Bond University

ePublications@bond Humanities & Social Sciences papers

Faculty of Humanities and Social Sciences

9-1-1994

Examination of the reliability and validity of the Personality Assessment Inventory Gregory J. Boyle Bond University, [email protected]

Tania J. Lennon University of Queensland

Follow this and additional works at: http://epublications.bond.edu.au/hss_pubs Part of the Psychology Commons Recommended Citation Gregory J. Boyle and Tania J. Lennon. (1994) "Examination of the reliability and validity of the Personality Assessment Inventory" Journal of Psychopathology and Behavioral Assessment, 16 (3), 173-187: ISSN 1573-3505. http://epublications.bond.edu.au/hss_pubs/792 This Journal Article is brought to you by the Faculty of Humanities and Social Sciences at ePublications@bond. It has been accepted for inclusion in Humanities & Social Sciences papers by an authorized administrator of ePublications@bond. For more information, please contact Bond University's Repository Coordinator.

Examination of the Reliability and Validity of the Personality Assessment Inventory

Gregory J. Boyle Department of Psychology, Bond University and Department of Psychology, University of Queensland

and

Tania J. Lennon Department of Psychology, University of Queensland

Abstract The reliability, discriminant validity, and construct validity of the Personality Assessment Inventory (PAI) -a multidimensional self-report measure of abnormal personality traits-were examined within the Australian context. Subjects included 151 normal individuals, 30 alcoholics, and 30 schizophrenic patients. A subsample of 70 non-psychiatric adults responded to the PAI items twice over a test-retest interval of 28 days. The resulting median retest coefficient was 0. 7, indicating less than optimal stability. The median alpha (KR21 coefficient was 0.8, suggesting rather narrow measurement scales. A significant multivariate main effect was obtained across groups after the effects of age and gender were removed. Multiple comparisons for each of the PAI scales revealed significant differences between the respective groups, as discussed. A higher-order scale factoring did not strongly support the purported PAI structure. In reanalyses of the correlation matrices included in the Professional Manual, the purported PAI factor structure was unable to be replicated for the standardization clinical sample (N = 1246), and a confirmatory factor analysis using the normative (validation) correlational data (N = 1000) revealed poor fit indices, raising further concerns about construct validity.

Introduction The Personality Assessment Inventory, or PAI (Morey, 1991), is a recently constructed multidimensional personality inventory designed "to provide information relevant to clinical diagnoses, treatment planning and screening for psychopathology" (Morey, 1991, p. 1). Construction of the PAI was intended to provide a superior alternative to the MMPI-2 instrument (Butcher, Dahlstrom, Graham, Tellegen, & Kaemmer, 1989), in view of the latter's well-documented psychometric limitations (e.g., Caldwell, 1991; Duckworth, 1991; Eysenck, 1991; Helmes & Reddon, 1993). According to White (1993, p. 23), "When compared with the MMPI-2, the PAI has a number of significant advantages. The PAI is more logically structured, it is better designed, it is more relevant to contemporary issues and terminologies, it is easier for subjects to complete, it is easier for clinicians to score and it is more straightforward to interpret." However, because of its relative newness, little psychometric data other than that contained in the Professional Manual is available. Consequently, the present study examines the test-retest reliability, discriminant validity, and construct (factor analytic) validity of the PAI using Australian samples of normal, alcoholic, and schizophrenic individuals. The PAI consists of 344 items and 22 scales, including validity, clinical, and interpersonal scales. The four validity scales are labelled: Inconsistency (to ascertain if the respondent is answering consistently), Infrequency (to detect careless or random responding). Negative Impression (to ascertain whether the individual is attempting to portray an unfavourable impression or is malingeringfaking bad), and Positive Impression (to determine if the respondent is attempting to present an unduly favourable impression- faking good). There are 11 clinical scales. Somatic Complaints focuses on preoccupation with health matters. Anxiety measures the phenomenology and observable signs of

