The MMPI-ARC and MMPI-AF Scales: Psychometric Properties of the MMPI-2-RF’s Restructured Clinical (RC) and Higher Order Factor Scales When Adapted for Use with Archival Adolescent MMPI Data

Objective: To explore the psychometric properties of MMPI-adapted versions of the MMPI-2-RF’s RC and factor scales when applied to archival adolescent MMPI data. Method: We explored the internal consistency, temporal stability, and convergent and criterion validity of newly derived MMPI-adapted RC scales and MMPI-adapted higher-order factors (MMPI-ARC and MMPI-AF) using data from the first normative adolescent MMPI samples: 3,971 ninth graders assessed in Minneapolis public schools in 1947-1948 and 11,329 ninth graders assessed statewide in 1954. Results: Most MMPI-ARC and MMPI-AF scales showed good internal reliability. Scale reliability did not vary uniformly with the proportion of items available in the original MMPI. The MMPI-ARC scales corresponded most strongly with the same-numbered original MMPI scales 1, 7, 8 and 9, and least with scale 3. ARC4 and ABXD were strong predictors of behavior, and ARC3 and 4 captured much of the covariation of reported problem behaviors with MMPI CYN and MMPI Scale 4, respectively. Conclusions: This first psychometric investigation of the MMPI-ARC and MMPI-AF scales suggests that they represent additional, powerful research tools for studies based on historically invaluable, MMPI-assessed samples. Introduction Few, if any, psychological assessment instruments have seen the worldwide research, clinical, and general uses of the Minnesota Multiphasic Personality Inventory (MMPI; Hathaway & McKinley, 1943) and its successors, the Minnesota Multiphasic Personality Inventory-2 (MMPI-2; Butcher, Dahlstrom, Graham, Tellegen, & Kaemmer, 1989), the Minnesota Multiphasic Personality Inventory-Adolescent (MMPI-A; Butcher, Williams, Graham, Archer, Tellegen, Ben-Porath, & Kaemmer, 1992), and the Minnesota Multiphasic Personality Inventory2 Restructured Form (MMPI-2-RF; Tellegen & Ben-Porath, 2008; Ben-Porath & Tellegen, 2008). In the decades since the original MMPI scales were formulated, genetic, neurophysiological, clinical, and personality research have revealed more clearly the underlying dimensional structures of personality and psychopathology vulnerability. The MMPI-2-RF reflects many of these advances and began with Tellegen, Ben-Porath, McNulty, Arbisi, Graham, and Kaemmer‘s (2003) initial Restructured Clinical (RC) Scales. The RC scales are not mere recapitulations of the original MMPI or MMPI-2 scales, but instead, extract “distinctive core features of each, after removing Demoralization variance” (Tellegen et al., 2003, p.1), which, Trumbetta, Bolinskey, and Gottesman ________________________________________________________________________ 24 shared among the original scales, had reduced their discriminant validity. The RC scales opened a new chapter in MMPI assessment and attracted considerable interest, including an entire issue of Personality Assessment (Archer, 2006; Caldwell, 2006; Meyer, 2006). The MMPI-2-RF’s RC scales and three second-order factors of Emotional Internalizing Dysfunction (EID), Behavioral Externalizing Dysfunction (BXD), and Thought Dysfunction (THD) represent useful assessment tools, which also may expand the research potential of archival MMPI samples, including various medical, military, forensic, school, and community samples, as well as some archival adoption and twin studies. The original MMPI was so widely translated for other languages and cultures that, worldwide, archival and longitudinal follow-up studies may be enhanced through the newer MMPI-2-RF scales adapted to archival MMPI items. The MMPI-2-RF scales, based in the MMPI-2 item pool, contain some items that postdate the original MMPI, which leads to varying degrees of scale attenuation, a potential limitation for archival data sets, and yet, the RC scales seem robust even in MMPI-2 samples for which up to 50% of responses were unscorable (Dragon, Ben-Porath, & Handel, 2012). This suggests more than sufficient robustness of the RC scales for use with the original MMPI item pool, which contains at least 72.7% of MMPI-2-RF scale items (ARC-4) and up to 96.4% (ARC9). Furthermore, despite the MMPI-2’s minor rewording of 87 original MMPI items to make them simpler, clearer, and more appropriate to contemporary culture (Butcher, Dahlstrom, Graham, Tellegen, & Kaemmer, 1989), these changes produced no reductions in item-to-scale correlation (Ben-Porath & Butcher, 1989). The vast majority of MMPI items critical to the MMPI-2-RF scales have remained unchanged (Ben-Porath & Tellegen, 2008; Tellegen & BenPorath, 2008). The RC scales already have shown promise when adapted to the original MMPI item pool, contributing to the discriminant prediction of adult antisocial personality and schizophrenia from adolescent samples (Bolinskey, Trumbetta, Hanson, & Gottesman, 2010). The potential utility for other archival