New Subscales for the MMPI-2 Social Introversion ( Si ) Scale

The development of three new content-homogeneous subscales for the revised Minnesota Multiphasic Personality Inventory (MMPI-2) Social Introversion ( Si ) scale designed to replace the Serkownek (1975) subscales, which were not included in the revised MMPI, is described. The subscales, termed Shyness/Self-Consciousness, Social Avoidance, and Self/Other Alienation, were developed with data provided by college students (515 men and 797 women). Data analyses with this and the MMPI-2 normative sample demonstrated that the new subscales independently contribute to the assessment of nearly 90% of the variance in the full Si scale, that they display both convergent and divergent validity, and that these attributes generalize beyond the sample with which they were developed.

With the introduction of the revised Minnesota Multiphasic Personality Inventory (MMPI-2; Butcher, Dahlstrom, Graham, Tellegen, & Kaemmer, 1989 ), a number of changes have occurred in the instrument. Among these is the deletion of the Serkownek (1975) subscales for clinical scales Mf (Masculinity-Femininity) and Si (Social Introversion). After a careful examination of the psychometric foundation and clinical utility of these scales, the MMPI restandardization committee determined that they did not perform adequately in either of these areas. ¹ Although the Serkownek scales were developed following item-level factor analyses of the Si scale ( Graham, Schroeder, & Lilly, 1971 ), they did not possess acceptable levels of internal consistency as would be expected of factorially based scales. Similarly, examination of the content of these scales showed instances of loosely related items within subscales and several cases where no apparent relation could be identified between item content and scale name.

As part of the MMPI restandardization, a number of items that were deemed by the restandardization committee to contain objectionable (e.g., religious) content were deleted from the booklet. The Si scale lost only one item in this process, leaving a total of 69 items. Of these, five items were slightly reworded to correct for sexist or otherwise inappropriate or outmoded language. Ben-Porath and Butcher (in press) demonstrated that item rewording had no appreciable effect on the psychometric characteristics of the MMPI-2 items.

We conducted the present study in an attempt to develop new Si subscales to replace the Serkownek subscales. As noted by Graham (1987) , when properly conducted, subscale interpretation can yield highly useful supplementary information when interpreting scores on a full scale. Subscales are most useful in attempting to "dissect" a full scale elevation and identify which of the components of that scale contributed most to the elevated score. Such information may be particularly useful in cases where the clinical history of the individual is not consistent with an elevation found on the full scale or where elevation is only marginal ( Graham, 1987 ). The current study was undertaken to provide subscales to aid in the interpretation of scores on the Si scale. ²

The data used for the initial development of the Si subscales were the responses of 515 college men and 797 college women who participated in a study conducted by Butcher, Graham, Dahlstrom, and Bowman (in press) . The mean age of the men was 19.77 years ( SD = 2.4, range = 17—37), and the mean age of the women was 19.81 years ( SD = 3.42, range = 17—48). The ethnic composition of the college sample was as follows: 84.7% White, 7.5% Black, 2.8% Native American, 2.7% Asian, and 2.1% of other ethnic backgrounds.

A second sample, used for the development of T scores for the subscales and to determine whether the psychometric characteristics of the scales generalize beyond the sample with which they were developed, consisted of the 1,138 men and l,462 women who made up the MMPI-2 normative sample ( Butcher et al., 1989 ). The mean age of the normative-sample men was 41.71 years ( SD = 15.32, range= 18—84), and the mean age of the normative-sample women was 40.52 years ( SD = 15.24, range = 17—85). As reported in the MMPI-2 manual, the ethnic composition of the normative sample closely matched that of the 1980 United States Census.

Our decision to use a college sample in the development of the Si subscales was guided by several considerations. First, the original Si scale, which is widely used in clinical applications, was developed with a college sample ( Drake, 1946 ). In addition, Butcher et al. (in press) demonstrated that the college sample used in the current study was psychometrically similar to the MMPI-2 normative sample. We chose not to use the normative sample itself in subscale development so as not to confound the developmental and normative samples for these scales.

All 3,912 subjects completed the experimental Form AX of the MMPI that was used in the revision and development of the MMPI-2. The experimental form consisted of 704 items including all 550 of the original MMPI items (82 were slightly reworded). A subsample of 1,644 of the subjects in the normative sample were tested as couples, and in addition to the MMPI Form AX they completed a behavioral rating form on which they rated the degree to which their spouses exhibited 1 10 different behaviors on a scale of 1 ( almost never ) to 4 ( almost always ). The behavioral ratings provided by these 822 couples were used to assess the validity of the new subscales in the current study. Further details concerning the couples can be found in the MMPI-2 manual ( Butcher et al., 1989 ).