anxiety. Anxiety-Related Disorders focuses on symptoms and behaviours related to specific anxiety disorders. Depression quantifies symptoms and phenomenology of depressive disorders. Mania assesses affective, cognitive, and behavioural symptoms of mania and hypomania. Paranoia indexes symptomatology and personality elements of the paranoid individual. Schizophrenia measures symptoms relating to various schizophrenic disorders. Borderline Features assesses a number of elements related to severe personality disorders. Antisocial Features focuses on behavioural and personality features relevant to antisocial personality and psychopathy. Alcohol Problems and Drug Problems assess behaviours and consequences related to the use, abuse, and dependence on alcohol and drugs, respectively. Five treatment scales are included. Aggression measures hostility to- ward others. Suicidal Ideation assesses thoughts and ideas about death and suicide. Stress indexes life stressors, such as relationship problems, financial hardships, and major life changes, currently or recently experienced. Non- support measures perceived lack and quality of social support. Treatment Rejection quantifies an individual's willingness to participate actively in treatment. Also, two interpersonal scales are included. Dominance taps the extent to which an individual is controlling, submissive, or autonomous in interpersonal relationships. Warmth measures the extent to which a person is empathic, engaging or reflecting in interpersonal relationships. The Schizophrenia scale comprises 24 items, with three subscales of 8 items each, labelled Psychotic Experiences (unusual perceptions, sensations, and delusional beliefs), Social Detachment (social isolation, poor social, and interpersonal competence), and Thought Disorder (confused thought processes and difficulties in concentration). The Alcohol Problems scale contains 12 items (without subscales), ranging from statements of total abstinence through frequent

use to severe consequences of drinking, loss of control, and alcohol-related cravings. Items considered offensive or potentially biased with respect to gender, ethnic, economic, religious, or other reasons have been avoided in construction of the instrument (Morey, 1991). However, the PAI does contain a number of items concerning sexual and bodily functions which possibly might be offensive to some sensitive individuals, such as Item 315 ("I have little interest in sex") and Item 312 ("I frequently have diarrhea"). Gender differences in mean PAI scale scores appear to be slight, except for the Antisocial scale and the Alcohol Problems scale, corresponding with the higher incidence of antisocial personality disorder and alcoholism reported in males (Morey, 1991). Differences in scale scores attributable to race also tend to be slight. However, age does appear to influence scale scores, with younger individuals obtaining higher mean T scores (cf. Tellegen & Ben-Porath, 1992) on Anxiety, Paranoia, Borderline Features, and Antisocial Features, probably reflecting less stable personality at younger ages (cf. Costa & McCrae, 1989). Education also seems to have a consistent effect with less educated individuals obtaining T scores below sample means, and vice versa. Consequently, the Alcohol Problems and Schizophrenia scales may suffer from the same shortcomings as their MMPI-2 counterparts (e.g., loss of effectiveness in older individuals). Also, there may be a problem with validity stemming from the polythetic nature of the Schizophrenia scale. Many possible endorsement patterns might elevate scale scores into the "schizophrenic" range. In view of the low correlation between reports of alcohol consumption by alcoholics and their significant others over time, Watson (1985) suggested that self-reports of alcoholism are questionable. The pervasiveness of denial in the verbal behaviour of alcoholics (Ward & Rothaus, 1991) is likely to decrease the

correlation, as the Alcohol Problems scale items in- quire directly about alcohol use. Another possible difficulty is that distinct subgroups exist within the general diagnostic category of alcoholism which are widespread in terms of character and symptom pathology (Corbiserio & Reznikoff, 1991; Pattison, 1984). While the Alcohol Problems scale was not designed to discriminate among alcoholic subtypes, hopefully, response patterns on the other 21 PAI scales would differentiate among subtypes. Further empirical evidence is required to determine whether the PAI scales can reliably discriminate between psychiatric patients and normal individuals. For example, according to Morey (1991), an elevation at or above T score 70 on the Alcohol Problems scale is probably indicative of alcohol abuse and concomitant difficulties in interpersonal relationships or work performance. For the PAI Schizophrenia scale, elevations at or above T score 90 suggest the occurrence of an active schizophrenic episode (Morey, 1991). Based on such considerations, a number of hypotheses are postulated: Hl: Alcoholics will obtain significantly higher scores on the Alcohol Problems scale than will non-alcoholic individuals. H2: Only individuals from the alcoholic group will obtain T scores greater than 70 on the Alcohol Problems scale. H3: Schizophrenic individuals will obtain significantly higher scores on the Schizophrenia scale than will non-schizophrenic individuals. H4: Only schizophrenic individuals will obtain T scores greater than 90 on the Schizophrenia scale. H5: Test-retest reliability coefficients equal to or greater than 0.8 will be obtained for each PAI scale. H6: The PAI higher-order factor structure will not differ significantly from that reported in the Professional Manual.

