Person. individ. D#
Vol. 6. NO.
1,pp.11-82.
1985
019lL8869/85
Printed in Great Britain. All rights reserved
%3.00+0.00
Copyright0 1985Pergamon Press Ltd
MULTIDIMENSIONAL SCALING OF RORSCHACH INKBLOTS: RELATIONSHIPS WITH STRUCTURED SELF-REPORT KAREN OSTERHOLM,’ DONALD J.WOODS'
and ARNOLD
LE UNES~
*Kilgore College, Kilgore, TX 75662, U.S.A. ‘Department
of Psychology,
University
of Delaware,
‘Texas A & M University,
220 Wolf Hall, Newark,
College Station,
DE 19711, U.S.A.
TX 77843, U.S.A.
(Receioed 4 June 1984)
Summary-This
study
attempted
a replication
of previous
niultidimensional
scaling
(MDS) results purpose was the investigation of relationships between individual Ss’ relative reliance on these salient features, and personality characteristics as assessed by the Jackson Personality Inventory (JPI). The two-dimension solution obtained in previous research, involving inkblot color and twoness of form, was closely replicated. Only very low correlations between Ss’ use of these dimensions, and personality characteristics as measured by the JPI, were obtained. The MDS-Rorschach approach may prove useful in delimiting peculiarities of attention and perception in clinical states. The usefulness of the approach in furthering study of personality constructs assessed through traditional Rorschach administration is currently in question.
concerning the most salient perceptual features of the Rorschach inkblots. A second
INTRODUCTION More than 60 years after its introduction, the Rorschach Inkblot Test is still considered one of the most controversial assessment procedures in wide usage, attracting both avid champions and harsh critics (Anatasi, 1983; Howes, 1981). One of the assumptions of various Rorschach scoring systems is that persistent use of certain perceptual features of the cards (e.g. form, color, shading) in forming responses is indicative of Ss’ relatively stable personality traits and dispositions. The purpose of the present study was to explore a nontraditional approach to investigating personality correlates of the salience to Ss of perceptual features of the inkblots. In recent years the exploration of perceptual characteristics of complex stimuli has been facilitated through the development of various multidimensional scaling (MDS) techniques (Nunnally, 1978; Torgerson, 1952). Based on the same mathematical models as factor analysis and discriminatory analysis, MDS involves judgment of stimuli in terms of similarities and differences among these stimuli. Typically, Ss are asked to rate the degree of similarity of pairs of stimuli, or are asked whether stimulus A is more similar to stimulus B or to stimulus C. Computation of the average ‘distances’ thus obtained between all possible pairs of stimuli allows one to estimate the location of each stimulus in a multidimensional ‘psychological space’. Analysis of multidimensional solutions with regard to fit considerations and predictive ability allows one to determine the number of dimensions making significant contribution. The schematic representation of interstimulus distances derived from the MDS procedure becomes psychologically meaningful under the assumption that salient dimensions serve an organizational role in the perception and processing of stimuli (Becker, 1980). In addition to allowing specification of salient dimensions of complex stimuli MDS procedures permit the investigation of the differential salience of stimulus features for individual Ss (Carroll and Chang, 1970; Tucker and Messick, 1963). This possibility makes MDS a potentially useful tool for studying personality correlates of individual differences in salience of Rorschach inkblot features. A successful application of one such individual differences MDS model to Rorschach inkblots is reported by Wainer, Hurt and Aiken (1976). All possible pairs of Rorschach cards were scaled for perceived similarity by a college student control group, a psychiatric group labeled ‘depressive’, and another psychiatric group described as ‘schizophrenic’. The similarity matrices thus obtained were used as input to INDSCAL (Chang and Carroll, 1968), a statistical procedure which decomposes the data into one matrix whose entries are the coordinates of each stimulus in a 77
78
KAKLN OS
rl:KHOLM
<‘I 01.
