A Caution on the Use of the MMPI K-Correction in Research on Psychosomatic Medicine

Transcription

1 A Caution on the Use of the MMPI K-Correction in Research on Psychosomatic Medicine ROBERT R. MCCRAE, PHD, PAUL T. COSTA, JR., PHD, W. GRANT DAHLSTROM, PHD, JOHN C. BAREFOOT, PHD, ILENE C. SIEGLER, PHD, MPH, AND REDFORD B. WILLIAMS, JR., MD The MMPI K scale is widely used to screen for invalid responses and to adjust substantive scale scores for defensiveness. In a normal volunteer sample, correlations of MMPI clinical scales and the Cook-Medley Hostility (HO) scale with self-reports and peer ratings on the NEO Personality Inventory (NEO-PI) were decreased rather than increased by K-correction. Similarly, in a medical sample, structured interview-based ratings of Potential for Hostility were better predicted by uncorrected HO scores than by K-corrected HO scores. Finally, in a prospective study of mortality among lawyers, uncorrected HO scores were a significant predictor of all-cause mortality; K-corrected scores were not. The data suggest that, under some circumstances, the K scale may measure substantive traits rather than defensiveness, and should be used and interpreted with caution Its use is probably contraindicated for most research on psychiatrically normal subjects. INTRODUCTION If individuals are defensive or try to present a falsely favorable picture of themselves, they will score high on the MMPI K scale. Some individuals score high on the K scale. Therefore, these individuals are presenting a falsely favorable picture of themselves. From the Gerontology Research Center, National Institute on Aging, National Institutes of Health, Baltimore, MD (R.R.M., P.T.C.), University of North Carolina, Chapel Hill, NC (W.G.D.), and Duke University Medical Center, Durham, NC (J.C.B., I.C.S., R.B.W.). Address reprint requests to: Robert R. McCrae, Ph.D., Personality, Stress and Coping Section, Gerontology Research Center/NIA/NIH, 4940 Eastern Avenue, Baltimore, MD Received August 5, 1988; revision received November 3, The first premise of this syllogism was empirically established in the derivation of the K scale (1) and has subsequently been replicated in other studies (2). The second premise can be substantiated by reference to published norms. The conclusion, however, does not follow either empirically or logically; it is a classic example of the fallacy of affirming the consequent. There are many other reasons why a high K score might be obtained, the most likely of which is that many of the items accurately describe the individual. Unless this possibility can be ruled out, it is unreasonable to infer defensiveness from a high K score, and the use of the K scale to screen subjects, to correct clinical scores, or to control for defensiveness or social desirability is questionable. This argument is of more than academic interest. The MMPI is the most widely used questionnaire for the assessment of personality and psychopathology, and has been a fixture in psychosomatic medicine for decades (e.g., Refs. 3 and 4). The designers of the instrument were par- 58 Psychosomatic Medicine 51:58-65 (1989) /80/51O1-OO58S0O 00/0 Copyright by the American Psychos<

2 CAUTION ON USE OF MMPI K-CORRECTION ticularly concerned with distortions that might be produced by malingering, lying, and more subtle defensive processes, and set out to create so-called validity scales to detect and correct for these artifacts. The K scale originated in studies contrasting normals with patients having low MMPI profiles (and thus presumed to be defensive), and specific weights were empirically determined to correct five of the clinical scales to maximize differentiation of patients from normals. K-corrected scores were introduced as part of the published profile sheet and have become standard in most clinical and research applications of the MMPI. Because standard correction weights are not available for the many MMPI research scales, some investigators use the K scale to screen subjects (5) or as a covariate (6) to minimize the effects of defensiveness. These practices are based on very little, and very mixed, evidence (2, 7) for the validity of the K-correction. Even in clinical settings, K-corrected scores do not necessarily lead to improved prediction of psychiatric diagnosis (8). In psychiatrically normal samples, where the base rate of defensiveness is presumably lower, the utility of K-correction becomes even more questionable (9). For researchers in psychosomatic medicine, this issue takes on increased importance given a recent paper which employed the K-correction in an attempt to clarify the meaning of the MMPI-based Hostility (HO) scale of Cook and Medley (10). Carmelli et al. (6) conducted a heritability analysis of the HO scale, partialling out the variance from K on the rationale that such residualization yielded a slightly larger association with independent behavioral ratings of hostility. Carmelli et al. (6) (p. 170) acknowledged that use of the K-correction was problematic and that the construct measured by K was complex, but concluded that K-correction was appropriate. The present study was designed to contribute additional evidence on this question by comparing validity coefficients for uncorrected and K- corrected scales. We present three studies of the K scale. In the first, we correlate corrected and uncorrected MMPI clinical scales and the Cook-Medley HO scale (10) with relevant personality traits as assessed both by selfreports and peer raters. In the second, we assess the value of adjusting for K scale scores in predicting Potential for Hostility ratings from HO scores. In the third, we compare uncorrected and K-corrected HO scores as predictors of mortality in a prospective study of lawyers. All three studies test the hypothesis that eliminating defensiveness by use of the K scale will enhance the accuracy of MMPI scores and thus increase validity coefficients. If the hypothesis is not supported, it will suggest that the K scale does not measure defensiveness or falsely favorable responding, at least in these samples. An examination of the correlates of the K scale may shed light on what it does measure, and make possible a substantive reinterpretation. STUDY 1 A scale shows evidence of validity when it is significantly correlated with theoretically relevant criteria, and when the correlations cannot be attributed to artifacts. Perhaps the most common method of assessing validity is by correlating the scale with another measure of the same construct or of theoretically re- Psychosomatic Medicine 51:58-65 (1989) 59

