MAR05386 Reciprocal Effects of Self-concept and Achievement: Competing Multidimensional and Unidimensional Perspectives

The rationale for this presentation is a theoretical model indicating that people who perceive themselves to be more effective, more confident, and more able accomplish more than people with less positive self-perceptions. Support for this prediction is strongest in academic self-concept research where a substantial body of research in support of the reciprocal effects model now exists. On this basis, Marsh and Craven (1997) claim that academic self-concept and achievement are mutually reinforcing constructs, each leading to gains in the other. In contrast, Baumeister, Campbell, Krueger and Vohs (2003) claim that self-esteem has no benefits beyond seductive pleasure and may even be detrimental to subsequent performance in high profile publications that have received extensive international attention. Here we review the theoretical and empirical bases for each set of claims, contrasting the older unidimensional perspectives that focus on global self-esteem and more recent, multidimensional perspectives that focus on specific components of self-concept. Juxtaposing these contrasting sets of results integrating the unidimensional and multidimensional perspectives into a common theoretical framework offers resolution to this apparent conflict, and has important implications for educational research, policy and practice. About the Authors: Professor Herb Marsh is Professor of Educational Psychology, founding Director of the SELF Research Centre and served as UWS’s inaugural Dean of Graduate Research Studies and Pro-Vice-Chancellor. He received UWS’s inaugural awards for Research, Postgraduate Supervision, and Doctorate of Science. Herb has published more than 250 peer-reviewed journal articles, 40 chapters, 10 monographs, and 225 conference papers. He is Australia’s most widely cited researcher in both education and psychology, and the 11 mostly widely cited researcher in the world across all disciplines of psychology. The Australian Research Council has funded all 14 of his ARC grant proposals and awarded him its ARC Special Investigator grant. Email: [email protected] Associate-Professor Rhonda Craven is the Deputy Director of the Self-concept Enhancement and Learning Facilitation (SELF) Research Centre, an Associate Professor in the School of Education and Early Childhood Studies, University of Western Sydney. She is a highly accomplished self-concept researcher having successfully secured over 1.4 million dollars in prestigious, external, nationally competitive funding for 19 research projects. She is also the editor of the bestselling text ‘Teaching Aboriginal Studies’ published by Allen and Unwin in 1999, a founding member of the national Aboriginal Studies Association (ASA), an associate member of NSW Aboriginal Education Consultative Group, and the inaugural recipient of the ASAs Life Achievement Award. E-Mail: [email protected] Self-concept is also widely presumed to make a causal difference in addressing some of the key social issues of our time. Attesting to this pervasive significance of the self-construct and the outcomes that are mediated by it, Nathaniel Branden (1994, p. xv) contends: I cannot think of a single psychological problem—from anxiety to depression, to underachievement at school or at work, to fear of intimacy, happiness or success, to alcohol or drug abuse, to spouse battering or child molestation, to co-dependency and sexual disorders, to passivity and chronic aimlessness, to suicide and crimes of violence—that is not traceable, at least in part, to the problem of deficient self-esteem (Branden, 1994, p. xv). Hence positive self-belief is valued as a hot variable that makes good things happen, facilitating the realization of full human potential in a range of settings. A theme emphasized here is that the most powerful effects of self-concept are based on specific components of self-concept most logically related to specific outcomes considered in a particular study (a multidimensional perspective) rather than the global component of self-concept represented in global measures of self-esteem (a unidimensional perspective). This pervasive role of the self-construct has recently been challenged, fuelling debate on its importance. In a highly influential review commissioned for Psychological Science in the Public Interest Baumeister, Campbell, Krueger, and Vohs (2003; also see Baumeister, Smart, & Boden, 1996) posed the question: “Does high self-esteem cause better performance, interpersonal success, happiness, or healthier lifestyles?” On the basis of their literature review, Baumeister et al. (2003) concluded that enhancing selfesteem only resulted in “seductive pleasure” (p. 39) and as such “self-esteem per se is not the social panacea that many people hoped it was” (p. 38). Although Baumeister et al. considered a variety of outcome measures, they particularly focused on school performance because of its importance in the self-esteem movement, and because of the quality and quantity of research in relation to school performance. More recently, Baumeister, Campbell, Krueger, and Vohs (2005) claimed to have exploded the self-esteem myth that self-esteem has benefits. Although conceding that high self-esteem is clearly associated with higher