jpa_19_2_142.pdf

F. Gurin et al.: Construct Validation of the Self-Description Questionnaire II with a French SampleEJPA 19 (2), 2003 Hogrefe & Huber Publishers

Construct Validation of theSelf-Description Questionnaire

II with a French SampleFlorence Gurin1, Herbert W. Marsh2, and Jean-Pierre Famose3

1Universit Paris XI, France, 2University of Western Sydney, Australia, 3Universit Paris XI, France

Keywords: Adolescent self-concept, French validation, multidimensionality, hierarchy

European Journal of Psychological Assessment, Vol. 19, Issue 2, pp. 142150

Summary: This investigation is a French validation of the Self-Description Questionnaire (SDQ) II, an instru-ment derived from the Marsh and Shavelson model and designed to measure adolescent self-concept. Previoustheoretical and methodological considerations in SDQ research provided guidelines for the instrument within-construct validity. 480 students completed the questionnaire. Reliability and confirmatory factor analyses wereused to demonstrate support for the good psychometric properties, the well-defined 11 facets of the multidimen-sional self-concept, the two remarkably distinct higher order academic areas, and the weak hierarchical ordering.Third-order HCFA models resulted in a hierarchical general self-concept modestly related to most nonacademicself-concepts and rather weakly to academic self-concepts. The present research strongly supports the multifac-eted nature of self-concept, but cannot supply clear evidence for the usefulness of hierarchical representationsof adolescent self-concept.

Self-concept research has developed considerably in re-cent decades. It is the objective product of W. Jamessfindings on self-consciousness (James, 1909), the modelsof intelligence proposed by British and American re-searchers such as Spearman, Thurstone, Guilford, or Cat-tell, and advances in the application of factor analysis andrelated statistical techniques. The enhancement of self-concept is the ultimate outcome in many disciplines suchas educational, developmental, clinical, social, and sportpsychology (Marsh & Hattie, 1996). Henceforth, the aimof the psychologist is to describe and explain the complexstructure of the self in various environments, to predict apersons behavior, to produce measures for interventionscores, and to relate self-concept to other constructs.

The Marsh/Shavelson Model

In the 1970s a great deal of criticism was formulatedabout the methodological shortcomings and the ques-tionable quality of self-concept measurement instru-ments (Burns, 1979; Wells & Marwell, 1976; Wylie,1974, 1979). Conducting their research in the education-

al field, Shavelson, Hubner, and Stanton (1976) re-viewed and criticized prior self-concept research andcontended that it lacked sophistication in theory, mea-surement, and methodology. They addressed these criti-cal deficiencies and subsequently developed a particular-ly complete and unified self-concept theory. They posit-ed a self-concept construct identifying seven featuresthat described the model: organized, multifaceted, hier-archical, stable, developmental, evaluative, and differen-tiable. They elaborated a possible graphic representationof self-concept which was dominated at its apex by aGeneral-self, which in turn was divided into Academicand Nonacademic self-concepts. The latter separated in-to several different facets: Physical, Social, and Emo-tional self-concepts. Nevertheless, this theoretical modelby Shavelson et al. consistent and heuristic though itwas had not been empirically validated by its authors.In order to address this concern, Marsh (Marsh, 1990c,1993; Marsh & Craven, 1997) designed the SDQ instru-ments to measure different areas of self-concept for pre-adolescents (SDQ I), adolescents (SDQ II), and late ad-olescents (SDQ III). The results provided strong supportfor the multidimensionality of self-concept. And al-

DOI: 10.1027//1015-5759.19.2.142EJPA 19 (2), 2003 Hogrefe & Huber Publishers

though they were consistent with Shavelson et al.s as-sumption (1976) that self-concept was hierarchically or-dered, these findings failed to demonstrate the strengthof the hierarchy. While supporting most self-concept dis-tinctive features assumed in the Shavelson et al. con-struct, Marsh was subsequently led to develop a differenthierarchical self-concept structure. In this model the stu-dents Verbal and Math self-concepts were so dis-tinct that the two constructs could not be collapsed intoa single academic self-concept and had to be separated(Marsh & Shavelson, 1985). To provide a plausible ex-planation for that near-zero correlation, Marsh (1986)developed an internal/external (I/E) frame of referencemodel that proposed that students Math and Verbal self-concepts were formed on the basis of an external com-parison (with other students) and an internal comparison(between his/her subject-specific domains). These twotypes of comparison could be expanded to most academ-ic subjects (including P.E. or sports), and therefore illus-trated the potential breadth of applicability of the I/Emodel (Marsh & Hattie, 1996). The SDQ self-conceptresearch posits a General-self split into two academicself-concepts (Maths/Academic and Verbal/Academic),as well as two nonacademic self-concepts (Physical/So-cial) for preadolescents, and three nonacademic self-concepts (Physical, Social and Moral) for late adoles-cents.

