🚧 Website Maintenance in Progress: Thank you for visiting! We are currently in the process of enhancing our website to serve you better. Please check back soon for our new and improved website.

Childhood victimization experiences, including abuse and neglect, are associated with short-and long-term impairments in cognitive, affective, and psychosocial performance (Bolger & Patterson, 2001; Rogosch, Oshri, & Cicchetti, 2010). Often studies addressing these associations focus on a specific type of victimization (e.g., sexual abuse) despite the tendency for different types of victimization experiences to co-occur (Adams et al., 2015; Charak & Koot, 2015; Ford, Elhai, Connor, & Frueh, 2010; Ford, Grasso, Hawke, & Chapman, 2013). Such accumulating victimization experiences have an aggregate effect on a child’s functioning and well-being (Charak, Koot, Dvorak, Elklit, & Elhai, 2015). Consequently, it is imperative that researchers examine the combined effect of co-occurring types of victimization on child/adolescent and adult psychopathology.

Problems researchers often encounter when investigating the co-occurrence of victimization experiences through variable-centered approaches (e.g., regression) include multicollinearity concerns and challenges with interpreting and understanding interactions between victimization variables (Roesch, Villodas, & Villodas, 2010). One potential solution to these problems is the implementation of latent class analysis (LCA). LCA is a person-centered approach, and with multiple statistical software programs (e.g., Mplus, Latent Gold, SAS) available for implementing this approach, its use today is easy and convenient. In this article, we begin by briefly explaining LCA, its assumptions, and ways to augment results from LCA with other statistical techniques. Next, we touch upon selected examples of recent research using LCA for examining childhood victimization.

What is Latent Class Analysis?

LCA is a subtype of finite mixture modeling that is employed to discover or confirm homogeneous subtypes/classes from a multivariate data. Simply put, it yields mutually exclusive classes of individuals who have similar response patterns (e.g., experiences/traits) within each class. For example, LCA might be used to identify discrete patterns of victimization such as co-occurring emotional and physical neglect or co-occurrence of moderate-severe levels of emotional, physical and sexual abuse and physical neglect (see Charak & Koot, 2015). When the variables under consideration are categorical in nature, this statistical technique is referred to as ‘latent class analysis,’ compared to when the variables are continuous in which case it is often referred to as ‘latent profile analysis.’ From this point forward, we will be using “LCA” to refer to applications involving categorical and/or continuous variables.

Assumptions of Latent Class Analysis

Like other statistical techniques, LCA has certain assumptions that must be met. First, LCA assumes that the population under study is a mix of qualitatively different individuals. Further, it assumes that between-subject variance is due differences in class-membership and that within class variance is absent. Second, it assumes that the measurement of all observed variables (e.g., sexual abuse, physical abuse, emotional abuse) is statistically independent within each latent class that is the occurrence of one observed variable is independent of the other. Third, it is assumed that, in the process of reaching the best class solution, the estimation algorithm converged at the globally best solution with the largest log-likelihood (i.e., there is clearly one set of parameter values that is superior to all others). This is referred to as the global maximum. Something that LCA does not assume is a normal distribution of each variable, an added advantage of using LCA when assessing victimization experiences that tend to have skewed distributions.

Augmenting Latent Class Analysis with Additional Analysis

Assessing varying patterns of victimization in the form of latent classes aides the identification of unique types of victimization experiences, thus facilitating the formulation of primary prevention interventions and policies. Building on the initial results of LCA, predictors of class membership can also be examined. For example, one can examine whether class membership depends on certain demographic characteristics (e.g., age, gender). Additionally, class membership can be examined as a predictor of certain outcomes such as psychopathology. For example, one could examine whether membership in a certain class (e.g., characterized by relatively severe levels of abuse and neglect) is associated with higher levels of PTSD or depressive symptoms. These extensions of LCA have the potential to inform tailor-made clinical interventions for individuals with similar victimization experiences.
After identifying latent classes, class-membership can be exported as an observed variable from LCA software packages to other popular statistical packages, such as SPSS for further analysis (e.g., Charak & Koot, 2015); however, one disadvantage of this method is that it does not take into account misclassification bias. An alternative, and arguably preferential approach, involves 3 steps (Asparouhov & Muthén, 2013; Vermunt, 2010) including (1) estimating the LCA model, (2) fixing the measurement error for the most likely class, and (3) estimating the auxiliary model where the latent class is measured by the most likely class and the measurement error is fixed. Fortunately, statistical software programs (e.g., Mplus, SAS) have been quick to include syntax (in Mplus v. 7 see Auxiliary option) or macros for this approach in more recent versions of their software programs. 

