(1) 199697871-001 199697871-001

8/12/2019 (1) 199697871-001 199697871-001

1/35

MeasurinaSelf-concipt:The ConceptualIssues

he study of selfconcept has a long history in the field of socialT cience research (see, e.g., Wells Marwell, 1976; Wylie, 1974).Itis valued as a desirable outcome in many psychological and educationalsituations and is frequently posited as a mediating variable that facili-tates the attainment of other desired outcomes, such as academic per-formance and social competence (Markus Wurf, 1987). Despite awealth of research bearing on substantive issues related to the topic,systematic reviews have revealed inconsistent and indeterminant find-ings, with methodological weaknesses being cited as the major problems(see, e.g., Byme, 1984; Hansford Hattie, 1982; Hughes, 1984; Shav-elson et al., 1976; West et al., 1980; Wylie, 1974, 1979).

The intent of this chapter is to examine these shortcomings in self-concept research. In particular, I first review problems of definition,related to the customary use of self terms. I then note how problemsof definition at the conceptual level lead ultimately to methodologicalproblems at the measurement level, thereby hindering construct validityresearch bearing on both the measuring instrument and the theoreticalmodel of selfconcept to which it is linked. Next, I describe seven em-pirically testable theoretical models of selfconcept and i d e n q whichselfconcept measures in the present volume are linked to each. Finally,I address the issue of importance/discrepancy ratings of selfconcept.We turn first to the issue of selfconcept definition.

Definitional ConsiderationsProblems of definition related to selfconcept research are now widelyknown. Essentially, these difficulties can be related to five major factors:

1

8/12/2019 (1) 199697871-001 199697871-001

2/35

2 MEASURINGLF

8/12/2019 (1) 199697871-001 199697871-001

3/35

CONCEPTUAL lSSUES

elements common to most research. First, academic self-concept reflectsdescriptive (e.g., I like most school subjects) as well as evaluative (e.g.,I do well in most school subjects) aspects of self-perception. Clearly,a delineation of the descriptive and evaluative components of self-concept would represent a major contribution to the field. Nonetheless,despite a 20-year hiatus since Shavelson et al. (1976) first noted that adistinction between the description and evaluation had not been clari-fied, either conceptually or empirically, the feat has not yet beenaccomplished. (For one attempt to disentangle these two aspects of self-concept measurement, see Shepard, 1979.) Of course, these character-istics (descriptive versus evaluative aspects) apply also to the specificsubject-matter (e.g., math, reading, and science) and artistic (e.g.,drama, dance, and music) domains of academic self-concept as well asto nonacademic (e.g., physical, social, and emotional) self-concepts.

Streins (1993) second point is that definitions of academic self-concept tend to focus on self-perceptions of behavior (e.g., I enjoymost school subjects) rather than on feelings (e.g., I feel bad aboutmyself in school). Thus, Strein (1993) contended that the termacademic self-concept is more related to ideas such as self-perceptionsof competence (Harter, 1990b) or selfefficacy (Bandura, 1982) than toconstructs that might be more accurately labeled selfesteem, self-acceptance, or self-worth (p. 273; but see the discussion of selfefficacy later).Self-concept Versus Self-Eff icacyBecause of the increasing pervasiveness of studies in the educationaland psychological literatures concerned with links between selfefficacyand various aspects of academic achievement (e.g., Marsh, Walker,Debus, 1991; Norwich, 1987; Pajares Miller, 1994; Skaalvik Rankin,1995; Zimmerman, Bandura, Martinez-Pons, 1992) and the fact thattheir conceptual differences have not always been made clear (for re-search examples that exem pw such lack of clarity, see Pajares Mil-ler), I now offer a brief comparison between the terms seZjk@cuq andself-concept.Indeed, Bandura (1986) contended that although beliefs ofones confidence are integral to both self-concept and selfefficacy, thetwo represent different phenomena and thus should not be regardedas synonymous constructs. Rooted in social learning theory (Bandura,1977) which emphasizes reciprocal relations among environmental,behavioral, and personal factors-seZfeficuq represents peoplesjudgements of their capabilities to organize and execute courses of ac-

8/12/2019 (1) 199697871-001 199697871-001

4/35

tion required to attain designated types of performances (Bandura,1986, p. 391). Based on inferential processes that bear on prior perfor-mance attainment, reflected experiences, verbal persuasion, and emo-tional arousal, selfefficacy judgments represent the personal aspect ofthis triumvirate of factors; more specifically, these judgments mediateinterplay between the environmental and behavioral factors (Norwich,1987).

In his earlier writings, Bandura (e.g., 1981 [cited in Norwich,19871) contended that the global nature of selfconcept not only servedto undermine its power to explain behavior in particular situations butalso failed to take fully into account the complexity and diversity of self-efficacy judgments across different activities and situations. However,given the wealth of research that has shown selfconcept domains, aswell as subdomains, to be multidimensionally structured, Norwich ar-gued that this early perspective is less relevant today. Despite this moremultifaceted notion of selfconcept, Pajares and Miller (1994) main-tained that the construct still fails to attain the same level of specificityas sekfficacy. They argued that whereas selfefficacy is a context-spe-cific assessment of competence to perform a specific task, a judgementof ones capabilities to execute specific behaviors in specific situations(p. 194), selfconcept is not measured a t the same level of specificity;moreover, selfconcept incorporates beliefs of self-worth associated withones perceived competence (see also Strein, 1993). Within the contextof academic endeavors, for example, Pajares and Miller contended thatalthough self-concept udgments may be subject specific (eg., pertinentto science, history, math, and the like), they are never task specific, inthat they represent selfevaluations of perceived competence. In thissense then, selfconcept judgments, compared with those of self-efficacy,are more global and less context dependent. For example, whereas asubject-specific selfconcept test item might require the respondent toreact to the statement I am a good science student, the selfefficacyitem would require reaction to the statement I can solve this particularscience problem. As such, the two items tap different cognitive andaffective processes (Pajares c Miller, 1994).

Social comparison and frame-of-reference effects, implicit to theinternal/external model proposed by Marsh (1986d; see later in thischapter), are likely to have differential influences on selfconcept andself-efficacy responses (Marsh, Walker, Debus, 1991). For example, inresponding to selfconcept measures that elicit evaluation of capabilitiesin relation to a particular school subject, students typically make such

8/12/2019 (1) 199697871-001 199697871-001

5/35

judgments by comparing their own performance with that of their class-mates (an external comparison), as well as with their own performancein other subject areas (an internal comparison) hese dualcomparativeprocesses represent frame-of-reference effects. In contrast, selfefficacyjudgments focus on the students capabilities in relation to the specificcriterion items presented; as a consequence, these fkame-of-referenceeffects are minimized.

Although Schunk (1985) suggested that students may use the per-formance of classmates as a benchmark against which to determinetheir own likelihood of success (i.e., selfefficacy), Marsh, Walker, andDebus (1991) noted that even this minimal impact of the external com-parison process is only likely when the performance being tapped issufficiently novel or ambiguous that previous experience with a similartask is not possible. Thus, Marsh, Walker, and Debus contended thatselfconcept responses, compared with selfefficacy responses, are likelyto be more influenced by fkame-of-reference effects, in particular, theinternal comparison process. Marsh, Walker, and Debus pointed to theway in which selfconcept and selfefficacy responses are measured asthe primary factor contributing to this distinction between the two con-structs. (For a further elaboration of conceptual and empirical differ-ences between selfconcept and self-efficacy within an academic frame-work, see Marsh, Walker, Debus, 1991; Norwich, 1987; PajaresMiller, 1994.)Self-concept Versus Self-EsteemPerhaps the definitional issue that has generated the most debate inthe literature is that bearing on the discrepancy between self-conceptandseq+ssteem In general, researchers would seem to agree that conceptually,the two terms represent different aspects of the self system (e.g., Brint-haupt c Erwin, 1992; Hattie, 1992; Juhasz, 1985). Whereas self-conceptconnotes a relatively broad definition of the construct that includescognitive, affective, and behavioral aspects, seZfesteem is thought to be amore limited evaluative component of the broader self-concept term(Blascovich Tomaka, 1991; Wells Marwell, 1976). Indeed, Brint-haupt and Erwin have likened this distinction to the difference betweenseljkkseri tion and se l jkva lua t im

For Hattie (1992) and others (eg ., Rosenberg, 1965),however, thekey element distinguishing selfconcept from selfesteem was the extentto which one considered the attribute under study to be important. For

8/12/2019 (1) 199697871-001 199697871-001

6/35

6 MEASURINGSELF-CONCEPT

example, a student may describe him- or herself as not being very goodat sports and thereby exhibit evidence of low physical selfconcept. How-ever, if the student considers sports to be of little importance, his orher self-esteem will be totally unaffected. Hattie therefore consideredselfesteem to be closely linked to ones sense of self-worth. Accordingly,he posited that whereas the body (beautiful) was of the highest impor-tance to some, for others it might be more important to be academi-cally able, to have a happy family life, to gain respect from others, orto have a desirable personality (Hattie, 1992, p. 55).