To date, the hierarchy has been thoroughly tested onthe SDQ I and III self-concept models but has not beenexplored on the SDQ II responses.

The goal of the present investigation is, first, to estab-lish a reliable French version of the SDQ II and then tovalidate the construct presented in the revised model(Marsh & Shavelson, 1985), focusing exclusively onwithin-network studies. The internal structure of self-concept will be examined in order to provide support forits multidimensionality, to confirm the clear separationof the two higher-order academic self-concepts, and totest the hierarchical hypothesis that has previously beenexamined for young and late adolescents (Marsh, 1987;Marsh & Hocevar, 1985).

Methods

Sample

The sample consisted of 480 talented French students(264 males, 216 females) enrolled in grades 10 to 12;ages ranged from 15 to 17. The respondents attended ahigh school specialized in sciences and tended to comefrom upper-middle-class families.

The SDQ II Instrument

The SDQ II (Marsh, 1990b) was developed in Australiafor adolescents aged 12 to 17 years. This 102-item ques-tionnaire yields 11 scale scores assessing seven nonaca-demic scales: Physical abilities (Pabl), Physical appear-ance (Appr), Opposite sex (Osex), Same sex (Ssex), Par-ents (Prnt), Honesty/Trustworthiness (Hnst), Emotionalstability (Emot); three academic scales: Maths (Math),Verbal (Verb), General school (Schl); and one GlobalSelf-Esteem scale (Estm). Its factor structure was vali-dated in numerous prior studies on general populations(Marsh, 1990a,b; Marsh, Parada, & Ayotte, 2002; Marsh,Parker, & Barnes, 1985) as well as on gifted and talentedpopulations (Marsh, Plucker, & Stocking, in press). Theresponse format is based on a 6-point Likert scale. Reli-ability estimates for the 11 scales ranged from .83 to .91.Each a priori factor was identified through factor analy-sis. As to the relationships to the external criteria, theSDQ II was correlated with academic achievement indi-cators, age, gender, locus of control, self-attribution forsuccess or failure, physical fitness, participation insports, self-concept enhancement interventions andmental health. That previous research provided strongsupport for the construct validity of responses to theSDQ II. The instrument has already been tested in theUnited States (Marsh, 1994) and in Canada (Marsh et al.,2002), thus demonstrating its generalizability. Byrne(1996) introduced the SDQ II as the most validated ado-lescent self-concept measurement instrument with soundpsychometric qualities.

Translation

The current study is the first validation of the SDQ II inFrance. We followed the different steps described in thetranslating approach for psychological tests developedby Vallerand and Halliwell (1983) and Vallires and Val-lerand (1990). First, the original English anchor versionwas translated into French (the target language) by a firsttranslator. This version was then backtranslated intoEnglish (the source language) by another translator. BothEnglish versions proved globally similar from a seman-tic point of view, indicating a significant degree of equiv-alence between the anchor and the target versions. Sec-ond, the French version was evaluated by the research-ers. The following steps tested the clarity of the items andminor cultural differences. Next, a committee formed bysix translators and the researchers examined the Frenchtranslation and developed a pilot version that was admin-istered to 20 11th-grade students who were later inter-viewed individually on various aspects of the instrument.A final version was then constructed.

F. Gurin et al.: Construct Validation of the Self-Description Questionnaire II with a French Sample 143

EJPA 19 (2), 2003 Hogrefe & Huber Publishers

Statistical Analyses

The precision of a measurement includes two orderedtypes of analyses: the estimation of reliability and con-struct validity. This latter study can be broadly subdividedinto internal and external validities (Cronbach & Meehl,1955). In the course of construct validation, the researcherseeks empirical evidence in support of hypothesized rela-tions both among facets of the same construct (within-net-work studies) and among different constructs or otherexternal criteria (between-network studies). The presentstudy focused solely on the within-construct analysis. Theprocedures we used were: confirmatory factor analysis(CFA), which identifies an a priori first-order factor struc-ture; and hierarchical confirmatory factor analysis(HCFA), which tests the ability of higher-order factors torepresent domain-specific self-concepts.

Estimation of Reliability

Reliability consists of estimating the internal consisten-cy, stability, or both (Fleishman & Benson, 1987). Inter-nal consistency was examined using the traditional coef-ficient estimate of reliability.

