Co-Development of Student Agency Components and Its Impact on Educational Attainment—Theoretical and Methodological Considerations

Abstract

Studies of co-developmental processes of student agency components, particularly in relation to educational attainment, are still rather scarce to date. To set the stage for the four respective contributions assembled in this issue, the introduction discusses divergent conceptualizations of co-development and shows how these concepts are translated into statistical modeling. Particular attention is paid to the role of the educational context in which the co-development of student agency components and educational attainment are embedded. Drawing on longitudinal data, the four contributions examine these processes from different angles and with regard to the educational contexts of Finland, Switzerland, the United Kingdom, and the United States.

Research has well documented the separate effects of several components of student agency for educational success (for an overview see Wigfield et al., 2015). Studies have also theorized and empirically examined the associations between agency components and their joint impact on student educational attainment (e.g., Green et al., 2012; Otis, Grouzet, & Pelletier, 2005; Wang & Eccles, 2011). Less attention, however, has been paid to their conjoint development across educational trajectories and, consequently, to the question of how this process relates to educational attainment. Pretty much missing is therefore research on dynamic developmental processes and how they shape educational success. This neglect is partly attributable to the scarceness of longitudinal data that would allow tracking student agency components and their co-developmental processes across the early life course. Partly the neglect is due to the insufficiently spelled-out theories, forwarding arguments about reciprocal effects and feedback loops (Li & Lerner, 2013).

Studies investigating a co-development process of two or more student agency components have done so mostly without examining its impact on an educational outcome of interest (e.g., Archambault, Eccles, & Vida, 2010; Denissen, Zarrett, & Eccles, 2007; Li & Lerner, 2013; Marsh, Trautwein, Lüdtke, Köller, & Baumert, 2005; Skinner, Furrer, Marchand, & Kindermann, 2008). There are, however, some studies that have investigated educational attainment as part of the co-developmental process (e.g., Gottfried, Marcoulides, Gottfried, Oliver, & Guerin, 2007; Hughes, Luo, Kwok, & Loyd, 2008). Importantly, most studies in this field draw on U.S. data. Consequently, the question of how particular educational contexts may shape this co-developmental process has not yet been part of the research agenda. Hence, the role of deadlines imposed by ability tracking or final examinations for shaping co-developmental processes has been largely neglected. We hardly know anything about how the social and institutional embeddedness may bend these dynamic developmental processes.

CO-DEVELOPMENTAL PROCESSES OF STUDENT AGENCY COMPONENTS: DIVERGENT CONCEPTUALIZATIONS

The notion of co-development, often used synonymously with coevolution or conjoint development, captures the idea of the dynamic associations and interrelations between two or more agency components of interest. The literature also refers to the notion of co-development when investigating the reciprocal influence of two actors on their respective agentic development as in the dyadic relationships of parents and children, for example (Nurmi, 2004; cf. also Salmela-Aro, 2009). Although these aspects are also of great importance, this special issue limits itself to examining the developmental processes of within-person agency components. Even this more narrow definition of co-development is not part of the standard conceptual repertoire in the research fields studying the agency-related components of educational attainment. It does therefore not come as a surprise that the specific meaning of co-development encountered in the respective literature does indeed vary. We may discern at least two different understandings of co-developmental processes together with their distinct applications in empirical research.

The first usage of this term applies to the longitudinal development of intraindividual associations of two agentic components of interest. The literature labels this type of developmental processes as “coupling” (e.g., Archambault et al., 2010; Denissen et al., 2007) or “alignment” (Meece, Hutchins, Byun, Irvin, & Weiss, 2013; Sabates, Harris, & Staff, 2011; Schneider & Stevenson, 1999). The analytical interest rests in the developmental patterns of these associations: Does coupling or alignment remain stable or does it become tighter or looser over time? Or does any other form of associational pattern prevail? An excellent example for this understanding of co-development is the study by Denissen et al. (2007). Using U.S. data from the longitudinal study Childhood and Beyond, they investigate the longitudinal coupling between domain-specific achievement, self-ability concept, and interest between Grades 1 and 12 (ages 6–17). The authors find the highest degree of coupling between interest and self-concept of ability, the lowest between interest and achievement. Across time an increase in coupling in all components was observed, however. In the study by Archambault et al. (2010), coupling assumes a somewhat different meaning in that the goal was to identify joint trajectories of ability self-concept and subjective value in literacy from Grades 1 through 12. Also drawing on the Childhood and Beyond longitudinal study, the authors attempted to detect groups of students characterized by distinct trajectories of change in subjective value in literacy and ability self-concepts. Seven distinct patterns of conjoint trajectories were identified, attesting to the astonishing variability of the co-development of a particular task value and ability self-concept.