3 R. R. McCRAE AND P. T. COSTA, JR. lated constructs. If the two measures were created independently, and if the criterion measure has previously shown evidence of its validity, then a significant association between the scales is taken as evidence for validity of the new scale. However, it might be argued that correlating MMPI scales with other self-report measures provides ambiguous evidence at best, because shared artifacts of defensiveness or socially desirable responding might inflate validity coefficients. Individuals who are defensive in their response to one scale may also be defensive in their response to another. However, when MMPI scales are correlated with ratings by external observers, defensiveness is much less plausible as an explanation for any observed association. Peer raters who have known the ratee for an extended period of time probably have a reasonably accurate idea of what the individual is really like, and provide a better test for the utility of K-correcting. If K-correction increases the validity of MMPI scores, corrected scales should show higher correlations with theoretically relevant traits assessed by external observers than do the uncorrected scales. Methods As part of the ongoing Baltimore Longitudinal Study of Aging (11), the MMPI was administered to 174 men and 100 women aged 20 to 89 during their visits to the Gerontology Research Center from 1981 to These subjects had also completed the NEO Personality Inventory (NEO-PI) (12) in 1986 (13). Furthermore, 112 of them had been rated by one to four peers on a rating version of the NEO-PI in 1983 (14). The NEO-PI provides global measures of each of the five major dimensions of personality: Neuroticism, Extraversion, Openness to Experience, Agreeableness, and Conscientiousness. Both self-report and rating forms of the NEO-PI have demonstrated high reliability, longitudinal stability, and construct validity as assessed by correlations with other questionnaires, observer ratings, and sentence completion tests (12). The MMPI clinical scales were intended to identify individuals with specific psychiatric diagnoses common in the 1940s, and there is no reason to expect a one-to-one relation with personality traits in a normal sample. It does seem, however, that measures of psychopathology in general should be positively related to Neuroticism (the tendency to experience distress and negative affect) and negatively related to Agreeableness and Conscientiousness. Individuals scoring high on the five K-corrected MMPI scales should be emotionally unstable, interpersonally hostile, and low in self-discipline and socialization. Results Table 1 presents correlations of NEO-PI scales with uncorrected and K-corrected MMPI scales. Three notable conclusions are suggested. First, as hypothesized, the MMPI scales are positively related to Neuroticism and negatively related to Agreeableness and Conscientiousness in self-reports. (The associations with Extraversion, although not predicted, are understandable: The Pt and Sc scales include items about social isolation and so should be inversely related to Extraversion, whereas the Ma scale measures in part the energy and excitement seeking of extraverts.) Second, the predicted associations with Neuroticism, Agreeableness, and Conscientiousness are replicated when peer ratings are employed in place of self-reports, suggesting that the correlations are not due to the operation of response sets within self-reports. Finally, the effect of K-correcting on correlations with Neuroticism, Agreeableness, and Conscientiousness is to diminish the magnitude of the correlations, both with self-reported and observer-rated criteria, 60 Psychosomatic Medicine 51:58-65 (1989)