Despite conceptual claims supporting the distinctiveness of self-concept and selfesteem, construct validity research to date (e.g., Marsh,1986b; Shepard, 1979) has been unsuccessful in providing empiricalevidence of such discriminability. Brinthaupt and Erwin 1992) andothers (e.g., Watkins 8c Dhawan, 1989) have linked this indeterminacyto the fact that most selfconcept research has relied on self-report mea-surements. Typically, self-report scales comprise items that elicit bothdescriptive and evaluative aspects of the self, thereby making it difficult,if not impossible, to tease apart the two constructs. As a consequenceof the inability to disentangle self-concept &om self-esteem, it is commonpractice for most researchers to use the two terms interchangeably(Hughes, 1984; Shavelson et al., 1976).

Recent research (Greenwald, Bellezza, 8c Banaji, 1988; Watkins 8cDhawan, 1989) has attempted to gain more insight into this discrimi-nability issue by using open-ended, in lieu of self-report, measures, be-cause they lend themselves to the examination of important aspects ofselfconcept that may be independent of selfesteem (W. J. McGuire 8cMcGuire, 1982). On the basis of responses from 101 college students,Greenwald et al. 1988) concluded that selfesteem is a pervasive com-ponent of measured selfconcept, even for measures that lack manifestesteem-related content (p. 43).

From a crosscultural perspective, collective findings reported byWatkins and Dhawan suggest that the degree of discriminability betweenselfconcept and selfesteem may be a function of the extent to whichparticular societal values have been internalized. Summarizing findingsfrom a comparison of American and Indian undergraduate students(Dhawan 8c Roseman, 1988), New Zealand (Watkins, Alabaster, Free-mantle, 1988), and Filipino (Watkins, 1988) high school students, Wat-kins and Dhawan contended that although the descriptive and evalua-tive aspects of self were strongly intertwined, they were definitely not

8/12/2019 (1) 199697871-001 199697871-001

7/35

synonymous; the distinction was more evident for the Asian (Indian andFilipino) than for the Western samples. This clearer distinction for thenon-Western samples was attributed to the cultural emphasis on group,rather than on individual, success. Encouraged by these findings, andin keeping with the argument that any construct loses its scientific utilityif it is defined too broadly (see Fleming Courtney, 1984), Watkinsand Dhawan argued for a clear need to distinguish between the termsself-concept and self-esteem Construct validity research concerned with thediscriminability of these two self terms is clearly in the embryonic stage;as such, it provides a rich source of challenges to be met through futurestudies of selfconcept measurement.

Prevalence of Informal Notions of Self-ConceptA final definitional problem bearing on self-concept measurement isthe tendency by researchers to substitute common-language notions(Blascovich Tomaka, 1991, p. 116) of the construct for more precise,academic connotations. Echoing this lament, Marsh (in press-a) veryaptly stated that selfconcept, like many other psychological constructs,suffers in that everyone knows what it is, so that many researchers donot feel compelled to provide any theoretical definition of what theyare measuring. Given that selfconcept is a hypothetical construct andtherefore not directly measurable, it is critical that its meaningfulnessand legitimacy be established through tests of its construct validity (atopic to be described in chapter 2).

Problems of definition at the conceptual level, of course, lead ul-timately to methodological difficulties at the measurement level. In-deed, definitional imprecision leads inevitably to at least three deficien-cies in the conduct of selfconcept research: (a) It retards scientificprogress by making the task of replication difficult, if not impossible;(b) it severely hinders the selection of appropriately valid self-conceptinstruments; and (c) it obscures the conceptual framework of the study,thereby making it difficult to link the research hypotheses and the in-terpretation of findings to a particular selfconcept theoretical model.Each of these factors bears importantly on the construct validity relatedto both the measuring instrument and the theoretical model. We turnour attention now to a review of postulated theoretical frameworks ofself-concept that can be tested empirically.

8/12/2019 (1) 199697871-001 199697871-001

8/35

8 MEASURINGELF-CONCEPT

Theoretical ConsiderationsIn any discussion of self-concept measurement, it is worthwhile to firstreflect on Harters (1990b) cogent caveat that we take care not to putthe methodological cart before the conceptual horse (p. 292). Be-cause there are many definitions of self-concept, as noted earlier, it iscritical that the researcher first clanfy the particular theoretical frame-work he or she wishes to adopt and then select the most appropriateself-concept measures accordingly. Perhaps as an offshoot of the increas-ing use of structural equation modeling in tests of Zonstruct validity(see, e.g., Bentler, 1978; Byme, 1994), it has now become commonpractice to refer to such conceptual frameworks as theoretical mo els (butsee Hattie, 1992, for a discussion of differentiation between the twoterms). Essentially, the term theoretical model conveys the notion that re-lations among the constructs embodying a particular theory (i.e., itsnornological network [Cronbach Meehl, 19551) can be modeled pic-torially, thereby providing a clearer conceptualization of the theory un-der study.

Although numerous theories of self-concept abound (for a com-prehensive review, see Hattie, 1992), he term theoretical model is typicallyreserved for those theories that have undergone empirical scrutiny.More specifically, the construct validity of the theory has been evaluatedstatistically by testing hypotheses that bear either on its postulated nclmological network or on the factorial structure of a measuring instru-ment developed within the framework of that particular theory. Before1980, very few selfconcept instruments could be linked to any concretetheory. Furthermore, of the self-concept instruments that were theoret-ically conceived, most were grounded in the notion that self-concept isunidimensionally structured (although see Sears, 1966, for an exampleof a multidimensional instrument: the Sears Self-Concept Inventory).As a consequence, most instruments were designed to measure only aglobal aspect of the construct (Harter, 1990b).

After 1980, however, and perhaps in answer to urgent pleas (Shav-elson et al., 1976; Wylie, 1974, 1979) for a cessation of substantive re-search albeit implementation of construct validity research in the areaof self-concept, there is now a wealth of evidence that substantiates themultidimensional nature of self-concept (e.g., Hattie, 1992; Marsh,1990b). In keeping with these findings, most measuring instrumentsdeveloped since 1980 are multidimensionally structured, and many areclosely tied to a particular theoretical model.

8/12/2019 (1) 199697871-001 199697871-001

9/35

In general, theoretical models of selfconcept can be polarized intotwo broad perspectives: those supporting the unidimensionality of theconstruct versus those supporting multidimensionality. Within these twobroad categories, and despite some inconsistency with respect to no-menclature, it seems safe to suggest that there are basically seven the-oretical models of self-concept (Byme, 1984; Harter, 1990b; Hattie,1992; Marsh Hattie, 1996; Strein, 1993); each, in turn, dictates arelated measurement strategy.