levels of happiness, the authors again concluded “that efforts to boost people’s self-esteem are of little value in fostering academic achievement or preventing undesirable behaviour” (p. 84). In dramatic contrast to Baumeister et al. conclusions, we (Marsh & Craven, 1997; in press) claim that enhancing self-concept is a vital goal in and of itself and that self-concept is an important mediating variable that causally impacts on a variety of desirable outcomes including academic achievement. Capitalizing on advances in theory, statistical methodology, and empirical research, we developed a reciprocal effects model (Marsh, 1990b, 1993a; Marsh, Byrne & Yeung, 1999; Marsh & Craven, 1997; in press) whereby the causal relation between academic self-concept and achievement is conceived as dynamic and reciprocal. The reciprocal effects model is underpinned by the notion that people who perceive themselves to be more effective, more confident, and more able, accomplish more than people with less positive self-perceptions (I believe, therefore I am). Support for the reciprocal effects model is particularly strong in relation to academic self-concept and school performance, but is also evident in other domains. Marsh and Craven (1997) and Baumeister et al. (2003) thus appear to draw contradictory conclusions about the role of self constructs in making good things happen; resolution of this debate has profound policy implications for a wide range of issues in the public interest. Furthermore, both sides of this debate argue that the strongest support for their claims comes from educational research, in which academic achievement is the critical outcome variable. How can such apparently discrepant views exist in an area where so much research has been done? The answer lies in the distinction between a multidimensional perspective of selfconcept and the unidimensional perspective that considers only self-esteem—the global component of selfconcept. Baumeister et al.’s review focused almost exclusively on global measures of self-esteem (a unidimensional conceptualization of self-concept) and specifically excluded from consideration, research that focused on domain-specific measures of self-concept (consistent with a multidimensional conceptualization of self-concept) that underpins the conclusions of Marsh and Craven (1997). Indeed, support for the necessity of the multidimensional perspective is particularly strong in educational research that was flagged as a key a focus of the Baumeister review and the major focus of research reviewed by Marsh and Craven. Hence, the key to understanding these apparently contradictory conclusions is to distinguish between the unidimensional perspective that underlies Baumeister et al.’s claim and the multidimensional perspective that is the basis of the Marsh and Craven claim. The overarching purpose of this paper is to disentangle the research bases upon which these seemingly contradictory conclusions are based. More importantly, we demonstrate that both viewpoints can be integrated into a single theoretical model that explains why both are accurate and consistent with recent developments in self-concept research. In the following review of relevant research stemming particularly from our self-concept research program, we: 1. Describe a multidimensional, hierarchical model of self-concept and present research support for the salience of this multidimensional perspective which demonstrates that different components of selfconcept (e.g., social, physical, emotional, academic, self-esteem) are highly differentiated and cannot be explained in terms of a single global component, as implied in a undimensional perspective; 2. Develop a construct validity approach to evaluate the external validity of our multidimensional perspective in relation to important criterion variables from a wide variety of psychological disciplines. In support of the convergent validity of interpretations based on a multidimensional perspective, we show that important criterion variables from many areas of psychological research are substantially related to specific components of self-concept to which they are most logically related (e.g., academic achievement and academic self-concept; physical fitness and physical self-concept). In support of the discriminant validity of interpretations based on a multidimensional perspective, we show that the same criterion variables are substantially less correlated with other domains of self-concept to which they are not logically related (e.g., academic achievement and physical self-concept, physical fitness and academic self-concept); 3. Develop a reciprocal effects model to evaluate the causal ordering of self-concept and important criterion variables based on longitudinal data from different psychological disciplines. In developing this model we argue for a multidimensional perspective in which we focus on relations between specific outcomes of importance and components of self-concept most closely related to those outcomes. In our research program we have focused particularly on the reciprocal effects of academic self-concept and academic achievement in longitudinal panel studies. Thus, for example, we demonstrate that there is good support for the reciprocal effects model when the causal relation of academic achievement and academic self-concept is