Child Problem Behavior in Kindergarten and in Primary School: A Comparison between Prediction Configural Frequency Analysis and Multiple Regression1
Mark Stemmler, Friedrich A. Lösel, Andreas Beelmann, Stefanie Jaursch, Barbara Zenkert · Psychology science · 2005
Summary This study compares a variable-oriented and a person-oriented analysis of stability of child problem behavior from kindergarten to primary school. The sample contained 448 children from Erlangen-Nuremberg Development and Prevention Study (Losel, Beelmann & Stemmler, 1998). Problem behavior in kindergarten was measured using preschool teachers' ratings on Social Behavior Questionnaire (Tremblay et al., 1992). Indicators of problem behavior in classroom were derived from a content analysis of teachers' remarks in school report cards of first grade. Multiple regression and prediction configuration frequency analysis (CFA) were used for methodological comparison. Male gender and externalizing problems in kindergarten proved to be significant predictors of problem behavior in school. In contrast to multiple regression, prediction CFA revealed significant interaction effects. Girls who did not exhibit behavioral problems in kindergarten were significantly under-represented among difficult children at school (antitype). In group of boys with intensive problems at school, boys with externalizing problems or with both externalizing problems and internalizing problems in kindergarten were significantly more frequent than expected (two types). The exclusion of these two small stable deviant groups from total sample, led to a substantial reduction in predictive validity of variable-oriented analysis. The results are discussed within context of developmental psychology and methodology. Key words: child behavior problems, externalizing and internalizing problems, prediction configural frequency analysis (P-CFA), multiple regression, person-oriented approach. 1. Introduction In developmental psychology, various authors have emphasized that dominating variable-oriented research strategy needs to be complemented by an individual-oriented approach (e.g., Bergman & Magnusson, 1997; Cairns, 1983; Magnusson & Bergmann, 1988; Thomae, 1988). Addressing patterns of individual characteristics is not only relevant for scientific reasons but also for practice that is dealing with the person as a whole (Magnusson & Alien, 1983). That variable-oriented approaches are over-represented in longitudinal research may be partially due to dominance of such methods in statistical training of young researchers. In contrast, pattern-oriented prediction techniques have been developed more recently and are not yet part of standard academic teaching. Such a method is configural frequency analysis (CFA; Lienert & Krauth, 1975; von Eye, 1990; Krauth, 1993; Lautsch & von Weber, 1995) and its more recent model of prediction CFA (von Eye, 2002). In e^sub ij^ is natural logarithm of expected frequencies, λ^sub 0^ is intercept, λ^sub i^ is parameter of variable A and λ^sub j^ is parameter of variable B. The lambda parameters can be interpreted similarly to beta weights in multiple regression (Cohen & Cohen, 1983). For each log-linear model goodness-of-fit between observed and expected frequencies may be calculated by using likelihood ratio (LR) and/or Pearson chi-square statistic (von Eye, 2002). A non-fit is necessary for identification of types or antitypes (Stemmler & Bingham, 2003). Therefore, types or antitypes in prediction CFA include interactions between any predictor and criterion. In addition, resulting type or antitype contains information of a directed relationship (von Eye, 2002). This refers to a class of persons or common types sharing same characteristics (von Eye, 2002; Stemmler, 2001). Such an approach is highly relevant for developmental psychopathology. For example, in field of aggression, variable-oriented approach revealed relatively large correlations of problem behavior over time (Olweus, 1979; Zumkley, 1994). On other hand, there is growing evidence for different developmental pathways (Loeber & Stouthamer-Loeber, 1998; Moffitt & Caspi, 2001). …