4 CAUTION ON USE OF MMPI K-CORRECTION TABLE 1. Correlations of Uncorrected and K-corrected MMPI Scales with Self-reports and Peer Ratings on the NEO Personality Inventory NEO-PI scale Self-Reports (N = 274) N E O A C Mean Peer Ratings (N= 112) N E O A C K -49 a a 12" -29 a a 36 a -20 a Hs 11-15" a a -26 a -17 a 40 a 12-35" -21 b Pd 06 12" 27 a a a - MMPI scale 69 a -31' a -33 a 31 a a Pt 43 a -32 a -32 a a -26 a a -30 a 30 a a Sc -20 a a " 19 a -33" 1 9t a -23" Ma - 45" 21" -26" a -18 Correlations with uncorrected scales are given in roman; correlations with K-corrected scales are given in boldface. Decimal points are omitted. Abbreviations: N = Neuroticism; E = Extraversion; O = Openness; A = Agreeableness; C = Conscientiousness. a p<0.. 6 p<0.. in 26 of 30 comparisons. The median absolute correlation between the Neuroticism, Agreeableness, and Conscientiousness scales and the uncorrected MMPI scales is 0.27; the median absolute correlation with the corrected scales is only Instead of enhancing the validity of MMPI scales, K-correction actually appears to reduce valid variance in this sample. An examination of the first column of Table 1 shows that the K scale itself is positively related to self-reported Agreeableness and Conscientiousness, and negatively related to self-reported Neuroticism. Two of these associations are replicated when peer ratings are examined, while the correlation of the K scale with peer-rated Agreeableness just fails to reach significance, p < These crossobserver replications strongly suggest that the association is substantive rather than artifactual. Individuals who score high on the K scale see themselves and are seen by their acquaintances as being well-adjusted, cooperative, and well-socialized. The Cook and Medley HO scale (10) is increasingly used in the prediction of coronary heart disease, and previous research has suggested that it measures aspects of Neuroticism and low Agreeableness (15). Are these correlations enhanced by K-correction? The first column of Table 2 gives uncorrected correlations of the HO scale with self-reported and peerrated Neuroticism and Agreeableness. The second column gives corrected correlations between residual HO scores and self- and peer-rated Neuroticism and Agreeableness. HO scores are residual- Psychosomatic Medicine 51:58-65 (1989) 61

5 ized statistically: regression analysis is used to predict HO from K (r = -0.78], and the predicted HO score is subtracted from the raw HO score to yield the residual HO score. The third and fourth columns give correlations with raw and residualized HO scores for a screened sample: subjects with K scores over 23 (a T-score of 70 according to published norms) were excluded from the sample. Again, the results are clear. The statistical removal of the effects of K substantially reduces the validity of the HO scale, according to both self-reports and peer ratings. Excluding subjects with high K scores has a much less pronounced effect but clearly does nothing to enhance the validity of the scales. Adjusting or screening on the basis of K scores does not appear to be advisable in the present volunteer sample. STUDY 2 Expert raters, who may be more objective and skilled in their assessments than peer raters, provide another way to examine the utility of the K scale for adjusting MMPI scores. Ratings of Potential for Hostility provide an appropriate criterion for comparing the predictive utility of uncorrected and K-corrected HO scores. Methods Between 1976 and 1981 the structured interview for the assessment of the Type A behavior pattern (16) was administered to over 2000 patients undergoing coronary angiography at Duke Medical Center. The MMPI was administered to nearly 1500 patients. Data collection procedures are described in detail elsewhere (17). Dembroski et al. (18) selected a subsample of 131 Sl-audiotapes to score Potential for Hostility. Details of the selection criteria and scoring procedure are R. R. McCRAE AND P. T. COSTA, JR. TABLE 2. Correlations of the Cook and Medley HO Scale with Self-reported And Peer-rated Neuroticism and Agreeableness NEO-PI Scale Self-reports Neuroticism Agreeableness Mean peer ratings Neuroticism Agreeableness Total sample 37" -53" 33" -22" HO scale - -36" " -52" 37" K<23-37" Correlations with raw HO scale are given in roman; correlations of residualized HO (with K removed) are given in boldface. Decimal points are omitted. For the total sample, N = 274 for self-reports and 112 for peer ratings. For the subsample with K scores <23, N = 251 for self-reports and 104 for peer ratings. "p<0.. "p<0.. presented in the paper describing their study (18). The Potential for Hostility ratings were significantly correlated with the severity of coronary atherosclerosis in these patients, attesting to the success of the assessment procedure. Results Complete data were available for 82 subjects. As in Study 1, a residualized HO score was created by subtracting from the HO score that portion which could be predicted from K scale scores. Potential for Hostility as assessed by expert raters based on structured interview responses was significantly correlated with the raw HO scale, r = 0.36, p < 0.0. It was substantially lower magnitude, r = 0.25, p < 0.. The "corrected" HO scale accounted for only half as much variance in ratings of Potential for Hostility as did the uncorrected HO scale. This result is understandable in substantive terms. There is a significant negative correlation between K and Potential 62 Psychosomatic Medicine 51:58-65 (1989)