Data Description

All questionnaires containing more than five missing re-sponses were discarded. Missing data (0.25% of totalresponses) were handled by means of the SerialMeantransformation method from the MVA procedure inSPSS 10.0. The SDQ II Test manual strongly advises theuse of item pairs in factor analysis (Marsh, 1990b) as thisincreases the subjects-to-variables ratio and consequent-ly results in more reliable variables (due to a smallercomponent of unique variance). In addition, factor load-ings are less affected by the idiosyncracies of individualitems, and this method proves most cost-efficient andtime-saving.

Factor Analyses

Confirmatory Factor Analysis (CFA) and HierarchicalCFAs (HCFAs)

A detailed presentation of the conduct of the CFA andHCFAs is beyond the scope of the present investigationand is being developed elsewhere (e.g., Bollen, 1989;Jreskog & Srbom, 1999). Analyses were conductedwith LISREL 8.30/Prelis 2 software (Jreskog & Sr-bom 2000), using the ML estimation method. LISRELprovides numerous overall goodness-of-fit indices. Fol-lowing Marsh, Balla, and McDonald (1988) and Mc-Donald and Marsh (1990), the present research empha-

sized the RMSEA, TLI, and RNI to evaluate the good-ness of fit. The test statistics and the /df ratio werealso presented. The statistics range between zero andinfinity, with zero indicating a perfect fit and a largenumber indicating an important lack of fit. Moreover, itis quite sensitive to sample size and the number of vari-ables and increases proportionally to these two elements.The TLI and RNI should vary between 0 and 1; valuesgreater than .90 and .95 reflect an acceptable and anexcellent fit, respectively. Acceptable values of theRMSEA range between .05 and .08. Browne and Cudeck(1993) suggest that a value of .05 indicates a close fit andvalues up to .08 represent reasonable errors of approxi-mation in the population. It contains, as does the TLI, apenalty for lack of parsimony.

Whereas the SDQ II first-order model has been re-peatedly replicated, its hierarchical structure has re-mained unexplored. That explains why our generalframework for testing structural equation models wasstrictly confirmatory (SC) for the first-order factor struc-ture, and somewhat in between the confirmatory and thealternative modeling (AM) approach for higher-orderstructure (Jreskog & Srbom, 1993). Model 1, a first-order model was initially posited to test the a priori struc-ture of the 11 SDQ II factors. Examination of correlationpatterns of the first-order model provided a strong basisfor the development of higher-order model fitting. Therationale for fitting hierarchical models was firmlygrounded in the Marsh/Shavelson model, which primar-ily supported the following assumptions:1. The separation of two self-concept areas (academic

and nonacademic domains as posited by Shavelson etal.),

2. The clear distinction between two uncorrelated aca-demic factors,

3. The diversity of nonacademic higher-order factors andthe complicated first-order cross-correlation patternexisting among the different nonacademic factors.

Following a parsimonious ordering, several competinghierarchical models were proposed to explain the rela-tions among the first-order structure. First, Model 2, asingle global factor structure, was posited to explain therelationships among the 11 lower-order factors. Model 3tested the superiority of the academic and nonacademicareas hypothesis over the structure of Model 2. The nextstructure (Model 4) assessed, at the academic level, thesuperiority of the Marsh/Shavelson model (separation oftwo academic factors) over the Shavelson et al. model(one single academic factor). Lastly, we developed twoalternative models proposing different higher-order fac-tors the nonacademic area (Models 5 and 6) in order totest the higher-order model that would better representthe complex first-order factor covariations. Complying

144 F. Gurin et al.: Construct Validation of the Self-Description Questionnaire II with a French Sample


with the rule of parsimony, when two models exhibitedan equivalent goodness-of-fit, the one requiring fewerparameters and having the larger degrees of freedom waspreferred, provided it allowed a balance between statis-tical and substantive model fit.

Heywood Cases

An important source of concern and consternation instructural covariance modeling is the appearance of un-desirable results such as negative variance estimates Heywood cases which necessarily yield improper so-lutions. These offending estimates typically mean that amodel is unidentified or unstable. Several solutions havebeen discussed for handling Heywood cases (Bentler &Weeks, 1980; Rindskopf, 1983; Jreskog & Srbom,1999; Marsh, 1987).