To facilitate a clearer conceptualization of these models, a sche-matic representation of those embracing a unidimensional perspectiveis shown in Figure 1.1 and of those embracing a multidimensional per-spective in Figure 1.2. Before turning to a descriptive review of each ofthese selfconcept models, a brief explanation of the symbolism used inboth figures is perhaps in order. In keeping with convention associatedwith covariance structure modeling, the ovals represent latent con-structs, whereas the boxes represent observed scores. Thus, in Figurel.la, for example, GSC depicts the latent construct of general self-concept, and the indexed letters Al,A2,A,, and ) represent observedscores on Items 1, 2,3, and 4, respectively, of the academic selfconceptsubscale. The single-headed arrows represent the influence of the un-derlying latent construct on the observed scores. In Figure l.la, then,the first arrow on the left represents the prediction of the observedscore on A, from general selfconcept. The tweheaded curved arrowsin Figure 1 . 1 ~ ~or example, represent correlations among the latentself-concept constructs. Finally, because the prediction of observedscores by underlying latent constructs is not without error, these resid-ual terms are typically indicated schematically by short, single-headedarrows pointing toward the related observed score. However, for sakeof simplicity and clarity in the present instance, these symbols have beenomitted in Figures 1.1 and 1.2. We turn first to a discussion of themodels that represent a unidimensional selfconcept structure.Unidimensional PerspectivesNomothetic ModelThe nomothetic model (Figure l.la) represents the oldest and mosttraditional view of self-concept, and was first labeled as such by L. MSoares and Soares (1983). Perhaps because the term nomothetic impliesgenerality or universality, Marsh and Hattie (1996) have chosen insteadto reference it as the unidimensional, general-factor model. Al

8/12/2019 (1) 199697871-001 199697871-001

10/35


Figure 1.1

Heuristic representations of self-concept models that embrace a unidimensional per-spective. GSC = eneral self-concept, ASC = academic self-concept, SSC = social self-scale, s1 s4 = Items 1 to 4 on a social self-concept scale, PI-P., = Items 1 to 4 on aphysical self-concept scale, GI G8 Items 1 to 8 on a general self-concept scale.concept, PSC = p ysical self-concept,A, -A4 = Items 1 to 4 on an academic self-concept

8/12/2019 (1) 199697871-001 199697871-001

11/35

CONCEPTUALSSUES1

Heuristic representationsof self-concept models that embrace a multidimensional per-spective. GSC = eneral self-concept, ASC = academic self-concept, SSC = social self-concept, PSC - aysical self-concept, ESC = emotionat self-concept, A,-% = Items 1to n an acaCr mic self-concept scale, S,-S4 = Items 1 to n a social self-conceptscale, PI-P4 = Items 1 to 4 on a physical self-concept scale, El-E4 = Items 1 to 4 onan emotional self-concept scale.

8/12/2019 (1) 199697871-001 199697871-001

12/35


Figure 1.2, continued

8/12/2019 (1) 199697871-001 199697871-001

13/35

though the nomothetic model considers selfconcept to be unidimen-sional, it argues for a unitary construct that is made up of overlappingfacets of information. In other words, this model dictates that items ona measuring instrument tap different content areas and that each begiven equal weight; item scores are then summed to yield one overallselfconcept score. In Figure l.la, for example, general selfconcept isshown to be a summative representation of academic, social, physical,and emotional selfconcepts. The nomothetic model thus makes theassumption that the single score derived from this additive combinationrepresents an accurate unidimensional reflection of an individualssense of self as it relates to the various facets of his or her life (Harter,1990b). Proponents of this perspective (e.g., Coopersmith, 1967; Marx

Winne, 1978; Piers Harris, 1964) have argued that given the heavydominance of an overall, global selfconcept factor, it is virtually impos-sible to differentiate among its subcomponents.

Since the early 1980s, the validity of the nomothetic model hasbeen challenged and severely criticized by researchers who support theopposing view that selfconcept is a multidimensional construct. Theircontention is that by simply combining all item scores together, onemasks important distinctions that individuals make in their self-perceptions of adequacy related to various aspects of their lives (Harter,1990b).Over the past decade, a substantial amount of construct validityresearch, most of which has been spearheaded by Marsh and colleagues(see Marsh 8c Hattie, 1996), has shown no support whatsoever for thenomothetic model. In contrast, these studies have demonstrated quiteclearly not only that selfconcept is a multidimensional construct butalso that it cannot be adequately understood unless this multidimen-sional structure is taken into account. Indeed, Marsh and Hattie havesuggested that conclusions drawn by supporters of the nomotheticmodel regarding the unidimensionality of selfconcept are rendered du-bious as a consequence of problematic measurement and statisticalanalyses.

In the light of a plethora of evidence that argues against the validityof the nomothetic model, it seems curious that this perspective shouldhave weathered the tests of time for so long. Strein (1993) has attributedthe longevity to its conceptual simplicity, its historical predominancein the field, and its support for the magic bullet view, that changes inselfconcept will alter a variety of situation-specific behaviors withouthaving to intervene in each specific situation (p. 274). In addition,the answer may lie with methodological factors that too are bound by

8/12/2019 (1) 199697871-001 199697871-001

14/35


the constraints of time. For example, it is only since the early 1980s thatapplications of highly sophisticated statistical procedures such as con-firmatory factor analyses have been used to test hypothesized factorialstructures of measuring instruments.

Self-concept instruments in this volume that are rooted in the no-mothetic model are the following: the Joseph Pre-School and PrimarySelf-concept Screening Test (Joseph, 1979, see chapter 4) ; the Piers-Harris Childrens Self-concept Scale (Piers Harris, 1964; chapter 5);the Dimensions of Self-concept scales (Form S: Michael Smith, 1976,chapter 6; Form H: Michael, Denny, KnappLee, Michael, 1984, chapter 7) ; and the Self-concept Scale for the Hearing Impaired (Oblowitz,Green, Heyns, 1991, chapter 8 .True Unidimensional ModelTo distinguish the unidimensional view of self-concept found in thework of Rosenberg (1979) fi-om that associated with the nomotheticmodel, I have termed the Rosenberg perspective as the true unidimen-sional model. Two schematic representations of this model are depictedhere. The first, Figure l . lb , represents a unidimensional structure astapped by items constituting a scale designed to measure only generalself-concept (e.g., the Rosenberg Self-Esteem Scale, 1965). The secondillustration (Figure 1. c) reflects the true unidimensional model withinthe context of a multidimensional scale (Harters, 1985b, Self-Perception Profile for Children); in contrast to the nomothetic model depictedin Figure l . la , general selfconcept is tapped by items comprising itsown subscale. Although Rosenberg acknowledged that individuals formselfevaluations of their adequacy in other areas of their life, in additionto possessing a general sense of worth as a person, he chose to focussolely on the latter. The Self-Esteem Scale (Rosenberg, 1965) was de-veloped within the framework of the true unidimensional model (seechapter 6, this volume). As such, the instrument yields a single scorethat represents only the level of an individuals overall (i.e., global) self-esteem (or general self-concept); it does not measure the extent towhich other dimensions of the construct filter into this global sense ofwell-being. Thus, as shown in Figure l.lb, each item GI-G,) is de-signed to measure general selfconcept.

As noted by Harter (1990b), it is critical to distinguish between thetwo unidimensional perspectives of self-concept as operationalized by,for example, Coopersmith (1967) and Rosenberg (1965) in the devel-opment of their measuring instruments. Whereas the Coopersmith mea-

8/12/2019 (1) 199697871-001 199697871-001

15/35

sure (the Coopersmith Self-Esteem Inventory) assumes that ones senseof global self-worth is a simple additive combination of item responsesthat tap attributes or competencies representing content-specific do-mains, the Rosenberg scale makes no such assumption. Rather, it mea-sures global selfesteem directly and makes no attempt to tap the morespecific self-perceptions, which, for Rosenberg, are quite likely com-bined in a very complex and obscure manner of which the individualis unaware. The Rosenberg Self-Esteem Scale is the only measuring in-strument in the present volume that is linked to the true unidimen-sional model.