examined, but not for the causal relation of academic achievement and self-esteem. Furthermore, in this section we demonstrate how this distinction allows us to reconcile and more fully explicate, both those results which focus on self-esteem, reviewed by Baumeister et al. (2003) and those of Marsh and Craven (1997), which emphasize academic self-concept; 4. Demonstrate how findings from our reciprocal effects model generalize to different settings. Historically, our work on the reciprocal effects model has been focused on relations between academic achievement and academic self-concept for adolescent students in Western school settings. In this section of our paper, we demonstrate that our more recent research extends the external validity of the reciprocal effects model. We evaluate developmental aspects of this support with young children, demonstrating support for the reciprocal effects model at younger ages than previously thought possible. We demonstrate the crosscultural generalizability of these results to non-Western school settings. Importantly, we demonstrate the generalizability of the reciprocal effects model to health settings, non-elite sport, and even to winning gold medals at international swimming championships; and 5. Conclude by mapping the important implications of our research for theory, research and public policy. A Multidimensional Perspective Of Self-Concept. Historically, self-concept measurement, theory, research, and application has been plagued by the poor quality of both theoretical models and self-concept measurement instruments (e.g., Shavelson, Hubner & Stanton, 1976; Wells & Marwell, 1976; Wylie, 1979). In an attempt to remedy this situation, Shavelson et al. (1976) reviewed existing self-concept research and instruments, proposed a new theoretical model of selfconcept, and provided a blueprint for the development of multidimensional self-concept instruments (see review by Marsh & Hattie, 1996). In the multidimensional, hierarchical model proposed by Shavelson et al. general self-concept at the apex of the model is divided into academic and non-academic components of selfconcept. The academic component is divided into self-concepts specific to general school subjects, and nonacademic self-concept is divided into physical, social, and emotional components, which are further divided into more specific components. Over the last two decades, instigated in part by the Shavelson model (see review by Marsh & Hattie, 1996), many subdisciplines of psychology have shifted from primary reliance on a global self-esteem measure to including domain-specific assessments of self-concept in addition to or instead of a global selfesteem measure. This research has consistently shown that the proposed hierarchy is weaker than anticipated by Shavelson et al. (1976) and that the specific components of self-concept (e.g., social, academic, physical, emotional) are highly differentiated (Marsh & Craven, 1997; also see Harter, 1998). Marsh (1993a; Marsh & Craven, 1997) argued logically—and demonstrated empirically—that if specific components of self-concept are highly differentiated, then much variation in the specific components cannot be explained in terms of a single global component such as self-esteem. Following from this result, they proposed a multidimensional perspective to self-concept in which specific components of self-concept most logically related to the aims of the research will typically be more useful—more strongly related to important criteria, more influenced by interventions, and more predictive of future behavior—than a single, global component of self-concept that is intended to provide an overall index of self-concept. Thus, Marsh and Craven (1997) argued that specific domains of self-concept logically related to goals of a particular study are more useful than a global measure of self-esteem. Based on this logic, a new generation of self-concept instruments stimulated by the Shavelson et al. (1976) model has provided overwhelming support for the multidimensionality of self-concept (Byrne, 1996b) that is increasingly being recognized in different psychological disciplines. Multidisciplinary Support for a Multidimensional Perspective. Educational psychology provides particularly compelling support for the multidimensional perspective (Marsh, 1993a, b). Many important academic outcomes are substantially related to academic self-concept but are relatively unrelated to self-esteem and nonacademic components of self-concept (e.g., Byrne, 1996a; Marsh, 1993a, b; Marsh, Trautwein, Ludtke, & Baumert, in press). For example, Marsh, Trautwein, et al. demonstrated a predictable pattern of substantial relations between eight academic criterion variables (grades, test scores, and coursework selection in different school subjects) and corresponding academic self-concepts, whereas self-esteem was nearly uncorrelated with all these criteria (rs = -.03 to .05). Similarly, Marsh (1992) established that relations between academic self-concepts in eight specific school subjects were substantially related to school grades in the matching school subjects (rs = .45 to .70), offering support for the external validity of specific facets of academic self-concept. In contrast, self-esteem was nearly uncorrelated with school grades in all the school subjects, indicating that it had