Sei sulla pagina 1di 13

Journal of Personality and Social Psychology 1986, Vol. 51, No.

6, 1224-1236

Copyright 1986 by the American Psychological Association, Inc. 0022-35l4/86/$00.75

Global Self-Esteem: Its Relation to Specific Facets of Self-Concept and Their Importance
Herbert W. Marsh
Department of Education, University of Sydney, Australia

Theory and common sense posit that the effect of a specific facet of self-concept on esteem will vary with the importance of the facet, but little support for this proposal was found in a study of late adolescents and young adults. Subjects, particularly those with high esteem, were more likely to have high self-concepts in facets they perceived to be more important, but their importance ratings did not contribute to the prediction of esteem. Unweighted averages of 12 distinct dimensions of selfconcept correlated about .7 with Esteem, but weighting each facet by the importance assigned to it by the entire group, by diverse subgroups, or by each individual resulted in little or no improvement. Neither self-concept/importance interactions nor self-concept/importance discrepancies were able to explain much variance in Esteem beyond that which could be explained by specific facets of selfconcept. Nevertheless, some support for the effect of importance was found for the Spiritual and Physical Abilities facets, and these were the two facets for which the perceived importance was most variable.

Historically, self-concept research has emphasized the general or total self-concept, relegating specific facets of the construct only a minor role. More recent theoretical and empirical research has emphasized the multiple dimensions of self-concept (e.g., Byrne, 1984; Dusek & Flaherty, 1981; Fleming & Courtney, 1984; Harter, 1982; Marsh, Barnes, Cairns, & Tidman, 1984; Marsh, Barnes, & Hocevar, 1985; Marsh & Shavelson, 1985; Shavelson, Hubner, & Stanton, 1976; Scares & Scares, 1982). In an extensive review of construct validation research, Byrne (1984, p. 427) concluded that self-concept "is a multidimensional construct, having one general construct and several specific facets." Shavelson and Marsh (1986; Marsh & Shavelson, 1985) reached similar conclusions in a review of research stimulated by Shavelson's model (Shavelson et al., 1976) and the Self Description Questionnaire (SDQ) instruments based on the model. They concluded that self-concept cannot be adequately understood if its multidimensionality is ignored. With the increased emphasis on the multidimensionality of self-concept, the specific facets have become more important and the role of general self-concept has become less clear. There is no widely accepted definition of how the general construct should be defined, and at least five operational definitions are common: (a) a hierarchical general self that appears at the apex of hierarchical models such as Shavelson's model; (b) a conglomerate general self that is the total score from a hodgepodge of self-referent items that attempt to sample broadly from a range of characteristics; (c) a global self-esteem scale that is rela-

lively unidimensional and content free in that it is composed of items that infer a general sense of serf-worth or self-confidence that could be applied to many specific areas (e.g., the General Self scale from the SDQ III and other scales described by Harter, 1982, and Rosenberg, 1965); (d) a discrepancy general self for which ratings of specific facets of self (actual ratings) are subtracted from ideal ratings (e.g., Higgins, Klein, & Strauman, 1985; but also see Wyiie, 1974, for a critical discussion); and (e) a. weighted average general self where specific facets Hoge& McCarthy, 1984; Watkins, 1978). are weighted according to their salience, value, or importance (e.g.,

Role of Facet Importance and the Interactive Hypothesis The alternative conceptualizations of the general self construct have implications for the role of the importance of a specific area of self-concept in determining general self-concept. Historically, William James (1890/1963) argued that failure in areas deemed to be unimportant has little impact on general self-esteem, and this contention has been reiterated by many theorists (e.g., Coopersmith, 1967;Harter, 1982,1983,1984,in press; Hoge & McCarthy, 1984; Rosenberg, 1965, 1979, 1982; Watkins, 1978; Wells & Marwell, 1976; Wylie, 1974). Coopersmith (1967, p. 6) indicated that an individual's self-appraisals might vary in different areas so that "his overall appraisal of his abilities would presumably weight these areas according to -their subjective importance enabling him to arrive at a general level of self-esteem" but that "objective evidence on the method of arriving at general appraisals is sparse." Wylie (1974, p. 48) stated, "The sum is simple expedient in the face of ignorance and should be so recognized. Steps should be taken to weight item ratings according to their perceived salience to S, but this has not yet been tried." Rosenberg (1965, 1979; also see Hoge & McCarthy, 1984) proposed an interactive hypothesis, and one possible model of
1224

I would like to thank Samuel Ball, Jenifer Barnes, Raymond Debus, Richard Shavelson, and Donald Spearritt for their helpful comments on earlier drafts of this article. Correspondence concerning this article should be addressed to Herbert W. Marsh, Faculty of Education, University of Sydney, NSW 2006, Australia.

SELF-CONCEPT

1225

in a fashion congenial to one's self-image is not, of course, without limit." Hoge and McCarthy (1984), and others, have emphasized the constraints that group and subgroup values place on the freedom to select one's values. These may force individuals to place a high value on some facets in which their self-perceptions are poor, thus leading to lower esteem.
LJJ

I -

in
UJ

Hoge and McCarthy 1984 Study


Importance Hoge and McCarthy (1984) examined relations among specific facets of self-concept, their perceived importance, and esteem (the Rosenberg scale) in a large sample of high school students. Using a variety of empirical and a priori weightings of the specific facets, they found that no weighted average correlated with Esteem higher than .45, which suggested to them that the weighted summations measure a construct different from esteem. An unweighted average of the specific self-concept facets correlated about .4 with esteem, and there was little or no improvement when the specific facets were weighted by the mean importance rating assigned by the entire group, by students within the same school, or by students within the same class. Paradoxically, when facets were weighted by the individual's own importance rating, the weighted average was significantly less correlated with esteem than with other weighted and unweighted averages. In further analyses, they found little or no support for Rosenberg's interactive hypothesis. These findings led the authors to conclude that, at least for this age group, group values are much more important than individual values in determining the value placed on specific facets, and they considered this to be their most important theoretical conclusion. However, they actually found that neither group nor individual ratings of importance were very useful, and that the weighted average based on individual importance ratings was significantly less correlated with esteem than with even the unweighted average. Because these findingsparticularly those based on individual importance ratingsare theoretically and logically implausible, alternative explanations, possible artifacts, and methodological weaknesses are examined here.

z o u
UJ LL U. UJ

Low

Medium

High

LEVEL OF SELF-CONCEPT FOR SPECIFIC FACETS


Figure 1. The interactive hypothesis. (According to the interactive hypothesis the effect of a specific facet of self on esteem varies as a function of the level of self-concept in the specific facet and the importance of the facet. For specific facets of self-concept that are very important, high levels of self-concept substantially enhance esteem, whereas low levels of self-concept substantially reduce esteem. For specific facets that are less important, the contributionpositive or negativeto esteem is much smaller.)

this hypothesis is presented in Figure 1. According to this model, having a positive self-concept in a particular facet will contribute positively to esteem, but the size of this positive contribution will depend on the facet's importance. The positive contribution to esteem will be larger when the level of the specific self-concept is more positive and the perceived importance is greater. Similarly, having a negative self-concept in a particular facet will detract from esteem, and the size of this negative contribution will be larger when the level of the specific selfconcept is more negative and its perceived importance is greater. Theorists also speculate about how individuals determine the importance of specific areas of self-concept and how this relates to esteem. Following Festingers social comparison theory, Wells and Marwell (1976, p. 55) described the dilemma of a person with a poor ability in a particular area who must balance a reality principle against the need to have a favorable self-evaluation. One possible compromise is to recognize the poor ability in the particular area but to give it little importance in the determination of esteem, Rosenberg (1982, p. 538) found support for a selectivity hypothesis in that an individual "will be disposed to value those things at which he considers himself to be good and to devalue those qualities at which he considers himself poor," but he also recognized that "the freedom to select one's values

Use of Single-Item Rating Scales


Hoge and McCarthy assessed esteem with a multi-item scale, but the specific facets and their importance were inferred from single-item responses. Their esteem measure was reasonably reliable (a = .75), but the reliability of the specific facets and the importance ratings was not reported. Marsh et al. (1985) measured specific components of self-concept with both multi-item scales and single-item responses, and they concluded that the single-item responses had poorer psychometric properties than did the multi-item scales (also see Rushton, Brainerd, & Pressley, 1983). Hence, the use of single-item scales may constitute an important methodological weakness in the Hoge and McCarthy study.

