A Test of a Multi - Faceted , Hierarchical Model of Self - Concept ®

This study tested a multi-faceted, hierarchical self-concept model for university students. Self-concept was conceptualized as composed of three 1 order facets, each with three 2 order facets: academic self-concept (capability, achievement and confidence), social self-concept (same-sex peer, opposite-sex peer and family) and self-concept presentation of self (personal confidence, physical and honest/trustworthy). Data from a convenience sample of 400 students were analyzed with the Extended Logistic Model of Rasch. The 45 How I actually am items, separately, fitted the model and formed a valid and reliable scale. The 45 How I actually am items, together with 21 How I would like to be items (66 items), fitted the model and formed a separate, valid and reliable scale. The How I would like to be items were all easier than their corresponding How I actually am items, as conceptualized. The results supported the multi-faceted, hierarchical model of self-concept as a unidimensional latent trait. A TEST OF A MULTI-FACETED, HIERARCHICAL MODEL OF SELF-CONCEPT Introduction It has been written often that the journal and book literature has many self-concept (and selfesteem) scales that are not based on a strong theoretical model that is multi-faceted and hierarchical (see for example, Bracken, 1996, 1992; Bryne, 1984; Hattie, 1992; Marsh & Shavelson, 1985; Marsh, 1994, 1993; Shavelson, Hubner & Stanton, 1976; Wylie, 1974, 1979). Shavelson, Hubner and Stanton (1976) reviewed the literature and proposed a multifaceted, hierarchical model of self-concept. It was suggested that general self-concept is composed of four 1 order facets: academic self-concept, social self-concept, emotional selfconcept and physical self-concept. The 1 order facets are composed of 2 order facets. Academic self-concept has aspects relating to each of the academic areas of English, History, Math and Science. Social self-concept is composed of peer self-concept and significant others self-concept. Emotional self-concept is composed of self-concept for particular emotional states. Physical self-concept is composed of physical ability selfconcept and physical appearance self-concept. There are a number of self-concept scales based on this or a similar model (see Marsh 1992a,b,c for example), and many other, nonmodel, based scales have been reviewed by Hattie (1992) and Bracken (1996). The Self Description Questionnaires (I for preadolescents, Marsh, 1992a, 1988; II for adolescents, Marsh, 1992b, 1990; and III for late-adolescents, Marsh, 1992c; Marsh & O'Neill, 1984) are based on, or extended from, the Shavelson, Hubner and Stanton (1976) model. The Self Description Questionnaire II consists of a general-self scale and ten subscales (102 items). There are four non-academic sub-scales relating to physical abilities, physical appearance, same-sex peer relations, opposite-sex peer relations, parent relations, emotional stability and honesty/truthfulness, and three academic sub-scales relating to reading, mathematics and general school. Each sub-scale contains eight or ten items in six response categories: false, mostly false, more false than true, more true than false, mostly true, and true. There is evidence from Sydney, Australia, and from the USA (Marsh & Yeung, 1997), using traditional measurement techniques, supporting various aspects of validity and reliability for this scale. According to Marsh (1990, p27, cited in Bracken, 1996), self-concept is a "person's perceptions regarding himself or herself; these perceptions are formed through experience with and interpretations of one's environment. They are especially influenced by evaluations by significant others, reinforcements, and attributions for one's own behavior." Song and Hattie (Hattie, 1992, p84) modified the Shavelson, Hubner and Stanton model (1976). They suggested that general self-concept is composed of three first-order orientations: academic self-concept, social self-concept and self-regard/presentation of self. Academic self-concept is composed of three 2 order orientations: ability self-concept, achievement self-concept and classroom self-concept. Social self-concept is composed of peer self-concept and family self-concept. Self-regard/presentation of self is composed of confidence and physical self-concept. Song and Hattie (Hattie, 1992, p162-163) developed a 35 item self-concept scale, with five items for each of the seven 2 order self-concept orientations, and applied the scale to adolescents. Adolescents responded in one of six categories, from strongly disagree to strongly agree. Evidence was presented from Australia and Korean data, using traditional factor analysis and measurement techniques, supporting its validity and reliability in various aspects, and confirming the conceptual structure of the scale. According to Hattie (1992, p117), self-concept is "both a structure and a structure/process". This means that, for some people, it is