follow-up studies of the MMPI-adapted RC and factor scales (henceforth called MMPI-ARC and MMPI-AF or ARC and AF scales) led us to conduct this first systematic investigation of the adapted scales’ psychometric properties, specifically examining the internal and temporal consistency and the convergent and criterion validity. Method Participants Hathaway and Monachesi (1963) described their citywide and statewide MMPI samples of ninth graders, respectively, assessed through Minneapolis public schools in 1947-1948 and Minnesota public and parochial schools in 1954. Most participants provided follow-up information in early adulthood. The current citywide sample is described in a previous study (Trumbetta, Seltzer, Gottesman, & McIntyre, 2010) in which profiles were eliminated as invalid if omitted or “cannot say” items>35 (hereafter referred to as ?-invalid) or if Raw F>25 or if participant age fell outside the normative range of 13 to 16 years old at first assessment. We used the same exclusion criteria for the statewide sample. Primarily northern European, the two samples’ representations of Native American and/or African American ancestry were estimated at 0.4% and 0.5%, respectively, from 1950 Minnesota census data (M. McMurry, Minnesota State Demographic Center, personal communication, 20 April 2009). Citywide Minneapolis (1947) sample. Of Hathaway and Monachesi’s (1963) reported 3971 citywide ninth graders (1958 males, 2013 females), their electronic archive contained 3872 cases (1903 males, 1969 females). After comparison with hard-copy records and exclusions for MMPI-Adapted Restructured Clinical (MMPI-ARC) scales ________________________________________________________________________ 25 age and invalidity, the citywide Minneapolis sample yielded 1812 male and 1889 female valid cases. Statewide Minnesota (1954) sample. From 11,338 student names (5,707 male, 5,631 female) in the original statewide lists, Hathaway and Monachesi (1963) reported a total of 5701 males and 5628 females in this statewide sample, or 11,329 cases, 11,322 of which (5695 males and 5627 females) had sufficient data for a study of urban-rural differences in adolescent personality (Hathaway, Monachesi, and Young, 1959). The electronic archive of MMPI data, however, recorded only 11,216 cases, 5630 male and 5586 female. After comparison with hardcopy records and exclusions for age and invalidity, the statewide sample yielded 10,627 cases (5261 male and 5366 female). In the statewide subsample with valid twelfth grade MMPIs, 1738 males and 1806 females also had valid ninth grade MMPIs, yielding 3544 valid cases for testretest comparison. Measures MMPI scoring. We converted MMPI-2 item numbers from RC, EID, BXD, and THD to those used in the original MMPI Group Form to create the MMPI-ARC and -AF scales, archival estimates of the MMPI-2-RF’s RC scales and higher-order factors. Whenever an item was repeated in the original MMPI, we used only the item’s first appearance in computing an ARC or AF scale. In the absence of adolescent norms for the MMPI-ARC and AF scales, we used raw scores and, for comparison, retained raw scores for all other MMPI measures. Electronic recomputation from MMPI items of scales’ raw scores improved reliability over Hathaway’s original hand-scoring. Other variables. We utilized archival data on delinquency, teacher-rated conduct and adjustment problems, and high school drop-out to examine their relative associations with the MMPI, MMPI-ARC, and MMPI-AF scales. Severity-coded on a scale of 1-4, delinquency ranged from minor offenses to felonies, with 0 added to represent non-delinquent cases. Ascertained once in the statewide sample and twice in the citywide sample, delinquency scores for citywide participants could be up to twice that for statewide participants. The statewide archive (but not the citywide archive) contained teacher and school personnel ratings of adjustment problems (such as excessive fearfulness, petulance, or other indicators of dysfunction) and conduct problems, respectively, each on a 0 to 4 scale, with 0 for no problems, 1 for minor problems, 2 to 3 for increasingly severe problems, and 4 for problems requiring intervention from professional services or legal authorities beyond the school. Analyses Internal and temporal reliability. We computed Cronbach’s alpha and plotted it by MMPI-available RC and higher-order factor items for each scale to examine the internal reliability of the ARC and AF scales according to their degree of attenuation. We also assessed the scales’ temporal stability between ninth and twelfth grade, first with the Pearson correlation (r), and then, after correcting for internal unreliability (stability r / internal consistency estimate). Convergent validity. We examined correlations of the ARC and AF scales with the original MMPI Wiggins Content scales and Harris-Lingoes subscales to see which older constructs were most strongly associated with each of them. As the