The alternative understanding of co-development prevalent in the literature attempts to capture the internal dynamics of agency across time (Martin & Liem, 2010; Wang & Fredricks, 2014). The analytical appeal of this conceptualization of co-development rests in deciphering the directionality of effects, identifying the drivers or catalysts of developmental processes and scrutinizing feedback loops. This approach offers the prospect of understanding the dynamic interactions between several agency components: Does change in one component trigger change in another, subsequently retroacting on the former one? Understanding the ways in which reciprocal influences play out offers glimpses into processes of coregulation or calibration, sometimes also referred to as “adaptation.” Exemplary studies within this framework are those by Skinner et al. (2008) as well as Li and Lerner (2013). Skinner and her colleagues examined the internal dynamics of emotional and behavioral components of school engagement in a sample of fourth to seventh graders in a rural-suburban school district in upstate New York. Although the results need to be interpreted with caution as the emotional and behavioral components of school engagement were measured at two time points only, results show that emotional engagement was a predictor of change in behavioral engagement. The authors did not find consistent feedback from behavior to change in emotions. Li and Lerner (2013) analyzed the co-development of cognitive, emotional, and behavioral components of school engagement of students participating in Grades 9 through 11 study waves of the 4-H Study of Positive Youth Development (PYD). They were able to decipher the internal dynamics “in which behavior, emotion, and cognition influence each other and amplify themselves over time” (Li & Lerner, 2013, p. 29). A crucial finding was that emotions were the “point of entry” for the unfolding of the process: “higher levels of emotional engagement lead to more active participation and also higher cognitive engagement” (Li & Lerner, 2013, p. 23).

TRANSLATING CONCEPTS OF CO-DEVELOPMENT INTO STATISTICAL MODELING

The question of interest is how the different concepts of co-development identified in the literature are translated into statistical modeling. Research on coupling/alignment focuses primarily on cooccurrence of particular states or developmental trajectories of two (or more) constructs. Such patterns can be extracted from the data to describe prevalent group differences in the cotrajectories of constructs (e.g., Archambault et al., 2010). Alternatively, they are constructed by classifying combinations of individual scores of agency components in terms of how well they fit together (e.g., Denissen et al., 2007; Sabates et al., 2011; Schmitt-Wilson & Faas, 2016). Based on such classifications researchers are in the position of analyzing the trajectory of fit/misfit across time. This analytic strategy thus allows describing within-person growth patterns of two (or more) simultaneously observed trajectories or of a combined score representing the degree of coupling/alignment. Useful statistical frameworks are offered by hierarchical linear modeling/multilevel modeling (see e.g., Garson, 2013) and structural equation modeling (see e.g., Little, 2013), including latent growth analysis, latent class growth analysis, growth mixture modeling, and others (see, e.g., Curran, Obeidat, & Losardo, 2010; Muthén & Muthén, 2000; Preacher, Wichman, MacCallum, & Briggs, 2008). Typically, the literature conceives the drivers of co-development in terms of coupling/alignment as external factors and not as a part of the costructure itself. When focusing on children and adolescents, the most important external drivers include age and maturation, apparently increasing consistency of students’ motivations and self-concepts (e.g., Denissen et al., 2007). With age, students’ aspirations are also likely to increasingly adjust to (chastening) real-life experiences (Uno, Mortimer, Kim, & Vuolo, 2010).

