The Devereux Adolescent Behavior Rating Scale: A tentative model of second-order factor structure across independent clinical samples

Abstract

The second-order factor structure of the Devereux Adolescent Behavior Rating Scale was examined in (a) a sample of 254 adolescents who were receiving residential treatment for severe emotionaVbehavior disorders and (b) an independent sample of 404 adolescents hospitalized for substance abuse. A plausible range of factors was estimated for each group through parallel and average partial analyses and suggested either a two-or threefactor solution. Subsequent congruence analyses provided tentative support for a three-factor model: (a) undercontrolled/disruptive behaviors; (b) withdrawn/psychotic behaviors; and (c) needs approvaVdependent behaviors. The third factor was poorly defined in the residential treatment sample, but extraction of the third factor increased the across-group replicability of the first two factors.

Structural analysis of correlated factors: Lessons from the verbal-performance dichotomy of the Wechsler scales.

“…The analyses of incremental validity reported above provide an empirical basis for the generation (and preliminary evaluation) of hypotheses regarding the number of differentiable dimensions within a set of measured variables, but the results are most appropriately viewed as a directional guide rather than a strict mechanical rule (e.g., Macmann et al, 1991; Macmann, Wilkins, & O,Malley, 1992; Velicer & Jackson, 1990a, 1990b). In the case of the Wechsler scales, there was no compelling evidence on the basis of these procedures to support the extraction of three or more factors.…”

“…Although these minor or poorly-defined components might have a substantive foundation (i.e., a hypothesis that might be explored through further scale development and research), the constructs as measured are unlikely to provide a satisfactory basis for the interpretation of scores. In sum, even though there may be some research contexts in which the extraction of minor factors may be desirable (e.g., Comrey, 1978;Gorsuch, 1983;Macmann, Wilkins, & O'Malley, 1992), an accurate estimate of the number of factors to retain can reduce both the (a) magnitude of across factor correlations and (b) likelihood of across method differences (i.e., solutions that are unique to a particular method of factor extraction or rotation). Some strategies for determining the number of factors to retain may be less accurate than others (e.g., Velicer & Jackson, 1990a;Zwick & Velicer, 1982.…”

“…The rationale for the PA criterion is straightforward; that is, the eigenvalues for “nontrivial,” factors should be larger than the mean eigenvalues obtained through the parallel analysis of random data. Presumably, the shared variance associated with any factors that exceed this criterion is greater than might be expected by chance, but “poorly-defined” or weak components nonetheless may be retained (e.g., Macmann, Plasket, Barnett, & Siler, 1991; Macmann, Wilkins, & O’Malley, 1992; Zwick & Velicer, 1986). To provide a more accurate index of greaterthan-chance expectations, evaluation of the 95th percentile of the distribution of eigenvalues for random data (rather than the mean of the distribution) has been recommended (Longman et al, 1989).…”

“…As a consequence, both poorly defined and unique (or “singlet”) factors generally will not be retained by the MAP criterion (e.g., Zwick & Velicer, 1986). Given the respective tendencies of the two techniques, Zwick and Velicer (1986) suggested that the PA and MAP criteria might be used in tandem to bracket a “plausible range” for the number of factors to retain (e.g., Macmann & Barnett, 1992; Macmann et al, 1991; Macmann, Wilkins, & O’Malley, 1992).…”

Using and misusing factor analysis to explore group cohesion

“…Details are given in Cota, Dion, and Evans (in press). how many principal components to retain (e.g., Clark, Abrams, Niaura, Eaton, & Rossi, 1991; Macmann & Barnett, 1992, in press;Macmann, Wilkins, & O'Malley, 1992).…”