Explained Variance Ratio (EVR; Table 3)

An EVR is designed to provide a summary of the vari-ance in first-order factors explained by higher-order fac-tors. The smaller the residual variance, the better thelower-order factor is explained by a higher-order factorand conversely. The EVR is estimated as one minus theratio of the residual variance to the factor variance; itvaries between 0 and 1.0. Zero indicates that none of thevariance is explained by higher-order factor, and onemeans that all of the variance is explained by higher-or-der factor (for details on EVR estimation, see Marsh1987).

ResultsReliability

Reliability estimates for the SDQ II were good (Table 1).Internal consistency for the total self-concept score was.94. Coefficients for the 11 facet scores varied from .85for Honesty to .95 for Math (median = .91).

Factor Analyses

Confirmatory Factor Analysis (CFA)

Model 1 (Table 2) was a simple structure model in whicheach measured variable was assigned a non-zero loadingon only the one factor it was supposed to measure. Thisrestrictive a priori factor structure provided a good fit tothe data (TLI = .925; Table 2). All factor loadings werestatistically significant (from .63 to .96; Md = .83) andclearly identified the 11 components the SDQ II instru-ment was designed to measure. The most positive corre-lations involved were the following: first, the global Es-teem factor that substantially correlated with Generalschool (.58), Appearance (.60), Opposite sex (.50),Emotion (.47), then General school, which correlat-ed with Math (.65) and Verbal (.43); and lastly, Op-posite sex with Appearance (.55) and Same sex(.51). As predicted, the Academic Math and Verbalself-concepts were nearly uncorrelated with each other(0.1). Correlations among the 11 factors, varying from.08 to .65, were moderate and revealed remarkably dis-tinct factors (Table 1). The evaluation of the parameter

Table 1. Self-concept factor structure.

FactorIndicator Math Verb Schl Pabl Appr Ssex Osex Prnt Hnst Emot Estm

Factor CorrelationsMath 1.00Verb .01 1.00Schl .65 .43 1.00Pabl .02 .08 .05 1.00Appr .14 .09 .22 .32 1.00Ssex .07 .04 .13 .24 .31 1.00Osex .10 .07 .14 .29 .55 .51 1.00Prnt .26 .12 .35 .15 .17 .22 .11 1.00Hnst .09 .24 .14 .06 .06 .08 .02 .37 1.00Emot .17 .06 .24 .13 .35 .31 .25 .24 .05 1.00Estm .38 .27 .58 .28 .60 .40 .50 .46 .18 .47 1.00

Reliability estimatesIndicator Math Verb Schl Pabl Appr Ssex Osex Prnt Hnst Emot Estm

.95 .89 .92 .91 .93 .92 .93 .91 .85 .86 .87

Maths (Math), Verbal (Verb), General school (Schl), Physical ability (Pabl), Physical appearance (Appr), Opposite sex (Osex), Same sex(Ssex), Parents (Prnt), Honesty (Hnst), Emotion (Emot), Global self-esteem (Estm)



estimates, the well-defined 11 factors, the reliability es-timates, and the ability of the a priori first-order modelto fit the data provided good support for the multidimen-sionality of adolescent self-concept, and for the frequent-ly reported need to separate Math and Verbal self-con-cepts. This finding confirmed the Marsh/Shavelson fun-damental hypothesis of a clear separation between thetwo first-order facets and provided further support for themultidimensionality of the academic self-concept.

Hierarchical Factor Analyses (HCFAs)

Heywood cases were encountered in four higher-ordermodels (models 3, 4, 5, 6; Figure 1). Improper solutionswere probably due to an insufficient number of indica-tors per factor. They were solved by constraining first-order loadings to equality in the academic area (seeFigure 1, note b). After elimination of the offendingestimates, these models yielded acceptable fit to the da-ta.

In Model 2, a single second-order factor explained the11 first-order factors. Most of them loaded significantlyon the General factor (with the exception of the Hones-ty factor: .231). The fit was barely acceptable (TLI =.898).

Model 3 was a two second-order factor model. It hada better fit than did Model 2 (TLI = .908; Table 2). EVRsshowed a much better representation of the three aca-demic facets than in the preceding model (Table 3).

Model 4 differed from Model 3 in that its academicarea was divided into two second-order factors, Math/ac-ademic and Verbal/academic, as posited by Marsh andShavelson (1985) and Marsh (1990c). The nonacademicarea remained unchanged with the exception of the Es-teem factor, which loaded directly on the third-orderGeneral self-concept factor because of its superordinatestatus. Moreover, the three second-order factors loadedon a third-order General factor (Figure 1). This modeldid better than Model 3 (TLI = .911); in addition, exam-ination of EVRs showed a better representation of first-order facets than second-order factors. And among sec-ond-order factors, Verbal/academic was more poorlyrepresented by a General self-concept factor thanMath/academic.