multidimensional space, and a second matrix whose entries are the weights each S used on the dimensions. Using the INDSCAL procedure, Wainer et al. (1976) isolated two dimensions: color vs black-andwhite, and twoness of form. The first dimension involved the amount of color on the cards, ranging from black figures on a white background to black-and-red figures on white background, to multicolored blots on white. Twoness of form involved whether the inkblot was perceived as a single entity or two separate but associated entities. Analysis of differential salience of dimensions for individual Ss indicated complete separation of the control and psychiatric samples, with members of the control group utilizing the isolated features to a greater degree. Within the psychiatric population, Wainer et al. found nearly complete separation of ‘depressive’ and ‘schizophrenic’ groups. Wainer et al.‘s two-dimension solution for the inkblots was replicated by Vestewig and Paradise (1977), in a study in which similarity judgments of pairs of selected Thematic Apperception (TAT) cards were also subjected to MDS. Undergraduates who were rated as high and low on need for achievement on the basis of standard scoring procedures for TAT stories differed significantly in their use of two TAT dimensions in making similarity judgments. Although MDS of Rorschach similarity judgments yielded the two dimensions described by Wainer et al. (color and twoness), individual weights for these dimensions did not correlate with the need for achievement ratings, The present study attempted a replication of Wainer et al.‘s (1976) and Vestewig and Paradise’s (1977) results with regard to the dominant perceptual characteristics of the Rorschach stimuli. Its second purpose was to begin a program of study exploring relationships between individual differences in the salience of perceptual features of the inkblots, and personality characteristics as measured by objective self-report. In this study, Ss completed the Jackson Personality Inventory (JPI; Jackson, 1976), a relatively recent rationally-derived, construct-oriented self-report measure, in addition to completing similarity ratings of pairs of Rorschach stimuli. METHOD Subjects Two hundred and nine Ss (100 females and 109 males) were drawn from the enrollment of Introductory Psychology classes. Participation in the study constituted partial fulfillment of course requirements. Materials Slides of stimuli. Two sets of photographic slides were prepared. The first consisted of reproductions of each individual Rorschach card. Next, photographic slides of all possible pairs of Rorschach inkblots were prepared. Stimuli were ordered in this set so as to allow an equal number of presentations of each inkblot on the right and left sides of a pai;, as well as to insure maximum separation across pairs of repetitions of any single card (Ross, 1934). Similarity rating sheets. Rating sheets were prepared, each consisting of 45 15-point scales, one corresponding to each of the 45 unique combinations of the 10 Rorschach cards. The endpoint of the scale having a value of 1 was labeled ‘completely different’, while the other endpoint, having a value of 15, was labeled ‘virtually the same’. Jackson Personality Inventory (JPI). The JPI (Jackson, 1976) is composed of 15 substantive scales: Anxiety, Breadth of Interest, Complexity, Conformity, Energy Level, Innovation, Interpersonal Affect, Organization, Responsibility, Risk Taking, Self Esteem, Social Adroitness, Social Participation, Tolerance and Value Orthodoxy. A sixteenth scale, a validity scale labeled Infrequency, is also included. Each of the 16 scales involves 20 items. Internal consistency reliability estimates for the JPI scales ranged from 0.84 to 0.95 with a median of 0.93 for 82 California college students, and from 0.75 to 0.93 with a median of 0.90 for 307 Pennsylvania college students (Jackson, 1976). Normative data, enabling T-score transformations of raw scores, were based on the results of administration to 2000 male and 2000 female students attending 43 1976). Evidence for the convergent and North American colleges and universities (Jackson, discriminant construct validity of the JPI subscales, including correlations between the JPI and peer ratings and selected psychological tests, is provided in the JPI manual.
Procedure
Subjects were tested in groups of from 20 to 25 individuals during nine testing sessions. The experimental procedure followed during the paired similarities portion of the study was patterned after that of Wainer et al. (1976). Initially, Ss were shown slides of the individual Rorschach cards in order to familiarize them with the inkblots. Following this presentation, participants were informed that they would be asked to rate the degree of similarity of pairs of the inkblots on a 15point scale. No guidelines were offered by which to judge degree of similarity. Rather, Ss were advised to use any criteria desired, but to be consistent in their use of the chosen criteria. Slides of the individual cards were then presented once again. Subsequently, slides of all possible pairs of inkblots were exposed for 6 set/slide, and Ss were asked to mark the point on the appropriately numbered scale which they felt was representative of the degree of similarity of the pair. All slides of pairs of inkblots were presented a second time in order that Ss might have the opportunity to review their scaling and make desired changes. After completion of similarity ratings for pairs of Rorschach inkblots, instructions for the completion of the JPI were given.