Response Variability in Different Facets


The relative contribution of each specific facet to the weighted and unweighted averages depends, in part, on response variabilityfacets with more variability contribute

1226

HERBERT W. MARSH importance of domains in which one is not performing competently, as well as endorse the importance of domains in which one is competent" (p. 23). However, instead of weighting each specific facet by its perceived importance (an interactive model), she determined the difference between each facet and its perceived importance (a discrepancy model). Harter found that importance ratings were nearly always higher than self-concept ratings when both sets of ratings were made on the same response scale, but she discovered that the discrepancy scores were close to zero for subjects with the highest esteem, moderately negative for subjects with intermediate esteem, and most negative for subjects with the poorest esteem. Although Harter's discrepancy scores were strongly correlated with esteem, she failed both to demonstrate that they contributed beyond the contribution of just the specific facets and to contrast the theoretical basis of her model with the more typical interactive model.

more. The self-concept and importance ratings tend to be very negatively skewed so that the greatest response variability occurs for those facets for which the ratings are lowest. However, the larger contribution of these facets runs counter to the intention to weight more heavily those specific facets that are more important. This potential problem is evident in the Hoge and McCarthy study inasmuch as their means and standard deviations are correlated about .9 (see their Table 1, p. 406) for both self-concept and importance ratings. The z-score transformation provides an alternative scaling that would remove this potential source of bias because it eliminates mean differences between the self-concept facets.

Nature ofSelf-Concept/Importance Cross-Product Scores


Hoge and McCarthy tested whether the perceived importance of a specific facet of self-concept influences how that facet relates to self-esteem by averaging the self-concept/importance cross products. However, these cross products may not provide an adequate test of the interactive hypothesis. Potential problems and possible solutions are illustrated with the following two examples. 1. Consider one subject who has a very positive self-concept in a facet that is perceived to be unimportant and a second subject who has a very negative self-concept in a facet that is perceived to be very important. Common sense and the interactive hypothesis (Figure 1) suggest that the first situation should contribute positively, albeit weakly, to esteem, whereas the second situation should detract substantially from esteem. However, the self-concept/importance cross products for both subjects will be about the same (depending on the characteristics of the response distributions), so that the cross products do not adequately represent the interactive hypothesis. One possible solution would be to transform the self-concept ratings into z scores (M = 0, SD = 1) so that the cross products will always be positive for self-concepts above the mean and negative for self-concepts below the mean. 2. Consider two possible response patterns: (a) subjects who have low self-concepts for some facets and view these facets to be unimportant, but who have high self-concepts for other facets and perceive these facets to be important; and (b) subjects who also have low self-concepts for some facets and high selfconcepts for others, but who perceive all facets to be important. The interactive hypothesis posits that the first group of subjects should have higher esteem scores, but the sum of self-concept/ importance cross products will actually be higher for the second group. Again the nature of the cross products does not adequately represent the interactive hypothesis. The problem is that the average of cross products will be larger for subjects who perceive all facets to be important. One possible solution is to rescale the importance ratings to be ipsative. For example, each subject's importance ratings could be divided by the sum of all of his or her importance ratings so that the transformed importance ratingscalled proportionalized importance ratings for purposes of this studysum to 1.0.

A General Test of the Interactive and Discrepancy Models


A problem with the use of self-concept/importance cross products is that the effect of the self-concept/importance interaction is confounded with the main effects of the self-concept and importance ratings. Similarly, a problem with the use of self-concept/importance differences in the Harter discrepancy model is that the difference scores confound the main effects of the self-concept ratings and the importance ratings. More suitable tests of both models can be conducted with analyses of variance (ANOVAS) or equivalent tests based on multiple regression (Cohen & Cohen, 1975; also see discussion of a related problem by Hall & Taylor, 1985; Lubinski, Tellegen, & Butcher, 1983; Marsh, in press-b; and Taylor & Hall, 1982). The interactive model as depicted in Figure 1 suggests that there is (a) a main effect of self-concept; (b) no main effect of importance (although the existence of such an effect may not contradict the interactive model), and (c) a self-concept/importance interaction. In terms of multiple regression, the self-concept facet and the corresponding self-concept/importance cross product should each contribute significantly, positively, and independently to the prediction of esteem. The discrepancy model posits main effects for both the self-concept and importance ratings, although the existence of a self-concept/importance interaction may not contradict the discrepancy model. In terms of multiple regression, the self-concept facet should contribute positively, and the importance rating negatively, to the prediction of esteem. The interactive and discrepancy models posit alternative conceptualizations of the influence of importance, but they are not antithetical. In particular, if both the main effect of importance and the importance/self-concept interaction are significant and in the right direction, then there is empirical support for both models.

Present Investigation
The purpose of the present investigation is to relate esteem to specific facets of self-concept and to the importance of each facet. The study is based on data from all my published studies that have used the SDQ III. Self-concept facets are represented

Harter Discrepancy Model


Harter (1984, in press) also hypothesized that "in order to maintain a positive sense of self-worth, one must discount the

SELF-CONCEPT by both multi-item scales and single-item responses, allowing a comparison between the two. Diverse subgroups of subjects who have completed the SDQ III are expected to differ widely in the importance they attach to different facets (e.g., Physical Ability for a group of athletes and Spiritual Values for students from a Catholic girls' senior school), providing a good basis for comparing the influence of individual, subgroup, and group values. Analyses are performed with both original and z-score self-concept ratings, and with both original and proportionalized (ipsative) importance ratings, thus providing tests of the transformations proposed in the discussion of the Hoge and McCarthy study. Although there are many differences between the present investigation and the Hoge-McCarthy study, the Esteem scale from the SDQ III was initially derived from the Rosenberg scale used in the earlier study, and many of the specific facets considered in the two studies are similar. Finally, general tests of both the interactive and discrepancy models are conducted with multiple-regression analyses.

1227

responses to the importance ratings for the 12 summary description items. 6. Proportionalized importance ratings: For each subject, these were derived by (a) summing the 12 importance ratings for that subject and (b) dividing the importance rating for each facet by the sum of the 12 importance ratings for that subject. Because the set of 12 proportionalized importance ratings sum to 1.0 for each subject, the ipsative scores are relative to the other facets for each subject rather than relative to responses by other subjects. 7. Self-concept/importance cross products: This is the product of each self-concept facet and its perceived importance. 8. Self-concept/importance discrepancy scores: This is the difference between each facet and its perceived importance.

