Reliability, Faking Susceptibility, and Discriminant Validity of the Basic Personality Inventory

The Minnesota Multiphasic Personality Inventory (MMPI) represents the most frequently used and researched personality instrument available today (Edwards and Abbott, 1973). Yet, despite its popularity as a diagnostic instrument in mental health settings, reviewers generally agree that the MMPI is a "psychometric monstrosity" (Rodgers, 1972). The recently developed Basic Personality Inventory or BPI (Jackson, 1974) represents a potential alternative to the presently popular MMPI. The reliability, susceptibility to dissimulation, and discriminant validity of the BPI scales were examined in this study. The BPI was administered to 168 university students and 224 community college students. Data from the first sample (N = 168) was used primarily to conduct an internal consistency and test-retest analysis of the BPI. It was hypothesized that the BPI scales would demonstrate adequate reliability (r >. 70). This hypothesis was supported for some scales but not for others. One explanation given for the lower than expected reliability coefficients was the restricted variability observed in the population studied. It was suggested that reliability coefficients based on a clinical population would probably be higher. The observed reliabilities, however, were generally higher than those reported for the MMPI scales. The second sample (N = 224) was used primarily to investigate

popularity as a diagnostic instrument in mental health settings, reviewers generally agree that the MMPI is a "psychometric monstrosity" (Rodgers, 1972).

The recently developed Basic
Personality Inventory or BPI (Jackson, 1974) represents a potential alternative to the presently popular MMPI. The reliability, susceptibility to dissimulation, and discriminant validity of the BPI scales were examined in this study. The BPI was administered to 168 university students and 224 community college students. Data from the first sample (N = 168) was used primarily to conduct an internal consistency and test-retest analysis of the BPI. It was hypothesized that the BPI scales would demonstrate adequate reliability (r >. 70).
This hypothesis was supported for some scales but not for others. One explanation given for the lower than expected reliability coefficients was the restricted variability observed in the population studied. It was suggested that reliability coefficients based on a clinical population would probably be higher. The observed reliabilities, however, were generally higher than those reported for the MMPI scales. The second sample (N = 224) was used primarily to investigate ii the faking susceptibility of the scales. It was predicted that mean scale scores for persons receiving standard instructions (n = 124) would be significantly higher than mean scale soores obtained £ran persons receiving "fake gooo adjustment" instructions (n = 50), and significantly l01er than mean scale soores obtained £ran persons receiving the "fake maladjustment" instructions (n = 50). In