Examples of Recent Studies on Childhood Victimization Employing the Technique of LCA

To illustrate the use of LCA when examining multiple types of childhood victimization experiences and associated psychopathology, we summarize results from two recent studies.* In the first study, after obtaining the latent classes, the class-membership variable is used as an observed variable to predict differences in personality pathology. The second study implemented the 3-step approach (i.e., directly modeling misclassification bias) to examine the effect of demographic variables on class-membership, and the effect of class-membership on psychopathology.

Study 1  

Using Class-membership as an Observed Variable: In a study of 702 school-going adolescents (age range 13-17 years; 41.5 percent girls) from India, Charak and Koot (2015) examined patterns of abuse and neglect using LCA. For this, five different types of maltreatment were examined, namely emotional abuse, physical abuse, sexual abuse, emotional neglect, and physical neglect. Additionally, three levels of severity of each type of maltreatment were investigated (i.e., minimal, low, moderate-severe).

Four discrete classes of maltreatment were identified and subjectively labeled as, Moderate-severe abuse and physical neglect (Class 1), Low to moderate-severe abuse (Class 2), Moderate-severe neglect (Class 3), and Minimal abuse or neglect (Class 4). A four-class solution was selected from a series of class-solutions based on a number of goodness-of-fit indices (Masyn, 2013; Nylund, Asparouhov, & Muthén, 2007). Specifically, the four-class solution had lower BIC and sample size adjusted BIC values. Further, no improvement in fit was detected between neighboring class models (i.e., comparing k-1 and the k class models) as indicated by the Lo-Mendell-Rubin likelihood ratio test (LMR) and Bootstrapped likelihood ratio test (BLRT). Next, after exporting the class-membership into SPSS software, authors examined class differences across 17-lower order dimensions of personality pathology (e.g., conduct problems, affect lability, self-harm), after controlling for the effects of age and gender, via multivariate analysis of covariance (MANCOVA). Classes with higher percentages of adolescents reporting abuse and neglect with higher severity (Classes 1 and 2) were higher on personality pathology than the other classes (Classes 3 and 4).

Study 2 

Using the 3-step Approach: In a study on 3,485 adolescents (ages 13-18 years; 63 percent girls) from the National Child Traumatic stress Network Core Data Set (NCTSN-CDS), Adams et al (2016) examined latent classes based on 14 types of potentially traumatic events (PTE, e.g., sexual abuse, neglect, injury or accidents, school violence) and number of developmental periods/epochs (i.e., 0-5years of age, 6-12 years, and 13-18 years) in which the trauma was experienced. Using the 3-step approach (Vermunt, 2010), the authors examined the effect of a number of demographic variables on class-membership, and the effect of class-membership on indicators of emotional and behavioral functioning.

A five-class solution was found to be optimal out of a series of class solutions ranging from 1-9 classes. The obtained five classes were labeled as High exposure subgroup (Class 1), Multi-epoch emotional abuse subgroup (Class 2), Emotional abuse subgroup (Class 3), Loss/violence exposure subgroup (Class 4), and Low exposure subgroup (Class 5). The 3-step approach revealed that gender, race, primary residence, and the number of PTE types that had been experienced were statistically significant predictors of class membership (e.g., boys were more likely to be in the loss/violence subgroup). Similarly, class-membership predicted emotional and behavioral functioning, including internalizing symptoms, externalizing symptoms, traumatic stress, and risk behaviors (e.g., alcohol/substance use, self-injurious behaviors), such that the High exposure subgroup was at greatest risk for negative outcomes. Classes 2-4 were more at-risk for internalizing and externalizing symptoms compared to those in Class 5.

