The three dimensions of a persons self-concept are complexity, consistency, and

The self has long been of great interest to psychologists, first appearing prominently in the writings of James, 1890. James distinguished between two notions: the self as known and the self as knower, which he, respectively, termed the me and the I. Recently, Jamess terms have been translated into modern cognitive terms (Linville and Carlston, 1994); the notion of known self is conceptualized as the declarative knowledge we have about ourselves, whereas the notion of the knower self is conceptualized as procedural knowledge that directs our actions, thoughts and feelings. The analysis of the cognitive-informational nature of the self (e.g. Klein and Loftus, 1993) falls mostly within the realm of declarative self-knowledge (Linville and Carlston, 1994), with studies focusing on the content and the structure of this knowledge. Thus, from the pioneering work of Kelly, 1955 to more recent social-cognitive studies of the self (e.g. Markus and Wurf, 1987; Klein and Loftus, 1993), the self has been discussed as the organized set of knowledge that is acquired by each person about himself or herself.

Investigation of any knowledge structure or set naturally includes attention to the content of this set. Accordingly, research on the self as a knowledge structure first examined the content of self-knowledge (Markus and Wurf, 1987). Studies of the cognitive content of the self explore the declarative or the evaluative components of the self-concept (Campbell et al., 1996). One example of a focus on cognitive content is the attention given to the positive or negative valence of self-knowledge; this focus has been central for investigators examining the construct of self-esteem (e.g. Rosenberg, 1965). Another example is the focus on the specific beliefs about one self that may differentiate depressed and non-depressed individuals (e.g. Beck et al., 1979).

Until two decades ago, the empirical study of the self was limited to questions concerning content. This content focus was due to the prevailing view of the self-concept as unitary (e.g. Rogers, 1951; Allport, 1955). This view gave rise to a large body of research examining self-esteem ..which assumed that people have a single, global self-concept about which they feel good or bad (Linville and Carlston, 1994). Newer alternative structural models of the self draw on the general trends within psychology (e.g. Markus and Wurf, 1987) and sociology (e.g. Thoits, 1983) that see the self as multi-faceted rather than as unitary. Although psychologists (e.g. James, 1890; Kelly, 1955) and symbolic interactionists (e.g. Mead, 1934)have long favored such a multi-faceted approach, it was only with the advent of social-cognitive theory and methodology that the empirical study of a multi-dimensional self began (cf. Markus and Wurf, 1987). Consequently, self researchers began attending not only to the content of the knowledge but also to its structure, exploring self-knowledge as the multi-faceted set of information it is assumed to be.

Studies of structural properties of the self . . . refer to how the knowledge components or specific self-beliefs are organized (Campbell et al., 1996, p. 141). The last decade has brought a host of proposed structural characteristics of self-knowledge, and of studies examining these characteristics. For example, Linville (Linville, 1985, Linville, 1987) coined the term self-complexity (SC), which she operationalized as the dimensionality underlying the self-concept. Showers, 1992 discussed the property of compartmentalization, or the degree to which one uses partitioning of differently valenced self-knowledge into distinct categories. Donahue et al., 1993 have examined the integration of self-dimensions. Finally, Higgins and his colleagues (e.g. Higgins et al., 1986; Strauman and Higgins, 1987) have explored self-discrepancies, or the degree to which the real-self diverges from the ideal-self or the ought-self.

In accord with a social-cognitive view of the self (Markus and Wurf, 1987), Linville, 1985 model of SC assumes that the self is composed of multiple aspects. These aspects can include social roles, relationships, goals, future and past selves and so on. The SC model posits that there are individual differences in the number of such self-aspects. It also posits individual differences in the extent to which each of these self-aspects overlaps with other self-aspects with respect to content or characteristics. Thus, Linvilles SC model suggests that there are individual differences in both the number of self-aspects and the degree to which they overlap.