Method Subjects The total sample comprised 211 subjects, divided into three separate groups. For the non-clinical group, 170 individuals were recruited from among college students and the general adult community at large. A sub-sample of 70 normal individuals was administered the PAI twice (28 days apart). Clinical patients were recruited from Wolston Park Psychiatric Hospital, Brisbane. Schizophrenic patients were recruited from the hospital's Clinical Studies Unit, and• alcoholics from the hospital's Rehabilitation Clinic. Diagnoses were defined in terms of DSM-III-R criteria and verified as reliable by independent psychiatric assessment. The fact that patients were currently receiving treatment for either alcoholism or schizophrenia further served to ensure the accuracy of assignment of patients into the two clinical groups. The total sample comprised 31 females and 180 males• whose overall mean age was 33.67 years. The number of subjects in each group was alcoholics (N = 30), schizophrenics (N = 30), and normal individuals (N = 151). Procedure Subjects read the questions and marked their answers on a 4-point scale (ranging from False, Slightly True, Mainly True, to Very True). If they experienced reading difficulties, the questions were read aloud to them, and the subjects marked their responses on the answer sheet provided. Subjects who experienced motor or reading difficulties, responded verbally to the PAI items choosing from among the four response options, while the test administrator marked their responses on the answer sheet. Difficulties were encountered with some clinical patients due to withdrawal from alcohol and medication effects. Prior

to completing the questionnaire, all volunteers were asked to complete an informed consent form.

Results and Discussion

A MANOVA was carried out across all PAI scales, with group (control, alcoholic, and schizophrenic) as the independent variable. Using Pillai's criterion, there was a• significant multivariate main effect for group [F(2,378) = 17.307, p < .001], thereby justifying examination of the univariate between group effects for each scale. T scores for each of the PAI scales (except for Inconsistency) are presented in Table 1, along with the corresponding test-retest (Pearson productmoment) reliability coefficients for each scale. Differences in scale scores due to gender and age (younger than 35 years, 35 years or older) were also investigated. Mean Schizophrenia scale scores for the older and younger age groups were 26.30 and 18.93, respectively, while for males and females, the mean scores were 22.12 and 18.35, respectively. Likewise, for the Alcohol Problems scale, mean scores were 17.63 and 13.06 respectively for the older and younger groups, and 14.80 and 13.19 respectively for males and females. Differences in scores due to age were significant on the Schizophrenia scale (F(1,209) = 24.06, p < .0001] and on the Alcohol Problems scale [F(1,209) = 15.99, p < .0001]. A significant effect was also obtained for gender on the Schizophrenia scale [F(I,209) = 6.57, p < .01].

Table 1 T-Scores and Reliability for Normal and Clinical Groups on PAI Scales

Scale

Normal Alcohol Schizo Stability KR20

Validity Infrequency

50

60

45

.67

.60

Negative Impression

62

100

79

.85

.77

Positive Impression

39

46

45

.81

.58

Somatic Complaints

57

78

74

.73

.90

Anxiety

54

76

71

.62

.88

Anxiety-Related Disorders

51

78

67

.76

.84

Depression

55

82

69

.80

.89

Mania

48

67

62

.72

.so

Paranoia

52

79

66

.65

.84

Schizophrenia

53

77

72

.79

.82

Borderline Features

54

81

71

.73

.88

Antisocial Features

55

81

67

.63

.85

Alcohol Problems

63

88

64

.78

.83

Drug Problems

59

89

71

.76

.82

Aggression

51

72

63

.77

.86

Suicidal Ideation

60

95

74

.86

.85

Stress

61

82

62

.73

.79

Nonsupport

59

82

58

.69

.89

Treatment Rejection

49

36

47

.69

.76

Dominance

42

49

45

.70

.63

Warmth

37

43

50

.74

.76

Clinical

Treatment

Interpersonal

Notes. Data for the Inconsistency scale not available. Stability coefficients based on a 28-day test-retest interval.

