g., Graña et al., 2014): previous arrest, frequency of serious interpersonal problems in last month and lifetime of substance.

     3. Regarding internal validation, it would be necessary to include the parameters and steps carried out before and after processing to choose the number of clusters, as well as include any measure of within-cluster homogeneity and between-clusters heterogeneity, and the dendrogram since is a part of output.

     4. Concerns to assess the stability of the clusters, it is not considered appropriate to include a different comparison sample (i.e., n= 256 PVMs). This sample and the initial sample (n= 483 PVMs) should ideally be independent. A more suitable approach could consist in randomly dividing the initial sample into two halves or subsamples (50%). The most appropriate number of clusters could then be identified in one subsample and then, it could be validated in the second subsample.

     5. It is incorrect to claim that Hostility (measured by the AQ) decreased in the UCA group since these results are not evident in the article. The only related psychological variables that were measured in order to analyze the change were both External Control and Trait Anger as measured by STAXI-2— showing a significant reduction—. This emphasized a decrease in anger rather than hostility, that are not interchangeable concepts. Instead, it would be accurate to say that AQ Hostility scores before treatment predicted recidivism regardless of the effects of treatment, as it was measured precisely before this.

   6. The term «individual perpetrator problem» was used to refer to the over-controlled or under-controlled category, which is inaccurate, since the clusters or subtypes are subgroups that were created based on the identification of characteristics within the groups (clusters) rather than individual problems. The statement «not all PVMs could benefit equally» is not empirically supported from the findings. In addition, they expected that the «individual’s typology» (see Cain et al., 2021; Leary, 1957), the degree of participation (treatment completion), and posttreatment External Anger Control scores (STAXI-2) would predict recidivism.

   One could erroneously assume a relationship between the intra-cluster complexity and recidivism. Starting from the premise that complexity has not been evaluated, but only differences in variables and the predictive ability of some variables over others have been measured to my best knowledge.

ORCID IDs

Pedro J. Horcajo-Gil https://orcid.org/0000-0003-4433-4005

References

Cain, N. M., Ansell, E. B., Wright, A. G., Hopwood, C. J., Thomas, K. M., Pinto, A., Markowitz, C., Sanislow, C. H., Zanarini, M. C., Shea, M. T., Morey, L. C., McGlashan, T. H., Skodol, A. E., & Grilo, C. M. (2012). Interpersonal pathoplasticity in the course of major depression. Journal of consulting and clinical psychology, 80(1), 78-86. https://doi.org/10.1037/a0026433

Clatworthy, J., Buick, D., Hankins, M., Weinman, J., & Horne, R. (2005). The use and reporting of cluster analysis in health psychology: A review. Trauma, Violence, & Abuse, British journal of health psychology, 10(3), 329–358. https://doi.org/10.1348/135910705X25697  

Gottman, J. M., Jacobson, N. S., Rushe, R. H., Shortt, J. W., Babcock, J., La Taillade, J. J., & Waltz, J. (1995). The relationship between heart rate reactivity, emotionally aggressive behavior, and general violence in batterers. Journal of Family Psychology, 9, 227-248.

Graña, J. L., Redondo, N., Muñoz-Rivas, M. J., & Cantos, A.  (2014). Subtypes of Batterers in Treatment: Empirical Support for a Distinction between Type I, Type II and Type III. PLoS ONE 9(10): e110651. https://doi.org/10.1371/journal.pone.0110651

Hershorn, M., & Rosenbaum, A. (1991). Over- vs. undercontrolled hostility: Application of the construct to the classification of maritally violent men. Violence and Victims, 6, 151-158.

Leary, T. (1957). Interpersonal diagnosis of personality. New York, NY: Ronald Press.

Megargee, E. I., Cook, P. E., & Mendelsohn, G. E. (1967). Development and validation of an MMPI scale of assaultiveness in overcontrolled individuals. Journal of Abnormal Psychology, 72, 519-528.      https://doi.org/10.1037/h0025242

Murphy, C. M., Taft, C. T., & Eckhardt, C. I. (2007). Anger problem profiles among partner-violent men: Differences in clinical presentation and treatment outcome. Journal of Counseling Psychology, 54(2), 189-200. https://doi.org/10.1037/0022-0167.54.2.189

Ornduff, S. R., Kelsey, R. M., & K. O’Leary, D. (1995). What Do We Know About Typologies of Batterers? Comment on Gottman et al. (1995). Journal of Family Psychology, 9(3), 249-252.

Redondo, N., Cantos, A., Graña, J. L., Muñoz-Rivas, & O’Leary, K. D.  (2019). Treatment-Induced Changes in Undercontrolled and Overcontrolled Anger Subtypes of Perpetrators of Intimate Partner Violence and 5-Year Recidivism. Criminal Justice and Behavior, 46(12), 1700-1718. https://doi.org/10.1177/0093854819879201

Subotnik, L. S. (1988). Men who batter women: From overcontrolled to undercontrolled in anger expression. In G. W. Russell (Ed.), Violence in interpersonal relationships: Sec. 2. Personality (pp. 203-213). Great Neck, NY: PMA.

Ullmann, T., Hennig, C., Boulesteix, A. (2022). Validation of cluster analysis results on validation data: A systematic framework. WIREs Data Mining and Knowledge Discovery, 12(3), e144 https://doi.org/10.1002/widm.144