Next, we conducted item analyses for five provisional Si subscales that were constructed by deleting from each factor scale any item loading .20 or higher on an additional factor and/or not loading at least .10 on the remaining four factors in both samples. The item analyses consisted of computing corrected (by deleting the contribution of the item to the total scale score) item—scale correlations and Cronbach's (1951) alpha coefficient of internal consistency for each provisional subscale for each gender. These analyses showed that two of the provisional subscales were extremely low in internal consistency (a < .50). They were dropped from further analyses, and the remaining three subscales were chosen to serve as a core for scale development and refinement. Any item whose corrected correlation with the total scale score in either gender was less than .15, or that was attenuating the scales' internal consistencies, was deleted from the provisional subscales. The total number of items in the provisional subscales at this point was 33. To determine whether there were any additional meaningful dimensions underlying the remaining 36 Si items, we factor analyzed those items separately by gender following the same procedures as just described for the entire set of items. One of the factors identified by these analyses appeared to be promising and served as a basis for developing a fourth provisional subscale containing six additional items.

The next step involved computing item-scale correlations between the remaining 30 Si items and the four provisional scales to identify any of the remaining items that might contribute to the assessment of a dimension measured by one of the provisional subscales. In order to be added to a provisional scale an item had to correlate .20 or higher with that subscale and less than .10 with the three remaining subscales. We conducted these analyses for each gender separately, and the criterion for inclusion had to be met in both men and women. No additional items were identified for the fourth provisional scale, which led to a decision to drop it from further consideration because it did not appear to have enough items. Its items were then added to the other 30 Si items not appearing on any of the provisional subscales, and the analyses just described were redone. This led to the identification of six additional items that were added to the three provisional subscales, resulting in a final list consisting of 39 items on the three Si subscales.

MMPI-2 booklet item numbers and scoring directions for the three subscales are presented in Table 1 . On the basis of an examination of their content following the final analysis, we named the subscales Shyness/Self-Consciousness ( Si 1 ), Social Avoidance ( Si 2), and Self/Other Alienation ( Si 3). Norms for converting raw scores on the three scales to linear T scores are based on the MMPI-2 normative sample and are presented in Table 2 . As is the case for the full MMPI-2 Si scale, the T scores are linear transformations of the raw scores into standard scores, with M = 50 and SD = 10. The MMPI restandardization committee is recommending that elevations on T scores based on the new norms be interpreted when the score exceeds 64 ( Butcher et al., 1989 ).

Table 3 presents the intercorrelations among the subscales as well as each subscale's correlation with the full Si scale for men and women separately in both the college and normative samples. Examination of the intercorrelations shows that the goal of developing relatively independent subscales was achieved. The similarity between results for the college and normative samples demonstrates that the independence among the subscales is not specific to the sample whose data were used to develop the scales.

To find out how much of the variance in the full Si scale could be accounted for by the three subscales in combination, we computed multiple correlations between the three subscales and the full Si scale. These were .93 for the college men and the college and normative-sample women and .92 for the normative-sample men. The adjusted (for shrinkage) squared multiple, correlations were .87 and .86, respectively, indicating that the subscales can account for nearly 90% of the variance in the full scale and that this finding generalizes beyond the sample used for scale development.

The reliability of the new subscales was assessed by examining their internal consistencies and test-retest reliabilities. The three scales' internal consistency was examined by computing Cronbach's (1951) coefficient alpha for men and women in the college and normative samples. For the college men, the coefficients for Si 1, Si 2, and Si 3 were .82, .77, and .77, respectively. For the college women they were .82, .75, and .77, respectively. There was no appreciable shrinkage when alpha was computed for the normative sample. For the men the three coefficients were .81, .77, and .75, and for the women they were .84, .75, and .78, respectively. All of the subscales' coefficients of internal consistency were within the range found for full scales in many measures of personality. Comparison of the coefficients found in the two samples demonstrates that the subscales' internal consistency generalizes beyond the developmental sample.

The test-retest reliability of the subscales was evaluated in a subsample of 82 men and 11 women from the normative sample who took the experimental form of the MMPI twice within 1 to 2 weeks. The test-retest correlations for Si 1, Si 2, and Si 3 were .91, .88, and .77, respectively, for the men and .90, .87, and .88 for the women. These coefficients demonstrate an impressive degree of reliability for all three subscales, although the coefficient for men on the Self/Other Alienation subscale is somewhat lower than the rest.