Samples
Data came from five groups that were expected to differ systematically with respect to the perceived importance of some specific facet of self-concept. Group 1 consisted of the sample of 151 Australian university students (mean age = 21.9, 79% female) described by Marsh et al. (1985; also see Marsh & O'Niell, 1984, Study 2). As part of that study, subjects asked the person who knew them the best to complete the SDQ HI, too. These significant others completed the survey as if they were the person who had given it to them (i.e., they were to predict what the subject had said). For purposes of the present investigation, the primary focus is on the self-ratings, but the self-other correlations are also considered. Group 2 consisted of the 361 Outward Bound participants (mean age = 21.3,76% male) described by Marsh et al. (1986). As part of that study, participants completed the SDQ HI one month before, the first day of, and the last day of a 26-day residential program. In the present investigation, the primary focus was on data from Time 1, but testretest correlations over the Time I/Time 2 control interval were also considered. Because of the nature of the Outward Bound program, it was anticipated that this group would have higher Physical self-concepts and would place more emphasis on this facet. Group 3 consisted of the 296 1 Ith-grade girls (mean age = 16.7) from two private Catholic girls' schools described by Marsh and O'Niell (1984). Because of the nature of this group, it was anticipated that these subjects would have higher Spiritual self-concepts and would place more emphasis on this facet. Group 4 consisted of 46 high-school girls (mean age = 16.6) who were selected as the nonalhlete control for subjects summarized in Group 5 (see Jackson, 1984; Marsh & Jackson, in press). Subjects in Groups 4 and 5 completed accuracy and importance ratings for the 12 summary items, but unlike subjects in Groups 1-3, they completed only four of the self-concept multi-item scales (Physical Ability, Physical Appearance, Emotional Stability, and Global Self-Esteem). For purposes of the present investigation, only the summary-rating items and their relation to Esteem were considered for these two groups. Nevertheless, results from the original study showed that subjects in this nonathlete group, compared with the athletes, were substantially lower in the self-concept of Physical Ability for both multi-item and single-item scales and the importance they placed on this facet, but they did not differ substantially in terms of other areas of self-concept or in the importance they placed on these other areas. Hence, these groups were particularly relevant to the purposes of the present investigation, Group 5 consisted of 76 female athletes (mean age = 20.1). Forty six of these women were high school athletes and 30 were finalists in the 1984 Australian women's powerlifting championships (Jackson, 1984; Marsh & Jackson, in press). These subjects completed the same materials as did the subjects in Group 4. Although the high school athletes and powerlifters differed slightly in some areas of self-concept in a way that appeared to be age related, the differences were small, and they did not

Method Self-Description Questionnaire HI


The present investigation is based on responses to the SDQ III. The theoretical rationale for its construction and a considerable body of empirical support for its dimensionality, reliability, and validity are summarized elsewhere (Marsh, in press-a; Marsh et al., 1985; Marsh & Jackson, in press; Marsh & O'Niell, 1984; Marsh, Richards, & Barnes, 1986). Each of the 13 SDQ HI scales is represented by 10 or 12 items, approximately half of which are negatively worded, and subjects respond on an 8-point response scale that ranges from 1 (definitely false) to 8 (definitely true). The 13 scales are as follows; Physical Ability, Physical Appearance, Opposite Sex Relations, Same Sex Relations, Relations With Parents, Spiritual Values/Religion, Honesty, Emotional Stability, Verbal, Math, General Academic, Problem Solving, General-Self (the Esteem scale for purposes of this study). Although not formally part of the SDQ III, the SDQ III studies have also asked subjects to respond to 12 single summary items designed to reflect 12 of the 13 scalesall but the Esteem scale. For each of these 12 single items, subjects indicate the item's accuracy (i.e., How accurate is this statement as a description of you?) and its importance (i.e., How important is this characteristic in determining how you feel about yourself?). Responses to these items are made on a scale ranging from 1 (very inaccurate/very unimportant) to 9 (very accurate/very important). Psychometric properties of the accuracy ratings and their relation to the multi-item scale scores that they are designed to reflect were examined by Marsh et al. (1985) and are further considered as part of the present investigation. For purposes of this investigation, the following variables, derived from the SDQ III responses, were considered: 1. Original (untransformed) Self-Concept scale scores: These scores are the mean response to the 10 or 12 SDQ III items designed to measure each scale. 2. Standardized (z score) Self-Concept scale scores: These scores are the same as the Original Self-Concept scale scores except that the scale scores are standardized (M - 0, SD - 1) across the total group. 3. Original (untransformed) summary ratings: These are the original responses to each of the accuracy ratings for the 12 single items. 4. Standardized (z score) summary ratings: These are the same as the original summary ratings except that each of the original ratings is standardized (M = 0, SD = 1) across the total group. 5. Original (untransformed) importance ratings: These are the raw

1228

HERBERT W. MARSH examined by relating them to responses by significant others. Subjects in Group 2 completed the self-concept scales, the summary ratings, and the importance ratings twice during a onemonth period. The test-retest correlations for this group appear in Table 2. Coefficients for the multi-item scales (mean r = .87) were substantial and similar in magnitude to the internal consistency estimates that were reported in the original study. The test-retest correlations for the corresponding single-item scales were lower (mean r = .70), but still moderate. The testretest coefficients for single-item importance ratings (mean r = .57) were lower than for either the multi-item or single-item selfconcept ratings. Significant others were selected by subjects in Group 1, and these significant others inferred responses (i.e., completed the SDQ III as if they were the subject) to the self-concept scales, the summary ratings, and the importance ratings. Correlations between the self-reports and the responses by the significant others were used to test the validity of responses to the SDQ III. The self-other correlations (Table 2) were substantial for the multi-item self-concept scales (mean r = .57), lower for the summary self-concept ratings (mean r - .40), and lowest for the importance ratings (mean r = .26). These findings provide more support for the validity of the self-concept scales and summary ratings than they do for the validity of the importance ratings. Correlations between each of the 12 multi-item self-concept scales and the corresponding single-item summary ratings (Table 2) were substantial, particularly given the apparent unreliability of the summary ratings. These findings suggest that the summary ratings may provide a reasonable estimate of the multi-item scales that they are designed to parallel, even though they are apparently less reliable and valid. The apparently modest reliability and validity of the importance ratings were discouraging. However, two qualifications to this conclusion must be noted. First, there was a wide variation in the estimates for the different facets. In particular, the testretest and self-other correlations were substantially higher for the Spiritual facet and, to a lesser extent, for the Physical Ability facet compared with the other facets. Second, the relatively poor psychometric properties of the importance ratings referred only to individual responses, and the reliability of the mean importance rating across all 930 subjects in the total group and the subjects in each subgroup was substantial even when the reliability of individual responses was modest. Hence, the psychometric properties may be reasonable for averages across group and subgroup responses and for individual ratings of some specific facets. Also, the Spiritual and Physical facets for which the importance ratings were postulated to have the greatest influence were the importance ratings with the best psychometric properties.

differ on ratings of the Physical Ability facet or on any of the importance ratings. For this reason, and also because of the small sample sizes, both groups of female athletes were combined to form Group 5 for purposes of this investigation. Across all the groups, a total of 930 subjects completed the multiitem Esteem scale, the 12 single-item self-concept facets, and the 12 single-item importance ratings. However, only 808 (all but Groups 4 and 5) completed the 12 multi-item self-concept scales. For the total group, ages varied from 15 to 60 (mean ages were 19.6 for the group of 930 subjects and 19.7 for the group of 808 subjects), but most were younger than 30 years of age. A majority of the subjects were female, but 306 (33% of the group of 930 subjects and 38% of the group of 808 subjects) were male. A majority of the subjects were either high school or university students, but approximately 30% were nonstudents.

Statistical Analysis
Relations between the self-concept scales and Esteem were examined across all subjects in Groups 1 through 3 as well as separately for each group, whereas relations between the summary ratings and Global SelfEsteem were examined across all subjects in Groups 1 through 5 as well as separately for each group. Except for the self-concept scales not completed by Groups 4 and 5, there were very few missing data for any of the subjects, and mean responses were substituted for the few missing values that did occur. The actual analyses are described in more detail in the Results section. It was anticipated that the diverse subgroups who completed the SDQ III would vary substantially in self-concept and importance ratings for some of the specific facets. Support for this contention was necessary in order to justify the weighting of self-concept facets by subgroup ratings of importance; if the subgroups did not differ in their importance ratings, then weighting self-concept ratings by the mean importance ratings for the subject's subgroup would make no sense. Furthermore, if the predicted group differences did existparticularly for the perceived importance of the Physical and Spiritual facetsthen there would be support for the validity of the importance ratings. In preliminary analyses, multivariate and univariate analyses confirmed the expectation of group differences. Because of the large sample size virtually every effect was statistically significant, but the etas from the one-way ANOVAS provided an estimate of the size of each effect (see Table 1). A detailed examination of these group differences is not the focus of this study, but of particular relevance is the demonstration that the groups do vary substantially for some but not for all facets (etas vary from .08 to .47). As anticipated, Physical self-concept and the importance of this facet were substantially higher for the group of female athletes (Group 5) and for the group of Outward Bound participants (Group 2) and substantially lower for the group of female nonathletes (Group 4) than they were for the total group. Similarly, the Spiritual self-concept and the importance placed on this facet were substantially higher for the sample of girls from the Catholic high school (Group 3) than they were for the total group. Despite the large group differences for some of the specific self-concept facets, group differences accounted foronly2%(t) = .14) of the variance in the multi-item Esteem scores.