MANCOVAs were carried out on the data with gender and age as covariates, in order to correct statistically for distorting effects due to those variables. The main effect across groups was still significant [F(2,374) = 15.22, p < .0001], after the effects of age and gender were partialled out. In order to counteract the effects of heterogeneity of variance due to unequal group sizes, a more stringent criterion of statistical significance was used as advocated by Keppel (1991, p. 283). In addition, because of the large number of PAI scales, Bonferroni corrections were applied to minimise finding significant between-group differences due to chance alone. Despite the use of this conservative approach, multiple comparisons revealed significant differences (all p's < .001) in mean scores between the normal and the combined clinical groups in every instance, except for the Infrequency scale. While the alcoholic and schizophrenic groups exhibited significant differences on several PAI scales, the results were less straightforward. Thus, there were no significant differences on Positive Impression, Somatic Complaints, Anxiety, Mania, Schizophrenia, Dominance, and Warmth scales. Anxiety-Related Disorders, and Aggression differed between the two clinical groups at the p < .01 level, while the remaining scales differed at the p < .001 level. In general, normal individuals obtained significantly lower mean scores than did the schizophrenic patients, who in turn, scored significantly lower than did the alcoholic individuals. The effects of antipsychotic medication may have attenuated scale scores for the schizophrenic group. The fact that the PAI Schizophrenia scale did not differentiate significantly between the two clinical groups suggests either invalidity of the scale or a medication-induced attenuation of Schizophrenia scale scores. The Alcohol Problems scale seems to have done a better job of discriminating the alcoholic and schizophrenic groups. However, the

alcoholic patients were not taking antipsychotic medication, so there is no reason to believe that their scores were attenuated. This interpretation certainly fits in with the finding that the Paranoia scale exhibited a high T score for the alcoholic group, and a low score for the schizophrenic group, where one of the most frequent diagnoses is paranoid schizophrenia. It is likely that antipsychotic medication reduced the Paranoia scale scores for the schizophrenic group. Had there been no medication, it is possible that the Schizophrenia scale scores would have been appreciably higher than those reported in Table I, as would be expected if the scale has adequate discriminative validity. In addition, there are several scales (e.g., Positive Impression, Dominance, and Warmth) where the respective groups scored relatively low and did not differ appreciably. Of these, the first scale is a validity scale, while the last two were interpersonal scales, which (according to the Professional Manual) were not expected to discriminate highly between normal and clinical groups. Regarding the Positive Impression scores, Morey (1991, pp. 12-13) reported that "marked elevations in clinical subjects are particularly rare… Both patients and normal individuals score considerably lower than research subjects completing the PAI under a positive-impression-enhancement instructional set." Indeed, a low T score on this scale suggests a valid PAI profile, lending weight to the present findings. Group membership of individuals scoring greater than 70T on the Alcohol Problems scale was also examined to determine rates of false positives. Of the total sample, 77 patients were classified as meeting the criteria for alcohol abuse. Of these, 35% were alcoholics, 15% were schizophrenics, and 50% were otherwise normal individuals. For the alcoholic group, 90% were correctly classified as meeting the criteria for alcohol abuse. Yet no less than 25% of the normal group was classified as meeting the criteria for alcohol abuse. Assuming that the base rate in Queensland of alcohol abuse (including alcoholic psychosis,

alcoholic dependence syndrome, nondependent abuse of alcohol, alcoholic liver disease, and admissions to psychiatric hospitals and alcoholism units for treatment of alcoholic psychosis and alcohol dependence syndrome) is only 0.18% (Department of Community Services and Health, Statistical Services Section, 1988), the majority of these individuals is likely to have been false positives. Scores on the Schizophrenia scale were also examined. The scale correctly identified 70% of schizophrenic patients. However, only three patients (1.5% of the sample) scored greater than 90T on the scale (indicating an active schizophrenic episode) and these were from the alcoholic group (as pointed out above, schizophrenic patients were taking antipsychotic medication). Some 76% of the alcoholic group scored greater than 70T on the Schizophrenia scale (as compared with only 4% of normal individuals). This suggests either inadequate discriminative validity or the possible comorbidity of schizophrenia and alcohol abuse in the present samples. However, the Thought Disorder subscale was highly elevated among the alcoholics suggesting reduced cognitive abilities in this group. In accord with Hl, the alcoholic group obtained a significantly higher mean score on the Alcohol Problems scale than did either the normal or schizophrenic groups (M = 26.60 vs. M = 12.12 and M = 12.77, respectively). However, H2 received only mild support, as subjects from the alcoholic sample were not the only ones to obtain T scores greater than 70; indeed, 65% of those who scored above this criterion were not alcoholics. In partial support of H3, the schizophrenic group exhibited a significantly higher mean score as compared with the normal group, but there was no significant difference compared with the alcoholic group. In fact, the alcoholic group obtained a marginally higher mean score on the Schizophrenia scale compared with the schizophrenic group (M = 34.97 vs. M = 30.87). As predicted, the mean score for normal individuals was considerably lower (M = 16.11). H4, that T scores greater than 90 would be obtained on the