a "set of beliefs that dominate processes and actions" and guide behaviour across situations. It also means that, for other people, it is a latent, "hierarchical and multi-faceted set of beliefs that mediate and regulate behavior in various social settings". "Self-concept relates to descriptions, expectations and prescriptions and can be actual, possible, ideal, evaluative, interpretative, and dynamic". Later however, Marsh and Hattie (1996, p77) wrote that "ideal ratings typically do not contribute beyond what can be explained by actual ratings alone, and mean discrepancy scores have no more and, perhaps, less explanatory power than the mean of actual ratings". According to Bracken (1992, p10), self-concept is "a multidimensional and contextdependent learned behavioral pattern that reflects an individual's evaluation of past behaviors and experiences, influences an individual's current behaviors, and predicts an individual's future behaviors." Bracken (1992) uses a Multidimensional Self Concept Scale comprising 150 self-report items with a Likert format. There are six domains each composed of 25 item sub-scales relating to social, competence, affect, family, physical and academic self-concept. Evidence is presented, using traditional measurement techniques, for various aspects of validity and reliability of this scale, with data representative of the 1990 USA Census by gender, race, ethnicity and geographic region. Problems with the current measures of self-concept Seven aspects of many of the main self-concept scales (Bracken, 1996, 1992; Song & Hattie, cited in Hattie, 1992, p84, and Marsh, 1992a,b,c) are called into question. First, adolescents are asked to respond to items in a Likert format (strongly disagree to strongly agree) and apply this format across all orientations or domains. This response format contains a discontinuity between the disagree and agree response categories. That is, the response measurement format is not ordered from low to high and those who are undecided, don't want to answer, are unclear or just neutral, will answer the middle (neutral) category. If a neutral category is not provided, they will be forced to answer either agree or disagree. This means there is a consequent measurement problem. Two, the scales only measure adolescent descriptions of themselves (how they are). It is likely that, because of the massive exposure of role models across many sports and occupations through television, self-concepts will be influenced by their views of what they could be like and, therefore, how they would like to be, as well as how they are. It is expected that the claim by Marsh and Hattie (1996, p77) that ideal ratings do not contribute anything beyond actual ratings is due to measurement problems. Only ordinal level scales were used in these studies (numbers were just added to form a total score), no check was made to reject items that are contributing 'noise' to the measures, and the ideal and realistic scales are not linked in measurement (except that the same items are used). However, both How they would like to be and How they actually are, ought to be measured at the same time and calibrated on the same interval level scale. Three, the self-concept items are not always separated into their sub-scales on the questionnaires, so that it is not clear to the respondents what is being measured. Four, nearly all the self-concept scales were administered to school students and relatively few were administered to other adolescents. Five, positively and negatively worded items are mixed to avoid the fixed response syndrome (a common procedure in traditional measurement). There is some evidence that this causes an interaction effect between items in modern measurement models (see Andrich & van Schoubroeck, 1989). Consequently, it is considered better to word all items in a positive sense when using modern measurement models. Six, the main analysis of the self-concept scales have only been performed with traditional measurement programs and ordinal level scales. A consequence of this is that, because there are low correlations between total academic scores and total non-academic scores on the Self Description Questionnaire II, Marsh suggests that users should only analyze specific self-concept areas (such as academic, social or physical), rather than total scores using all self-concept orientations (Marsh, 1990, cited in Bracken, 1996 p146). This line of reasoning calls into question the concept of a general self-concept latent trait and the concept of the hierarchical model postulated, including the proposed unidimensional concept of the scale. Modern measurement programs are now available to create interval level measures in which item difficulties and person self-concept measures can be calibrated on the same scale. Their use would test the multi-faceted structure of self-concept, including its dimensional nature (see Andrich, 1988a, 1988b; Andrich, Sheridan, Lyne & Luo, 1998; Rasch, 1980/1960; Waugh, 1999, 1998).