We turn now to selfxoncept models representative of a multidi-mensional perspective, as schematically portrayed in Figure 1.2.Mu1 dimensional PerspectivesIndependent-Factor NodelThe theoretical perspective underpinning this model (see Figure 1.2a)was first proposed by A. T. Soares and Soares 1980); to date, theseresearchers appear to be its only supporters. Although they originallytagged it the taxonomic model, this label runs counter to a taxonomicmodel as interpreted by Marsh and Hattie 1996) on the basis oftheories of intelligence. The major thesis of the A. T. Soares and Soares1980) research was that the selfconcept is composed of multiple facets,

each of which is independent of all other dimensions; at worst, theyshould be only weakly correlated. According to A. T. Soares and Soares,multiple selfconcepts develop independently as a consequence of onesdaily experiences, capabilities, and interaction with significant others.Furthermore, they (L. M. Soares 8c Soares, 1986) argued against theexistence of both a global (i.e., general) self-concept and a correlatedhierarchical structure, as proposed by Shavelson et al. 1976), which isdescribed later.

The independent-factor model of selfxoncept was derived as a the-oretical framework within which to develop the Affective PerceptionInventory (API; A. T. Soares Soares, 1979), a +point semantic differ-ential scale designed to measure perceptions of self as a person, a stu-dent, and with respect to competency in English, math, science,social sciences, the arts (i.e., music and a rt), and physical education. Aninth subscale tapped perceptions of school, and it could be de-

1 . The Coopersmith Self-Esteem Inventory is currently published by Consulting PsychologistsPress.

8/12/2019 (1) 199697871-001 199697871-001

16/35

16 MEASURINGflPCONCEPT

scribed more as a measure of attitudes toward school than perceptionsof self within the school environment. Interestingly, little if any justifi-cation of a strong independent-factor model can be found in the self-concept literature; rather, recent research has provided ample evidenceto dispute all claims in its support. For example, not only have substan-tial correlations been identified among the API subscales (ByrneShavelson, 1986; Hattie, 1992; Marsh Hattie, 1996), but a secondaryanalysis of the correlation matrix reported by L. M. Soares and Soares(1983) yielded a hierarchical factor structure comprising two higherorder factors (Hattie, 1992). These findings, together with strong criti-cism of the statistical analyses used in the Soares and Soares research(Hattie, 1992; Marsh Hattie, 1996), argue strongly against the validityof the independent model of selfconcept. Indeed, Marsh and Hattie(1996) concluded that there was little or no support for the strongversion of the multidimensional, independent model. No selfconceptmeasure in the present volume is grounded in the independent-factormodel.Correlated-Factor ModelIn direct contrast to the independent-factor model, this theoreticalstructure allows the multiple, domain-specific self-concepts to be cor-related both among themselves (as shown in Figure 1.2b) and with aseparate facet of global selfconcept (as indicated in Figure 1 .1 ~ ) . nthe basis of developmental theory and findings from empirical research(see, e.g., Harter, 1983, 199Oc; Marsh, 1989, 1990b), he postulated spec-ificity of selfconcept facets will vary with age. The Self-perception Pro-file for Children (as well as the other Harter instruments, some of whichare described in chapters 4-8) exemplifies an assessment instrumentdeveloped within the framework of the correlated-factor model.

Harter (1990b) argued that an important aspect of the conceptualand empirical separation of domain-specific facetsof self-perceived com-petence from a global self-worth construct is that it enables the deter-mination of relations that the specific competencies bear to global self-worth. In other words, it allows one to specify the manner by which thevarious dimensions are weighted and combined in yielding an overallsense of self-worth (i.e., general selfconcept). Within the framework ofthe correlated-factor model, then, selfconcept is conceptualized alonga continuum of very specific to very global perceptions of ones com-petency, and these approaches are not necessarily mutually exclusive(Harter, 1990b).

8/12/2019 (1) 199697871-001 199697871-001

17/35

To the best of my knowlege, no construct validity research has en-deavored to test hypotheses directly bearing on the structure of the cor-related-factor model. Nonetheless, findings from some validity studiesof the Shavelson et al. (1976) hierarchical model (e.g., Marsh, 199Oc;Marsh, Byme, Shavelson, 1988; Marsh 8c Shavelson, 1985), as well asextended versions of that model (see Vispoel, 1995c) demonstratedweak evidence of hierarchical structure, thereby offering indirect sup-port for the correlated model. Furthermore, substantial construct valid-ity research abounds that has indirectly tested its postulated networkthrough the validation of assessment measures developed from thismodel. Research bearing on the selfconcept instruments developed byHarter and colleagues, for example, are a case in point.

Instruments in the present volume that are based on the corre-lated-factor model comprise the self-perception profiles developed byHarter and colleagues for preschool children (Harter 8c Pike, 1983;chapter 4) preadolescents (Harter, 1985b; chapter 5) adolescents(Harter, 1988d; chapter 6) ; college students (Neemann 8c Harter, 1986;chapter 7); adults (Messer 8c Harter, 1986; chapter 7); and childrenwith learning disabilities (Renick 8c Harter, 1988; chapter 8); as well asthe Body Esteem Scale (Franzoi 8c Shields, 1984; chapter 7).Compensatory ModelThis model (Figure 1 . 2 ~ nitially proposed by Marx and Winne (1980)argues that once a global selfconcept has been accounted for, remain-ing variation is explained by multiple bipolar facets (e.g., academic,social, and physical) that are inversely rather than proportionally re-lated; that is to say, these domain-specific facets can be negatively ratherthan positively correlated. For example, in Figure 1 2 ~ ~he correlationsof social and physical self-concepts with academic selfconcept are neg-ative, whereas all other correlations are positive. These negative corre-lations indicate either that low scores on academic selfconcept are as-sociated with high scores on social and physical selfconcepts (typicallythe case in research bearing on low academically tracked students) orthat high scores on academic selfconcept are associated with low scoreson social and physical selfconcepts (often found in research related toacademically pf ted students).

The major thrust of the compensatory model is that in the uncon-scious attempt to maintain ones sense of well-being, self-perceptions oflow status in one domain will be compensated by self-perceptions ofhigh status in other domains. With respect to low-ability students, for

8/12/2019 (1) 199697871-001 199697871-001

18/35


example, it would be argued that whereas they are likely to perceivethemselves as less capable academically, their perceptions of social com-petence are likely to be high, thereby balancing out their overall per-ceptions of self.

Although the basic rationale underlying the compensatory modelseems intuitively sound, research in support of this model (MarxWinne, 1978, 1980; Winne Marx, 1981; Winne, Marx, c Taylor, 1977)has been strongly challenged. Some researchers (Hattie, 1992; Shavel-son, Bolus, Keesling, 1983) have severely criticized their statisticalanalyses; others (e.g., Marsh Hattie, 1996) have denounced their useof the same data to support both unidimensional and compensatorymodels of selfconcept. Taken together, Marsh and Hattie concludedthat support for the compensatory model, as claimed by Marx, Winne,and colleagues, was more an artifact of the use of ipsative rating (i.e.,rank order) scales than a compensatory process underlying selfconceptstructure.As these authors (Marsh Hattie) noted, ipsative scores arenecessarily negatively correlated because ranking yourself more highlyon one scale inevitably means that rankings on the other scales mustbe lower.

Research bearing on Marshs (1986d) internal/external frame ofreference (I/E) model appears to provide substantially stronger andmore rigorous support for the compensatory model. This model sup-ports the notion that in addition to a natural inclination to comparetheir academic abilities with those of their peers in the same schoolenvironment, students also compare their own ability in o ne academicsubject with that in other school subjects. The model was originallyderived from construct validation of the Shavelson et al. (1976) modelof selfconcept, wherein Marsh (1986d; Marsh et al., 1988) uncoveredintriguing relations between English and math selfconcepts and be-tween English and math academic achievement. Specifically, Englishand math achievement correlated at approximately .5 to .8, whereas thecorrelations between respective selfconcepts were near zero. The I/Emodel was designed specifically to explain why English and math self-concepts are so distinct.