Model 5 proposed the same academic structure asModel 4. Except for the Esteem factor loading directlyon the General self-concept, the nonacademic area wasdivided into three second-order factors (Physical, Social,and Moral). All five higher-order factors loaded on aGeneral third-order factor (Figure 1). This model wasable to fit the data better than Model 4 (TLI = .916).EVRs revealed that first- and second-order factors werebetter represented by higher-order structure than in thepreceding model.

Model 6 is different from Model 5 in that the Physicaland Social second-order factors were collapsed into asingle facet. Comparison of these two models (Table 2)showed, for Model 6, a slightly better and better good-ness-of-fit indices (TLI = .917). EVRs suggested a nearly

Table 2. Summary of goodness-of-fit for SDQ II models (N = 480).

MODEL DF /DF RNI TLI RMSEA

0 20150.34 1275 15.80 .266First-Order (FO)1A 2472.47 1169 2.11 .931 .925 .049Higher Order (HO)2 3047.30 1213 2.51 .903 .898 .0583 2862.86 1213 2.36 .913 .908 .0564 2803.26 1211 2.31 .916 .911 .0545 2722.85 1210 2.25 .920 .916 .0536 2705.49 1211 2.23 .921 .917 .053

Note. RNI = Relative noncentrality index, TLI = Tucker-Lewis index, RMSEA = root mean square error of approximation.Model Description0 Null model; 51 uncorrelated FO factors1 11 correlated first-order factors, no higher order factors.2 One higher order factor defined by all 11 first-order factors.3 Two correlated higher order (Academic/Nonacademic) factors.4 Three 2nd order factors (Math/academic, Verbal/academic, Nonacademic) and one 3rd order (General self) factor. (First-order

Esteem loading directly on General self)5 Five 2nd order factors (Math/academic, Verbal/academic, Physical, Social, Moral) and one 3rd order (General self) factor.

(First-order Esteem loading directly on General self)6 Four 2nd order factors (Math/academic, Verbal/academic, Physical/social, Moral) and one 3rd order (General self) factor.

(First-order Esteem loading directly on General self)



similar explained variance for the two models. Consis-tent with the rule of parsimony, Model 6, having oneparameter less than Model 5 (115 parameters), and onemore degree of freedom (df = 1211), was preferable toModel 5 as the best structure to explain responses to theSDQ II.

In summary, though not the best solution, Model 4,which was designed to explain the organization of aca-demic facets, demonstrated its superiority over moreparsimonious Models 2 and 3; it also provided supportfor the Marsh and Shavelson model (Marsh, 1990c).Models 5 and 6, explaining the structure of the nonaca-demic area, provided a reasonable improvement overModel 4.

DiscussionPrevious SDQ research carried out within the Marsh/Shavelson theoretical framework posited a multidimen-sional self-concept for preadolescents (SDQ I; Marsh &Hocevar, 1985), adolescents (SDQ II; Marsh, Parker &Barnes, 1985), and late adolescents (SDQ III; Marsh,1987). The findings of the present investigation, basedon responses by adolescents to the French SDQ II, sub-stantiated the multifaceted nature of self-concept.

Prior SDQ research, based on responses by preadoles-cents and late adolescents, also posited a complicated

Figure 1. The five HCFA models.Note (a): G = General self-concept; A = Academic self-concept; N = Nonacademic self-concept; M/A = Math/Academic self-concept; V/A =Verbal/Academic self-concept; Phy = Physical self-concept; Socl = Social self-concept; Morl = Moral self-concept; P/S = Physical/Social self-concept.Note (b): In the academic area, in models 4, 5, and 6, proper solutions were obtained by constraining first-order General School (Schl) to load equally oneach of the two higher-order academic factors. In model 3, General School was constrained to a fixed value. In the nonacademic area, in models 4, 5,and 6, first-order factor Esteem was constrained to a fixed value.

Table 3. Explained variance ratios (EVRs) for first-order and sec-ond-order factors in models 2, 3, 4, 5, 6.

