Testing the Construct Validity of Dixon and Johnson's (2007) Gambling Functional Assessment

Abstract

The Gambling Functional Assessment (GFA; Dixon & Johnson, 2007) is a 20-item self-report inventory identifying four potential consequences maintaining gambling behavior. Exploratory and confirmatory factor analyses are performed for two large, nonclinical samples of university undergraduates. For the exploratory analysis, the optimal model yields two factors: Positive Reinforcement (correlated with GFA Sensory, Attention, and Tangible scores) and Negative Reinforcement (correlated with GFA Escape scores). One GFA item fails to load on either factor adequately. Factor loadings are confirmed using structural equation modeling for the second sample. The resulting model yields a mix of adequate and suboptimal fit indicators. Although the 2-factor model of the GFA has great theoretical utility and shows significant promise, confirmation within clinical samples of gamblers will be necessary to further validate the model. GFA Escape scores are uniquely distributed in the two samples and may represent functions most likely to maintain pathological gambling.

Assessing the Reliability of the Gambling Functional Assessment: Revised

“…The lowest observed Cronbach's alpha was 0.87. These results were superior to those reported for the original GFA (see Miller et al 2009a). Furthermore, to put these results in perspective, internal consistency measures for the SOGS have ranged from 0.69 (Stinchfield 2002) to 0.97 (Lesieur and Blume 1987).…”

“…With a 4-week interim between administrations, the lowest observed correlation coefficient was 0.74, which was for the scores on the positive reinforcement items after the ''non-gamblers'' had been excluded from the analysis. Miller et al (2009a) did not test the reliability of the original GFA at 4 weeks, so one cannot conclude whether the GFA-R performed better or worse than the GFA over this time interval. However, over this same time interval, SOGS scores correlated at 0.89 or higher, which was superior to the GFA-R. Millon et al (1997) reported that test-retest reliability of MCMI-III ranged between 0.82 and 0.91 over 5-14 days with a similarly sized sample.…”

“…Furthermore, it was not cleanly parsing these contingencies. Miller et al (2009a) tested the reliability of the GFA. They found that the internal consistency of the GFA overall (i.e., across all 20 items), as measured by Cronbach's alpha, was excellent (i.e., a = 0.92) when data from all respondents in the sample were included in the analysis.…”

“…Internal consistency for the items intended to measure the four potential maintaining contingencies were also adequate using the subsample of respondents who were likely gamblers (a = 0.71-0.79). Miller et al (2009a) also had 124 participants complete the GFA a second time, 12 weeks after they first completed it. For the overall GFA score (i.e., all 20 items), testretest reliability was good (i.e., r = 0.735).…”

“…The contingency of social attention showed the highest test-retest reliability (i.e., r = 0.712), while the contingency of escape showed relatively poor test-retest reliability (i.e., r = 0.403). Miller et al (2009a) concluded that the internal consistency of the GFA was good, but that the test-retest reliability results suggested that items intended to measure escape were likely testing state, rather than trait, factors.…”