Linvilles earlier work on in-group heterogeneity and out-group homogeneity (e.g. Linville and Jones, 1980; Linville, 1982) led her to conclude that complex (i.e. heterogeneous) concepts evoke more moderate affective reactions or evaluative judgments, whereas concepts that are more simplistic (i.e. homogeneous) evoke evaluations and reactions that are more extreme. For example, Linville, 1982 found that the extremity of ratings of both favorable and unfavorable vignettes about an elderly target was significantly and negatively correlated with the complexity of the raters concept of elderly males. In two subsequent influential papers, Linville (Linville, 1985, Linville, 1987) offered evidence for the existence of this type of complexity effect in the field of self-knowledge or self-concept. In a two-week mood-diary study, Linville, 1985 found that participants whose self-concept was more complex did not differ from participants whose self-concept was less complex with respect to average mood rating, but did differ in the variance of their moods: participants who were higher in self-complexity had less variable moods.

Linville (Linville, 1985, Linville, 1987) suggested that these effects of complexity on mood or affective extremity could be explained by two emotional mechanisms that stem from individual differences in complexity. Specifically, these mechanisms reflect the two components of SC: quantity of self-aspects and overlap among them. First, Linville, 1985 assumed that when a person experiences a stressful event, it affects the self-aspect that is most pertinent to the stressor. She argued that for a person with numerous self-aspects (high quantity), the affected self-aspect is but one of many aspects and, therefore, that a relatively small proportion of the total self is affected. In contrast, a stressor will affect a great proportion of the self in persons who have fewer aspects in their self-concept. Thus, greater quantity is one mechanism that moderates the impact of stressors.

Individuals can also differ in the degree to which they maintain distinctions among their various self-aspects. Linville (Linville, 1987, p. 664) posited that higher overlap (and thus, lesser distinction) allows for a spill-over effect: feelings and inferences associated with the originally activated self-aspect spill over and color feelings and inferences regarding associated self-aspects. A stressor that affects one self-aspect will initiate a process of spill-over of activation to overlapping self-aspects; it cannot do so, however, if no overlapping aspects exist. Thus, little or no overlap is a second moderator of the impact of stressors. Defining SC as the confluence of high quantity of self-aspects and low overlap among them, Linville, 1987 suggested that SC serves as a buffer against stress-related illness and depression. Indeed, she proposed its antithesis, self-simplicity, as a diathesis for such ailments or disorders.

Linville (Linville, 1985, Linville, 1987) operationalized the SC construct using a trait-sorting task. In this task, participants are instructed to sort a set of trait words into meaningful categories or groups, such that each group consists of those traits that are descriptive of the participant in some aspect of his or her life. The choice of categories is therefore idiosyncratic, as is each individuals sorting of the traits into these categories. An individuals SC score (SC-D) is then computed using the dimensionality statistic, H (Attneave, 1959; Scott, 1969;see Appendix Afor a detailed explanation of this measure). Scott proposed this statistic as a measure of the number of dimensions a person uses to describe a knowledge domain, in this case the domain of ones knowledge about oneself.

Several studies have used SC-D to test the predictions of the Linville, 1987 SC theory. In particular, studies have investigated the role of self-simplicity, or low SC, as a possible diathesis for stress-related depression. Because of the two mechanisms discussed above, the proportion of aspects that are affected by a stressor and the magnitude of the spill-over of activation among the aspects, SC theory posits that individuals who are more complex should be buffered against strong reactivity to stress. In contrast, individuals who have a cognitively simpler concept of themselves should exhibit more extreme reactions to stressors or negative events.

Although the Linville, 1985 mood diary study demonstrated the cognitive simplicity–affective extremity effect, it was her subsequent study (Linville, 1987), adopting the more appropriate prospective panel design (cf. Barnett and Gotlib, 1988), that gave the most support to SC theory. The study found the hypothesized buffering effect of SC; the impact of an accumulation of stressors on stress-related physical and psychological distress, controlling for earlier symptoms, was moderated by high levels of self-concept dimensionality (Linville, 1987). Other investigators (e.g. Dixon and Baumeister, 1991; Niedenthal et al., 1992;Kalthoff and Neimeyer, 1993, study 1) have reached similar conclusions regarding the predictive validity of dimensionality. (Note that two other studies by Kalthoff and Neimeyer (1992, studies 2 and 3) failed to support Linvilles claims when using conceptually similar methodologies).