EVRs for each model2 3 4 5 6

First-order factorsMath 16.74 49.68 74.10 71 72.1Verb 7.14 17.27 70.11 67.39 67.7Schl 34.74 99.09 91.72 92.34 92.7Pabl 9.41 10.89 15.00 15.65 16.8App 37.76 40.94 49.29 67.47 53.9Ssex 19.64 21.75 30.29 40.88 31.6Osex 28.52 31.71 44.73 63.35 57.2Prnt 21.50 20.54 15.78 29.10 30.3Hnst 3.33 2.93 1.37 6.05 7.0Emot 24.33 25.44 25.85 29.41 28.4Estm 92.00 90.09 93.57 96.19 97.3

Second-order factorsMTAC 43.90 37.71 41.8VBAC 8.15 8.94 8.98ACADNACAD 67.41PHYS 54.51SOCL 41.83MORL 73.31 73.17PHSO 49.20

Note. EVR is defined as: [(1 (Residual Variance/Variance esti-mate)) 100]; Maths (Math), Verbal (Verb), General school (Schl),Physical ability (Pabl), Physical appearance (Appr), Opposite sex(Osex), Same sex (Ssex), Parents (Prnt), Honesty (Hnst), Emotion(Emot), General self-esteem (Estm).



and relatively weak hierarchical ordering. Our studyconfirmed this result and expanded this finding to ado-lescents as a new area of self-concept. It also provided astrong test of the generality of SDQ II results to a Frenchpopulation.

In the academic area, we supported the clear distinc-tion between the two second-order academic facets pro-posed in the Marsh and Shavelson revision (Marsh,1990c). In addition, in the present study, these two facetswere not equally represented. General self-concept wasbetter defined by Math/academic than by Verb/academic(Table 3). This could be explained as a typical education-al setting effect (importance of Math in French highschools) reinforced by an even greater predominance ofMath in this science-oriented high school. However, fur-ther research would be needed on this aspect.

In the nonacademic domain, the reason for favoringModel 6 over Model 5 stemmed from the study of theirPhysical and Social facets. In Model 5, Physical (Physi-cal appearance, Physical ability) and Social (Same sex,Opposite sex) factors were distinct. Yet, correlations be-tween Physical appearance and Opposite sex were higherthan correlations among the Physical and among the So-cial facets. A similar feature had been found in SDQ I(Marsh, 1988), SDQ II (Marsh, Parker, & Barnes, 1985),and SDQ III (Marsh, 1987) research. Moreover, the cor-relation between Physical and Social factors (.477) wasthe highest among the second-order facets. All these rea-sons argued in favor of the single-Physical/Social factorthat was tested in Model 6. It revealed a slightly better fitthan Model 5. It also corresponded to the best fittingmodel proposed by Marsh and Hocevar in 1985 for pre-adolescents and was quite similar to the authors second-best solution for late adolescents (Marsh, 1987). Thischoice of a single factor seemed intuitively consistentwith an adolescent population.

The last second-order factor, Moral, in the nonaca-demic area was composed of first-order Parents, Hones-ty, and Emotion self-concepts (Table 2). The very poorcorrelation between Emotion and Honesty (.05; Table 1)was a definite weakness in this higher-order factor andcontributed to the weakness of the hierarchical represen-tation.

In Models 4 to 6, the first-order factor Esteem loadedon the General self-concept (the Esteem self-concept isan adaptation of the Rosenberg Self-Esteem Scale(1979) incorporated in all three SDQ instruments). Itwas hypothesized that Esteem, as a superordinate fac-tor, would contribute directly to the hierarchical Gener-al self-concept. For that reason, overall correlations ofEsteem self-concept among first-order factors, as wellas with the General self-concept, were found to be thehighest.

In summary, the present investigation demonstrates

the existence of a clearly defined multidimensionalstructure of self-concept, which is the essential and re-current finding in SDQ research. However, the negativeevidence because much of the variance in many lower-order factors was not explained by higher-order factors,the generalized improper solutions found in Models 3,4, 5, 6 and hence, the need for so much constraining(especially in the academic facet) for solving the of-fending issue did not argue in favor of a strong hier-archy. It seemed very clear that the best-fitting modelwas by far the CFA first-order model. Each of the HCFAmodels did significantly poorer than the first-ordermodel that was the basis of comparison for each HCFAmodel, which argues against an overemphasis on theHCFA structure. In previous studies, similar results onthe hierarchy of general self-concept led Marsh and col-leagues to the conclusion that General self-concept wasnot a particularly useful construct (Marsh, Byrne, &Shavelson, 1992). In fact, it could not adequately reflectthe diversity of specific self-facets. In a different agerange, recent self-concept research on preschool chil-dren (SDQP) brought further support in favor of thefirst-order model (Marsh, Craven, & Debus, 1991,1998; Marsh, Ellis, & Craven, 2002). Their resultsshowed that self-concept hierarchy was so weak that itwould have been useless to use a higher-order represen-tation instead of a first-order structure. Moreover, in theSDQ academic area, it had been found that each self-concept domain was more highly correlated with exter-nal criteria relevant to its corresponding facet ratherthan with other self-concept domains (Marsh, 1993).These considerations strongly support the particular rel-evance of specific self-concept research in various ar-eas: Academic (Marsh, 1990c; 1992; 1993), Physical(Marsh, Hey, Johnson, & Perry, 1997; Marsh, Richards,Johnson, Roche, & Tremayne, 1994), Social (Byrne,1996) or, performing Arts (Vispoel, 1995), especially ifresearchers wish to understand, explain and predict thepersons specific behavior in order to help implementappropriate interventions.