no validity in relation to this criterion. Marsh and Yeung (1997a, b) also demonstrated that whereas self-concepts in school subjects and matching school grades were substantially correlated, the specific components of academic self-concept predicted subsequent coursework selection much better than did school grades or more general components of self-concept. These results provide empirical evidence calling into question the usefulness of a general self-esteem construct in educational psychology research, and offer strong support for the multidimensional perspective. Developmental psychology research shows that children as young as five can reliably differentiate between multiple dimensions of self-concept with appropriately constructed self-concept instruments (Marsh, Craven & Debus, 1991; Marsh, Ellis & Craven, 2002; Marsh, Debus & Bornholt, 2005) and that the different facets of self-concept become increasingly distinct with age (Marsh & Ayotte, 2003). Thus, for example, factor analysis of adolescent responses to a recent adaptation of the SDQ III clearly supported the 17 self-concept factors that the instrument was designed to measure (Marsh, Trautwein et al., in press). The average correlation among the 17 self-concept factors, even after controlling for unreliability, was only .14, attesting to the distinctiveness of these factors. In hierarchical factor analyses the correlations between responses to the self-esteem scale and the highest-order self-concept factor are consistently greater than .9, thus supporting the construct validity of both these conceptualizations of global self-concept (see Marsh & Craven, 1997; Marsh, Parada, Craven & Finger, 2005). Obviously, however, neither the single higher-order factor nor the self-esteem factor is able to provide an adequate summary of such distinct factors so that provide much useful information beyond that provided by self-esteem. In mental health research, Marsh, Parada and Ayotte (2004; also see Marsh, Parada, Craven & Finger, 2005) demonstrated that relations between 11 self-concept factors and seven mental health problems varied substantially (+.11 to -.83; mean r = -.35) and formed an a priori multivariate pattern of relations that supported a multidimensional perspective. Support for the multidimensional perspective was particularly strong for the externalizing (e.g. delinquent and aggressive behaviors) mental health factor. It was modestly negatively correlated with self-esteem (r = -.34), substantially negatively correlated with some specific components of self-concept (e.g., parent relations, r = -.70), and nearly uncorrelated or even positively correlated with Physical, Appearance, Same-Sex, and Opposite-Sex SDQII self-concept factors. Self-esteem was able to uniquely explain only 3% of the covariation between mental health and self-concept factors, whereas specific components of self-concept explained 97% of this covariation. Based on higher-order factor analyses, Marsh, Parada and Ayotte noted that single higher-order factors could not explain relations among the self-concept factors, among the mental health factors, or between the self-concept and mental health factors; a unidimensional perspective was not viable. In personality research, Marsh, Trautwein, et al. (in press) demonstrated a well-defined multivariate pattern of relations between multiple dimensions of self-concept and personality (e.g., Big Five personality factors, positive and negative affect, life satisfaction). Seven higher-order factors resulted from the 25 firstorder (17 self-concept, 8 personality) factors. Each Big Five personality factor loaded primarily on one higher-order factor, along with a distinct set of self-concept factors to which it was most logically related. Importantly, self-esteem contributed substantially to only one of the seven higher-order factors, and even for this one higher-order factor, self-esteem was not the highest loading self-concept factor. In this same study, academic outcomes were logically and substantially related to the academic self-concept factors, but nearly unrelated to self-esteem, other non-academic components of self-concept, and the eight personality factors (except, perhaps, Openness). This highly differentiated multivariate pattern of relations argues against the unidimensional perspective of self-concept still prevalent in personality research. In sports and exercise psychology, there is also broad acceptance of the multidimensional perspective of self-concept (Marsh, 1997, 2002). For example, Marsh and Peart (1988) reported that results of a physical fitness intervention, and physical fitness indicators, were substantially related to physical selfconcept but nearly uncorrelated with non-physical components of self-concept. Marsh (1997, 2002) demonstrated that objectively measured components of physical fitness are substantially correlated to the specific components of physical self-concept to which they are most logically related, but are substantially less correlated with selfesteem. Gender studies also support the multidimensional perspective. Although gender differences in selfesteem are small (Wylie, 1979), there are modest differences favoring boys that grow larger through high school and then decline in adulthood (Kling, Hyde, Showers, & Buswell, 1999). However, these small gender differences