Results and Discussion Psychometric Properties of Responses to the Summary and Importance Ratings
Single-item scales may have limited reliability, validity, and generalizability, which may affect conclusions based on them, but these limitations are often disregarded. In the present investigation test-retest correlations provided one indication of the reliability of the single-item responses, and their validity was

Selectivity Hypothesis: Self-Concept/Importance

Correlations

Rosenberg's(1965,1982) selectivity hypothesis posits that individuals with high self-concepts in particular facets will attach higher importance to these facets and that this mechanism will allow individuals to enhance or protect their esteem. Across all respondents, correlations between the importance ratings and the self-concept measures (Table 3) were moderately positive,

SELF-CONCEPT Table 1

1229

Mean Ratings for Multi-Item Self-Concept Scales (Scale), Single-Item Summary Self-Concept Ratings (Sum), and Importance Ratings (Impt) as Well as Etas Representing Differences Between Groups on These Scores Facet' Score

10

11

12

13

Group 1 Scale Sum Impt


5.42 5.33 5.47 5.18 5.75 6.40 4.70 6.41 7.29 4.80 6.59 7.27 5.62 6.33 7.64 5.15 6.24 7.33 4.90 4.95 5.49 6.17 7.13 7.93 5.75 6.54 7.10 4.54 5.20 5.49 5.83 6.20 6.93 5.06 5.94 6.63 6.00

Group 2 Scale Sum Impt


6.13 6.16 6.61 5.25 5.40 5.90 4.28 5.72 7.24 4.67 6.32 6.99 5.91 6.60 7.52 5.40 6.53 7.34 4.41 4.33 4.83 6.17 7.17 8.04 5.52 6.32 7.14 5.10 5.89 6.24 5.68 6.19 6.47 5.36 5.95 6.87 5.73

Group 3 Scale Sum Impt 5.28 4.04 5.17 4.29 5.13 5.64 4.17 5.34 6.60 4.96 6.84 7.32 5.81 6.82 7.87 4.92 6.28 7.21 5.41 5.49 6.28 6.02 6.84 7.86 5.25 5.92 6.64 4.81 5.40 6.01 5.37 5.84 6.83 4.82 5.57 6.24 5.53

Group 4

Sum Impt

4.04 5.46

4.54 6.28

5.37 7.24

6.33 7.15

6.85 7.78

6.26 7.91

4.09 4.78

6.78 8.13

6.37 7.47

5.50 6.54

6.00 7.26

5.80 6.85

Group 5 Scale Sum Impt _ 6.97 7.45 _ 5.29 6.26 _ 6.46 6.90 _ 6.67 6.82 _ 6.71 7.46 _ 6.80 7.50 _ 3.92 4.20 _ 7.34 8.13 _ _ _ _ 5.96 6.80 6.15 6.37 6.37 7.43 6.55 6.76 6.87

Total group Scale M SD Sum M SD Impt M SD


5.72 1.42 5.67 1.91 5.96 2.03 4.78 1.23 5.05 1.59 5.95 1.82 4.32 1.11 5.75 1.85 7.01 1.53 4.80 0.84 6.56 1.47 7.13 1.50 5.82 1.20 6.65 1.85 7.66 1.52 5.18 1.25 6.41 1.73 7.34 1.48 4.87 1.66 4.75 2.42 5.34 2.52 6.11 0.72 7.06 1.27 7.97 1.25 5.47 1.06 6.27 1.44 7.01 1.45 4.89 1.56 5.63 1.87 6.08 1.75 5.59 1.10 6.08 1.47 6.72 1.56 5.11 5.71 0.96 1.16 5.85 1.53 6.63 1.62

Etasb Scale Sum Impt

.29 .36 .39

.47 .36 .15

.17 .23 .20

.16 .16 .12

.09 .09 .11

.18 .10 .10

.27 .24 .28

.10 .14 .08

.17 .19 .19

.14 .17 .18

.16 .12 .14

.25 .15 .17

.14

1 = Physical Ability; 2 = Physical Appearance; 3 = Opposite Sex Relations; 4 = Same Sex Relations; 5 = Parent Relations; 6 = Emotional; 7 = Spiritual Values; 8 = Honesty; 9 = Verbal; 10 = Math; 11 = General Academic; 12 = Problem Solving; 13 = Esteem. b One-way ANOVAS of differences between groups are summarized in terms of the eta values. For the scale scores, the etas are based on 808 sets of responses from Groups 1-3 (tj > .08 is statistically significant at p < .05). For summary and importance ratings the etas are based on the 930 sets of responses from Groups 1 -5 (TI > .09 is statistically significant at p < .05).

thus supporting the selectivity hypothesis. Furthermore, when subjects were divided into three groups on the basis of their Esteem scores, importance/self-concept correlations were largest for subjects high in esteem and lowest for subjects low in esteem (Table 3). The self-concept/importance correlations were consistently higher for Spiritual Values and, to a lesser extent, for Physical Ability than they were for the other facets. Consistent with earlier observations, these may also be those facets in

which subjects have the most freedom in determining the relative importance that is attached to them and the facets for which the importance ratings and self-concept ratings are most variable (see Table 1). These findings demonstrate that (a) a subject's self-concept in a specific area is moderately correlated with his or her perceived importance of the area, (b) for some self-concept facets this tendency is very strong, and (c) this tendency is stronger for subjects with higher esteem.

1230
Table 2

HERBERT W. MARSH tively poor and higher for facets in which self-concepts are more favorablethan for subjects low in esteem. However, the standard deviation of the 12 (original) importance ratings by each subject was not significantly correlated (r = .03) with the subject's esteem. Third, even though
rs between0

Psychometric Properties of the Multi-Item Self-Concept Scales (Scale), Single-Item Summary Ratings (Sum), and Importance Ratings (Impt)for the 12 Self Facets and Esteem

self-concept/importance

cross products and self-concept/importance discrepancies were moderately correlated with Esteem, these correlations were smaller than the correlations between the self-concept ratings and Esteem for every one of the 12 facets (Table 4). This last observation is important because other researchers seem to interpret self-concept/importance correlations such as those observed here as support for either the interactive model or the discrepancy model, but such interpretations may not be justi-

Test-retest rs* Facet Physical Scale

Self-other rs" Scale

Sum .77 .72 .76 .70 .70 .66 .84 .54 .66 .80 .69 .68 .70 .71

Impt

Sum .65 .48 .31 .32 .55 .28 .77 .15 .33 .63 .22 .15 .32 .40

Impt

Scale and sum

Appearance
Opposite Sex

Same Sex
Parents Emotional Spiritual Honesty Verbal Math Academic Problem Solving Esteem'1 Median r Meanr

.88 .85 .89 .85 .88 .85 .94 .75 .88 .91 .87 .86 .88 .87 .87

.61 .65 .55 .61 .66 .50 .86 .50 .46 .57 .57 .49 .57 .58

.77 .47 .51 .45 .76 .62 .82 .37 .62 .75 .36 .46 .45 .57 .57

.45 .18 .33 .20 .19 .19 .70 .19 .09 .23 .26 .17 .19 .26

.81 .73 .80 .65 .77 .64 .87 .49 .62 .83 .68 .68 .71 .71

fied.