The validity of the new subscales was assessed by examining their correlations with behavioral ratings provided by 822 couples from the normative sample. Spouses rated each other on 110 variables, some of which were related to the content of the new subscales. Table 4 presents these correlations for any of the 110 items that correlated .18 (absolute value) or higher with at least one scale. ³ Examination of the correlations in Table 4 demonstrates the scales' convergent and divergent validity. In evaluating the correlations, one needs to consider that none of these items was written specifically to assess any of the characteristics associated with the new subscales and that these are essentially correlations between single items and scales. They are of the same magnitude as item—scale correlations found in many measures of personality (including the item-scale correlations found for the Si subscales themselves). The higher number of correlations found for the women may reflect a greater association with the general psychopathology dimension of the MMPI in the Self/Other Alienation subscale for women.

The present results present good evidence for the independence and reliability, and initial support for the validity, of the new Si subscales for the MMPI-2. The independence of the subscales will enable users to "dissect" the Si scale in cases where it is elevated and to identify which of the three facets assessed by the subscales accounts for elevation in the full scale. As emphasized by Graham (1987) , subscale interpretation should be conducted only in cases where the full scale is elevated and as a supplement to the interpretation of the score on the full scale.

On the basis of their content and their external correlates ( Table 4 ), the Si subscales are related to the following characteristics: Shyness ( Si 1)–feeling shy around others, feeling easily embarrassed, feeling ill at ease in social situations, and feeling uncomfortable in new situations; Social Avoidance ( Si 2)—great dislike of group activities, concerns about group participation, active efforts to avoid being in a crowd, dislike for parties and social events, and avoidance of contact with other people; Self/Other Alienation ( Si 3)–low self-esteem, lack of self-confidence, a tendency to be self-critical, questioning of one's own judgment, and feeling ineffective at determining one's fate. In women, high scores may also reflect nervousness, fearfulness, indecisiveness, suspicion of others, viewing others as malevolent, and suspicion of being talked about by others. These external correlates need to be replicated and expanded. Interpretation at this point should be based primarily on subscale content.

By examining overlap between the new scales and previously developed MMPI scales, we may gain further understanding of what it is that they measure. The Shyness / Self-Consciousness subscale shares a great deal of its items with various MMPI scales of introversion-extraversion including Giedt and Downing's (1961) Extraversion scale, Gocka and Holloway's (1962) Introversion-Extraversion scale, Stein's (1968) Social-Introversion scale, and Wiggins's (1969) Social Maladjustment scale. It also shares several items with Comrey's (1957) Shyness scale. Among the Serkownek Si subscales, it greatly overlaps with the Inferiority-Personal Discomfort scale. The Social Avoidance subscale also shares many items with previous measures of introversion-extraversion (e.g., Giedt & Downing, 1961 ; Gocka & Holloway, 1962 , Wiggins, 1969 ). Among the Serkownek subscales, it shows the greatest overlap with the Discomfort with Others scale. The Self/Other Alienation subscale shares little or no items with the introversion-extraversion scales just mentioned, shares some items with the Wiggins (1969) Poor Morale scale, and shares a substantial number of items with the Serkownek (1975) Distrust scale.

The data presented in this article suggest that the new Si subscales can be quite useful to clinicians as aids in interpreting elevations in Si . Although they include only slightly more than half of the full scale's items, they account for nearly all of its variance. Further research into the external correlates of these subscales will be necessary to provide a more comprehensive picture of their ability to assess these three facets of Social Introversion as measured by the MMPI-2 and will furnish clinicians with additional interpretive information concerning individuals who produce elevated scores on these subscales.

1

The MMPI restandardization committee was made up of James N. Butcher, W. Grant Dahlstrom, John R. Graham, and Auke Tellegen. Its work began in 1982.

A failed attempt to develop new subscales for Mf is described elsewhere ( Hostetler, Ben-Porath, Butcher, & Graham, 1989 ).

Table 1. Compositionxs of Subscales

Table 2. T-Score Conversion Tables for Social Introversion (Si) Subscales

Table 3. Correlations With Full Social Introversion (Si) Scale and Intercorrelations Among Si Subscales

Table 4. Social Introversion (Si) Subscales' Correlations With Spouses' Ratings