ReferencesBentler, P.M., & Weeks, D.G. (1980). Linear structural equations

with latent variables. Psychometrika, 45, 289308.Bollen, K.A. (1989). Structural equations with latent variables.

New York: Wiley.Browne, M.W., & Cudeck, R. (1993). Alternative ways of assess-

ing model fit. In K.A. Bollen & J.S. Long (Eds.), Testing struc-tural equation models (pp. 445455). Beverly Hills, CA: Sage.

Burns, R.B. (1979). The Self-concept: Theory, measurement, de-velopment and behavior. London: Longman.

Byrne, B. (1996). Measuring self-concept across the life span:



Issues and instrumentation. Washington, DC: American Psy-chological Association.

Cronbach, L.J., & Meehl, P.E. (1955). Construct validity in psy-chological tests. Psychological Bulletin, 52, 281302.

Fleishman, J., & Benson, J. (1987). Using LISREL to evaluatemeasurement models and scale reliability. Educational andPsychological Measurement, 47, 925939.

James, W. (1909). Prcis de psychologie [Principles of psycholo-gy]. Paris: Marcel Rivire.

Jreskog, K., & Srbom, D. (1993). LISREL 8: Structural equa-tion modeling with simplis command language. Chicago: SSI:Scientific Software international.

Jreskog, K., & Srbom, D. (2000). LISREL 8.30/Prelis 2. Chica-go: SSI: Scientific Software international.

Jreskog, K., & Srbom, D. (1999). LISREL 8.30: Users refer-ence guide. Chicago: SSI: Scientific Software International.

Marsh, H.W. (1986). Verbal and Math self-concepts. An inter-nal/external frame of reference model. American EducationalResearch Journal, 23, 129149.

Marsh, H.W. (1987). The hierarchical structure of self-conceptand the application of HCFA. Journal of Educational Measure-ment, 24, 1739.

Marsh, H.W. (1988). Self-Description Questionnaire-I: Manualand research monograph. San Antonio, TX: PsychologicalCorp.

Marsh, H.W. (1990a). A multidimensional, hierarchical self-con-cept: Theoretical and empirical justification. Educational Psy-chology Review, 2, 77172.

Marsh, H.W. (1990b). Self-Description Questionnaire-II: Manualand research monograph. San Antonio, TX: PsychologicalCorp.

Marsh, H.W. (1990c). The structure of academic self-concept: TheMarsh/Shavelson model. Journal of Educational Psychology,82, 623636.

Marsh, H.W. (1992). The content specificity of relations betweenacademic achievement and academic self-concept. Journal ofEducational Psychology, 84, 4350.

Marsh, H.W. (1993). Academic self-concept: Theory measure-ment and research. In J. Suls (Ed.), Psychological perspectiveson the self (Vol. 4, pp. 5998). Hillsdale, NJ: Erlbaum.

Marsh, H.W. (1994). Using the national longitudinal study of 1988to evaluate theoretical models of self-concept: The Self-De-scription Questionnaire. Journal of Educational Psychology,86, 439456.

Marsh, H.W., Balla, J.R., & McDonald, R.P. (1988). Goodness-of-fit indexes in confirmatory factor analysis: The effect of samplesize. Psychological Bulletin, 103, 391410.

Marsh, H.W., Byrne, B., & Shavelson, R.J. (1992). A multidimen-sional, hierarchical self-concept. In T.M. Brinthaupt & R.P.Lipka (Eds), The self: Definitional and methodological issues(pp. 4495). Albany, NY: State University of New York Press.

Marsh, H.W., Craven, R.G. (1997). Academic self-concept: Be-yond the dustbowl. In G. Phye (Ed.), Handbook of classroomassessment: Learning, achievement, and adjustment (pp. 131198). San Diego, CA: Academic Press.