Schizophrenia scale only for schizophrenic individuals, was not supported. With regard to H5, test-retest reliability coefficients generally were a little below predicted values. H6, that the higher-order factor structure of the PAI would correspond closely with that reported in the Professional Manual, was not supported, in the main. For Alcohol Problems, classificatory accuracy was generally good, with 90% of alcoholics correctly classified. However, as has been found with the MMPI Mac alcoholism scale (cf. Davis, Offord, Colligan, & Morse, 1991; Gottesman & Prescott, 1989), a high rate of false positives (25%) occurred. This result may have been influenced by the reduced accuracy of the Alcohol Problems scale with younger individuals, as the mean age for the normal group was only 27.17 years. Classificatory accuracy of the Schizophrenia scale is unclear. No Schizophrenic patients obtained a T score greater than 90. Although 70% of the schizophrenic sample was accurately classified in terms of scores greater than no, a slightly greater percentage of alcoholics (76%) was classified as schizophrenic on this criterion. Thus, the Schizophrenia scale differentiated well between clinical patients and normal individuals but not between the clinical groups. As some of the schizophrenic patients were taking antipsychotic medication, this necessarily reduced overall scores on the Schizophrenia scale. Although the Schizophrenia scale was based on DSM-III-R criteria, some symptoms, such as delusions, hallucinations, social withdrawal, and impairment of personal hygiene, also may be common to individuals suffering from withdrawal of alcohol dependence. Preliminary evidence for the influence of moderator variables (age and gender) was also found for several PAI scales. Males scored higher on the Schizophrenia scale and the Alcohol Problems scale than did females. Significant age differences were also observed with older individuals obtaining higher scores. However, these analyses were limited by the low number of females and younger

individuals in the clinical groups. The effect of these variables on scale scores warrants further investigation to deter- mine if these differences impact on scale discriminant validity. Test-retest reliability coefficients were moderate in most instances, with a median stability coefficient of 0.73, ranging from 0.62 (Anxiety) to 0.86 (Suicidal Ideation). Thus, the reliability of the PAI is about as claimed in the Professional Manual, and about as found for most personality inventories assessing psychopathology (cf. Fernandez, 1990). Nevertheless, since no treatment occurred between the two measurement occasions, test- retest coefficients were expected to be 0.8 or above (cf. Ben-Porath & Butcher, 1989; Boyle, 1985). Indeed, Morey's (1991) test-retest results over a similar retest interval provided somewhat stronger support for the stability of most of the PAI scales (except for the Infrequency scale). Morey (1991) reported a median stability coefficient of 0.81, suggesting that the PAI scales are relatively stable over time. Alternatively, the PAI scales tap dynamic traits, which are relatively less stable than normal personality traits, in general. On present findings though, the stability of a small number of PAI scale scores over time (In-frequency, Anxiety, Paranoia, Antisocial Features, Nonsupport, and Treatment Rejection) is somewhat in doubt. Cronbach alpha coefficients were mostly quite high (median = 0.83), suggesting the possibility of rather narrow scales, with excessive item redundancy (cf. Boyle, 1991). To check on the construct validity of the PAI instrument, an exploratory higher-order factor analysis was undertaken on the scale intercorrelations for all subjects combined. This approach of combining both normal and clinical groups ensured a good spread of variance, thereby enhancing the likelihood of obtaining sizeable intercorrelations and a resulting optimal factor solution. In accord with appropriate factor analytic methodology (Gorsuch, 1983), an iterative maximum