By contrast, analyses of the internal dynamics of agency conceive the drivers of co-development themselves as constitutive components of the process under study. Co-developing components are specified as reciprocal predictors of each other’s levels or changes, occasionally specifying at least one of them as predictor of the other (Li & Lerner, 2013; Marsh et al., 2005; Martin & Liem, 2010; Skinner et al., 2008). Such conceptions require statistical modeling of structural relations between sequential developmental stages and/or trajectories. Interestingly enough, some researchers speak of “interactive” or “dynamic coupling” between two (or more) constructs (Hawley, Ho, Zuroff, & Blatt, 2006, p. 933). Although the repertoire of pertinent statistical tools is manifold, applications in the realm of structural equation modeling (SEM) are among the most powerful ones. SEM “makes the fewest and allows you to test the most assumptions” (Little, 2013, p. 2), providing options to simultaneously test and compare hypotheses about measurement invariance, structural hypotheses about means and covariance, and temporal hypotheses about changes within and across constructs (Paleari & Fincham, 2015, p. 8). Procedures such as cross-lagged panel modeling and multivariate models of latent change bear particular potential to test assumptions about temporal precedence of constructs and associated growth trajectories (Ferrer & McArdle, 2003; Grimm, An, McArdle, Zonderman, & Resnick, 2012; McArdle, 2009; Paleari & Fincham, 2015).

Differences between methodological approaches in the study of coupling/alignment and internal dynamics tend to mirror the contrast of person-centered versus variable-centered analyses (see, e.g., Laursen & Hoff, 2006). Person-centered approaches capture differences in how constructs relate to each other within persons. Variable-centered approaches grasp interindividual differences in the positions on two (or more) constructs and their interrelations. Here, the basic assumption is that constructs relate to each other in a generalizable fashion. Combining features of person- and variable-centered analyses have become increasingly common, however. An example is the specification of cotrajectory group membership as outcome of individual and social background factors (Archambault et al., 2010). Another one refers to a score of alignment specified as predictor of socioeconomic outcomes (Sabates et al., 2011; Schmitt-Wilson & Faas, 2016). Finally, multigroup analyses are used to examine potential diversity of patterns of interrelations between variables in a population (see e.g., Verboom, Sijtsema, Verhulst, Penninx, & Ormel, 2014).

THE EMBEDDEDNESS OF CO-DEVELOPMENTAL PROCESSES: THE ROLE OF THE EDUCATIONAL CONTEXT

From a conceptual and methodological point of view, co-developmental processes may be approached in different ways. The choice of a particular statistical procedure also depends strongly upon the precise formulation of the research questions. Particularly when interested in the impact of co-developmental processes of agentic components on educational attainment, the accurate phrasing of the research questions is in turn much influenced by the features of the educational system in which these processes take place. Given that most respective studies were conducted with U.S. data, the framing of the research questions in light of the educational system did not advance to the limelight and has hence been much neglected in previous research. It is easy to imagine, however, that co-developmental processes of student agency components differ in the strength of association over time (i.e., coupling), in the timing when level or change are decisive for educational success or in the directionality of effects. These issues need to be put on the research agenda.

Based on the comparative study by Blossfeld, Buchholz, Skopek, and Triventi (2016) we assume that institutional features of educational systems likely to matter for co-developmental processes and their impact on attainment include whether, and if so when, important transitions in educational trajectories take place. Equally relevant is how decisions about students’ allocation to respective educational tracks are taken. In comprehensive educational systems, characterized by weak stratification, co-development of agency components may be most adequately modeled as generalizable and running processes. By contrast, educational systems relying on strong and early-ability tracking tend to foreclose or open up subsequent educational opportunities at the respective transition points. This amounts to saying that the role of agency components for subsequent educational attainment may vary across the educational trajectory. Early manifestations of agency may appear more decisive than later ones due to the channeling of educational trajectories based on early decisions. Further, the co-developmental process may itself be shaped by this channeling.

To consider the role of educational tracking for the co-development of agency components and their power to predict subsequent educational outcomes, longitudinal data are required and modeling procedures need to account for path dependency of tracking. Life-course research has a good understanding of the value of path-modeling techniques, enabling the identification of attainment trajectories, and mediating mechanisms linking early resources and life events to later life outcomes (see, e.g., Baeckman & Nilsson, 2011; Caspi, Elder, & Bem, 1988; see also Cole & Maxwell, 2003). There are also reasons to assume that differences in learning environments after tracking events (see, e.g., Maaz, Trautwein, Lüdtke, & Baumert, 2008) may confound the interrelations between two (or more) agency components. To test such assumptions, statistical tools of choice include multigroup modeling strategies (cf. e.g., Bask & Salmela-Aro, 2013) or multilevel models with students nested in educational tracks, schools, or classes (cf. e.g., Neumann et al., 2007).