RESULTS MDS I$ similarity ratings
Data obtained from similarity ratings of pairs of Rorschach cards consisted of 209 lower-half matrices (without diagonals), each representing a different data source or S. Cell entries in the similarity matrices were the values corresponding to points indicated on the 15-point rating scales used to indicate Ss’ perceptions of degree of similarity. The 209 similarity matrices were used as input to SINDSCAL, a computer program utilizing the individual differences model for MDS developed by Carroll and Chang (1970). SINDSCAL is an adaptation of the INDSCAL procedure used by Wainer et al. (1976), requiring that input matrices be symmetric. The methods of analysis employed in INDSCAL and SINDSCAL are identical. Differences between the two programs involve computational procedures, which are simplified in SINDSCAL by the restriction to use with symmetric data, and user options (Pruzansky, 1975). The model developed by Carroll and Chang upon which the SINDSCAL program is based assumes a set of some unknown number of dimensions or factors underlying the set of stimuli being studied. These dimensions are assumed to be common to all individuals, but it is expected that different dimensions will manifest differential saliences across Ss. Each stimulus is conceptualized as occupying a unique position in a multidimensional ‘psychological space’ involving all stimuli in the set. Examination of relative positions of stimuli along individual axes hopefully gives an indication of the nature of the underlying dimensions. A second matrix consisting of weights of each S on each dimension is also generated. Thus, it is possible to determine the interspatial relationships among the stimuli as indicated by an ‘averaging’ of the similarity ratings for each pair across Ss, as well as the amount of emphasis each rater placed on the different dimensions in making his/her similarity judgments. In addition to these matrices, the SINDSCAL program output provides several values which may be used as aids in the judgment of the goodness-of-fit of various solutions to the data. As discussed previously, Wainer et al. (1976) reported isolation of a meaningful two-dimensional solution. Therefore, the initial analytical task involved a determination of whether two salient dimensions emerged from the data currently under investigation. SINDSCAL computes correlations between predicted and observed values of each individual S for each separate solution requested by the researcher. These correlations provide an indication of the goodness-of-fit of the weighted solution for each S. Wainer et al. compared those correlations occupying the median position in the lists for solutions of different dimensionalities. They reported that the median correlation of 0.59 for the two-dimensional solution was a significant improvement over the median correlation for the one-dimensional solution. The three-dimensional solution yielded a median correlation of 0.66, but Wainer et al. stated that this solution was much less easily interpreted than the twodimensional one. The present investigators requested solutions from SINDSCAL involving five, four, three, two and one dimensions in order to insure a sufficient range from which to make judgments regarding
KAKEN
x0
OSTEKHOLM
L’I trl
asymptotic trends. Median correlations between predicted values and observed ratings were examined in order to ascertain whether dramatic differences were evidenced between any two adjacent solutions. The median correlations for all five solutions are presented in Table 1. For purposes of comparison, the correlations reported by Wainer et al. (1976) are included as well. Regarding each of the data sets, the greatest difference between median correlations was seen in comparing the one- and two-dimensional solutions. Comparisons between other adjacent pairs indicated an asymptotic trend toward no difference.
Table
I. Median
correlations
between
predicted
and observed
values for
solutions
of differing
dimensionalities Median
correlations
Median correlatmns (Wainer cr al.)