in self-esteem mask larger, counter-balancing gender-stereotypic differences in specific components of self-concept (e.g., boys have high math, physical, and emotional self-concepts; girls have higher verbal, honesty/trustworthiness, and social self-concepts) and this pattern of gender differences is reasonably consistent from early childhood to adulthood (e.g., Crain, 1996; Eccles, Wigfield, Harold & Blumenfeld, 1993; Marsh, 1989; 1993a, b). This rich pattern of gender differences in multiple dimensions of self-concept could not be understood from a unidimensional perspective. Intervention Studies: A Multidimensional Perspective To Construct Validation Following from a multidimensional perspective, Marsh (1993a, b; Marsh & Craven, 1997) argued for a construct validity approach to self-concept interventions in which the specific dimensions of selfconcept most relevant to the intervention should be most affected, whilst less relevant dimensions should be less affected and should serve as a control for response biases. This approach was demonstrated in a series of studies based on the Outward Bound program. The Outward Bound standard course is a 26-day residential program based on physically and mentally demanding outdoor activities (Marsh, Richards & Barnes, 1986a, b; also see Hattie, Marsh, Neill & Richards, 1997). Consistent with the primarily nonacademic goals: (a) gains were significantly larger for the SDQIII scales predicted a priori to be most relevant to the goals of the program, compared to less relevant SDQIII scales; (b) the effect sizes were consistent across 27 different Outward Bound groups run by different instructors at different times and in different locations, and (c) the size and pattern of the gains were maintained over an 18-month follow-up period. In contrast, the Outward Bound bridging course is a 6-week residential program designed to produce significant gains in the academic domain for underachieving adolescent males through an integrated program of remedial teaching, normal schoolwork and experiences likely to influence particularly academic self-concept (Marsh & Richards, 1988). Consistent with the primarily academic goals: (a) academic self-concept effects were substantial and significantly larger than nonacademic self-concept effects; and (b) there were also corresponding effects on reading and math achievement. If only self-esteem had been measured in these studies, the interventions would have been concluded to be much weaker and the richness of understanding the match between specific intended goals and actual outcomes would have been lost. The juxtaposition of these two interventions and their contrasting predictions provides a powerful demonstration of the importance of a multidimensional perspective of self-concept. In a meta-analysis of self-concept intervention studies, Haney and Durlak (1998) found modest—but significantly positive—effect sizes. However, reflecting the prevailing unidimensional perspective in many studies included in their meta-analysis, they considered only one effect size per intervention—the mean effect size averaged across different self-concept dimensions if more than one was considered. In contrast, O’Mara, Craven and Marsh (2003) updated and extended this meta-analysis to embrace a multidimensional perspective, coding the nature of the self-concept outcomes in relation to the intervention. Consistent with our multidimensional perspective, effect sizes were substantially larger for specific components of selfconcept logically related to the intended outcomes of the intervention than for self-esteem and other less relevant components of self-concept. Studies designed to enhance global self-esteem were not very successful compared to studies that focused on more specific components of self-concept that were most relevant to goals of the intervention. In summary, intervention research supports the usefulness of a multidimensional perspective of self-concept. Summary and Implications: Multidimensional vs. Unidimensional Perspectives In research reviewed here, specific components of self-concept and global self-esteem have been integrated into a multidimensional, hierarchical model of self-concept. However, based on research comparing the multidimensional and unidimensional perspectives of self-concept, appropriately selected specific domains of self-concept are more useful than self-esteem in many research settings. Clearly it follows that a multidimensional perspective, which incorporates specific components of self-concept and self-esteem, is more useful than a unidimensional perspective that relies solely on self-esteem. Self-esteem can be ephemeral in that it is more affected by short-term response biases, situation-specific context effects, short-term mood fluctuations, and other short-term, time-specific influences. Self-esteem apparently cannot adequately reflect the diversity of specific self domains. Indeed, as emphasized by Marsh and Yeung (1999), it is worrisome that a construct so central to the self seems to be so easily influenced by apparently trivial laboratory manipulations, bogus feedback, and short-term mood fluctuations. In fact, according to modern ethical requirements, such manipulations would probably be unethical if they did have lasting effects on selfesteem. Despite the overwhelming empirical support for