INTRODUCTION
In clinical psychology, the role of assessment is obvious and crucial. Practitioners rely heavily on assessment procedures to aid in diagnostic decision making (Lanyon and Goodstein, 1971). Thus, we expect the assessment techniques used to make these judgements to be the best available.
One popular source of diagnostic information is the Minnesota Multiphasic Personality Inventory or MMPI (Hathaway and McKinley, 1967). The MMPI is generally considered to be a convenient, economical, and time saving method for screening patients (Sundberg, 1977), describing abnormal behavior patterns (Little and Shneidman, 1954), and aiding in diagnostic decision making (Martin, 1977). Is the MMPI, however, the best self-report personality inventory available?
The research reported here is related primarily to this question.
As McReynolds (1977) points out, behavioral scientists and practitioners must keep abreast of new developments in the area of assessment. New instruments must be extensively researched and compared to existing measures. When better assessment measures exist, 2 older ones must be replaced. What follows is a critical review considering the present status of the MMPI, a description of a potential alternative - Jackson's (1974) Basic Personality Inventory (BPI), and an initial empirical evaluation of some psychometric properties of Jackson's instrument. Edwards and Pbbott ( 1973) reported that the Minnesota M ul tiphasic Personality Inventory (MMPI) has been, and continues to be the most frequently used personality measure. Since its publication in 1943, it has been involved in over 4,000 studies. Today, the MMPI is primarily used to aid in psychiatric diagnosis and treatment planning; but is also widely used to determine job suitability and admission to educational programs (Harvey and Sipprelle, 1976). Despite its popularity, however, most reviewers concede that the MMPI suffers from a number of serious psychometric problems.
The most frequently voiced criticisms concern its poor reliability and excessive redundancy (Anastasi, 1976;Butcher, 1969;Cronbach, 1970;Lanyon and Goodstein, 1971). Estimated reliability coefficients from both normal and deviant samples have been generally poor. In one test-retest study, Gilliland and Golgin ( 1951) (Blanton and Landsman, 1952;Cottle, 1950;McQuary and Truax, 1952 (Block, 1965;Kassenbaum, Couch, and Slater, 1959;Messick and Jackson, 1961;Welsh, 1956). This is not surprising, h::Mever, considering that many of the lvMPI scales are highly correlated (Dahlstran, Welsh, and Dahlstran, 1972). One reason for this is that many of the pathological items are very similar (Block, 1965 Another area of concern involves the MMPI's sensitivity to response distortion; specifically, social desirability bias, acquiescence bias, and faking. At one point, Edwards (1964) and Messick and Jackson (1961) argued that the two principal factors underlying the MMPI scales reflected nothing more than social desirability and acquiescence response bias. The MMPI did not, in their opinion, measure psychopathology. In fact, a number of studies have reported a high significant correlation between desirability ratings of MMPI items and the primary factor underlying the scales (Edwards, 1967;Jackson and Messick, 1961, 1962a, 1962b (Jackson, 1973). In any case, many of the MMPI items are confounded by their desirability value.
Although it is probably impossible to eliminate desirability influences in inventories like the MMPI, confounding can be effectively minimized by careful item selection procedures. Jackson and Lay ( 1968) , for example, demonstrated that when test i terns were originally selected with a low level of association with desirability responding, it was possible to distinquish content factors from a desirability factor and thus measure each separately. This procedure has been described in detail by Jackson ( 1970) and used successfully in the construction of three recently published and promising personality instruments: the Personality Research Form (Jackson, 1967); the Differential Personality Inventory (Jackson, 1972); and the Jackson Personality Scales (1976). Although the influence of acquiescence is not as important as originally claimed, it can have a significant effect on scales with a disporportionate number of i terns keyed in a single direction ( Campbell et al. , 1967). Since many of the MMPI scales have this characteristic, specifically, the Hysteria, Psychopathic Deviate, Paranoia, and Hypomania scales (Campbell et al., 1967), acquiescence remains a problem. Messick and Jackson (1961) recommend balancing scales for true and false keying to restrick the influence of acquiescence. This strategy has since been adopted by most test developers.
Considerable research effort has been spent investigating the -6 problem of dissimulation on personality scales and inventories. The ability of subjects to consciously fake responses on the MMPI is well --established.
Considering that the MMPI appears to be a "psychometric monstrosity", why then its continued use? Its popularity indicates that there is an apparent need for such an instrunent. Social Deviation High Scorer Express attitudes markedly different fran cx:mnon social codes; is prone to depart fran the truth and behave in an unethical and 1.mtrustworthy manner; feels little or oo guilt. r..a.,, Scorer Ordinarily displays ethical and socially resJ:X)l'lSible attitudes and behavior; reports a sense of obligaticn toward society and its laws.
Persecutory Ideas High Scorer Believes that aertain people are against him and are trying to make his life difficult and 1.mpleasant. Inclined to brood. r..a.,, Sa::>rer Trusts others and <besn't feel threatened. ~epts resfX)l'lsibility for the events in his life and doesn't attribute maliciousness to others Anxiety High Scorer Easily scared.
Little things, even an idea, can throw him into a frenzy of anxiety.
Afraid of novelty and of the possibility of ftlysical or interpersonal danger. r..a.,, Scorer Remains caJm and ll'lruffled even when oonfronted by 1.mexpected occurrences. Maintains self control even in a crisis situation.
Thinking Disorder High Scorer Is markedly oonfused, distractable and disorganized. Cannot remember even simple things fran day to day. Reports that he feels he is living in a dream-like \\Orld, that people appear different to him and that he feels different fran them. IDN Sa::>rer AJ;:pears to be even-tempered and level-headed; carefully oonsiders the future before acting; generally has the patienae to oope with a lengthy and tedious task. inter-scale correlations.

Reliability of the BPI Scales
The       The group centroids representing the mean discriminant scores for each group on the two functions or dimensions are reported in Table 9, and plotted in Figure 3. The centroids summarize the group locations in the two dimensional space defined by the discriminant functions.
The first function serves to distinguish the "fake maladjustment" group from the other two, while the second function primarily differentiates the "fake good adjustment" group from the others.
Further evidence relating to group differences is evident in the plot of cases presented in Appendix D. The asterisks represent the group centroids and the numbers represent cases corresponding to groups 1 (standard instructions), 2 (fake good adjustment), and 3 (fake makadjustment).
The relative contributions of the three discriminatng variables to the two significant functions is presented in Table 10. The first dimension is most highly weighted with the Deviation scale, followed by the Self Depreciation scale. Both scales are weighted in a negative direction, indicating that high scores on Deviation and Self Depreciation result in low scores on Dimension I, and vice versa. The "fake maladjustment" group scored in the negative direction on this demension while the standard instructions group and the "fake good adjustment" group scored in the positive direction ( Figure 3).
The second dimension is highly weighted, in a positive directon, with only the Denial scale. Those who score high on the Denial scale will tend to also score high on Dimension II, while those who score low on Denial will also tend to score low on the second dimension. The  fake good adjustment" group scored high on this dimension while the "fake maladjustment" and standard instructions groups scored low.
Finally, the prediction results are presented in Table 11. Using the two significant discriminant functions, it was possible to correctly classify 78.17% of the cases into their respective group.
It was easier to predict ( or detect) a person's group membership if they faked maladjustment ( 88%) rather than good adjustment ( 70%), or did not fake at all (77.4%).
In summary, it appears that three scales -the Deviation scale, the Self Depreciation scale, and the Denial scale, can successfully account for most of the group differences due to faking.

