|
|
Footnote 1: For an advanced discussion of measurement, measurement error, reliability and validity, the reader is referred to Bohrnstedt, G.W. (1983). Measurement. In P.H. Rossi, J.D. Wright, & A. B. Anderson (Eds.), Handbook of Survey Research. San Diego: Academic Press, Inc. Footnote 2: Measures under consideration include the Quality of Life Interview (QOLI) and other scales contained in A Lehman, Evaluating Quality of Life for Persons with Severe Mental Illness, Evaluation Center @ HSRI, Cambridge, MA. The QOLI, with eight (8) life domains, is not easily connected to costs since the scales are not additive and there is no overall summary score. Costs might be associated with primary program goals such as social contacts or family/social relations (e.g., an objective measure such as frequency of social contacts or a subjective measure such as satisfaction with family or social relations), but the linkage of costs to the QOLI is ill defined. The paper by Bigelow, McFarland, and Olson (1991) is also useful. Footnote 3: For information contact Behavioral Measurement Database Services, PO Box 110287, Pittsburgh, PA 15232-0787; telephone 412.687.6850 or fax 412.687.5213. Footnote 4: Statistical assessment of outcomes can be a complex issue. Simple gain scores (viz., time 1 - time 2) are subject to much deserved criticism. If pre- and post-scores are correlated at reasonable levels (e.g., .3 to .4) and are linear, then analysis of co-variance (ANCOVA) with time 1 as the covariate may be explored. The results have to be interpreted with caution, however, since those with higher initial scores can be expected to improve at a higher rate than those with lower scores. By relating the actual gain to a potential gain and analyzing the percentages with ANCOVA is somewhat more defensible. The analysis uses the form: Time 2-Time 1/ Ideal -Time 1= %. ANCOVA is problematic in any event. First, the statistical assumption that the treatment and the covariate do not interact systematically is not met since entry levels of a mental health condition (e.g., depression) and treatment approaches do have a systematic interaction. Second, since the interaction of treatment and entry level is of concern along with the main effect of treatment, any statistical control procedure to partition or subtract out information typically used in clinical decision-making should be viewed with caution. Analysis of variance with repeated measures poses similar problems. Footnote 5: Most parametric statistical analyses pose problems in comparing the effectiveness of two approaches to mental health treatment. The ( (Theta) technique (with a x2 statistic) can analyze two outcome matrices by comparing the two approaches against an ideal matrix. The test is sensitive to the magnitude of differences in treatment effects and represents a measure of the differences in patterns of client outcomes for two treatments at measured levels of intake functioning ... relative to a hypothesized pattern of outcomes. See Newman and Sorensen (1985) and Ross and Klein (1979). Other approaches include structural equations that are beyond the scope of this paper.
Frontier Mental Health Resource Network
|
|||