likelihood (ML) procedure was employed (using SMCs as initial communality estimates), together with oblique (direct oblimin) simple structure rotation. Five factors were extracted on the basis of the Scree test (Cattell, 1978), as shown in Table 2. The five higher-order factors provide only partial support for the purported PAI structure (see Morey, 1991, pp. 136-138). Factor 3 resembles Morey's first factor, with high loadings on Depression, Anxiety, Anxiety- Related Disorders, Schizophrenia, and Borderline Features. However, the remaining factors reported in the Professional Manual are not well sup- ported by the present findings (cf. Boyle, 1987). In part, this lack of concordance may be accounted for by Morey's use of inappropriate factor analytic procedures. Morey employed a "Little Jiffy" procedure wherein principal components with eigenvalues 1.0 were rotated to an orthogonal (varimax) solution. Such "short-cut" procedures are known to produce less than optimal factor solutions, which fail to meet simple structure requirements (McDonald, 1985). Perusal of the factor correlations shown in Tables 2 and 3 indicates that the higher-order PAI factors exhibit sizeable intercorrelations in most instances and that Morey's use of orthogonal rotation was inappropriate. A maximum-likelihood confirmatory scale factor analysis using the Australian data indicated a poor fit to Morey's (1991) four-factor model. The adjusted goodness-of-fit index (AGFI) was only 0.69, and the corresponding root mean square residual (RMR) was 0.08, raising further doubts about the adequacy of the claimed PAl factor structure. Consequently, re-analyses of the PAl scale correlations included in the Professional Manual (Morey, 1991, p. 134) were undertaken in order to check on the accuracy of Morey's factors. For the censusmatched normative sample (N = 1000), an exploratory factor analysis was carried out. Extraction and rotation of four factors (using the dubious Little Jiffy approach) replicated the factor solution reported by Morey (pp. 136-138).

However, in line with the exploratory factor analysis of the Australian data, the Scree test suggested a five-factor oblique solution was more appropriate (see Table Ill). Only eight iterations were required for convergence of initial communality estimates.

Table 2 Oblique Higher-Order PAI Factor Pattern (Australian Data)

Scale Aggression Borderline Features Paranoia Drug Problems Suicidal Ideation Infrequency

Factor I

Factor 2

Factor 3

Factor 4

.72 .61 .48 .45 .43 .17

Dominance Warmth Positive Impression Mania

.72

.56 .63 .35 .78

.71 .57 .37 -.90 -.70 -.69 -.68 -.54 .25

Antisocial Features

.89 .76 .89 .83 .78 .28 -.53

Nonsupport Stress Alcohol Problems Negative Impression

Eigenvalue

.84 .74 .69 .44 .38

51.9 10.9

hl .72 .87 .78 .67 .81 .06

Anxiety Somatic Complaints Anxiety-Related Disorders Depression Schizophrenia Treatment Rejection

% variance

Factor 5

9.5 2.0

5.3 1.1

5.0 1.0

.55

.64 .55

.69

3.9 0.8

Notes. Data for the Inconsistency scale unavailable; Factor loadings shown to two decimal places only.' Factor 1 correlates .29 with Factor 2, -.66 with Factor 3, -.30 with Factor 4, and .53 with Factor 5; Factor 2 correlates -.22 with Factor 3, -.24 with Factor 4, and .13 with Factor 5; Factor 3 correlates .30 with Factor 4 and -.57 with Factor 5; Factor 4 correlates .19 with Factor 5.

With regard to Morey's (1991) clinical data (N = 1246), even though both the eigenvalues greater than unity criterion and the Scree test indicated five factors, Morey reported only a four-factor solution. In an effort to exactly replicate his findings, a four-factor solution was attempted (again using the Little Jiffy approach). Remarkably, Morey's factor solution could not be reproduced, the SPSS printout gave a warning that the correlation matrix was "ill conditioned," and Bartlett's test of sphericity could not be calculated, indicating that the correlation matrix did not meet the underlying assumption of multivariate normality, necessary to undertake a valid factor analysis. Likewise, attempts to produce a five-factor solution also failed to produce a valid output. Therefore, either the matrix for the clinical sample included in the Professional Manual is the wrong one or Morey's factor analytic results for the clinical sample are erroneous (factor analytic methodology aside).

Table 3 Oblique Five-Factor Solution (Morey’s Normative Standardization Sample)

Scale Factor Factor 2 Factor 3 Factor 4 Factor 5 h2 __________________________________________________________________________ Somatic Complaints