6 CAUTION ON USE OF MMPI K-CORRECTION for Hostility, r = -0.25, p < 0., which is predictable from the item content of the K scale. Low scorers on the K scale endorse items such as, "At times I feel like swearing" and "At times I feel like smashing things"; these same individuals are judged by raters as being high in Potential for Hostility. Item content, rather than a defensive style of responding, appears to determine K scale scores. Rather than strengthening the relation to behavioral ratings of Potential for Hostility from the SI, residualizing HO for K weakened the predictive validity of the HO scale. None of the 82 subjects in this study had K scale scores greater than 23, so it was not possible to test the effect of screening for high K scores in this sample (as had been done in Study 1). STUDY 3 Method The MMPI was administered to 128 law students at the University of North Carolina during the 1956 and 1957 school years; subjects ranged in age from 20 to 45 years, with a mean of years at the time of testing (15). A mortality follow-up of the students was conducted in 1985; of the 118 subjects located, 13 had died, most of coronary heart disease or cancer. Results As reported elsewhere, raw HO scores were associated with a significantly higher risk of mortality (15), using a Cox proportional hazards analysis (19), a technique that assesses the risk of earlier time to failure in this case, death. HO scores were linearly related to survivorship, X 2 (l, N = 118) = 6.37, p < 0., after controlling for age. To test the hypothesis that K-correction would enhance the validity of HO scores, and thus their ability to predict mortality, residualized HO scores were created as in Studies 1 and 2. When these scores were entered in a proportional hazards model controlling for age, they failed to predict survival significantly, x 2 (l. N = 118) = 1.31, NS. Again, "correcting" for K appears to lower the predictive validity of the HO scale in this sample. DISCUSSION K-correcting is supposed to improve the validity of MMPI clinical scales; to find that, in some cases, it actually diminishes validity relative to external criteria is very sobering. Efforts made to improve test accuracy may in fact worsen it. Clearly, clinicians who use the MMPI to assess individuals as well as researchers who seek relations between personality and health variables need to approach the use of the K correction with great caution. The results of the present studies extend to the area of psychosomatic research previous work on the K scale (2) which called into question the utility of employing the K-correction. They are also consistent with recent evaluations of other measures of defensiveness or socially.desirable responding (20, 21) which suggest that such scales usually measure substantive traits rather than artifacts of impression management. Our results appear to contradict those of Carmelli et al. (6), who found that K- correcting HO scores slightly increased rather than decreased correlations of the HO scale with independent behavioral ratings of hostility. However, they also found that intraclass correlations between Psychosomatic Medicine 51:58-65 (1989) 63

