Bridging Disciplinary Boundaries (January 11 - 14, 2007)

Purpose: Accurate identification of functional impairment is a primary component of sound diagnosis. The Columbia Impairment Scale (CIS) is a 13-item instrument intended to measure global impairment. Although existing psychometric testing supports the value of the CIS, it is nevertheless a new and important enough instrument to warrant further validation. Specifically, there have been no published confirmatory factor analyses. We report on an analysis of the factor structure of the CIS with a disturbed clinical population. Methods: Data were obtained from participants in studies 2 and 3 (N = 279). Trained non-clinical research staff administered the CIS to all parents and to children ages 10 and older. The factor structure was analyzed using Confirmatory Factor Analysis was performed using EQS 6.1, and Exploratory Factor Analysis (EFA) using Primary Axis Factoring (PAF) was performed using SPSS 13. Results: Results showed that both the established 1-factor model and the originally hypothesized correlated 4-factor model provided poor fit with the sample data on child and parent measures. For the parent 1-factor model, χ² (65, n = 279) = 538.59, p<.001, GFI = .75, AGFI = .65, CFI = 0.58, RMSEA = .16; for the parent 4-factor model, χ² (59, n = 279) = 418.03, p<.001, GFI = .81, AGFI = .71, CFI = .40, RMSEA = .15. Child versions had similar goodness-of-fit. Subsequent model modifications failed to significantly improve the model fit for either the parent or child versions. Although the child sample was too small for the next analysis, in an attempt to identify a possible factor structure, SPSS was used to choose a random sample of approximately 50% of the parent sample (n = 135) and to perform a split-half analysis. Extracted communalities and scree plot suggested a 12-item, three-factor solution. PAF with Direct Oblimin rotation identified three statistically and conceptually coherent factors with simple structure, accounting for 48% of the variance. A series of CFAs were run on the split-half data. Although the model modifications improved goodness-of-fit, the final model was unsatisfactory (χ² (44, n = 135) = 107.2, p<.001, GFI = .89, AGFI = .81, CFI = .88, RMSEA = .10). Implications: Although prior research supports the use of CIS in community samples, in the absence of a clear underlying factor structure, caution is advised before using the CIS with clinical populations. Further psychometric testing with disturbed clinical populations is needed. Research by social workers to evaluate measures using commonly-served populations will help bridge the gap between instrument development and clinical implementation. This is warranted because of the many potential uses of the CIS, including: 1) cost-effective screening tool; 2) way to train graduate students in assessing impairment; 3) empirical method for establishing impairment for Supplemental Security Income claims. However, as the present study suggests, the transfer of assessment and evaluation technology between disciplines requires continued psychometric testing to establish that instruments are usable with the populations served by social workers.