According to the I/E frame-of-referenceperspective, the formationof students perceptions of their own academic competence is based ontwo sets of comparisons: an external comparison by which students com-pare their abilities in particular subjects with the abilities of other stu-dents in their classroom or school and an internal (ipsative-like) com-parison by which students compare their own abilities in one subject

8/12/2019 (1) 199697871-001 199697871-001

19/35

CONCEPTUAl lSSUES 19

in relation to their ability in another subject. (Development of the I/Emodel, however, grew out of research related only to the areas of mathand English.)

For example (see Marsh, 1993a), suppose a student accurately per-ceives him- or herself as performing below average in both math andEnglish but is better at math than English. Although her or his mathskills are below average in relation to those of other students (externalcomparison), but higher than average in relation to her or his ownEnglish (internal comparison), the student may exhibit an average orabove-average math self-concept, depending on how these two factorsare weighted. Because math and English achievement test scores aresubstantially correlated, the external comparison should yield a positivecorrelation between their matching self-concepts. In contrast, becausemath and English achievements are compared with each other and itis the differential between these two scores that results in a higher self-concept in one area or another, the internal comparison should leadto a negative correlation. Depending on the relative strength of each,then, the join t operation of both sets of comparisons will lead to a near-zero correlation between English and math self-concepts.

A second aspect of the I/E model argues that the impact of mathachievement on English self-concept should be negative, and vice versa.That is to say, a high math self-concept is expected when a studentsmath skills are good compared with those of other students (external)and high compared with ones own English skills (in ternal). Holdingmath achievement constant, then, it is the difference between math andEnglish achievement that is predictive of math self-concept; high En-glish achievement serves only to weaken a high math selfconcept(Marsh, 1993a).

Marsh and colleagues have empirically tested the basic tenets ofthe I/E model with data from Australian (Marsh, Richards, 8c Barnes,1986b), American (Marsh, 1990a), and Canadian (Marsh et al., 1988)students at various levels of the educative system; near-zero correlationsbetween math and English selfconcepts were reported in all studies,thereby providing ample evidence in support of the I/E model.Although the consistently small correlation between math and Englishselfconcepts is inconsistent with the original Shavelson model, it isnonetheless consonant with the revised Marsh/Shavelson model(Marsh, 1993a). (For a more extended discussion of the I/E model,

2. The I/E model does not require that the correlation between math and English self-concepts be zero; however, the correlation must be substantially ess than the math-Englishachievement correlation Marsh, 1993a).

8/12/2019 (1) 199697871-001 199697871-001

20/35

20 MEASURINGELF CONCEPT

see Byrne, 1996; Marsh, 1990a, 1990b, 1993a.) Thus far, however, re-search bearing on the I/E model has been limited to relations betweenacademic self-concepts and academic achievement. Although this bodyof research holds great promise as a replacement for the earlier workof Marx and Winne (1978, 1980; Winne Marx, 1981; Winne et al.,1977) in support of the compensatory model, substantially more re-search that incorporates nonacademic facets of selfconcept is needed.

TaxonomicModelThis paradigm of self-concept structure was inspired by Guilfords(1969) model of intelligence. A unique aspect of this model is that thecomponents of intellect reflect the intersection of two or more facets,each of which has at least two levels (Marsh 8c Hattie, 1996). Given thecomplexity of theoretical structures that can result as a consequence ofthe variant number of self-concept dimensions, facets, and facetal levels,two possible formulations of the taxonomic model are illustrated (seeFigure 1.2d and e)

Although the term taxonomic was first used by A. T. Soares andSoares (1980) in labeling their proposed model of selfconcept, it seemsclear that their description of self-concept structure matched theindependent-factor model rather than the taxonomic model thatderived from Guilfords (1969) work. In sharp contrast to A. T. Soaresand Soaress (1980) description of the taxonomic model, Guilford(1985) recently noted that factors constituting his proposed structureof intelligence are postulated to be correlated and hierarchicallyordered. Marsh and Hattie (1996), in a review of selfconcept models,concluded that the A. T. Soares and Soares (1980) interpretation of thetaxonomic model was inappropriate.

Before trying to either explain or comprehend the structure of thetaxonomic model, I will highlight a critical distinction between the usualmeaning of the term multif ceted and its meaning in relation to thisparticular model. Using terminology associated with an analysis of var-iance ANOVA) design, Marsh and Hattie (1996) likened this distinc-tion to the difference between a one-way design (which reflects mostself-concept models) and a factorial design (which reflects thetaxonomic model). Within the context of the one-way design, therewould be one facet (eg., general self-concept) that has several levels(i.e., multiple selfconcept facets such as academic, social, and physical).In a factorial design, on the other hand, there would be at least twoselfconcept facets, each of which would have two or more levels.

8/12/2019 (1) 199697871-001 199697871-001

21/35

Two self-concept instruments that can be linked to the taxonomicmodel are the Tennessee Self-conceptScale (TSCS; Roid Fitts, 1988/1994) and the Multidimensional Self Concept Scale (MSCS; Bracken,1992); the TSCS is described in chapter 6, and the MSCS in chapter 5,of this volume. Turning first to the TSCS, we find an underlying struc-ture that reflects a taxonomic model having three facets. Applied to theClinical and Research Form of the scale (see chapter 6), these facetscan be further described as embracing a 5 (external frame of reference)X 3 (internal frame of reference) X 2 (positively/negatively wordeditems) design. The five levels of the external frame-of-reference facetare physical, moral, personal, family, and social self-concepts. Each ofthese traits is hypothesized as being manifested in relation to three in-ternal frames of reference (the second facet): identity (eg., what Iam), satisfaction (e.g., how I feel about myself), and behavior (e.g.,what I do, or how I act) . Identity represents the private internal self;satisfaction reflects a discrepancy between the actual and the ideal; andbehavior represents the external, observable self. The third facetrepresents the wording of the items, thereby providing a mechanismfor controlling various response biases; it may or may not be substan-tively important.The MSCS is also structured on a three-facet taxonomic model.The domain facet, reflecting context-specific self-concepts, has s x levels:social, competence, affect, academic, family, and physical. The secondfacet, evaluative perspectives, has two evels: a personal perspective (ac-quired directly through an individuals evaluations of his o r her ownbehavior) and a significant-other perspective (acquired indirectlythrough inferences about evaluations by others). The third facet, eval-uative standards performance, has four levels: absolute, ipsative, com-parative, and ideal standards. Additionally, the MSCS contains both pos-itively and negatively worded items that could be used to reflect a fourthfacet. Thus, in ANOVA terms, the MSCS exemplifies a 6 X 2 X 4 tax-onomic design.

Marsh and Hattie (1996) noted that scores that are based on thetaxonomic model are not always consistent with the design of theinstrument. The TSCS and MSCS serve to e x em p w this point. Forexample, the design of the TSCS allows for three manifestations of so-cial self-concept ( dentity, satisfaction, and behavior). However, becausethe combining of these three scores into a single social self-conceptscore results in a confounding of the three manifestations, this mode

8/12/2019 (1) 199697871-001 199697871-001

22/35


of scoring may be inappropriate (Marsh Hattie, 1996; Marsh Rich-ards, 1988b). In contrast, Marsh and Hattie (1996) noted that althoughBracken (1992) discusses all eight multiplicative combinations of thetwo perspectives and four standards, and despite the theoretical impor-tance of the perspectives and standards facets in the design of theMSCS, these facets are ignored in scoring the instrument; only s x scoresreflecting the domain facet are used.