7 monozygotic twins decreased from 0.34 to 0. when the variance attributable to the K scale was statistically removed. In this latter case, it appears likely that valid variance was removed by K-correction, as in our studies. The finding that K-correcting is counterproductive in many cases is easily explained. The K scale includes such items as "I have very few quarrels with my family" and "I worry over money and business [reversed]." Among normal samples, and in the context of volunteer research participation, most individuals will give truthful answers to these questions. As a result, agreeable and welladjusted individuals will receive high K scores. Indeed, as Table 1 shows, the K scale itself is negatively related to selfreported Neuroticism and positively related to Agreeableness in our normal, community-dwelling Baltimore Longitudinal Study of Aging sample (Study 1). Put simply, the K scale measures agreeableness and adjustment rather than defensiveness in this group. When a measure of adjustment is added to a measure of R. R. McCRAE AND P. T. COSTA, JR. psychopathology, the scores tend to cancel each other out, leaving a less meaningful and valid score. In particular, when Cook-Medley HO scores are K-corrected, the resultant score is no longer predictive of its original health outcome, namely mortality. Much existing literature has reported results based on K-corrected scores, and for purposes of comparisons, it will often be necessary to continue to report analyses of K-corrected scales. However, it would appear prudent to analyze and report uncorrected scales as well. For psychiatric subjects, under certain conditions of administration, the K scale could reflect defensiveness, and in these cases it might be useful as a correction. For other subjects and administration conditions, interpretation of the K scale as a measure of defensiveness is suspect. In normal populations, the weight of evidence is against this use of the K scale. Thus, from the present findings it appears that K-correcting MMPI research scale scores gathered on psychiatrically normal subjects is contraindicated. REFERENCES 1. Meehl PE, Hathaway SR: The K factor as a suppressor variable in the MMPI. J Appl Psychol 30: , Dahlstrom WG, Welsh GS, Dahlstrom LE: An MMPI Handbook, Vol 1: Clinical Interpretation (rev ed). Minneapolis, University of Minnesota, Dahlstrom WG, Welsh GS, Dahlstrom LE. AN MMPI Handbook, Vol 2. Research Applications (rev ed). Minneapolis, University of Minnesota, Bradley LA, Van der Heide, LH: Pain-related correlates of MMPI profile subgroups among back pain patients. Health Psychol 3: , Lebovits B, Lichter E, Moses VK: Personality correlates of coronary heart disease: A re-examination of the MMPI. Soc Sci Med 9: , Carmelli D, Rosenman RH, Swan GE: The Cook and Medley HO scale: A heritability analysis in adult male twins. Psychosom Med 50: , Wooten AJ: Effectiveness of the K correction in the detection of psychopathology and its impact on profile height and configuration among young adult men. J Consult Clin Psychol 52: , Silver R[, Sines LK: Diagnostic efficiency of the MMPI with and without the K correction. J Clin Psychol 18: , Psychosomatic Medicine 51:58-65 (1989)

8 CAUTION ON USE OF MMPI K-CORRECTION 9. Hsu LM: Implications of differences in elevations of K-corrected and non-k-corrected MMPI T scores. J Consult Clin Psychol 54: Cook W, Medley D: Proposed hostility and pharisaic-virtue scales for the MMPI. J Appl Psychol 38: , Shock NW, Greulich RC, Andres R, Arenberg D, Costa PT. Jr, Lakatta EG, Tobin JD: Normal Human Aging: The Baltimore Longitudinal Study of Aging (NIH Publication No ). Bethesda, MD, National Institutes of Health, Costa PT, Jr, McCrae RR: Manual for the NEO Personality Inventory. Odessa, FL, Psychological Assessment Resources, Costa PT, Jr, McCrae RR: Personality in adulthood: A six-year longitudinal study of self-reports and spouse ratings on the NEO Personality Inventory. J Pers Soc Psychol 54: , McCrae RR, Costa PT, Jr: Validation of the five-factor model of personality across instruments and observers. J Pers Soc Psychol 52:81-91, Barefoot JC, Dodge KA, Peterson BL, Dahlstrom WG, Williams RB, Jr: The Cook-Medley Hostility Scale: item content and ability to predict survival. Psychosom Med 51:46-57, Rosenman RH: The interview method of assessment of the coronary-prone behavior pattern. In Dembroski TM, Weiss SM, Shields JL, Haynes SG, Feinleib M (eds), Coronary-Prone Behavior. New York, Springer-Verlag, 1978 pp Williams R, Haney T, Lee K, Kong Y, Blumenthal J, Whalen R: Type A behavior, hostility and coronary heart disease. Psychosom Med 42: , Dembroski TM, MacDougall JM, Williams RB, Jr, Haney TL, Blumenthal JA: Components of Type A, hostility and anger-in: Relationship to angiographic findings. Psychosom Med 47: , Kalbfleisch JD, Prentice RL: The Statistical Analysis of Failure Time Data. New York, Wiley, McCrae RR, Costa PT, Jr: Social desirability scales: More substance than style. J Consult Clin Psychol 51: , McCrae RR: Well-being scales do not measure social desirability. J Gerontol 41: , 1986 Psychosomatic Medicine 51:58-65 (1989) 65