Discriminant yalidity of the BPI Scales
Correlations between scales were computed from the URI test 1, URI test 2, and CCR! standard instructions data and are presented in Tables 12 and 13. with respect to the prediction of near zero inter-scale correlations the results are mixed.
Correlations between scales were generally consistent across URI test administrations ( Table 12). These correlations are in the low positive to moderate positive range with the exception of scale 3.
Correlations for this scale, although in the same range, are consistently negative in direction. Thus, with respect to the URI data the prediction of near zero correlations between scales was not supported.    There are a number of possible explanations for these results.
First, since the greatest amount of error seems to be associated with the internal consistency estimates, it may be that many of the scales are factorally complex. If this is the case, the KR-20 coefficients would be expected to be low, but the stability coefficients would be relatively unaffected (Edwards, 1970 reveals that the former resembles the F scale of the MMPI (Hathaway and McKinley, 1943) while the latter resembles the Lie and K scale of the MMPI and the Lie scale found in the Eysenck Personality Questionnaire (Eysenck and Eysenck, 1975 Table 15. These coefficients are based on data collected from 97 college students (Gilliland and Colgin, 1951  From Gilliland and Colgin, 1951 44 coefficients are also generally higher than the stability coefficients reported for the M-fl?I in the Gilliland and Colgin (1951)  Our results indicate that it is easier for "oormals" to fake in a mala:::ijusted directicn oo the BPI. The differences found between the "fake maladjustment" group and the standard instructions group in terms of profile elevaticn were much greater than differences found between the "fake gcod adjustment" group and the "standard instroctions" group (Table 6 and Figure 2). The discriminant analysis results supported this ronclusion. Prediction results presented in Table 11 irrlicate that it was generally easier to predict a person's group membership if they faked maladjustment (88%) rather than gcod a:::ijustment (70%) or did not fake at all (77 .4%). Furthermore, the plot of indvidual cases (Appendix D) indicates nore overlap between 45 the standard instructions group and the "fake good adjustment" group, than between the "fake maladjustment" group and the others.
The fact that faking maladjustment produced greater change in profile elevation is related in part to the rather low mean scores on the BPI produced by normals in general ( Table 2). The "fake good adjustment" group could lower their scores only minimumly on most scales while the "fake maladjustment" group had considerable room to vary in the elevated or maladjusted direction. This probably also best explains why two scales, the Persecutory Ideas scale and the Self Depreciation scale, could not be faked in the more adjusted direction. Questionnaire ( Eysenck and Eysenck, 1975). Both the L scale and the Lie scale were designed to detect faking in a desirable direction.
Thus, we would expect the "fake good adjustment" group to score high on the BPI' s Denial Scale. Generally, these people are denying that they have even minor faults, or experience any negative emotions.
What is surprising, however, is that the "fake maladjustment" group also Profiles with high Deviation, Self Depreciation, and Denial scores suggest faking in a pathological direction. These persons freely admit to unusual symptoms and modes of thought, tend to be self depreciating, and report being relatively unresponsive emotionally.
Profiles presenting suppressed Deviation and Self Depreciation scores along with an elevated Denial score suggest faking in a desirable, more adjusted direction. Persons presenting this type of profile report being free from symptoms and unusual modes of thought, high in self-confidence, admit to few if any faults, and report experiencing few if any negative emotions.

Discriminant ~alidity of the BPI Scales
Although we were able to provide some initial evidence regarding the discriminant validity of the BPI scales, our results were mixed.
With respect to the CCRI data ( Our results may be due to the nature of the populations studied.
In any case, the discriminant validity of the BPI scales needs to be assessed further, especially in the context of a clinical population.

Conclusion
The reliability of the BPI appears to be generally superior to that of the MMPI. This involves scale reliabilities based on internal consistency and stability estimates. _Further, we have reason to believe that these coefficients would be even higher in clinical populations.
The BPI scales can be easily faked, both in a "more adjusted" and maladjusted" direction.
Our results further indicate that three scales -the Deviation scale, the Self Depreciation scale, and the Denial scale can be effectively used as "fake detectors", or validity scales. Certain profiles based on these scales were shown to clearly indicate faking on the part of the respondent.
Finally, the results concerning the discriminant validity of the BPI scales were mixed. This warrants concern generally but especially since the observed scale intercorrelations, being based on normal population data, were most likely conservative in nature.
Further research needs to be conducted examining the reliability, validity, and effects of dissimulation on the BPI scales when other relevant populations are studied, particularly clinical populations.
Sex differences, age, social class, and racial differences also need to be ascertained. We suggest the first step would be to conduct a