On the basis of their evaluation of the TSCS and MSCS, then,Marsh and Hattie (1996) concluded that although the taxonomic modelmay be consistent with the underlying structure of some multidimen-sional self-concept measures, it may be quite inconsistent with theiryielded scores. Nonetheless, because the taxonomic model appearspromising as a potentially fruitful means to combining structural (i-e.,domain) and process components of self-concept, Marsh and Hattiehave called for future construct validity research that identifies ways bywhich to determine: (a) the most appropriate structure, (b) corre-sponding scores related to this structure, and (c) the manner in whichscores reflecting various combinations of the facets are differentiallyrelated to external criteria.Hierarchical ModelThe theoretical notion underpinning this model is that general self-concept is a higher order factor that comprises multiple, domain-specific self-concepts, which, although correlated, can be interpreted asseparate constructs.As such, general self-concept and each of its relateddomain-specific facets are tapped by items constituting each of theirseparate subscales. The heuristic representation of the hierarchicalmodel is shown in Figure 1.2f.

Although the conceptualization of self-concept as a hierarchicallyordered structure has been implicit in the writings of several theorists(e.g., S. Epstein, 1973; Kelly, 1955; LEcuyer, 1981; for a review, seeHarter, 1985a),Shavelson et al. (1976) were the first to propose a modelof self-concept that could be tested empirically. This model (commonlycited as the Shavelson model) portrayed a multidimensional and hier-archically ordered structure, with global perceptions of self as a person(i.e., general self-concept) at the apex and actual behavior at the base;moving from the top to the bottom of the hierarchy, the structure be-came increasingly differentiated. More specifically, global self-conceptwas shown to split into two branches: academic and nonacademic self-concepts. The nonacademic branch comprised three facets: social,

8/12/2019 (1) 199697871-001 199697871-001

23/35

CONCPTUAL ISSUES 3

physical, and emotional selfconcepts. Each of these four facets, in turn,was shown to branch in to separate and more specific self-concepts (e.g.,math, peers, and physical appearance). A schematic representation ofShavelson et al.s originally hypothesized hierarchical model is shownin Figure 1.3. However, because (a) the Shavelson model is the mostextensively validated model of self-concept to date and (b) the devel-opment of many recent measures of selfconcept is theoretically linkedto this hierarchical model, I consider it important that readers be in-formed (as much as is possible at this time) regarding the current statusof construct validity research related to this model. Accordingly, severaladditional schematic figures representing proposed adaptations of thismodel accompany this discussion.

Since its inception, the Shavelson model has undergone extensiveconstruct validation. However, until recent tests of the social (Byme cShavelson, 1996) and physical (Fox c Corbin, 1989; Marsh c Red-mayne, 1994) selfconcept components of the model, virtually all con-struct validity research designed to directly test its theoretical structurehas focused on the academic branch of the model. In contrast, manyconstruct validity studies have indirectly tested hypotheses related to theShavelson model. These have included tests for the validity of (a) mea-suring instruments theoretically linked to the Shavelson model (eg.,Bracken c Howell, 1991; Marsh c ONeill, 1984) and (b) multidimen-sional and hierarchical structure relevant to an extension of the modelthat included visual arts and music selfconcepts (Vispoel, 1 9 9 5 ~ ) .

Because construct validation of the Shavelson model has been a pproached from different perspectives, particularly with respect to thetesting of hierarchical structure, it is important to cia+ Shavelson etal.s (1976) intended meaning in their postulation of a multidimen-sional and hierarchically ordered selfconcept structure. Accordingly,multidimensionality implies that selfconcept facets, although intercorre-lated, can be interpreted as separate constructs. For example, with ref-erence to the academic portion of the model (see Figure 1.3), althoughgeneral selfconcept is expected to correlate with academic selfconcept,academic selfconcept with subject-specific sekoncepts, and generalselfconcept with subject-specific self-concepts, each of these dimensionsoperates as a separately interpretable entity. Hierarchical structure, onthe other hand, suggests that the strength of correlations between self-concept facets varies in a systematic pattern so that (a) general self-concept correlates highest with academic self-concept, next highest withsubject-specific self-concepts, and lowest with academic achievement

8/12/2019 (1) 199697871-001 199697871-001

24/35

Fge13

Gea

SC

Ga

N

micSc

A

mica

N

mic

Sc

-

_--________--_-

___________________----

Seo

Sc

Euoo

B

on

S

c

Suo

A

mic

Sa

Emoo

Sc

;

SC

Sc

P

c

Sc

Pca

Emoo

Sae

Theaccsuueosc

po

bS

sH

aSaoI9

omSC

Vdooc

u

inepeaobRJS

sJJH

aGCSao1

ewofE

oR

c44C

g1

bhAmecE

oR

cA

aoRnewhpmiso

8/12/2019 (1) 199697871-001 199697871-001

25/35

(i.e., actual behavior); (b) academic self-concept correlates higher withsubject-specific selfconcepts than with academic achievement; and (c)subject-specific selfconcepts (e.g., math selfconcept) correlate higherwith their matching academic achievement scores (e.g., math) than withnonmatching academic achievement scores (e.g., history).

In testing for the hierarchical ordering of academic selfconceptstructure, researchers have examined the pattern of latent factor cor-relations among selfconcept facets (Byrne, 1986; Byrne Shavelson,1986; Byrne Worth Gavin, in press; Shavelson Bolus, 1982) andhave compared the fit of differently specified higher order factor struc-tures (e.g., Marsh, 199Oc; Marsh et al., 1988; Marsh Shavelson, 1985).With respect to these two different approaches to testing for hierarchi-cal structure, note that the use of higher order factor analyses is onlystatistically possible when there are multiple first-order factors, as in thecase of self-concepts in relation to several academic subject areas (see,e.g., Marsh, 1990~).his contingency is directly related to the issue ofstatistical identification in the analyses of covariance structures (for anelaboration of this topic, see Bollen, 1989; Byme, 1994). Because statis-tical identification must be satisfied at each level of higher order modelspecification, third-order models are rare.Four primary conclusions from this validation work of academicself-concept are of particular import here. First, on the basis of an abun-dance of evidence across various populations, it seems clear that aca-demic selfconcept is multidimensionally structured. One interestingcomplication, however, is the replicated finding of a negligible corre-lation between math and English selfconcepts, albeit both facets arepresumed to be explained by the higher order factor of academic self-concept (which, by definition, assumes their substantial correlation).Paradoxically, whereas the correlation between math and English self-concepts has been found to be close to zero, the correlation betweenmath and English achievement has been relatively high. These findingshave been consistent across all three Self Description Questionnaireinstruments (Marsh, 1990b) as well as across other self-concept mea-sures (Byrne Shavelson, 1986; Marsh et al., 1988). In search of an-swers to this conundrum, Marsh proposed (a) a revised academic struc-

3. To evaluate the appropriateness of a third-order model and, thus, the validity of interpre-tations that are based on its structure, readers need to familiarize themselves with at leastthe basic elementso statistical identification.For a nonmathematical and simp lied intm-duction to the topic, and for an application of the basic assessment criteria t a second-order model, see Byrne (1994).

8/12/2019 (1) 199697871-001 199697871-001

26/35

26 MEASURINGELF-CONCEPI

ture for the Shavelson model, which he labeled the Mursh/Shuvelsonmodd (Marsh 8c Shavelson, 1985), and (b ) the I/E frame-of-referencemodel (Marsh, 1986d), which was described earlier.

Second, although the hierarchical structure of academic self-concept, as originally postulated by Shavelson et al. (1976) and notedearlier, has been shown to basically hold across age when the patternof latent correlations among selfconcept facets has been examined(e.g., Byrne 8c Shavelson, 1986; Byrne 8c Worth Gavin, in press; Shav-elson Bolus, 1982), this structure has been shown to differ slightlywhen multiple academic subjects are included in the model and it istested as a higher order structure (see Marsh, 199Oc; Marsh et al., 1988;Marsh Shavelson, 1985). This revised structure was incorporated intothe Marsh/Shavelson model. For the sake of clarification, both theorginal (Shavelson et al., 1976) and modified (Marsh Shavelson,1985) academic components of the hierarchical model are portrayed inFigure 1.4.