Nevertheless, several attempts to replicate the Linville, 1987 findings have failed to support the conclusion that low SC is a diathesis for depression. These studies have identified limitations in the validity, the reliability, or the internal consistency of the SC construct as it has been operationalized by Linville. For example, the validity of Linvilles SC-D measure and of the SC construct were called into question by the findings of both Koenig (1989) and Hershberger, 1990. Using prospective panel designs similar to that used by Linville, 1987,these studies failed to find the buffering effect of SC on the level of depressive symptoms experienced after exposure to a stressor. Such failures to replicate raise questions concerning the predictive validity of selfsimplicity as a diathesis for depression.

The reliability of SC-D has been challenged by the findings of several studies. Salovey, 1992 has shown SC-D to be susceptible to both negative and positive affective states. In addition, Salovey explicitly used SC-D as a measure of the (temporary) state of self-focused attention. Showers, 1992 found the SC-D score to vary with different compositions of the trait-word lists supplied to the participant, in the same self-descriptive sorting task conducted by Linville (Linville, 1985, Linville, 1987). Hence, the Linville, 1985 SC-D measure may simply be unreliable, reflecting fluctuations of mood or of self-focused attention (Salovey, 1992)rather than assessing a stable individual difference in the structure of self-knowledge. Although Linville, 1987 herself reported a relative temporal stability (test–retest r=0.7), these studies raise concerns about Linvilles claims of dispositional individual differences on the SC construct.

Last and most detrimental are the findings of several studies that call into question the internal consistency of the Linville, 1987 measure, SC-D, and the degree to which it actually reflects a single latent property (i.e. the complexity of the self-concept structure). Contrary to Linvilles claim of such a single latent property, several investigators have found poor internal consistency for SC-D and have argued for the existence of at least two latent factors. For example, Woolfolk et al., 1995 found the dimensionality of positive self-knowledge to differ from that of negative self-knowledge. In their study, measures of positive-SC and negative-SC were computed, each reflecting the dimensionality of knowledge of one valence while ignoring the information of the other valence. The two measures were found to play different roles in participants reactivity to stress. In fact, Woolfolk et al. found the complexity of negative knowledge to be positively, not negatively, related to affective extremity. (Woolfolk et al. also found strong effects of word-list composition on positive SC but not on negative SC; this can be seen as additional evidence for poor reliability and poor internal consistency of this measure.) In a similar vein, Morgan and Janoff-Bulman, 1994reported that only the complexity of positive knowledge buffered stress. In sum, it seems the assumptions that structure and content are orthogonal and, specifically, that the valence of the content is unrelated to complexity, have not been supported by a number of studies that have used SC-D.

The findings of poor internal consistency are particularly harmful to SC theory because the existence of a general factor underlying SC is central to Linvilles (Linville, 1985, Linville, 1987) other tenets. Other models of structural properties of the self (e.g. compartmentalization, Showers, 1992; self-discrepancies, Strauman and Higgins, 1987) posit an interaction between the actual knowledge and the way it is structured, and do not maintain that structural features have an effect on mood that is independent of content. In contrast, SC theory divorces structure from content altogether. Linville, 1987 hypothesizes that complex structure as measured by SC-D, buffers strong affective reaction, regardless of the actual traits used, and regardless of the traits valences. Yet as demonstrated by Woolfolk et al., 1995, among others, valence might have a strong impact on levels of SC. It is conceivable that other features of the knowledge (apart from valence) may affect the structure of the self-concept or at least interact with it.

