Melbourne School of Psychological Sciences - Theses
Permanent URI for this collection
Now showing 1 - 1 of 1
ItemMeasurement invariance testing of the MMPI-2 when used with patients suffering a traumatic brain injuryALKEMADE, NATHAN ( 2013)The MMPI-2 is one of the most widely used tests of personality and psychopathology in both clinical and research settings (Archer & Newsom, 2000; Butcher, Rouse & Steven, 1996; Smith, Gorske, Wiggins & Little, 2010). It is suggested the neurological damage from a traumatic brain injury (TBI) can falsely inflate MMPI-2 profiles. Following this theory, the Gass (1991) correction procedure removes 14 items from the MMPI-2. Widespread use of the correction procedure continues despite conflicting results from replication studies. In this study measurement invariance analysis was completed separately on MMPI-2 scales Hs1, Hy3 and Sc8 to assess the Gass correction procedure. A TBI sample (n=254) and a sample generated from the MMPI-2 normative data (n=2600) was used for measurement invariance testing. In measurement invariance Test 1 (baseline model test) all residuals and one item loading for each factor were held to equality. In Test 2 (strict invariance test), all parameters were held to equality across groups. The requirement of invariance is that CFI decrease by equal to or less than .002 (Meade, Johnson and Braddy, 2008). If a model failed the test of strict invariance then a partial invariance model was defined using the backwards elimination procedure. Practical impact analysis was completed using the Millsap and Kwok (2004) procedure to assess the clinical effect from a failure to establish strict invariance. Prior to measurement invariance testing, exploratory factor analysis and confirmatory factor analysis were employed to define a factor model in Hs1, Hy3 and Sc8. A 4-Factor model was selected as best representing the 32 items from Hs1. In Test 1 the model produced reasonable fit indices (RMSEA = .023, CFI = .947, TLI = .943). In Test 2 the decrease in CFI was above the threshold of invariance (RMSEA = .025, CFI = .932, TLI = .930). A partial invariance model was defined with the parameters for four items freed (RMSEA = .023, CFI = .946, TLI = .945). All items passed the tests of no practical impact. In Hy3 a 4-Factor model was selected as best representing the 40 items from this scale not previously analysed in Hs1. In Test 1 the model produced reasonable fit indices (RMSEA .026, CFI = .921, TLI =.915). In Test 2 the decrease in CFI was above the threshold of invariance (RMSEA .043, CFI = .767, TLI =.761). A partial invariance model was defined with the parameters for 20 items freed (RMSEA .026, CFI = .919, TLI =.915). Items 161 and 185 failed the tests of no practical impact. In Sc8 a 5-factor model was selected as best representing the 68 items not previously analysed in Hs1 or Hy3. In Test 1 the model produced reasonable fit indices (RMSEA .015, CFI = .934, TLI = .932). In Test 2 the decrease in CFI was above the threshold of invariance (RMSEA .023, CFI = .838, TLI = .837). A partial invariance model was defined with the parameters for 28 items freed (RMSEA .015, CFI = .932, TLI = .930). Items 17, 92, 190, 278, 281, 291 and 303 failed the tests of no practical impact. Eleven of the 14 items from the Gass correction procedure passed the test of strict invariance, with the other three passing the tests of no practical impact. This finding fails to support continued use of the Gass correction procedure. Additionally the finding is contrary to the hypothesis that neurological content will bias MMPI-2 profiles in specific populations, such as the traumatic brain injury. However, some items were failed the tests of no practical impact and were identified as concerning. The implications from these findings are discussed.