RC scales were designed to create distinct constructs, implicitly reducing scale overlap, we also compared the intercorrelations of the MMPI-ARC scales with those of the original MMPI’s clinical scales. Trumbetta, Bolinskey, and Gottesman ________________________________________________________________________ 26 We evaluated the predictive validity of the ARC and AF scales by computing direct correlations with behavioral measures, specifically adolescent delinquency, conduct problems, adjustment problems, and high school graduation. We did the same with the original MMPI scales, and then compared these MMPI-behavior correlations with their partial correlations after removing MMPI-ARC scale-related covariation. Theoretically, if an ARC scale completely accounts for the corresponding MMPI numbered scale’s covariance with a behavior, then the partial correlation of the original MMPI scale with that behavior would drop to zero after accounting for the MMPI-ARC scale. Lower partial correlations would indicate scales in which ARC component closely mirrored the predictive content of the original scale. Higher partial correlations would indicate scales that did not fully capture the original scale’s covariance with the behavior, and would be expected wherever a new scale, for improved discriminant validity, has focused on a distinctive component from the more heterogeneous original scale. Original scale heterogeneity stemmed both from unidentified Demoralization variance and from the inclusion of multiple symptom dimensions (as seen clearly in original scales 2 and 8), as well as from nonspecific items (such as “I forget right away what people say to me” from original scale 7), which could represent any number of conditions, and yet, still predict the same behavioral outcome. As RC3 (Cynicism) represents a different construct from MMPI scale3 (Hysteria), we also computed direct and partial correlations of the behavioral measures with the MMPI-2 Cynicism content scale using the MMPI’s available 21 of 23 MMPI-2 items in the scale. Results Composition and internal and temporal reliability of MMPI-adapted MMPI-ARC scales The proportions of RC scale items available in the original MMPI Group Form were 79.2% for Demoralization (RCd; 19/24 items), 96.3% for Somatic Complaints (RC1; 26/27), 82.4% for Low Positive Emotions (RC2; 14/17), 80.0% for Cynicism (RC3; 12/15), 72.7% for Antisocial Behavior (RC4; 16/22), 88.2% for Ideas of Persecution (RC6; 15/17), 87.5% for Dysfunctional Negative Emotion (RC7; 21/24), 88.9% for Aberrant Experiences (RC8; 16/18), and 96.4% for Hypomanic Activation (RC9; 27/28). For factor scales, the proportions of items available were 87.8% for Emotional Internalizing Dysfunction (AEID; 36/41), 82.6% for Behavioral Externalizing Dysfunction (ABXD; 19/23), and 92.3% for Thought Dysfunction (ATHD; 24/26). Statistically significant differences between citywide and statewide samples emerged for all MMPI-ARC and MMPI-–AF raw score means (Table 1), as expected, given the large n of each sample. Rural-urban differences probably contributed to these findings (Gottesman et al., 1987). Sex differences were found for all of the MMPI–AF raw score means (p<.001 state and city), except for Thought Dysfunction (ATHD, p=.74; city), with effect sizes ranging from Cohen’s d=.01 and .13 for Thought Dysfunction in city and state, respectively, to d=.23 and .20 for Emotional Internalizing Dysfunction, to d=.86 and .82 for Behavioral Externalizing Dysfunction. Most MMPI-ARC raw score means also showed significant sex differences within sample, with smaller effects seen for Demoralization (ARCd, Cohen’s d=.10 and .03, city, p=.002; state, p<.001), Somatic Complaints (ARC1, d=.14, p=.004; state only), Ideas of Persecution (ARC6; d=.17, p<.001, state only), and Aberrant Experiences (ARC8; d=.07, p=.01, state only). Sex differences were larger for Cynicism (ARC3; d=.33 and .38), Hypomanic Activation (ARC9; d=.34 and .31), Dysfunctional Negative Emotion (ARC7; d=.40 and .32), and Antisocial Behavior (ARC4; d=.56 and .53), all significant at p<.001. Only Low Positive MMPI-Adapted Restructured Clinical (MMPI-ARC) scales ________________________________________________________________________ 27 Emotions (ARC2; state, p=.90; city, p=.34) showed no significant sex differences in either sample. Table 1. MMPI-ARC scales’ raw score means and standard deviations by sample and sex Citywide Minneapolis Sample All (n=3701) Male (n=1812) Female (n=1889) ARCd 4.61 (3.42) 4.43 (3.25) 4.78 (3.56) ARC1 3.34 (2.84) 3.33 (2.75) 3.35 (2.93) ARC2 3.38 (1.82) 3.41 (1.83) 3.36 (1.81) ARC3 6.28 (2.80) 6.74 (2.69) 5.84 (2.82) ARC4 3.46 (2.63) 4.20 (2.81) 2.74 (2.22) ARC6 1.64 (1.78) 1.66 (1.79) 1.62 (1.77) ARC7 7.73 (4.39) 6.86 (4.20) 8.56 (4.41) ARC8 2.82 (2.49) 2.83 (2.49) 2.82 (2.49) ARC9 12.96 (4.85) 13.78 (4.91) 12.17 (4.66) EID 10.12 (5.12) 9.51 (4.84) 10.70 (5.31) BXD 5.77 (3.47) 7.17 (3.54) 4.43 (2.81) THD 3.08 (2.63) 3.09 (2.62) 3.06 (2.63)