a multidimensional perspective on self-concept, we do not argue that researchers should abandon the self-esteem measures that have been used so widely. In fact, consistent with a multidimensional perspective, self-esteem is one of the scales in each of the SDQI, SDQII, and SDQIII instruments that are the basis of much of the research considered here. Rather, researchers should be encouraged to consider multiple dimensions of self-concept particularly relevant to the concerns of their research, supplemented, perhaps, by self-esteem responses. Here we have critiqued research relevant to the ongoing debate about the relative usefulness of a unidimensional perspective that emphasizes a single, global domain of self-concept, typically referred to as self-esteem, compared to a multidimensional perspective based on multiple, relatively distinct components of self-concept with a weak hierarchical ordering. Analogous debates reverberate across different psychological disciplines, where researchers are increasingly recognizing the value of multidimensional perspectives (e.g., multiple intelligences vs. a global measure of IQ to characterize a profile of intellectual abilities). The case for a multidimensional self-concept perspective is particularly strong because the multiple dimensions of self-concept are so distinct that they cannot be explained in terms of a single global component and because they display dramatically different patterns of relations with different background variables, outcomes, and experimental manipulations. These results attest to the powerful explanatory power of specific facets of selfconcept to influence and explain relations among a wide range of constructs of practical significance that serve to inform theory and practice. Although support for this multidimensional perspective is evident in many areas of psychological research, it is particularly strong in educational research where academic outcomes are substantially related to academic self-concept but nearly unrelated to global measures of self-esteem. This emphasis on a multidimensional perspective on relations between self-concept and academic achievement is critical to studies attempting to establish the causal ordering of self-concept and achievement in longitudinal panel studies. Following from the logic of a multidimensional perspective, longitudinal paths models of relations between self-concept and subsequent performance should focus on specific components of self-concept most logically related to the criterion performance rather than, or in addition to, other components of self-concept or self-esteem. Having established the importance of a multidimensional perspective to self-concept, we now demonstrate how a multidimensional perspective is the key to understanding the apparent conflict between Baumeister et al. (2003, 2005) and Marsh and Craven (1997) on the causal ordering of self-concept and subsequent performance. Reciprocal Effects of Academic Self-concept and Performance Theoretical Models of the Reciprocal Effects of Self-concept and Performance Do changes in academic self-concept lead to changes in subsequent academic achievement? The causal ordering of academic self-concept and academic achievement is, perhaps, the most vexing question in academic self-concept research. This critical question has important theoretical and practical implications, and has been the focus of considerable research. Byrne (1984) emphasized that much of the interest in the self-concept/achievement relation stems from the belief that academic self-concept has motivational properties such that changes in academic self-concept will lead to changes in subsequent academic achievement. Calsyn and Kenny (1977) contrasted self-enhancement and skill development models of the self-concept/achievement relation. According to the self-enhancement model, self-concept is a primary determinant of academic achievement. Support for this model would provide a strong justification for self-concept enhancement interventions explicit or implicit in many educational programs. In contrast, the skill development model implies that academic self-concept emerges principally as a consequence of academic achievement. According to this model, the best way to enhance academic self-concept is to develop stronger academic skills. Due largely to limitations in statistical techniques to test these models in the 1980s, researchers argued for “either-or” conclusions. In a review and critique of this research, Marsh (1990a, b, 1993a, b; also see Marsh, Byrne & Yeung, 1999) argued that much of this research was methodologically unsound and inconsistent with the academic self-concept theory. He emphasised that it was widely accepted that prior academic achievement was one determinant of academic self-concept. Hence, the critical question is whether there also exists a causal link from prior academic self-concept to subsequent achievement. The statistical significance and size of this path is of critical importance, whether or not it was larger than the path from prior academic achievement to subsequent academic self-concept. Marsh (1990a; Marsh, Byrne & Yeung, 1999) further argued that a more realistic compromise between the self-enhancement and skill-development models was a “reciprocal effects model” in which, prior self-concept affects subsequent