shy anxious sad steals truancy threats aggressive expected values. Results. According to the BIC, a solution with seven latent classes is favoured (BIC= 37617,6 ...
Classifying aggressive behaviour with latent class approaches
Kinik für Kinder- und Jugendpsychiatrie/Psychotherapie
Ferdinand Keller and Klaus Schmeck Department of Child and Adolescent Psychiatry/Psychotherapy, University of Ulm, Germany
Introduction / Objective Latent Class Analysis (LCA)
Aggressive behaviour can be differentiated in various subtypes according to clinical experience and to results of factor analytic studies (Schmeck, 2004; Vitiello & Stoff, 1997). Probabilistic methods like latent class approaches have only rarely been used for subtyping up to now. Van Lier et al. (2003) analysed disruptive behaviour in young children; Wadsworth et al. (2001) focused on the anxiety/depression complex in the CBCL (Achenbach, 1991). Both groups found no typical classes; instead, a continuum was suggested. The aim of this study is to classify CBCL symptom patterns of aggressive behaviour in a large clinical sample, with inclusion of core items of emotional problems, to improve the description of aggressive subtypes.
LCA has a long history in the social sciences (beginning probably with P. Lazarsfeld). Since then, latent class and latent trait models have been continuously developed (Rost & Langeheine, 1997). Basically, LCA allows to detect homogenous subgroups in a heterogenous group through evaluating and then minimizing associations among responses across a set of indicators. Latent class analyses were performed with the latent cluster option of LatentGOLD 3.0. Decisions concerning the optimal number of latent classes were based on the Best Information Criterion (BIC). Sex and age group (4-10 vs. 11-18) were included in the models as inactive or active covariates.
Composition of items In the final composition presented here, two core items of each subscale were used: for „withdrawn“the items 75 (shy) and 111 (withdrawn), for „anxiety/depression“the items 50 (anxious) and 103 (sad), for „delinquent behaviour“ the items 82 (steals) and 101 (truancy), and for „aggressive behaviour“the items 57 (physical aggression) and 97 (threats others).
Methods Data Collection Assessment of the CBCLs took place in three different clinics (Frankfurt, Berlin, Köln). The whole sample consisted of CBCL ratings of 3503 child and adolescent psychiatric patients (age 4-18; 2405 boys, 1098 girls).
Results 2
According to the BIC, a solution with seven latent classes is favoured (BIC= 37617,6; 6 classes reveal BIC=37672,9, 8 classes: BIC=37650,3). Inclusion of the covariate sex or the covariate age group improved the BIC, and the combination of both yields the lowest BIC=37121,0. Results displayed below are based on this solution.
expected values
1,5
latent class #: short description
1
0,5
size
boys:girls
LC1: low symptomatic
27,9%
68 : 32
4-10y : 11-18y 70 : 30
LC2: shy – anxious
24,8%
61 : 39
65 : 35
LC3: pure aggressive young boys
15,9%
93 : 7
83 : 17
LC4: sad - truant
11,7%
56 : 44
9 : 91
LC5: emotionally disturbed
8,6%
50 : 50
18 : 82
LC6: highly aggressive - delinquent
6,0%
80 : 20
33 : 67
LC7: aggressive - emot. disturbed
5,1%
83 : 17
57 : 43
base rates (1-class solution)
100%
69 : 31
56 : 44
th re at s ag gr es siv e
tru an cy
ste als
sa d
wi th dr aw n
sh y an xio us
0
Description of the latent classes Inspection of the profile plots (see Figure) reveals 3 classes with high aggressive behaviour. LC6 is seen as highly aggressive and delinquent, with sadness above average of this referred group (cf. the dashed line indicating the mean values for the whole sample). The class is predominated by adolescent boys (see Table). LC7 is aggressive and has highly elevated scores in emotional disturbance, but only moderate delinquency. Boys are predominant, but of both age groups. LC3 shows a high amount of physical aggression, with low scores on delinquency and low scores on emotional disturbance. Demographically, the class consists of boys with age < 11. The other 4 classes are characterised by absence of aggressive behaviour. LC2 is a shy-anxious group, LC5 is emotionally disturbed and includes mainly adolescent girls, LC4 shows sadness above average combined with an elevated score in truancy. Finally, there is a large low symptomatic class (LC1).
Table: Membership percent of diagnostic groups in the classes (if > 10%)
Conclusion
Latent Class # Diagnostic group
1
2
3
no diagnosis F2x F3x F4x F5x F6x F8x F90.x F91.x, F90.1 F92.x F93.x, F94.0 Class size
46%
30% 41% 20% 33% 26% 26% 42% 22%
11%
21% 27% 15% 13% 32% 16% 18% 25% 27,1%
18% 42% 27,1%
10%
13% 34% 35% 23% 12% 17,1%
4
5
22% 35% 17% 16% 25%
33% 22% 17% 30% 17% 17%
11% 13% 10,4%
6
10% 19% 14%
8,1%
7
5,6%
10% 4,5%
group size 616 27 79 310 81 65 48 292 503 191 539 2751
1) Application of LCA revealed clinically relevant classes. 2) The distinction of aggressive behaviour in three different classes seems to be more useful than the existing diagnostic categories; however, we have to take into account that CBCL relies on parent ratings. 3) Further modeling seems promising (e.g., a model with three latent factors has a lower BIC and might improve classification errors, but without changing the meaning of the results); also, additional items will be included to clarify the clinical descriptions.
References Achenbach TM (1991) Manual for the Child Behavior Checklist/4-18 and 1991 Profile. Department of Psychiatry, Burlington, University of Vermont Rost, J. & Langeheine, R. (Eds.) (1997). Applications of latent trait and latent class models in the social sciences. Münster/New York: Waxmann. Schmeck, K. (2004): Störungen des Sozialverhaltens. in: Eggers, C., Fegert, J.M. & Resch, F. (Hrsg.): Lehrbuch der Kinder- und Jugendpsychiatrie. Springer Verlag Heidelberg New York, S. 849-873 Van Lier, P., Verhulst, F., van der Ende, J. & Crijnen, A. (2003). Classes of disruptive behaviour in a sample of elementary school children. J Child Psychology Psychiatry, 44, 377-387 Vitiello, B., Stoff, D.M. (1997). Subtypes of Aggression and their relevance to child psychiatry. J Am Acad Child Adolesc Psychiatry, 36, 307-15 Wadsworth, M., Hudziak, J., Heath, A. & Achenbach, T. (2001). Latent class analysis of CBCL anxiety/depression in children and adolescents. J Am Acad Child Adolesc Psychiatry, 40, 106-114