Relations Among Specific Facets, Importance Ratings, and Esteem


Unweighted averages of specific self-concepts. Three un-

weighted averages used to summarize responses to the 12 specific facets of self-conceptthe average of the raw responses, the average of the z-score standardized responses, and the first principal component derived from the set of self-concept ratingswere correlated with Esteem (Table 5). For the multiitem scales, all the unweighted averages correlated about .7 with

Test-retest correlations (interval = 1 month, = 361) from Marsh, Richards, and Barnes (1986). b Self-other correlations ( = 151) are based on self-report responses and self-concepts inferred by a significant other from Marsh, Barnes, and Hocevar (1985; also see Marsh & O'Niell, 1984, Study 2). Correlations based on the total sample (n = 930 for sum and importance ratings and n = 808 for scales). d The Esteem scale was only measured with a multi-item scale.

Table 3 Selectivity Hypothesis: Self-Concept/Importance Correlations Based on Multi-Item Self-Concept Scales (Scale) and SingleItem Self-Concept Summaries (Sum) for the Tola! Group and Groups of Subjects With Low, Medium, and High Levels of Esteem Group Medium esteem Scale

Implicit in the selectivity hypothesis is the assumption that higher self-concepts in specific facets cause subjects to place more importance on these facets, which, in turn, causes higher levels of esteem. However, the correlational nature of the data in the present investigation and in Rosenberg's research does not provide an adequate test of these causal relations, and alternative explanations are plausible. For example, subjects may strive to improve their skills and self-perceptions in areas that they and significant others view to be important, and to the extent that they are successful, this may lead to higher levels of esteem. Also, such a tendency might be stronger for high-esteem individuals. Although the data in the present investigation are not appropriate for testing these alternative explanations of the empirical relation, several observations fail to support the selectivity hypothesis. First, importance ratings for every facet were nearly uncorrelated with esteem (Table 4). In particular, the facets for which all subjects tended to have the highest self-concepts and perceived to be most important were not more valued by subjects high in esteem, nor did subjects high in esteem devalue those facets for which all subjects tended to have the lowest self-concepts and/or perceived to be least important. Instead, it was the specific facets of self-concept that were substantially correlated with esteem, particularly those facets judged to be most important by all subjects. Second, the selectivity hypothesis suggests that importance ratings by subjects high in esteem will be much more variablelower for facets in which self-concepts are rela-

Total Facet Physical Appearance Opposite Sex Same Sex Parents Emotional Spiritual Honesty Verbal Math Academic Problem Solving Median r Meanr Scale Sum

Low esteem Scale

High esteem Scale

Sum .45 .20 .34 .37 .49 .08 .75 .38 .21 .33 .34 .39 .36 .36

Sum .60 .24 .31 .42 .52 .31 .89 .45 .33 .45 .35 .43 .42 .44

Sum
.71-

.62 .07 .29 .27 .43 .08 .86 .36 .13 .46 .33 .35 .34 .35

.59 .23 .37 .48 .54 .28 .85 .44 .34 .43 .36 .47 .44 .45

.45 .01 .17 .18 .29


-.07

.79 .31 .02 .40 .25 .30 .27 .26

.68 .15 .27 .25 .48 .08 .87 .33 .10 .45 .36 .30 .32 .30

.72 .06 .39 .33 .54 .11 .89 .40 .19 .53 .37 .48 .39 .42

.33 .46 .65 .60 .36 .89 .44 .47 .52 .39 .61 .49 .54

Note. Each correlation represents the relation between self-concept ratings and importance ratings for the same facet. For the total group, correlations between self-concept and importance ratings are based on 930 (for single-item self-concept ratings) and 808 (for multi-item self-concept scales) sets of responses. Correlations for the low-, medium-, and high-esteem groups are each based on about one third of the sets of responses*

SELF-CONCEPT

1231

Table 4
Correlation of Esteem With Self-Concept Ratings, Importance Ratings, Self-Concept/Importance and Self-Concept/Importance Cross Products, Differences Selfconcept3/ importance cross products Scale

older than subjects in their study. Hoge and McCarthy considered only 9 specific components, compared with 12 used here, so their unweighted average may have been less reliable and generalizable. Finally, the psychometric properties of multi-item scales are superior to those of single-item scales. The results of Selfconcept/ importance1* differences Scale the present investigation, as well as these observations, suggest that the findings by Hoge and McCarthy, and particularly their implications, must be viewed cautiously. Empirically weighted averages. For the total group, multiple correlations relating Esteem to the 12 specific facets of self (Table 5) were .78 and .67 for the self-concept scales and summary ratings, respectively. Similar analyses were performed for each subgroup, and these multiple correlations were somewhat higher. However, when the multiple correlations were corrected for sampling bias (see Nie, Hull, Jenkins, Steinbrenner, & Bent, 1975), the differences were much smaller (Table 5). These multiple correlations represent an absolute upper limit for correlations between Esteem and weighted averages of the specific facets based on group or subgroup ratings of importance. Although these multiple correlations are clearly higher than those obtained with the unweighted averages, the size of the differences is modest. Hence, any improvement in weighted averages of the self-concept facets by total group or subgroup importance ratings must also be modest. Specific facets weighted by group and subgroup importance ratings. For these analyses, each self-concept score was weighted by the mean importance weighting either for the entire group or for the subgroup to which the subject belonged (the means appear in Table 1). Different procedures for scaling the specific self-concept and importance ratings were considered, but these make virtually no difference and, consequently, are not discussed further. For the multi-item self-concept scales, correlations between all the weighted averages and Esteem were close to .7, whereas those for the summary items were close to .6. In each instance, averages weighted by total group impor-

Selfconcept ratings Facet Physical Appearance Opposite Sex Same Sex Parents Emotional Spiritual Honesty Verbal Math Academic Problem Solving Total0 Scale

Importance ratings

Sum .25 .41 .41 .35 .30 .51 .08 .28 .34 .14 .25 .31 .59

Sum .08
-.04

Sum .22 .26 .34 .27 .26 .45 .09 .26 .28 .11 .20 .23 .48

Sum .17 .34 .33 .27 .23 .35 .09 .14 .21 .13 .15 .24 .44

.25 .48 .50 .46 .34 .60 .06 .23 .34 .12 .32 .41 .67

.09 .07 .12 .15 .03 .14 .09 .01 .07 .05 .13

.20 .28 .41 .34 .29 .53 .07 .22 .29 .09 .23 .26 .51

.11 .32 .26 .17 .15 .29 .03 .00 .14 .11 .18 .21 .33

Note. Scale = multi-item self-concept. Sum = single-item summary rating. " Self-concept/importance cross products were calculated by multiplying the original (unstandardized) self-concept rating for each facet (multi-item scales and single-item summary ratings) by the corresponding original (unproportionalized) importance rating. b Self-concept/importance differences were calculated by subtracting the original (unstandardized) self-concept rating for each facet (multiitem scales and single-item summary ratings) from the corresponding original (unproportionalized) importance rating. c Total scores are the sum of the 12 scores for each of the specific facets.

Esteem, whereas correlations for the summary ratings were close to .6. Although there were minor differences among the three unweighted averages and among the different subgroups, the results are quite consistent. These correlations based on unweighted averages provide a lower limit for testing the usefulness of the various weighted averages. To the extent that the weighted averages perform no better than the unweighted averages, the rule of parsimony dictates that the unweighted averages are preferable. Hoge and McCarthy (1984) reported a correlation of only .4 between their unweighted average of single-item self-concept scales and Esteem, and they noted important implications based on this low correlation. However, the corresponding correlations in the present investigation, .6 for single-item ratings and .7 for multi-item scales, were much higher. Although many possible explanations may account for the different results, the most likely explanations include the reliability of the esteem measures, perhaps the ages of the subjects, the number of specific components that were considered, and the use of multiitem scales. Responses to the SDQ III Esteem scale were more reliable than responses to the Rosenberg scale in the Hoge and McCarthy study. Hoge and McCarthy found that the specific components were more highly correlated with the Esteem scale for their oldest subjects, and subjects in the present study were

tance ratings were virtually identical to those weighted by subgroup importance ratings and did not differ from those based on unweighted averages. The correlations reported here are higher than those in the Hoge and McCarthy study, probably for the same reasons as indicated earlier, but the similarity of correlations based on weighted and unweighted responses is consistent with their findings. These findings provide no support for the usefulness of the group or subgroup importance ratings in the weighting of specific components of self-concept to predict Esteem. Specific facets weighted by individual importance ratings. For these analyses, each self-concept facet for each subject was weighted by the importance rating assigned to that facet by the subject. These analyses are particularly important because (a) this is the type of weighting typically implied by researchers who advocate the use of importance ratings and (b) this is the type of weighting that Hoge and McCarthy found to perform more poorly than the unweighted average (and which I suggested might be an artifact of scaling problems). For these analyses, the way that the self-concept scores and the importance ratings were scaled makes a substantial difference (Table 5). When raw self-concept scores were weighted by raw importance ratings, as in the Hoge and McCarthy study, the weighted averages were substantially less correlated with Es-

