Recent studies indicate that the prevalence of psychopathology in children and youth populations is 13.4% worldwide (Polanczyk et al., 2015). Specifically, behavioral problems are the most frequently diagnosed disorders, followed by emotional disorders and attention-deficit hyperactivity disorder in Europe and other developed countries (Kovess-Masfety et al., 2016), Polanczyk et al., 2015). However, there have been few studies on child psychiatric disorders in developing countries like Bangladesh, suggesting that the prevalence there may be at least as high (Nikapota, 1991), which is perhaps unsurprising since so many children in the developing world are exposed to poverty, malnutrition, infectious diseases, violence and social disintegration.
In order to detect emotional and behavioral symptoms in children, (Goodman, 2001) developed an assessment instrument named The Strength and Difficulties Questionnaire (SDQ). This self-report questionnaire measures internalizing and externalizing problems in children and adolescents, as well as prosocial behavior. The scale consists of five subscales: Emotional issues, Conduct problems, Hyperactivity/Inattention, Peer Problems, and Prosocial Behavior. It is well known that both internalizing and externalizing behaviors are reactions to a children's experience of stress.
Those with internalizing behaviors may react inwards towards the self (e.g., social withdrawal, negative perceptions of self-worth). In contrast, those with externalizing behaviors may react outwards toward others (e.g., aggression, disruption). For each child, an internalizing score is produced by adding together the scores of the subscales: emotional symptoms and peer relationship problems. An externalizing score is produced by adding together the scores of the subscales: conduct problems and hyperactivity/inattention. Both the internalizing and externalizing score can range from 0–20 and the total difficulty scores is obtained by summing both internalizing and externalizing scores (i.e., summing all scores, except the prosocial behavior). The SDQ has been adapted and validated in multiple countries with its translated version into over 40 languages, both in community and clinical populations, and showed good psychometric properties (Goodman, 2001; Ortuño-Sierra et al., 2021). (Mullick & Goodman, 2001) argued that the Bengali version of SDQ might be useful as a screening tool for psychiatric disorders in community or pediatric clinics in Bangladesh. In this study, locally derived SDQ cut points (Mullick & Goodman, 2001) will be used to define the latent groups both for internalizing and externalizing problems among children. The SDQ is used in both community and clinical practice as a screening measure to identify high-risk symptom groups in children and adults.
Person-based methods, such as latent class analysis (LCA), examine the heterogeneity of symptomatology and group the population by focusing on individual differences. There are three types of subgroup identification methods and algorithms: nonparametric, parametric, and semi-parametric. Because nonparametric techniques make no assumptions about data distribution, a person's subgroup assignment is based on a dissimilarity measure (such as the cluster analysis technique). Parametric and semi-parametric techniques, such as the growth mixture model (GMM), latent transition analysis (LTA), and latent class analysis (LCA), on the other hand, require that data is created from a finite mixture of distributions. As a result, an individual's participation in a subgroup is determined by the conditional probability of that subgroup's membership. Traditionally, researchers in latent class analysis estimated LC models with various numbers of classes and chose the optimal model for categorical variables using likelihood-based statistics, which weigh model fit and complexity. In practice, it is frequently regarded as a challenge, particularly when dealing with big data sets such as community datasets. One issue is that the number of classes chosen may be quite huge. As a result, the classes pick up on extremely specific characteristics of the data that have nothing to do with the study issue. Furthermore, these distinct groups face interpretability and substantively similar challenges. A second issue arises from the fact that, depending on the model-selection criterion utilized, one would often select a different number of classes. To circumvent the issues mentioned above, (van den Bergh & Vermunt, 2019) proposed the Latent Class Trees (LCT) modeling approach, which is based on a divisive algorithm of latent class-based density estimation by (van der Palm et al., 2016). LCTs are a variant of model-based recursive partitioning, and as such related to decision trees (see Fig. 1 as an illustration).
There are very few studies on the use of LCT as a screening tool of child mental problems, where a large dataset is frequently collected at the community level. The aim of this paper is to apply a new tree-based latent class analysis method in determining the subgroup of internalizing and externalizing symptomology in community sample of Bangladeshi children aged 9 to 13 years and to explore the relationship between symptomology and some demographic variables such as gender and having siblings. To determine the existence of significant groups of children with similar internalizing and externalizing problems, the self-report SDQ was administered to those children and to their teachers.
The rest of the article is structured as follows. In Section 2, we describe the methodology in detail such as participants, data collection procedure, and measures. In Section 3, we discussed in detailed latent class trees (LCT) model. We apply LCT model to the motivating dataset in Section 4. Finally, Section 5 and 6 gives discussion and concluding remarks as well as identifies the scope for future research, respectively.