The aforementioned studies involved the implementation of LCA that reliably identified varying patterns of childhood victimization (i.e., different latent classes of victimization), and found support for the cumulative and detrimental effect of multiple childhood victimization experiences. Specifically, in each study, researchers identified a relatively homogenous class of adolescents who had experienced a distinct pattern of multiple types of victimization relative to other adolescents in the sample (e.g., Class 1), and demonstrated that members of that class were at greater risk for various forms of personality pathology, risk behavior, and traumatic stress.
To summarize, LCA is a statistical tool that yields mutually exclusive classes of individuals who have similar experiences/traits within each class. LCA has demonstrated tremendous potential for identifying meaningful patterns of childhood victimization, and has numerous advantages over traditional variable-centered approaches. Thus, it is not surprising that LCA has gained momentum in recent years as a valuable tool in the investigation of victimization. Future studies implementing LCA will add to the existing literature and help clarify underlying patterns in childhood victimization. The implications are far-reaching, with results of LCA having the potential to inform the development and refinement of tailor-made preventive and tertiary interventions for individuals with specific victimization experiences and related psychopathology.

*Selected seminal article on latent class analysis
**Note these studies are a selection from many possible alternatives.

About the Authors

Ruby Charak, PhD, is a Postdoctoral Research Associate at the Department of Psychology, University of Nebraska-Lincoln. She works on a NICHD project on sexual revictimization in young adult women. Her research interests and expertise pertain to childhood adversities, including childhood abuse and neglect and poly-victimization and their effect on adolescent and adult psychopathology. She has provided statistical consultation on many research projects including one with UNICEF.

Rebecca L. Brock, PhD, is an Assistant Professor in the Department of Psychology at the University of Nebraska-Lincoln. Her program of research is aimed at understanding the development of psychopathology (e.g., mood and anxiety disorders) across the lifespan with a focus on the family context and its etiological significance. She teaches graduate level courses in quantitative methods (e.g., Structural Equation Modeling) and serves as a statistical consultant in the department.


Adams, Z., Moreland, A., Cohen, J. R., Lee, R. C., Hanson, R. F., Danielson, C. K…& Briggs, E. C. (2016). Poly-victimization: Latent profiles and mental health outcomes in a clinical sample of adolescents. Psychology of Violence, 6, 145-155. doi: org/10.1037/a0039713

*Asparouhov, T., & Muthén, B. O. (2013). Auxiliary variables in mixture modeling: A 3-step approach using Mplus. Mplus Web Notes No: 15. Retrieved from https://www.statmodel.com/download/webnotes/webnote15.pdf

Bolger, K. E., & Patterson, C. J. (2001). Developmental pathways from child maltreatment to peer rejection. Child Development, 72, 549-568.

Charak, R., & Koot, H. M. (2015). Severity of maltreatment and personality pathology in adolescents of Jammu, India: A latent class approach. Child Abuse and Neglect, 50, 56-66. doi:10.1016/j.chiabu.2015.05.010

Charak, R., Koot, H. M., Dvorak, R. D., Elklit, A., & Elhai, J. D. (2015). Unique versus cumulative effects of physical and sexual assault on patterns of adolescent substance use. Psychiatry Research, 230, 763-769. doi: 10.1016/j.psychres.2015.11.014

Ford, J. D., Elhai, J. D., Connor, D. F., & Frueh, B. C. (2010). Poly-victimization and risk of posttraumatic, depressive, and substance use disorders and involvement in delinquency in a national sample of adolescents. Journal of Adolescent Health, 46, 545-552.

Ford, J. D., Grasso, D. J., Hawke, J., & Chapman, J. F. (2013). Poly-victimization among juvenile justice-involved youths. Child Abuse and Neglect, 37, 788-800.

*Masyn, K. E. (2013). Latent class analysis and finite mixture modeling. In T. D. Little (Ed.), The Oxford handbook of quantitative methods (pp. 551-611). New York, NY: Oxford University Press.

*Nylund, K. L., Asparouhov, T., & Muthén, B. O. (2007). Deciding on the number of classes in latent class analysis and growth mixture modeling: A Monte Carlo simulation study. Structural Equation Modeling, 14, 535-569.

Roesch, S. C., Villodas, M., & Villodas, F. (2010). Latent class/profile analysis in maltreatment research: A commentary on Nooner et al., Peras et al., and looking beyond. Child Abuse and Neglect, 10, 155-160. doi: 10.1016/j.chiabu.2010.01.003

Rogosch, F. A., Oshri, A., & Cicchetti, D. (2010). From child maltreatment to adolescent cannabis abuse and dependence: A developmental cascade model. Development and Psychopathology, 22, 883-897.