1232

HERBERT W. MARSH

Table 5
Correlations Between Weighted Averages of the 12 Self-Concept Scores (Scales and Summary Ratings) and Esteem for the TotalGroup and Each Subgroup Multi-item scales for groups Item Total Single-item summary ratings for groups

Total

Average self-concept scores' Total raw score Total standard score Principal component Uncorrected/{ Corrected R
.673 .688 .702 .777 .773 .658 .678 .707 .790 .768 .699 .708 .715 .800 .792 .646 .666 .695 .778 .767 .586 .591 .588 .673 .669 .551 .560 .565 .677 .640 .631 .635 .635 .721 .709 .518 .525 .522 .660 .641 .579 .587 .583 .836 .768 .559 .552 .541 .762 .708

Weighted by total group importance ratingsb Raw score/raw impt Stand score/raw impt
.685 .693 .672 .683 .709 .712 .659 .671 .595 .595 .564 .568 .641 .641 .532 .535 .585 .591 .559 .540

Weighted by subgroup importance ratings" Raw score/raw impt Raw score/prop impt Stand score/raw impt
.680 .685 .693 .676 .677 .686 .710 .711 .713 .656 .650 .669 .593 .596 .597 .568 .568 .572 .643 .643 .642 .530 .530 .534 .588 .589 .593 .554 .554 .544

Weighted by individual importance ratings'1 Raw score/raw impt Raw score/prop impt Stand score/raw impt Stand score/prop impt
.514 .701 .703 .706 .480 .700 .693 .698 .523 .722 .718 .722 .503 .670 .683 .680 .478 .615 .606 .611 .439 .595 .584 .588 .514 .664 .653 .658 .432 .544 .540 .543 .454 .632 .609 .633 .455 .549 .555 .538

Weighted by individual random weights6 Raw score/raw impt Raw score/prop impt Stand score/raw impt Stand score/prop impt
.379 .632 .652 .657 .293 .620 .652 .654 .392 .657 .672 .677 .360 .602 .621 .629 .378 .557 .560 .567 .358 .542 .559 .559 .410 .600 .589 .606 .337 .483 .486 .492 .347 .551 .531 .568 .308 .524 .513 .524

Note. Impt = importance rating. Stand = standardized. Prop = proportionalized. " Each of these mean scores was determined without taking into account the importance ratings. The commercially available Statistical Package for the Social Sciences program (Nie et al., 1975) was used to determine the first principal component of the set of 12 ratings. b These weighted averages were determined by multiplying self-concept ratings by the corresponding unsealed importance ratings. Similar analyses were conducted with the proportionalized importance ratings, but because the results did not differ from those reported here they are not presented. 0 These weighted averages were denned as in b except that the importance weightings were obtained from the subgroup ratings rather than from individual or total group importance ratings. Correlations based on z-score standardized self-concepts and proportionalized importance ratings were the same as those labeled stand score/raw impt and so they have not been presented. d These weighted averages were denned as in Footnotes b and c except that the self-concept scores were weighted by importance ratings assigned by each individual. Separate averages were derived from the original importance ratings (raw impt) and the proportionalized importance (prop impt) ratings. " For purposes of these analyses, the importance ratings were replaced with random numbers that varied between 0 and 1 on a uniform distribution, and these weighted averages were defined as in Footnote d.

teem than they were with any of the other weighted or unweighted averages. These findings were consistent for self-concept scales and summary items, were consistent across subgroups, and were consistent with Hoge and McCarthy's findings. Substantially larger correlations occurred when the self-concept scores were standardized or the importance ratings were proportionalized, but these correlations differed little from those based on only the unweighted average of self-concept ratings. Hence, these findings provide no support for the usefulness of importance ratings by individual subjects in the weighting of specific components of self-concept to predict Esteem. In order to further explore the apparent artifact of using un-

sealed scores, each subject was assigned a set of 12 random numbers that varied between 0 and 1. These were used to weight the specific self-concepts. Except for the use of these random numbers in place of the importance ratings by individual subjects, the same analyses were performed. Again the correlations based on the raw (random) weights and the raw self-concept scores resulted in substantially lower correlations than when either the self-concept scores were standardized or the (random) weights were proportionalized (Table 5). This demonstrates that the findings were not due to some serendipitous relation between the importance ratings and self-concept scores that is idiosyncratic to the present investigation. It is also gratifying to note that the randomly weighted averages were somewhat less

SELF-CONCEPT correlated with Esteem than were previously considered averages, although even these differences were surprisingly small. Summary, Three general conclusions resulted from these analyses of weighted and unweighted averages of self-concept ratings. First, the unweighted average of specific self-concept facets was much more positively correlated with esteem than was suggested by Hoge and McCarthy. Second, the Hoge-McCarthy conclusion that these weighted averages, based on individual importance ratings, perform more poorly than any other weighted or unweighted average is apparently an artifact of using unsealed scores that do not adequately represent the interactive hypothesis, which is the theoretical justification for their use. Third, the weighted averages based on individual importance ratings, subgroup importance ratings, and total group importance ratings perform no better than the unweighted averages. Hence, these analyses fail to support the usefulness of importance ratings in the weighting of specific components of self-concept to predict Esteem or theoretical models that propose that they will be useful.

1233

ings were entered on the first step, the 12 importance ratings were entered on the second step, and the 12 self-concept/importance cross products were entered on the third step. For the analysis of the multi-item scales, 63.5% (R = .797) of the variance in Esteem could be explained by the 36 variables, whereas 60.4% of the variance (R = .777) could be explained by only the set of 12 self-concept ratings. The set of 12 importance ratings entered on the second step contributed an additional 1.3%, and the 12 self-concept/importance cross products entered on the third step accounted for an additional 1.8% of the variance in Esteem. For the single-item scales, the set of 36 variables accounted for 49.9% (R = .706) of the variance in Esteem; the contribution of the self-concept ratings was substantial (45.3%), and the contributions of the importance ratings (2.7%) and serfconcept/importance interaction (1.9%) were small. These analyses provide statistically significantalbeit very weaksupport for both the discrepancy and interactive models.

Summary and Implications Influence of a Facet's Importance on Its Relation to Esteem


A variety of theoretical hypotheses, as well as common sense, posit that the effect of a specific facet of self-concept on esteem will depend on the facet's importance. William James (1890/ 1963) first proposed the hypothesis 100 years ago; it has been restated frequently; it seems intuitively plausible; and it appears to be easy to test. Nevertheless, rigorous tests of the hypothesisindeed, even clearly articulated accounts of how it should be testedare surprisingly rare. Hoge and McCarthy, in one of the most recent attempts, found little or no support for the hypothesis, but an examination of their study suggested meth* odological problems. The present investigation verified that methodological problems did exist and devised solutions to the problems, but it still found little support for the hypothesis. However, the theoretical notion has too much intuitive appeal to be completely rejected, and so further examination of the issues is needed. One potential limitation of the present investigation is that subjects were mostly late adolescents and young adults, and more than 60% were female. The findings are generally consistent with Hoge-McCarthy findings based on a younger sample of high school students (45% female), but the study needs to be replicated with a sample of older adults.

