Society for Social Work and Research

Sixteenth Annual Conference Research That Makes A Difference: Advancing Practice and Shaping Public Policy
11-15 January 2012 I Grand Hyatt Washington I Washington, DC

127P Examining the Psychometric Properties of the Family Assessment Measure-III (FAM-III) Among Alcoholic Families

Saturday, January 14, 2012
Independence F - I (Grand Hyatt Washington)
* noted as presenting author
Ya-Ling Chen, Ph.D. Candidate, State University of New York at Buffalo, Amherst, NY
Thomas Nochajski, PhD, Associate Professor, State University of New York at Buffalo, Buffalo, NY
Eugene Maguin, PHD, Research Associate, State University of New York at Buffalo, Buffalo, NY
Purpose: Nearly 20% of youth aged 12 to 17 live in an alcoholic family (Office of Applied Studies, 2005). Relative to non-alcoholic families, alcoholic families show poorer family functioning (FF) and more adverse impacts on offspring (Johnson, 2001; Nurnberger, et al., 2004). Family functioning is important to consider when working with alcoholic families. The FAM-III, which is based on the Process Model of family functioning (Steinhauer, Santa-Barbera, & Skinner, 1984), is a well-known measure. Although reliability and validity data have been established by the developers, no published study has examined the psychometric properties of the FAM-III. This study addresses that omission. Method: Baseline data from a bi-national, multi-site randomized control trial of the Strengthening Families Program for children of alcoholics was used. Parents (N= 674) in Buffalo, New York and Ontario, Canada completed the FAM-III, a 50 item self-administered measure with a 4-point response format that is scored for 7 substantive (35 items) and 2 validity scales (15 items). Only the substantive items are used in this analysis. Fifteen reverse-scored items describe negative/conflictual behavior patterns and 20 describe positive patterns. All scales have varying numbers of positive and negative items. Mplus 6.1 was used to estimate the EFA and CFA models for ordinal indicator variables. Results: Item-within-scale polychoric correlations ranged from about .09 to about .69 and alpha reliabilities ranged from .61 to .71. A CFA of the seven substantive subscales yielded an inadmissible solution, suggesting a possibly incorrect structural model. The scree plot of eigenvalues suggested two to eight factors. Inspection of the oblique-rotated two-factor solution revealed that reverse-scored negative items mainly formed one factor and positive items mainly formed the other factor. Each substantive scale contributed items to both factors. Extracting more factors, although yielding better exact and approximate fit statistics, simply formed smaller, including doublet and singleton, groups of positive or negative items. With a loading cutoff of .40, nineteen items loaded on the first factor (alpha = .87) and 11 items on the second (alpha = .86). A CFA model was estimated using these factor definitions. Its fit was X2 = 1435.298, df=404, p < .001; CFI=.926; TLI=.920; RMSEA=.062; WRMR=.572. The correlation between factors was .80. Inspection of modification indices indicated the presents of very large values for both possible cross-loading and residual covariance modifications. Likely, the high factor correlation was, in part, a consequence of unmodeled cross-loadings or residual covariances. Conclusions and Implications: Results indicated that people tended to respond according to whether the item described positive or negative behavior rather than to the underlying class of behavior described irrespective of whether stated in positive or negative terms. Although respondents may also respond to the substantive area of behavior, this can not be determined unless the positive-negative items response bias can be modeled. The current results suggest that the published scoring procedure is not supported. Instead, a simple two scale scoring is preferred.