The present study examined two issues. First, it provided an analysis of the internal consistency of SC-D. Second, the present study proposed two alternative measures of the mechanisms posited by SC theory and compared these alternative measures with SC-D. An explicit evaluation of the internal consistency of SC-D was deemed appropriate for several reasons. First, there have been no explicit reports of internal consistency estimates in the published SC literature. Second, as noted above, some of the replication attempts (e.g. Woolfolk et al., 1995, studies 1 and 5) have indicated a poor internal consistency; nevertheless, none of these studies discuss this phenomenon in psychometric terms. Even the clearest analysis to date of the internal nature of SC-D (the one carried out by Woolfolk et al., 1995) stops short of providing a detailed analysis of internal consistency and reports only the data regarding discrepancies between positive SC and negative SC. The third rationale for an internal consistency analysis of SC-D involves the poor showing of SC studies utilizing that measure, with respect to both reliability and predictive validity. Predictive validity for a purportedly unitary construct depends (among other factors) on internal consistency (Nunnally and Bernstein, 1994). An analysis of the internal consistency might help in the construction of alternative measures with which a greater validity would be achieved. Finally, and most importantly, an analysis of the internal consistency of the SC-D measure is a test of SC theory itself. Such a test will be able to examine the Linville (Linville, 1985, Linville, 1987) assertion concerning the orthogonal effects of content and structure within the self-concept and, in particular, the orthogonality of the effects of valence and of complexity.

SC theory (Linville, 1987) states that the complexity of the self-concept, irrespective of the specific content, plays a unique role in affect and judgment. If SC-D is a valid measure of structure (irrespective of content), its internal consistency should be high. Given the results of the studies described above, the first hypothesis of the present study was that internal consistency of this measure would be poor. In particular, the findings of Woolfolk et al., 1995 were expected to be replicated: the valence of self-knowledge was expected to hinder the internal consistency of the SC-D measure.

Although this first hypothesis examined the choice of the SC-D statistic suggested by SC theory (Linville, 1985), it did not directly explore the hypotheses of that theory. A second purpose of this study was to propose two alternative measures, both obtained from the same self-descriptive sorting task, and each reflecting one of the components of the SC construct, the quantity of self-aspects and the overlap among them. These two measures were obtained for comparison with the dimensionality measure used by Linville. They are seen as reflective of the two underlying processes which are thought to account for the self-complexity moderation effects. The quantity measure has been used in previous studies (e.g. Linville, 1985, Linville, 1987). The overlap, reflecting the spill-over process, was new to this study. The inclusion of these two measures together allowed us to examine the interrelation of the two component processes and to provide measures for tests of the interaction of the numerosity of aspects with the overlap among them.

Inter-correlations between the two component measures and the SC-D statistic were computed, and the internal consistency of the two component measures were compared to that of the SC-D statistic. It was hypothesized that the correlations of SC-D with the more specific measures of the underlying components (quantity and overlap of self-aspects) would not fit Linvilles (Linville, 1985, Linville, 1987) assumptions. It was also hypothesized that both specific measures would have greater internal consistencies than would SC-D.

Measures of structure

Three measures of structural properties, SC-D, NASPECTS and OL, were computed for each participant using the results of the trait sorting task. As noted above, SC-D is a measure of dimensionality (Scott, 1969) that has been used by Linville, 1987 to represent SC as a whole. The scores on SC-D in the present study (M=2.80, S.D.=0.58) were comparable to those reported by Linville, 1987. NASPECTS and OL are two alternative measures designed to represent the two components of SC: quantity of self

Discussion

The findings of this study support a measurement explanation of the problems in the consistency, reliability and validity surrounding self-complexity theory (Linville, 1985, Linville, 1987); they also suggest a measurement solution to these problems. The results indicate poor internal consistency and insufficient convergent validity for the frequently used dimensionality measure, SC-D. In turn, the study proposes two alternative measures of the components of SC and offers psychometric support

Acknowledgements

We would like to thank Gary Brown, Doug Medin and Jeff Sherman for helpful comments during the preparation of this manuscript. We are grateful to Cassandra Baxter and Kristen Davidson for help in conducting the study.