General Test of the Interactive and Discrepancy Models


In order to better test the interactive and discrepancy models, a series of hierarchical multiple regressions was used to predict Esteem from the self-concept ratings, importance ratings, and the self-concept/importance cross products. In separate analyses of each self-facet, the self-concept rating (if statistically significant) was entered on the first step, the importance rating (if statistically significant) was entered on the second step, and the self-concept/importance cross product (if statistically significant) was entered on the third step. The results of these multiple regressions (Table 6) are summarized in terms of the change in R at each step of the analysis and the part correlations (see Nie et al., 1975) between each variable and Esteem in the final regression equation. Consistent with earlier findings, the main effect of self-concept on Esteem was substantial, positive, and statistically significant for most of the facets. Despite the large sample size, the main effect of the importance ratings was statistically significant (p < .05) for only 10 of the 24 analyses (based on the 12 multi-items scales and the 12 single-item scales), and no part correlation was larger than .21. However, consistent with the discrepancy model, all the significant effects were negative. The self-concept/importance interaction was statistically significant for only 7 of the 24 analyses, and none of these part correlations exceeded .18. However, consistent with the interactive model, the direction of all these significant effects was positive. For only 2 of the 12 facets, the Physical and Spiritual facets, the effects of the importance ratings and the self-concept/importance interactions were consistently larger than was the main effect of self-concept ratings (which failed to reach statistical significance after taking into account the other two effects). Hence, although support for the interactive and discrepancy models was weak across the set of 12 facets, support was stronger for the two facets in which importance ratings were expected to have the largest impact.' A similar hierarchical multiple regression was conducted to assess the interactive and discrepancy models across all 12 selffacets simultaneously. In this analysis, the 12 self-concept rat2

In an alternative set of tests of the interactive and discrepancy models, all the self-concept scores and the importance ratings were divided into four categories. Using esteem as the dependent variable, a 4 (level of self-concept) by 4 (level of importance) ANOVA was conducted separately for each facet. Even though the degrees of freedom for main effects and for the interaction effect in this ANOVA were larger than for the multiple regression analyses summarized in Table 5, the percentage of variance explained by the ANOVA was smaller for nearly every facet. This suggests that systematic variance was lost in translating the scores into discrete categories in order to perform the ANOVA, and it also suggests that no appreciable amount of variance could be explained by nonlinear terms in either the main effects or the interaction term (such nonlinear effects were included as part of the variance explained by the ANOVA but not by the regression analysis).

1234
Table 6 The Effect

HERBERT W. MARSH

of Each Self-Concept Score (Scales and Summary Ratings), Each Importance Rating, Interaction on Esteem Self-concept Importance Partr Change in R2 Part r Cross product Change in R2

and the Self-Concept/Importance

Facet Physical Scale Sum Appearance Scale Sum Opposite Sex Scale Sum Same Sex Scale Sum Parents Scale Sum Emotional Scale Sum Spiritual'1 Scale Sum Honesty Scale Sum Verbal Scale Sum Math Scale Sum Academic Scale Sum Problem Solving Scale Sum All 12 facets0 Scale Sum

Rirtr"

Change in R2 "

R2'

-.08 -.03
49*

.066** .062** .233** .167** .248** .166** .216** .120** .113** .092** .362** .264**

-.21** -.18** -.08** -.14**

.009** .006** .006* .020**

.18** .16**

.031** .026**

.106** .094** .240** .187** .248** .171** .216** .143** .113** .092** .370** .264"
.027" .027** .052" .081" .118" .113" .015" .020" .106" .063" .178" .111** .635" .499**

.43** .50** .41**


.46"

-.07* -.14**
-.15** -.14**

.005* .013**

.010**

.03
.34** .30** .30** .51**

.11**

.09**

.009**


.23** .28** .34** .34**
.12** .14**


.052** .081" .118** .113** .015** .020** .106** .063** .166** .096** .604** .453*

.022** .019**

.16** .16**

.005* .008**


-.12** -.10**


.012** .012** .013** .027**

.06*

.004*

.32** .25** .419** .038

.018** .019**

Note. A series of hierarchical multiple regressions was conducted in which each self-concept, the corresponding importance rating, and their cross product was used to predict esteem. The self-concept score was entered in the first step, and the statistical significance of the change in R2 attributable to the inclusion of the importance rating was tested in the second step. In the third step, the change in R2 owing to the cross product (from the R2 attributable to the self-concept and importance rating, or just attributable to the self-concept rating when the contribution of the importance rating was not statistically significant) was tested. The change in R2 and part correlation are presented for each step that was statistically significant (see Table 4 for corresponding correlation coefficients). The part correlation between the specific self-concept rating and esteem when all statistically significant variables are in the regression equation. b The increment in R2 that occurs for each particular step (i.e., the change in R2 values sum to the R2 for the entire regression equation). c The R2 value based on all variables that were statistically significant. " Only for multiple regressions based on the single-item and multi-item Spiritual facets did the self-concepts fail to correlate significantly with Esteem, and thus, they were not included in the regression equation. The importance ratings for the Spiritual facet also failed to enter significantly on the first step. However, once the self-concept/importance cross product entered the equation the importance ratings accounted for a statistically significant portion of the variance. 'These Rs resulted from a hierarchical multiple regression in which all 12 self-concept scores were entered first, followed by the 12 importance ratings on the second step, and then the 12 cross products on the third step. Significance tests were conducted for the change in R2 at each step. *p<.05. **p<.OI.

A potential problem with the present investigation was its use of single-item responses to assess the importance of each facet. The importance ratings were psychometrically weak. Asking subjects to rate the importance of each item in the multi-item

scales would provide better importance scores. However, none of the weightings considered in the present study, not even the random weighting, really made much difference. Furthermore, importance ratings averaged across all the responses in various

SELF-CONCEPT subgroups were more reliable and there were systematic group differences in the importance ratings, but weighted averages based on these subgroup importance ratings did not perform better than the unweighted averages. Hence, it seems unlikely that obtaining better importance ratings would substantially alter the weighted averages or improve support for the hypothesis. Ironically, another possible problem is that the SDQ III is designed too well; it only includes facets that are at least reasonably important to most people, and it represents most of the facets considered to be important by the respondents. None of the facets was given a mean importance rating of less than 5, and all but one had means between 6 and 8 on a 9-point response scale. In contrast to the broadly denned, generally important facets that appear on the SDQ III, the two facets originally suggested by William James (1890/1963) to illustrate his proposal were skills as a psychologist, his profession, and skills at Greek language, an area in which he had "no pretensions." Consistent with James's original formulation, support for the proposal might be stronger if the "facets" were composed of narrowly denned characteristics that most respondents judged to be unimportant but that a few respondents found to be very important. Of the SDQ III facets, the Spiritual and Physical Ability facets seem closest to this description; these facets had the most variable importance ratings and were also the facets that provided the best support for the interactive proposal. Although this counterexplanation is worth pursuing, it has problems as well. First, an instrument composed of many diverse, generally unimportant facets would probably make a poor selfconcept instrument (see the discussion of the conglomerate approach to defining general self that was discussed earlier and that is presented in more detail by Marsh & Shavelson, 1985). Furthermore, unless there were a huge number of such narrowly denned facets, each assessed with multiple items, then neither the weighted nor the unweighted averages of responses to such an instrument would be likely to correlate with Esteem as highly as the unweighted average of responses to more broadly defined facets, such as those from the SDQ III. Second, this "problem" is predicated on the assumption that there is insufficient response variability in the importance ratings. In fact, the variability in single-item importance ratings for each facet is nearly the same as the variability in the corresponding self-concept ratings, and the self-concept ratings do not appear to suffer from inadequate response variability.