.76

.56

Anxiety

.67

.76

Depression

.65

.80

Anxiety-Related Disorders

.59

.58

Negative Impression

.57

.66

Suicidal Ideation

.41

.50

Schizophrenia

.40

.68

Dominance

.71

.57

Mania

.68

.64

Aggression

.35

.53

Drug Problems

.78

Alcohol Problems

.71

Antisocial Features

.62

Inconsistency

.40

.62 .49 .70 .44

Nonsupport

-.76

Paranoia

-.63

Warmth

.58

Borderline Features

-.40

Stress

-.37

Positive Impression Treatment Rejection Infrequency %variance

41.0

9.7

8.3

4.8

Eigenvalue

9.0

2.1

1.8

1.1

.63 .68 .43 .81 .45

.77

.68

.40

.27

.35

.40

4.3 1.0

Notes. Factor loadings shown to two decimal places only. Factor I correlates -.01

with Factor 2, .43 with Factor 3, -.54 with Factor 4, and -.36 with Factor 5; Factor 2 correlates .13 with Factor 3; .01 with Factor 4, and -.20 with Factor 5; Factor 3 correlates -.55 with Factor 4 and -.13 with Factor 5; Factor 4 correlates .28 with Factor 5.

Table 4 Confirmatory Factor Analysis of Morey’s Higher-Order Clinical Factors Standardized LISREL estimates (ML), Factor No.

ξ1

Morey’s Factor

ξ2

ξ3

ξ4

R2

Factor 1 (ξ1) Anxiety

.864

.75

Anxiety-Related Disorders

.716

.51

Depression

.824

.68

Schizophrenia

.800

.64

Borderline Features

.838

.70

Factor 2 (ξ2) Antisocial Features Alcohol Problems

.

Drug Problems

Factor

.808

.65

.680

.46

.695

.48

3

( ξ3) Mania

.582

.34

Aggression

.813

.66

Dominance

.194

.04

Fac:tor 4 (ξ4) Inconsistency

.693

.48

Infrequency

.563

.32

Notes. GFI = 0.802; AGFI = 0.694; RMR = 0.120 Communality estimates shown to two decimal places only. Factor loadings shown to three decimal places only. Correlation matrix for Morey's normative standardization sample (N = 1000) used. All lambda-X loadings are significant (p < .01 or better} using two-tailed tests.

20

As a further check on the accuracy of the purported PAI factor structure, a confirmatory maximum-likelihood factor analysis of the four-factor solution reported by Morey (1991) for the clinical sample was undertaken using LISREL (Jöreskog & Sörbom, 1989) on the normative (validation) data. Even with removal of 10 "noisy" scale variables from the analysis (having the lowest loadings in Morey's factor solution for the clinical sample), the results failed to support the claimed factor structure (the AGFI was only 0.69, while the RMR was 0.12). The standardized LISREL estimates for the CFA are shown in Table 4, while the corresponding covariances (phi coefficients) are provided in Table 5. Some other limitations also seem applicable to the PAI (Boyle, 1994). The PAI has several items asking about potentially sensitive issues. Even within medical clinical settings, some items may go beyond what many patients are willing to accept as legitimate inquiry. Second, the moderator variables, age and gender, may have systematic effects on the discriminant validity of these scales. Third, the efficacy of the Somatic Complaints, Anxiety, Mania, Schizophrenia, Dominance, and Warmth scales in discriminating between different clinical groups is open to question, although these scales appear to differentiate well between clinical and normal groups. These scales did not perform very well, the basic problem being too many false positives, especially among the alcoholic sample, for the clinical scales. Profiles for the three groups seem problematic. For instance, the very high mean scores on Negative Impression and on Suicidal Ideation suggest the occurrence of ceiling effects (cf. Fernandez, Nygren, & Thorn, 1991) and the possible invalidity of the protocols of the alcoholic group. The latter symptom, by

21

itself, is not pathognomonic for alcoholism. This and other peculiarities in the profile for alcoholics suggest that the sample studied may have been somewhat atypical.

Table 5 Covariances between Exogenous Latent Traits (Phi Martrix)

Morey's clinical factor

1

2

3

Factor 1 ( 1 )

Factor 2 ( 2)

.581

Factor 3 ( 3)

.598

.724

Factor 4 ( 4)

.699

.761

.410

Note. Covariances shown to three decimal places only.

Clearly, further research into the psychometric adequacy of the PAI is required, particularly focusing on the influence of moderator variables on the separate scales and the ability of the instrument to discriminate between different psychopathological groups.