No. of dimensions
(Present study)
I
2 3 4
0.43 0.61 0.68 0.74
Not reported
5
0.76
Not
0.32 0.59 0.66 reported
An alternative procedure for evaluating goodness-of-fit, achieved with solutions of varying numbers of dimensions, involves examination of variance accounted for by each solution. In the present data, the total amount of variance in original judgments accounted for by the fivedimensional solution was 0.59. An 8 % drop in variance accounted for occurred when the fifth dimension was deleted from the solution (0.54 for four-dimensional solution). A 12% decline (to 0.48) was reported when the solution was reduced to three dimensions. Removal of the third dimension resulted in a 16% decrease (to 0.40). When comparing the two- and one-dimensional solutions, however, an impressive 41 o/0 decrease in total variance accounted for was seen with the deletion of the second dimension (from 0.40 to 0.24). Thus, it appeared that the second dimension made a substantial contribution to accounting for variance, while the inclusion of additional dimensions resulted in less dramatic increases. As will be seen, examination of the two-dimensional solution indicated ordering along axes with rather obvious relationships to attributes of the Rorschach inkblots. There appeared to be no easily discernible pattern to the order of the stimulus cards along the axis representing the third dimension, however. Thus, it was determined that the two-dimensional solution was the most complex solution affording a meaningful interpretation of the data. Before concluding replication of the two-dimensional solution proposed by Wainer et al. (1976), it was necessary to determine whether the two dimensions isolated in the present investigation were in fact related to the colorfulness of the cards and twoness of form, the dimensions Wainer et al. recovered from their data. Using the coordinates of each stimulus in multidimensional space provided as part of the SINDSCAL solution, the two-dimensional blot configuration of Fig. 1 was plotted. Visual inspection of the ordering of stimuli along each of the two axes suggested that, indeed, the dimensions reported by Wainer et a/. had been recovered from the data. Along Dimension I, the x-axis, the most colorful cards (8, 9 and 10) are located at one end of the dimension. The all-black cards (1,4, 5,6 and 7) are grouped at the opposite end of the dimension. Card 2, a red-and-black card, occupies a central position along the axis. Only Card 3, also a redand-black card, appears slightly out of position with regard to clear-cut ordering of the cards based on colorfulness. Along Dimension II, the y-axis, cards again appear to be ordered appropriately, with those cards (2,3 and 7) conveying the impression of two separate but associated figures at one end of the dimension, and those perceived as a single entity (4,5 and 6) at the other end of the scale. The remaining cards (1, 8, 9 and 10) represent inkblots that are more ambiguous with regard to involvement of a single entity or two conjoint figures. Personality
correlates
of dimension weights
Stepwise multiple-regression individual dimension weights,
analyses were conducted to explore the relationships between Ss’ and their JPI scale scores. TWO sets of analyses were performed in
Multidimensional
SCdillg
Of
Korschach Inkblots
6 -02u, I5 2 Q
-04
-
-06
I
-08
1
-06
Fig. 1. Two-dimensional
-04
-02 DIMENSION
blot configuration
0
I
I
I
02 I-COLOR
04
06
derived from Ss’ similarity judgments
which each of the two dimension weights served as the criterion and the 15 transformed T-scores from the JPI served as predictor variables. Two female and 3 male Ss were deleted, based on Infrequency scale elevations > 7OT, leaving a total of 98 female and 106 male Ss in the analysis. In each analysis set, multiple regression was performed for both male and female Ss combined, and then for males and females separately. In the analysis for males and females combined, none of the JPI scales met the minimum specified degree of association with color (P < 0.05) and a significant R2 did not result. When male Ss’ data were analyzed, significant relationships between JPI scales and color dimension weights were again not found. However, when female Ss’ data were analyzed separately a significant R2 of 0.046 was obtained [F(l, 96) = 4.61, P < 0.041. Only the subscale Social Participation contributed to this equation, with a zero-order correlation with the color dimension for females of 0.21. For the dimension twoness of form, stepwise regression of combined male and female data yielded a significant R2 = 0.027 [F(l, 202) = 5.72, P = 0.021. Only the Organization subscale contributed to this equation, with a zero-order correlation with twoness of 0.166. When male and female Ss’ data were analyzed separately, the minimum specified degree of association between individual scores and twoness (P= 0.05) was not reached, and significant R2sdid not result.