Third, substantial research has demonstrated that although aca-demic selfconcept becomes increasingly differentiated during preadelescence (Grades 2-5), these facets tend not to become more so in theadolescent years (e.g., Byrne 8c Worth Gavin, in press; Marsh, 1989,1990b).Finally, findings related to developmental changes in the hierar-chical structure of academic selfconcept are somewhat mixed. WhereasMarsh has reported this pattern to be relatively clear for preadolescents,albeit to gradually weaken with age (Marsh, 1990b), findings reportedby Byrne and Worth Gavin showed only minor deterioration in fit. How-ever, these differences are quite likely a function of the two differentapproaches taken in analyzing hierarchical structure, as noted earlier.Given multiple subject-specific self-concepts at the first-order level, themost rigorous test of hierarchical structure is to use higher order factoranalysis. Typically, findings from these analyses will render intercorre-lations among selfconcept facets to be smaller, hence the findings of aweaker pattern of hierarchical structure in the Marsh studies.

Construct validity research bearing on the nonacademic branch ofthe Shavelson et al. (1976) hierarchical model is only now beginningto appear in the literature. Focusing only on social self-concept, Byrneand Shavelson (1996) recently tested the validity of a multidimensionaland hierarchical structure for pre-, early, and late adolescents. Becausesocial self-concept structure, as originally depicted by Shavelson et al.(see Figure 1.3 , was intended only as a general representation of di-mensional structure, it was necessary to mod+ this portion of the

8/12/2019 (1) 199697871-001 199697871-001

27/35

CO) IPTUAL ISSUES7

ENGLISH HISTORYS.C. S.C.

Figure 1.4A

MATH SCIENCES.C. S.C.

GENERALACADEMIC

SELF-CONCEPT

B

MATH/ACADEMIC VERBAUACADEMICSELF-CONCEPT SELF-CONCEPT

PHYSICAL BIOLOGICAL ECONOMICS FOREIGNMATH SCIENCE SCIENCE BUSINESSS.C. s.c. s. c. s.c.SCHOOL GEOGRAPHY HISTORY LANGUAGES ENGLISH, ( , , , s S. ,I s . [ ,The original academic self-concept component (A) and modified academic self-conce tcomponent B) of the Shavelson, Hubner, and Stanton (1976) hierarchical model (Mars[,Byrne, Shavelson, 1988).Reprinted from Journal of Educational Psychology, 80,378.Copyright 1988 by the American Psychological Association. (S.C. = self-concept.)model to enable the testing of conceptually viable hypotheses. As such,global social selfconcept occupied the apex of the pyramid; it thenbranched into two subcomponents: social selfconcept (school) and so-cial selfconcept (family). The third tier of the hierarchy comprised the

more specific facets of social selfconcept; these were classmate andteacher social selfconcepts and sibling and parent social selfconceptsas subcomponents of the higher order facets of school and family, re-

8/12/2019 (1) 199697871-001 199697871-001

28/35


Figure 1.5

SELFCONCEPTGENEW)

SELFCONCEPTSCHOOC)SELFCONCEPTFAMILY)

Byrne and Shovelson s (in press) revised social self-concept component of the Shovelson,Hubner, and Stonton (1976 ierarchical model. Reprinted from Journalof Personalitynd Social Psychology (in press). Copyright 1996 by the American PsychologicalAssociation.

spectively. A pictorial display of this revised social selfconcept portionof the hierarchical model is shown in Figure 1.5.

On the basis of self, peer, teacher, and parent ratings of perceivedsocial competence for each of three age groups-preadolescents(Grade 3), early adolescents (Grade 7), and late adolescents (Grade11)-findings from this initial study of the social selfconcept compo-nent of the Shavelson model revealed a multidimensional social self-concept structure that becomes increasingly differentiated and a hier-

8/12/2019 (1) 199697871-001 199697871-001

29/35

mmkISSUES9

archical ordering that becomes better defined with age. Overall,findings were in keeping with both the Shavelson et al. (1976) conceptualization of self-concept structure and developmental processes thatunderlie self-concept formation (see, eg., Harter, 1983, 1988a, 1990b).

Recently, Marsh and Redmayne have begun a series of constructvalidity studies bearing on the physical selfconcept portion of the Shav-elson hierarchical model (Marsh, 1994a, in press-a; Marsh Redmayne,1994). Of importance in this research is Foxs (1990; Fox Corbin,1989) postulated hierarchical model of physical selfconcept, which in-corporates aspects of the Shavelson et al. (1976) model. In keeping withall applications of the latter, Fox (1990; Fox Corbin, 1989) proposeda hierarchical structure composed of general selfconcept at the apexand global physical self-concept at the next level, with specific facets ofphysical self-concept on the lower tier. Formal testing of this hypothe-sized structure is an important component in the present constructvalidity work being conducted by Marsh and Redmayne.

From studies of Australian boys and girls age 9-15 years (Marsh,1993c), adolescent girls age 13-14 years (Marsh c Redmayne, 1994),and high school students age 12-18 years (Marsh, 1994a), Marsh andRedmayne reported findings that supported the hierarchical model ofself-concept. That is, they validated that global physical self-conceptcould be subdivided into components analogous to the subdivision ofglobal academic self-concept into specific school subjects (see Figure1.3) he specfic physical self-concept components related to endurance,balance, flexibility, strength, and appearance.

Although another revision of the Shavelson model that involvedboth the academic and nonacademic branches of the model was origi-nally proposed by Song and Hattie (1984), it has only recently beentested by Hattie (1992)4 Song and Hattie made twomajor modificationsto the Shavelson model; these involved the division of (a) academic self-concept into achievement self-concept (perceived actual achievement),ability self-concept (perceived capabilities for achievement), and class-room self-concept (confidence in classroom activities); and (b ) non-academic self-concept into social selfconcept and self-presentation (self-regard) selfconcept; these, in turn, branched into family and peer self-

4. The focus of Song and Hatties study was t investigate relations among home environment,self-concept, and academic achievement, within the framework of a structural equationmodel. More specifically, they tested the impact of family structure and socioeconomicstatus on social self-concept, academic self-concept, presentation of self, and academicachievement, mediated through a composite variable termed psychological dra7acfmistics

8/12/2019 (1) 199697871-001 199697871-001

30/35

concepts and confidence in self and physical selfconcepts, respectively.In describing the Song and Hattie revision, Marsh and Hattie (1996)underscored two important differences from the original Shavelsonet al. (1976) formulation. First, physical self-concept is intended toemphasize physical appearance. Second, confidence, although relatedto emotional selfconcept, overlaps with general selfesteem. A sche-matic presentation of the Song and Hattie (1984) model is shown inFigure 1.6.

On the basis of self-concept responses from Korean and Australianadolescents and of a comparison of several second-order factor models,Hattie (1992) reported strong support for the postulated seven first-order factors; his findings therefore further substantiate the multidi-mensionality of selfconcept. Furthermore, although his analyses yieldedsupport for three second-order self-concept facets, this structure dif-fered from the one originally postulated. As such, classroom self-concept related more to the second-order social, rather than the aca-demic, selfconcept facet. In keeping with the original Shavelson model,however, Hattie found that the ability and achievement facets could beextended to include specific subject matter selfconcepts; in addition,the physical domain was found to represent both appearance and abilitycomponents (Hattie, 1992; Marsh 8c Hattie, 1996). (For an extendeddiscussion of each of the theoretical models discussed in this section,readers are referred to Marsh Hattie, 1996).

Finally, interesting research has recently begun in which the Shav-elson model is extended to incorporate self-concepts bearing on artisticareas such as dance, dramatic art, visual art, and music skills (see, e.g.,Vispoel, 1 9 9 5 ~ ) . ispoel compared several alternatively specified mod-els, with the initial intents (a) to determine the model that most opti-mally allowed for the integration of the artistic domains into the Shav-elson hierarchy and (b) to assess relations between artistic and otherself-concept facets. On the basis of work with self-concept responses for831 college students, he reported that dance, dramatic art, visual art,and music skills were best integrated into the hierarchy as componentsof a higher order artistic selfconcept factor that was distinct from aca-demic (math and verbal) and other nonacademic (physical, social, andmoral) higher order factors. Overall, findings from this research sub-stantiated the multidimensionality of self-concept structure but pro-vided only a modicum of support for a hierarchical ordering of self-concept facets.