22

References Ben-Porath, Y. S., & Butcher, J. M. (1989). Psychometric stability of rewritten MMPI items. Journal of Personality Assessment, 51, 645-653. Boyle, G. J. (1985). Self-report measures of depression: Some psychometric considerations. British Journal of Clinical Psychology, 24, 45-59. Boyle, G. J. (1987). Psychopathological depression superfactors in the Clinical Analysis Questionnaire. Personality & Individual Differences, 8, 609-614. Boyle, G. J. (1991). Does item homogeneity indicate internal consistency or item redundancy in psychometric scales? Personality and Individual Differences, 12, 291-294. Boyle, G. J. (1994). Review of the Personality Assessment Inventory. In J. C. Conoley & J. lmpara (Eds.), Twelfth mental measurements yearbook. Lincoln, NE: Buras Institute of Mental Measurements. Butcher, J. N., Dahlstrom, W. G., Graham, J. R., Tellegen, A., & Kaemmer, B. (1989). Manual for the restandardized Minnesota Multiphasic Personality Inventory: MMPI-2. An interpretative and administrative guide. Minneapolis: University of Minnesota Press. Caldwell, A. (1991). Commentary on The Minnesota Multiphasic Personality Inventory-2: A review. Journal of Counselling & Development, 69, 568569. Cattell, R. B. (1978). The scientific use of factor analysis in behavioral and life sciences. New York: Plenum. Corbisiero, J. R., & Reznikoff, M. (1991). The relationship between personality type and problem drinking. Journal of Clinical Psychology, 47, 291-298.

23

Costa, P. T., Jr., & McCrae, R. R. (1989). The NEO-PI/NEO-FFI manual supplement Odessa, FL: Psychological Assessment Resources. Davis, L. J., Offord, K. P., Colligan, R. L., & Morse, R. M. (1991). The CAL: An MMPI Alcoholism scale for General Medical Patients. Journal of Clinical Psychology, 47, 632-646. Department of Community Services and Health Statistical Services Section (1988). Alcohol in Australia: A summary of related statistics. Canberra: Commonwealth Government. Duckworth, J. C. (1991). The Minnesota Multiphasic Personality lnventory-2: A review. Journal of Counselling and Development, 69, 564-567. Eysenck, H. J. (1991). Dimensions of personality: 16, 5, 3? Personality and Individual Differences, 12, 773-790. Fernandez, E. (1990). Artifact in pain ratings, its implications for test-retest reliability, and correction by a new scaling procedure. Journal of Psychopathology and Behavioural Assessment, 12, 1-15. Fernandez, E., Nygren, T. E., & Thorn, B. E. (1991). An "open transformed scale" for correcting ceiling effects and enhancing retest reliability: The example of pain. Perception and Psychophysics, 49, 572-578. Gorsuch, R. L. (1983). Factor analysis (rev 2nd ed.). Hillsdale, NJ: Erlbaum. Gottesman, I. I., & Prescott, G. A. {1989). Abuses of the MacAndrew MMPI Alcoholism Scale. A critical review. Clinical Psychol. Review, 9, 223-242. Helmes, E., & Reddon, J. R. {1993). A perspective on developments in assessing psychopathology: A critical review of the MMPI and MMPI-2. Psychological Bulletin, 113, 453-471.

24

Jöreskog, K. G., & Sörbom, D. (1989). LJSREL 7: A guide to the program and applications (2nd ed.), Chicago, IL: SPSS Inc. Keppel, G. (1991). Design and analysis: A researcher's handbook (rev. 3rd ed.). Englewood Cliffs, NJ: Prentice Hall. McDonald, R. P. (1985). Factor analysis and related methods. Hillsdale, NJ: Erlbaum. Morey, L. C. (1985). A psychometric analysis of live DSM-III categories. Personality and Individual Differences, 6, 323-329. Morey, L. C. (1991). The Personality Assessment Inventory: Professional manual. Odessa, FL: Personality Assessment Resources. Rogers, R., Ornduff, S. R., & Sewell, K. W. (1993). Feigning specific disorders: A study of the Personality Assessment Inventory. Journal of Personality Assessment, 60, 554-560. Tellegen, A., & Ben-Porath, Y. S. (1992). The new uniform T scores for the MMPI-2: Rationale, derivation and appraisal. Psychological Assessment, 4, 145-155. Ward, L. C., & Rothaus, P. (1991). The measurement of denial and rationalization in male alcoholics. Journal of Clinical Psychology, 47, 465-468. Watson, C. G. (1985). More reasons for moratorium: A reply to Maisto & O'Farrell. Journal of Studies on Alcohol, 46, 450-453. White, J. (1993). The Personality Assessment Inventory in a forensic setting. Paper presented at the 28th Annual Convention or the American Psychological Association, Los Angeles.