Development of discriminant functions to detect dissimulation for the Millon Clinical Multiaxial Inventory (3rd edition)

Abstract

This study investigated the ability of the MCMI-III to discriminate student dissimulators from psychiatric inpatients by re-analyzing data previously obtained by the authors. The sample consisted of 181 psychiatric inpatients and 218 college undergraduates. Students were randomized to either a fake-bad (FB) or standard instruction (SI) condition. Discriminant analyses were used to determine whether combining Scales X, Y, and Z with the clinical scales could improve the ability of the MCMI-III to identify dissimulators beyond using single scale cutoff scores. Two functions were developed. Function A yielded a positive predictive power (PPP) of 72%, a negative predictive power (NPP) of 75%, and an overall hit rate (HR) of 76%. Function B provided a PPP of 71%, a NPP of 80%, and a HR of 77%. The algorithms improved the ability of the MCMI-III correctly to identify dissimulators over any single scale cutoff score. The benefits and limitations of the discriminant functions are discussed.

Keywords:

• MCMI-III
• Millon
• malingering
• deception
• dissimulation
• fake-bad

Acknowledgements

These data were derived from the doctoral dissertation of Mike R. Schoenberg.

This dissertation was supported by a Seed Grant from the University of Kansas School of Medicine Endowment Association and a dissertation grant from National Computer Systems (NCS), without which this research would not have been possible.

We thank Lyle Baade and Michael Bagby for their thoughtful comments during this project and for assisting in the development of the experimental procedure.

Notes

1 The present study represents a re-analysis of the data originally presented in Schoenberg et al. (2003). The subsequent analysis in the current study supplements earlier analyses with a multivariate function. The psychiatric inpatient sample included patients with diagnoses of major depression and other affective disorders, schizophrenia and other psychotic disorders, and substance dependence disorders. In addition, over half of the sample was diagnosed with a comorbid personality disorder. None of the psychiatric patient sample were known to be in compensation-seeking circumstances, but roughly 10% had filled for Medicaid support. The experimental procedure is detailed by Schoenberg et al. (2003) and interested readers are referred to this study for a complete description of the experimental procedure. The odds of winning the lottery reported to participants were 1 in 100.

2 A copy of the Debriefing Questionnaire can be obtained from the first author.

3 To evaluate the effects of the small sample size of the development group, discriminant analyses were also completed using the total sample of students in the FB (n = 106) and PI (n = 181) groups. The classification accuracy yielded by the discriminant functions derived from the total sample did not differ significantly from the discriminant functions that were developed using the development group alone. Because of the statistical advantages of validating the algorithms on a sample that is separate from the group used to develop the algorithms, the discriminant functions obtained from the development group are presented.

4 To aid clinical use, a scoring program to compute the discriminant function values can be obtained from the first author.