1235

use of the entire SDQ III, then the use of the most relevant scales, perhaps the general Esteem scale, and the set of 12 summary rating items may be an expedient compromise. Selectivity hypothesis. According to the selectivity hypothesis, subjects selectively assign the greatest importance to those areas of self-concept in which their self-perceptions are most positive and assign the least importance to areas in which their self-perceptions are poorest. The positive correlation between self-concept and importance ratings (Table 4) is consistent with the hypothesis even though other explanations seem to be viable. Although a substantial relation between importance ratings and self-concept ratings is often interpreted as support for using the importance ratings to weight the self-concept responses, this conclusion was shown to be unjustified. Further research is clearly needed to test the assumption of causality that is implicit in the selectivity hypothesis and the implications of the hypothesis for the interactive and discrepancy models. Role of specific self-concept facets. Implicit in the Hoge and McCarthy study, and in the theoretical underpinnings that stimulated the present study, is the assumption that general selfconcept is more important than specific facets. The assumed role of specific self-concepts, weighted or not, was to provide a basis for defining general self-concept. However, this appears to be a historical weakness in self-concept research. Elsewhere (Marsh & Shavelson, 1985) I have argued that self-concept cannot be adequately understood if its multidimensionality is ignored. If the role of self-concept research is to better understand the complexity of the self, to predict diverse behaviors, and to relate self-concept to other constructs, then measures of multiple facets are more useful than a general facet. Although specific facets such as those measured by the SDQ III are hierarchically ordered, the hierarchy is so weak that the general self that appears at the apex accounts for only a small portion of the variance in the specific facets (Marsh, in press-a; Marsh & Shavelson, 1985). This situation appears to be analogous to the one in intelligence/ability testing where many researchers now argue against the sole reliance on IQ. Although the general constructs in both areas of research may be relevant for the very young, specific facets become more important as individuals grow older and the specific facets become more differentiated. Ironically, because the hierarchy of specific facets in self-concept research appears to be weaker than in intelligence/ability research, the reliance on a general construct is less justifiable in self-concept research even though it appears to be more prevalent.

Other Issues
Several other issues, although not the primary focus of the present investigation, were also addressed. These include the use of single-item scales, support for the selectivity hypothesis, and status of the general self construct. Use of single-item scales. Single-item scales designed to represent the SDQ III facets were found to have poorer psychometric properties than did the multi-item scales. This is consistent with findings in other areas of research. Nevertheless, there was modest support for the psychometric properties of the singleitem self-concept ratings. Thus, although it is better to use the 136 SDQ III items than the 12 summary items, it may be better to use the 12 summary items than not to consider multiple dimensions of self-concept. If external constraints preclude the

References
Byrne, B. (1984). The general/academic self-concept nomological network: A review of construct validation research. Review of Educational Research, 54, 427-456. Cohen, J., & Cohen, P. (1975). Applied multiple regression/correlation analysis for the behavioral sciences. Hillsdale, NJ: Erlhaum. Coopersmith, S. (1967). The antecedents of self-esteem. San Francisco: Freeman. Dusek, J. B., & Flaherty, J. F. (1981). The development of self-concept during the adolescent years. Monographs of the Society for Research in Child Development, 46(4, Serial No. 191). Fleming, J. S., & Courtney, B. E. (1984). The dimensionality of selfesteem: II: Hierarchical facet model for revised measurement scales. Journal of Personality and Social Psychology, 46, 404-421.

1236

HERBERT W. MARSH masculinity and femininity as a function of women's involvement in athletics. Sex Roles. Marsh, H. W, & O'Niell, R. (1984). Self-Description Questionnaire III (SDQ III): The construct validity of multidimensional self-concept ratings by late-adolescents. Journal of Educational Measurement, 21, 153-174. Marsh, H. W., Richards, G., & Barnes, J. (1986). Multidimensional selfconcepts: The effect of participation in an Outward Bound program. Journal of Personality and Social Psychology, 50, 195-204. Marsh, H. W, & Shavelson, R. J. (1985). Self-concept: Its multifaceted, hierarchical structure. Educational Psychologist, 20, 107-125. Nie, N. H., Hull, C. H., Jenkins, J. G., Steinbrenner, K., & Bent, D. H. (1975). Statistical package for the social sciences. New York: McGraw-Hill. Rosenberg, M. (1965). Society and the adolescent self-image. Princeton, NJ: Princeton University Press. Rosenberg, M. (1979). Conceiving the self. New York: Basic Books. Rosenberg, M. (1982). Psychological selectivity in self-esteem formation. In M. Rosenberg & H. B. Kaplan (Eds.), Social psychology of the self-concept (pp. 535-546). Arlington Heights, IV. Harlan Davidson. Rushton, J. P., Brainerd, C. J., & Pressley, M. (1983). Behavioral development and construct validity: The principle of aggregation. Psychological Bulletin, 94, 18-38. Shavelson, R. J., Hubner, J. J., & Stanton, G. C. (1976). Validation of construct interpretations. Review of Educational Research, 46, 407441. Shavelson, R. J., & Marsh, H. W. (1986). On the structure of self-concept. In R. Schwarzer (Ed.), Anxiety and cognitions (pp. 305-330). Hillsdale, NJ: Erlbaum. Scares, L. M., & Scares, A. T. (1982, July). Convergence and discrimination in academic self-concepts. Paper presented at the meeting of the 20th Congress of the International Association of Applied Psychology, Edinburgh, Scotland. Taylor, M. C., & Hall, J. A. (1982). Psychological androgyny: A review and reformulation of theories, methods and conclusions. Psychological Bulletin, 92, 347-366. Watkins, D. (1978). The development and evaluation of self-esteem measuring instruments. Journal of Personality Assessment, 42, 171182. Wells, L. E., & Marwell, G. (1976). Self-esteem: Its conceptualization and measurement. Beverly Hills, CA: Sage. Wylie, R. C. (1974). The self-concept (rev. ed., Vol. 1) Lincoln: University of Nebraska Press. Received January 1985 Revision received March 7, 1986

Hall, J. A., & Taylor, M C. (1985). Psychological androgyny and the Masculinity X Femininity interaction. Journal of Personality and Social Psychology, 49, 429-435. Harter, S. (1982). The Perceived Competence Scale for Children. Child Development, 53, 87-97. Harter, S. (1983). Developmental perspectives on the self-system. In P. H. Mussen (Ed.), Handbook of child psychology. (Vol. 4, 4th ed., pp. 275-385). New York: Wiley. Harter, S. (1984). Processes underlying self-concept formation in children. Unpublished manuscript. Harter, S. (in press). Processes underlying self-concept formation in children. In J. Suls & A. Greenwald (Eds.), Psychological perspectives on the self (Vol. 3). Hillsdale, NJ: Erlbaum. Higgins, E. T, Klein, R., & Strauman, T. (1985). Self-concept discrepancy theory: A psychological model for distinguishing among different aspects of depression and anxiety. Social Cognition, 3, 51-76. Hoge, D. R., & McCarthy, J. D. (1984). Influence of individual and group identity salience in the global self-esteem of youth. Journal of Personality and Social Psychology, 47,403-414. Jackson, S. (1984). Athletic or antisocial: The female sports experience. Psychological determinants and consequences of female sports involvement. Unpublished bachelor's thesis, University of Sydney, Australia. James, W. (1963). The principles a/psychology. New York: Holt, Rinehart & Winston. (Original work published in 1890) Lubinski, D., Tellegen, A., & Butcher, J. N. (1983). Masculinity, femininity, and androgyny viewed and assessed as distinct concepts. Journal of Personality and Social Psychology, 44.428-439. Marsh, H. W. (in press-a). The hierarchical structure of self-concept and the application of confirmatory hierarchical factor analysis. Journal of Educational Measurement. Marsh, H. W. (in press-b). Masculinity, femininity and androgyny: Their relations with multiple dimensions of self-concept. Multivariate Behavioral Research. Marsh, H. W, Barnes, J., Cairns, L., & Tidman, M. (1984). The SelfDescription Questionnaire (SDQ): Age effects in the structure and level of self-concept for preadolescent children. Journal of Educational Psychology, 75, 940-956. Marsh, H. W, Barnes, J., & Hocevar, D. (1985). Self-other agreement on multidimensional self-concept ratings: Factor analysis and multitrait-multimethod analysis. Journal of Personality and Social Psychology, 49, 1360-1377. Marsh, H. W., & Jackson, S. (in press). Multidimensional self-concepts,

Potrebbero piacerti anche