Marsh, H.W., Craven, R.G., & Debus, R.L. (1991). Self-conceptsof young children aged 5 to 8: Their measurement and multi-dimensional structure. Journal of Educational Psychology, 83,377392.

Marsh, H.W., Craven, R.G., & Debus, R.L. (1998). Structure,stability, and development of young childrens self-concepts:

A multicohort-multioccasion study. Child Development, 69,10301053.

Marsh, H.W., Ellis, L., & Craven, R.G. (2002). How do preschoolchildren feel about themselves? Unravelling measurement andmultidimensional self-concept structure. Developmental Psy-chology, 38, 376393.

Marsh, H.W., & Hattie, J. (1996). Theoretical perspectives on thestructure of self-concept. In B.A. Bracken (Ed.), Handbook ofself-concept (pp. 3890). New York: Wiley.

Marsh, H.W., Hey, J., Johnson, S., & Perry, C. (1997). Elite AthleteSelf-Description Questionnaire: Hierarchical confirmatoryfactor analysis of responses by two distinct groups of elite.International Journal of Sport Psychology, 28, 237258.

Marsh, H.W., & Hocevar, D. (1985). Application of confirmatoryfactor analysis to the study of self-concept: First and higher-or-der factor models and their invariance across groups. Psycho-logical Bulletin, 3, 562582.

Marsh, H.W., Parada, R.H., Ayotte, V. (2002). Multidimensionalself-concepts and adolescents mental health: A new Frenchversion of the Self-Description Questionnaire II. Manuscriptsubmitted for publication.

Marsh, H.W., Parker, J., & Barnes, J. (1985). Multidimensionaladolescent self-concepts: Their relationship to age, sex, andacademic measures. American Educational Research Journal,22, 422444.

Marsh, H.W., Plucker, J.A., & Stocking, V.B. (2001). The Self-De-scription Questionnaire II and gifted students: Another look atPlucker, Taylor, Callahan & Tomchins (1997) Mirror, Mirroron the Wall. Educational and Psychological Measurement,93, 87102.

Marsh, H.W., Richards, G.E., Johnson, S., Roche, L., & Tre-mayne, P. (1994). Physical self-description questionnaire: Psy-chometric properties and a multitrait-multimethod analysis ofrelations to existing instruments. Journal of Sport and ExercisePsychology, 16, 270305.

Marsh, H.W., & Shavelson, R.J. (1985). Self-concept: Its multi-faceted hierarchical structure. Educational Psychologist, 20,107123.

Mc Donald, R.P., & Marsh, H.W. (1990). Choosing a multivariatemodel: Noncentrality and goodness of fit. Psychological Bul-letin, 2, 247255.

Rindskopf, D. (1983). Parameterizing inequality constraints onunique variances in linear structural models. Psychometrika,48, 7383.

Rosenberg, M. (1979). Conceiving the self. New York: BasicBooks.

Shavelson, R.J., Hubner, J.J., & Stanton, G.C. (1976). Self-con-cept: Validation of construct interpretations. Review of educa-tional research, 46, 407441.

Vallerand, R.J., & Halliwell, W.R. (1983). Vers une mthodologiede validation trans-culturelle de questionnaires psycholo-giques: Implications pour la psychologie du sport [Toward across-cultural validation methodology for psychological in-ventories: Implications in sport psychology]. Canadian Jour-nal of Applied Sport Sciences, 8, 918.

Vallires, E.F., & Vallerand, R.J. (1990). Traduction et validationCannadienne-Franaise de lchelle de lestime de soi de Ro-senberg [Translation and French-Canadian validation of theRosenberg self-esteem scale]. International Journal of Psy-chology, 25, 305316.

Vispoel, W.P. (1995). Self-concept in artistic domains: An exten-



sion of the Shavelson, Hubner and Stanton (1976) model. Jour-nal of Educational Psychology, 87, 134153.

Wells, L.E., & Marwell, G. (1976). Self-esteem: Its conceptualiza-tion and measurement. Beverly Hills, CA: Sage.

Wylie, R.C. (1974). The self-concept. Volume 1: A review of meth-odological considerations and measuring instruments (reviseded., Vol. 1). Lincoln: University of Nebraska Press.

Wylie, R.C. (1979). The self-concept. Volume 2: Theory and re-search on selected topics. Lincoln: University of NebraskaPress.

Florence GurinUniversit Paris-SudU.F.R. S.T.A.P.SF-91405 ORSAY CedexFranceTel. +33 1 6915-4309E-mail [email protected]



jpa_19_2_142.pdf

Documents

Transcript of jpa_19_2_142.pdf