Institutional differences of educational systems also require a precise understanding of whether a temporary manifestation or the process of change in agency components is decisive for educational attainment. For example, if transition outcomes are mainly the result of student decision making, the strength of the association between agency components shortly before transitioning may be crucial in guiding a well-informed choice. A statistical model would take this into account by specifying the state of coupling/alignment at a given time point (e.g., the moment of course selection in high school) as predictor of educational attainment. In educational systems where students are tracked (early) by ability, considering not only levels, but also trajectories of student agency prior to the tracking event is crucial. Researchers may want to specify developmental trajectories (the process as such) as predictors of educational outcomes.

Although educational systems vary greatly in regards to whether, how, and when students are tracked, there are systems that include a combination of different features. Research has thus made use of growth analysis to test the role of engagement trajectories for educational achievement (Wang & Eccles, 2011). Others have simultaneously tested and compared the importance of different agency components observed at meaningful time points in student educational trajectory, such as school entry and school examination (Hughes et al., 2008). Particularly when developmental processes are specified as predictors of educational outcomes, the choice of a meaningful starting point of the trajectory (e.g., school entry or the beginning of “urgent” phases of engagement; cf. Heckhausen, Wrosch, & Fleeson, 2001)) is crucial to derive a reasonable interpretation of respective results (cf. McArdle, 2009).

In conclusion, this elaboration highlights that well-informed decisions about modeling procedures and model specifications require the consideration of possible implications educational contexts may have for the ways in which particular agency components relate to educational attainment. The following introduction into the main issues deliberated in the studies assembled in this special issue shows how this may translate into research practice.

THE SPECIAL ISSUE: THE EMBEDDEDNESS OF CO-DEVELOPMENTAL PROCESSES OF STUDENT AGENCY COMPONENTS AND EDUCATIONAL ATTAINMENT

The four studies assembled in this special issue investigate co-developmental processes of young people’s academic agency components from different angles and with regard to four particular educational contexts. Common to all contributions is that they focus on an educational outcome in adolescence or early adulthood that is of particular significance for students in the respective educational systems.

The Finnish educational system, the context of Salmela-Aro and Upadyaya’s contribution, is known for its comprehensive schooling until the end of compulsory education followed by the grade dependent bifurcation into vocational and academic education. For those following the academic track, the one high-stake test is the matriculation examination for university entry. Salmela-Aro and Upadyaya investigate the highest educational degree attained at age 25. The contribution by Steinhoff and Buchmann applies to the Swiss context, known for an educational system characterized by strong ability tracking at early ages. In light of the sequential tracking by ability the Swiss contribution examines whether students are landing the academic track in upper-secondary education. Although the Finnish and Swiss educational systems differ greatly in their institutional arrangements, they nonetheless show a similarity in upper-secondary education as they are characterized by a bifurcation of vocational and general education.

The contribution by Schneider and her coauthors focuses on college enrollment during the fall after high school graduation based on the representative national U.S. Database High School Longitudinal Study 2009. In the U.S. context college education has become a de-facto normative expectation in the “college-for-all era” (Domina, Conley, & Farkas, 2011). Given that the U.S. educational system is not structured around ability tracking but relies more on course selection and grades (Grade Point Average), the question of interest is whether educational aspirations are over- or undershooting occupational aspirations and how this relates to college enrolment. To close the issue, Schoon and Ng-Knight also examine university entry as the educational outcome of interest in the United Kingdom, a country whose system has come to allocate educational opportunities over the past decades primarily based on curricular differentiation and grades (i.e., General Certificate of Secondary Education (GCSE) examinations; A-levels). The U.K. and U.S. educational systems may thus be characterized as the “individual choice model” (see Triventi, Kulic, Skopek, & Blossfeld, 2016, p. 19).

The agency components chosen for the analyses of the co-developmental processes in the four studies show some overlap and some distinctive features. Three articles (Salmela-Aro & Upadyaya, Schoon & Ng-Knight, Steinhoff & Buchmann) examine the co-development of a motivational component (educational goals and aspirations, academic interest) and an effort-related component (willingness to exert effort, effortful engagement or the negative component of burnout). The analyses focus on the reciprocal influence of these two components, the predominant directionality of influence, and the joint impact on the educational outcome of interest. The contribution by Schneider and coauthors, by contrast, examined educational and occupational aspirations in the light of their alignment (i.e., aligned, under- or overaligned). Overall, this special issue assembles contributions that refer to the different conceptualizations of co-development identified in the literature.