As a follow-up to this important research, Vispoel (1995b) concen-

8/12/2019 (1) 199697871-001 199697871-001

31/35

Figure16

Caom

sc

Gea

sc

Ay

Aeme

Cd

P

c

sc

sc

sc

sc

ins

sc

sc

Mahmac

L

SaSue

NuaSe

sc

os

SaHes19resoohS

sH

aSao19heaccmoRneomJournalokuo

P

o71

C

g

bhAmecPocA

ao

i= 5E v

8/12/2019 (1) 199697871-001 199697871-001

32/35

3 MEASURIWGELF CONCEPT

bated on the music component of his initial model and subsequentlyproposed a theoretical model of music selfconcept that could be usedto guide future investigative inquiry related to this aspect of the finearts. With this postulated model, Vispoel hypothesized that music self-concept was organized, multifaceted, and hierarchically ordered andthat relations between music and higher order facets of selfconceptwere mediated to some degree by domain importance. As such, Vis-poels music self-concept model embraces the features of the Shavelsonmodel as well as the notions of domain importance advocated by Harter(1985b).

Although the Shavelson hierarchical model remains the most em-pirically tested and validated of all selfconcept models, it nonethelesshas not been without some modicum of criticism (see, e . g . , Harter,1983, 1985a, 1986; Hattie, 1992). Hattie suggested that the form of thehierarchical model was still equivocal and suggested two directions thatfuture research should undertake. The first of these relates to the matchbetween homogeneity of sample and appropriateness of theoreticalmodel. He argued, for example, that although a multidimensional, hi-erarchical structure of self-concept might be optimal for one group ofindividuals, a unidimensional structure might better represent anothergroup. Hatties second concern represented an extension of his firstone but related more directly to developmental considerations. For ex-ample, arguing that selfconcept was more unitary before the childreaches adolescence, Hattie (1992, p. 87) called for research that pre-dicted, from theoretical concerns, which persons were likely to haveselfconcepts that were multidimensional and hierarchical and whichones were likely to have selfconcepts that were unitary. In the light ofthe substantial volume of research that has addressed such developmental concerns in relation to the hierarchical model, however, thelatter call seems rather curious. Finally, Harters (1983, 1985a, 1986,199Oc) concerns would appear to echo those of Hattie, that the hier-archical model cannot be generalized to all. Arguing that individualsdiffer in the extent to which a particular structure of selfconcept isoptimal, she supports the incorporation of information that reflects theperceived importance of domains being targeted by the researcher.

In summary, the hierarchical model, to some extent, can incor-porate one of four models described earlier as a special case (Marsh,in press-a). For example, in keeping with the true unidimensionalmodel, it hypothesizes a global self-concept at the apex of the hierarchy.Thus, Marsh (in press-a) argued that support for the true unidimen-

8/12/2019 (1) 199697871-001 199697871-001

33/35

sional model can also reflect concomitant support for the hierarchicalmodel when the hierarchical structure is very strong. Conversely, su pport for a weak version of the independent model could be interpretedas support for the hierarchical model when the hierarchy is very weak.In fact, Marsh (in press-a) stated that only in extreme cases in whichcorrelations among the selfconcept factors approach the reliabilities ofthe factors, or are consistently close to zero, would support for the hi-erarchical model be dubious. Relatedly, support for the correlated-factor model, by implication (in terms of higher order factoring),means support for the hierarchical model. Finally, given the possibilityof a hierarchical ordering of levels in one or more facets (Marsh cHattie, 1996), support for the taxonomic model implies support for thehierarchical model. As noted by Marsh (in press-a), however, this ver-satility of the hierarchical model is both its strength and its weakness.It is a strength in the sense that it provides a broad framework withinwhich to study the structure of selfconcept; it is a weakness because itmay not be falsifiable. It is critically important, therefore, that any hy-pothesized hierarchical model of self-concept be specified in sufficientdetail to allow for a rigorous testing of its postulated structure (Marsh,in press-a).Because (a) the Shavelson et al. (1976) model, together withall its later revisions (e.g., Byrne Shavelson, in press; Marsh Shav-elson, 1985; Marsh et al., 1988; Song c Hattie, 1984), has met thesedemands of specification and (b) virtually all construct validity researchhas directly tested its theoretical structure using statistically sophisti-cated and rigorous methodological strategies, one c n feel confident inconstruct interpretations arising from this research.

Selfconcept instruments in the present volume that have been de-veloped from the Shavelson hierarchical model comprise the following:all Self Description Questionnaires (Marsh, 199Oc, 199213; chapter 5;Marsh, 199Oc, 1992c; Marsh, Richards, Johnson, Roche,. c Tremayne,1994; chapter 6; Marsh, 1992d; chapter 7); the Perception of AbilityScale for Students (Boersma c Chapman, 1992; chapter 5 , and thePhysical Self-Percep ion Profile Fox c Corbin, 1989).

importance-Discrepancy Ratings of Self-conceptHarter has been steadfast in her contention that any model of self-concept should incorporate the importance (or salience), as well as thevalence and strength of the domains, about which the individual is

8/12/2019 (1) 199697871-001 199697871-001

34/35

34 MEASURINGELF CONCEPI

asked to make a self-judgment.Thus, in the development of their seriesof selfconcept measures, Harter (1985b, 1 9 8 8 ~ )nd colleagues (HarterPike, 1984; Messer c Harter, 1986; Neemann c Harter, 1986; Renick

Harter, 1988) addressed this concern through the introduction ofimportance ratings and the actual/importance discrepancy approach toselfconcept scoring. Although this aspect of selfconcept measurementcertainly has its proponents (e.g., Vispoel, 1992, 1993b, 1995a), it alsohas its critics (e.g., Cronbach Furby, 1970; Marsh, 1986b; Wylie, 1974,1979, 1989). In the light of substantial debate on the topic over at leastthe past two decades, we turn now to a brief review of the issues, as theyrelate to this particular approach to self-concept measurement. How-ever, for a more extended discussion of the actual/importance discrepancy, as well as other approaches to the formulation of selfconceptscores, see Marsh (1986b, 1993e), Marsh and Hattie (1996), and Wellsand Marwell (1976); for a summary of empirical research designed totest these approaches, see Harter (1988a, 1988b, 1990a) Hattie (1992)and Marsh (198613, 1993e).

The notion of importance ratings and discrepancy scores is rootedin Jamess (1890/1963) contention that domain-specific sekvaluations,as the building blocks of global self-esteem, are integrated according totheir perceived importance, salience, certainty, and relations to ideals.On the basis of this theoretical perspective, one could conclude thatrelations between general and domain-specific selfconcepts (e.g., mathselfconcept) should be a function of (a) the importance that an indi-vidual associates with each domain, (b) the standard of excellence (orideal) that the individual sets for each domain, and (c) the certaintywith which the individual views each domain. Researchers who endorsethis view typically have respondents complete a global selfconcept (i.e.,self-esteem) scale (e.g., Rosenbergs [19651 Self-Esteem Scale) or sub-scale (e.g., Global Worth subscale of Harters [1985bl Self-perceptionProfile for Children [SPPC]) a multidimensional, domain-specific scale(e.g., Harters SPPC); and a rating scale designed to tap the perceivedimportance of each domain (e.g., the Importance Rating Form accom-panying Harters SPPC).

Taken together, the essence of Jamess perspective is that for in-dividuals who perceive themselves as competent in areas that theydeem to be important, general self-concept will be high; for those per-ceiving themselves as incompetent in these same areas, genera1 self-concept will be low. Thus, for domains regarded as unimportant, itmakes little difference whether perceptions of competence are high or

8/12/2019 (1) 199697871-001 199697871-001

35/35

(1) 199697871-001 199697871-001

Documents