DISCUSSION The present study replicated closely the two-dimensional solution for Rorschach similarity judgments obtained by Wainer er al. (1976) and Vestewig and Paradise (1977). There would seem to be consensus among researchers that dimensions most meaningfully labeled color and twoness of form are the most utilized perceptual characteristics when groups of unselected college students are asked to render similarity judgments of inkblot pairs. What is left in considerable doubt by our research is whether degree of reliance on these characteristics by individual Ss may be related to personality dispositions, at least as assessed by structured self-report inventories. The present investigation found low, statistically significant correlations between Social Participation and use of color, for females only, and between Organization and use of twoness, for the combined male and female sample. Though statistically significant, these results represent two out of a total of 90 correlation coefficients and should be viewed with some allowance for the possibility of Type I errors until further research is conducted. It is of interest to compare Jackson’s (1976) description of the Social Participation dimension to comments in the Rorschach literature regarding the intepretive significance of use of color. Jackson describes a higher scorer on the Social Participation scale as an eager joiner of social groups, funloving, extraverted and gregarious. Low scorers are described as avoiding social activities,
uncommunicative and withdrawn. Although in conventional Rorschach interpretation use of color as a determinant of responses is classified according to its type of occurrence in conjunction with other determinants, notably Form, there is some agreement among authors that color scores relate to the respondent’s responsiveness to environmental stimuli, with interpersonal stimuli being of particular importance (Beck, 1949, p. 29; Klopfer and Davidson, 1962, p. 137). In the absence of significant differences between male and female Ss in the present study on their individual weights for either of the two dimensions, the occurrence of a significant relationship between Social Participation and use of color for females only remains an issue in need of further clarification. High scorers on the Organization subscale are described as orderly, disciplined and systematic, with low scorers being seen as inefficient, easily distracted and disorganized. It is intriguing to speculate on the relationship between these traits and the use of such a simplistic, objective heuristic as degree of apparent separateness of two entities in the inkblots in forming similarity judgments. Given the high degree of precision with which depressed patients, schizophrenic patients and nonpsychiatric controls were discriminated in the Wainer et al. (1976) study, the present set of minimal findings is surprising. Successful replication of Wainer et d’s result would suggest that perceptual anomalies related to clinical disorder, and not necessarily related to premorbid personality characteristics of Ss, may have been responsible for the earlier findings. Thus, the MDS-Rorschach approach may yet prove useful in delimiting pecularities of attention and perception in clinical states. The usefulness of this approach in quantifying response tendencies operating in the traditional Rorschach administration, and which have been shown to relate to respondents’ personality dispositions (Parker, 1983), remains yet to be demonstrated. Ack,tow/rdyefll~flt.~-This article is based on a Masters Thesis conducted by the first author at Texas A&M University. under the supervision of the second and third authors. We would like to acknowlege Larry Ringer, John Riskind, Deb Sowa and Brian Stagner for their help.
REFERENCES Anastasi A. (1983) Psychological testing. In 77~ Handbook of Clinical Psycholog?: Theory, Research. and Practice, Vol. 1 (Edited by Walker C. E.). pp. 420-445. Dow Jones-Irwin, Homewood, III. Beck S. J. (1949) Rorschach’s Test: II. A Variefy ojPersonalit~ Pictures. Grune & Stratton. New York. Becker J. (1980) Stimulus dimensions or stimulus aspects: a comparison of both concepts by means of similarity judgments. Psycho/. Res. 42, 227-240. Carroll J. D. and Chang C. C. (1970) Analysis of individual dilTerences in multidimensional scaling via on N-way generalization of “Eckert-Young” decomposition. Psychometrika 35, 283-320. Chang C. C. and Carroll J. D. (1968) How to use INDSCAL-a computer program for canonical decomposition of N-way tables and individual differences in multidimensional scaling. Unpublished manuscript, Bell Telephone Labs, Murray Hill. N.J. Howes R. J. (1981) The Rorschach: does it have a future? J. Person Assess. 45, 339-351. Jackson D. N. (1976) Juckson Persmaliry Inc;emory Manual. Research Psychologists Press, Goshen, N.Y. Klopfer B. and Davidson H. H. (1962) 7hr Rorschach Technique: art Inrroducrory Manud. Harcourt, Brace & World. New York. Nunnally J. C. (1978) Psychometric Theory. McGraw-Hill, New York. Parker K. (1983) A meta-analysis of the reliability and validity of the Rorschach. J. Person. Assess. 47, 227-231. Pruzansky S. (1975) How to use SINDSCAL-a computer program for individual differences in multidimensional scaling. Unpublished manuscript, Bell Telephone Labs, Murray Hill. N.J. Ross R. T. (1934) Optimum order for the presentation of pairs in the method of paired comparisons. J. educ. Psychol. 25, 375-382.
Torgerson W. S. (1952) Multidimensional scaling: theory and method. PsJchomerrika 17, 401419. Tucker L. R. and Messick S. (1963) An individual differences model for multidimensional scaling. 333-368,
Psychometrika
28,
1963.
Vestewig R. E. and Paradise C. A. (1977) Multidimensional motivation. J. Person. Assess. 41, 595603, 1977. Wainer H., Hurt S. and Aiken L. (1976) Rorschach revisited:
scaling
of the TAT and the measurement
of achievement
a new look at an old test. J. consulr. &I. Psycho/. 44, 390-399.