Against this background we find difference in the methodological procedures that were chosen to investigate co-developmental processes of various components of student agency in the four different educational contexts. To grasp the development of aligned and misaligned aspirations for educational trajectories in the U.S., Schneider and colleagues analyze changes in the prevalence of (mis-) aligned educational and occupational aspirations from Grades 9 to 11. Using logistic regression analyses, the authors further examine, among other aspects, the consequences of temporary manifestations and trajectories of alignment for students’ likelihood to get enrolled in college.

The other three contributions in this special issue focus on reciprocal effects between two agency components across three or more time points. Schoon and Ng-Knight examine the stability of aspirations and effort across time as well as reciprocal dynamics between the two by specifying a cross-lagged panel model within a structural equation modeling framework. In accordance with the system of general education in the United Kingdom, co-development is specified as a generalizable running process. The authors conceive the level of agency observed in midadolescence as outcome of the foregoing co-developmental process and, in turn, as predictor of educational attainment (university entry).

Similarly, Salmela-Aro and Upadyaya model stability and mutual effects of aspirations and engagement using a cross-lagged path model. Other than the U.K. system, the Finish system includes a vocational and an academic school track in upper secondary schooling. To take the associated implications for individual development into account, the authors conduct a multigroup analysis to identify differences in how agency components relate to each other in groups of students enrolled in vocational and general education, respectively.

Steinhoff and Buchmann specify a reinterpretation of the cross-lagged panel model to include latent change scores of two agency components. In these models, intraindividual changes in one agency component are conceived as self-feedback processes as well as a function of the other component’s previous level (cf. Hawley et al., 2006). This model can be mathematically transformed into a model like the cross-lagged panel model employed by Schoon and Ng-Knight and Salmela-Aro and Upadyaya. However, the idea of modeling latent changes is grounded in another line of historical development and associated with different analytic purposes (Usami, Hayes, & McArdle, 2015). Most importantly, the model directly assesses relative degrees of intra-individual changes in agency. Steinhoff and Buchmann make use of this to decompose the role of students’ initial levels and subsequent changes in motivation and effort for educational attainment across two sequential periods logically divided by “hard” educational tracking deadlines.

WHERE DO WE GO FROM HERE?

The contributions in this special issue present a range of conceptualizations of the co-development of student agency components in the context of distinct educational systems. Together they demonstrate nicely the complexity of agency dynamics in view of intra-individual change, interindividual differences, and education systemic contexts. However, the research presented here represents only a small sample of national educational contexts and offers just a glimpse into the broad spectrum of agency components potentially interacting and complementing each other during the adolescent life course.

The studies also make use of a variety of statistical modeling. However, the full potential of respective tools is by far not yet exhausted. A case in point is that, though models of temporal sequence in the realm of structural equation modeling are powerful tools to examine the directionality of effects between constructs, caution needs to be exercised against prematurely interpreting predictive paths as indication of causality. Adequate conclusions in this regard require careful comparison of parameters derived from models based on different specifications (Usami et al., 2015). They also require preclusion of ignoring potentially important third variables that may be the “true beginning” of a respective time ordered sequence (Paleari & Fincham, 2015, p. 20). Comparative use of different methods and model specifications in the analysis of co-developmental processes is needed, based on diverse samples and including examination of different age spans. Further progress in the study of the internal dynamics of student agency is also dependent upon greater sophistication in strong theory about interrelations of particular agency components and their joint, time-dependent, or consecutive impact on educational attainment in the context of different educational systems. Particularly underdeveloped are assumptions about the mechanisms driving conjoint intra-individual growth patterns of two (or more) agency components. Research is thus needed that decomposes age-related, context-driven, and internal dynamics of student agency trajectories.

In conclusion, there is a huge array of novice research questions looming. To set a stage, this special issue provides an insightful collection of studies highlighting different aspects that compose, trigger, and shape co-development of student agency components. We hope that this collection encourages researchers to get and stay engaged in advancing our understanding of the complexity and the relevance of agency development across the adolescent life course.