Factor Structure of the Comprehensive Adolescent Severity Inventory (CASI): Results of Reliability, Validity, and Generalizability Analyses

Abstract

This article describes the results of psychometric work conducted on the Comprehensive Adolescent Severity Inventory (CASI) among 205 in-treatment substance-abusing adolescents. Four dimensions, each composed of component subscales, resulted from standard psychometric analyses: Chemical Dependency, Psychosocial Functioning, Delinquency, and Risk Behavior. Each dimension had high internal consistency (alpha coefficients for the component subscales comprising each clinical dimension range from. 78 to. 96) and test-retest reliability (intraclass correlation coefficients range from. 88 to. 96 and all are significant at p < .0001.). Concurrent validity and specificity of the CASI dimensions also were found: significant and substantial variance in NIMH Diagnostic Interview Schedule for Children-Revised (DISC-IV) and Brief Symptom Inventory (BSI) scores was associated with relevant CASI dimensions; CASI dimensions that theoretically should show no significant relationship with divergent pathology were not associated. The dimensions forecasted substantial variance in adolescent functioning posttreatment discharge, supporting predictive validity. Finally, the dimensional clinical structure was found to be generalizable over male and female adolescents, younger and older adolescents, and adolescents from different ethnic groups. These results provide further evidence for the CASI's promise in research and practice as an adolescent-specific assessment instrument that comprehensively assesses multidimensional areas of functioning within a developmental context of measurement. Limitations of the study along with future work currently being conducted on the CASI are discussed.

Keywords:

• Adolescent assessment
• adolescent substance abuse
• instrument development
• substance abuse assessment

Notes

¹Devised by Hakstian ([55]), hyperplane count refers to the number of near-zero pattern loadings. In the present case, loadings ≥ .40 are deemed salient, those ranging − .10 to. 10 are considered in the hyperplane (per Gorsuch's ([42]) recommendation), and those between. 40 and. 10 are considered in the ambiguous zone. When exploratory factor analysis has reached the stage where competing models are being considered, the model with the highest hyperplane count will tend to define simple structure ([56]). This is because, as the number of loadings in the hyperplane increases, the number in the ambiguous zone decreases, revealing a clear separation between the salient and near-zero loadings. In present analyses, hyperplane count is used to compare the best orthogonal solution with competing oblique solutions attendant upon increased exponentialization of loadings in promax rotation.

²Ubder parallel analysis, the plot of eigenvalues resulting from the reduced common factors matrix is compared to the plot of random variables for an identically large sample. The plot based on random variables is actually the average of many replications (300 in the present case) and the plots for the real and random variables are superimposed. The position prior to where the two plots first intersect suggests the maximum number of nonrandom factors that may be extracted from the correlation matrix of real data. Buja and Eyuboglu ([45]) have shown that parallel factoring may tend to overidentify the number of useful factors and therefore the technique tends to suggest an upper bound estimate of the correct number of factors.

³See especially page 250, formula 51.