achievement and prior achievement affects subsequent self-concept. Marsh’s reciprocal effects model has major implications for the importance placed on academic self-concept as a means of facilitating other desirable outcomes, as well as being an important outcome variable. Because self-concept and academic achievement are not readily amenable to experimental manipulations, most research relies on longitudinal panel data in which both self-concept and achievement are measured on at least two occasions (i.e., a 2-wave, 2-variable design). With hindsight and 15 years’ experience, Marsh, et al (1999) offered commentary on potential problems and how they can be avoided in future research; demonstrated new, more defensible models of these data; emphasised more generally the role of researcher as substantive data detective; and updated Byrne’s (1984) standards of an “ideal” study and directions for future research. Ideally, studies will: (a) measure academic selfconcept and academic achievement (school performance, standardized test scores, or preferably both) at least twice (i.e., a 2-wave study) and preferably more frequently; (b) infer all latent constructs on the basis of multiple indicators; (c) consider a sufficiently large and diverse sample to justify the use of CFA and the generality of the findings, and; (d) fit the data to a variety of CFA models that incorporate measurement error and test for likely residual covariation among measured variables. In Figure 1 we offer a prototype for a causal ordering study. Despite Byrne’s guidelines and the growing popularity of the SEM techniques, Marsh (1990a) was able to find only three studies meeting Byrne’s criteria (Byrne, 1986; Newman, 1984; Shavelson & Bolus, 1982). Despite some methodological inconsistency in the results of these studies, Marsh suggested that they seemed to be consistent with his earlier suggestion that the effect of prior academic self-concept on subsequent achievement is likely to be greater when achievement is based on school grades. Marsh (1990a) tested the causal ordering of academic self-concept and academic achievement with four waves of data (last 3 years of high school and 1 year after graduation) based on standardized test scores, school grades, and academic self-concept. He found support for reciprocal effects in which the largest paths were from prior academic self-concept to subsequent school grades. Marsh and Yeung (1997a; also Byrne, 1996a updated previous reviews to include new research, but found only nine relevant longitudinal causal modeling studies and again judged none to be fully adequate. They concluded, however, that this research provided reasonably consistent support for a reciprocal effects model. Classic Causal Ordering Study Marsh (1990a) tested the causal ordering of academic self-concept and academic achievement with data from the large, nationally representative (of the US) Youth in Transition study (see Figure 2). He considered data from Times 1 (early 10th grade), 2 (late 11th grade), 3 (late 12th grade), and 4 (one year after normal high school graduation). Three latent constructs were considered: academic ability (T1 only) inferred on the basis of four standardized test scores, academic self-concept (T1, T2, and T4) inferred from responses to 2 (T4) or 3 (T1 and T2) self-rating items, and school grades (T1, T2, T3). Analyses were conducted on responses from the 1,456 students who had complete data at T1, T2, and T3. The initial a priori model (Figure 2) was based primarily on the temporal ordering of the data collection (i.e., T1 variables precede T2 variables). At T1, there were three constructs: academic ability, school grades, and academic self-concept. Academic ability was posited to precede school grades because students were asked to report their grades from the previous year. Similarly, at T2, school grades preceded academic self-concept. At T3 and at T4, only one construct was considered and no casual ordering was necessary. Figure 1. Prototype causal ordering model to test self-enhancement, skill development, and reciprocal effects Models. [Note: In Figure 1 we present a full-forward multiwave-multivariable model in which multiple indicators of academic self-concept and achievement are collected in three successive waves (T1, T2, and T3). Boxes represent multiple indicators (of self-concept or achievement at each wave), ovals represent latent constructs (self-concept or achievement factors), straight, single-headed arrows represent “causal” paths, and curved lines represent covariances. In the full-forward model, each latent construct has paths leading to all latent constructs in subsequent waves. Within each wave, academic self-concept and achievement are assumed to be correlated; in the first wave this is a covariance between two latent constructs; in subsequent waves this is a covariance between residual factors. Correlated uniquenesses (covariances between measured variable residuals) associated with each measured variable are included between occurrences of the same indicator in different waves. Paths connecting the same variable on multiple occasions reflect stability (the solid gray paths in Figure 4), but will typically differ from the corresponding test-retest correlations (which do not include the effects of other variables).] T1-ASC T2-ASC T3-ASC T1-ACH T2-ACH T3-ACH