Self-Concept Research: A substantive-quantitative synergy
Invited talk by Herb Marsh, Department of Educational Studies, University of Oxford, UK.
Chair: Tim Croudace, Wednesday 22nd July, 17.15 - 18.00, Uppercroft, School of Pythagoras.
My self-concept research programme represents a substantive-quantitative synergy, applying and developing new quantitative approaches to better address substantive issues with important policy implications. Contemporary theory and research shows self-concept to be a multidimensional hierarchical construct with highly differentiated components (academic, social, physical, emotional self-concepts) as well as a global self-esteem; this multidimensionality is a central theme in my research. Self-concept is an important mediating factor that facilitates attainment of many desirable outcomes. In education, for example, a positive academic self-concept (ASC) is reciprocally related to subsequent academic accomplishments. I begin with an overview of my self-concept research in which I address diverse theoretical and methodological issues with practical implications, and then focus on the big-fish-little-pond effect (BFLPE).
The BFLPE predicts that equally able students have lower ASCs when attending schools or classes where the school-average ability is high, and higher ASCs when the school-average ability is low. The effect of individual achievement on ASC is positive, but the effect of school-average achievement is negative. BFLPE effects are remarkably robust, generalizing over a wide variety of different individual student characteristics, school settings, countries, and long-term follow-ups. Here I briefly summarize alternative tests of the BFLPE with particular emphasis on evolving quantitative approaches used over the past 25 years. Early research applied inappropriate single-level models based on manifest measures that ignored the multilevel structure and unreliability. Recently developed doubly-latent models simultaneously control and unconfound unreliability due to measurement error at the individual (L1) and group (L2) levels, and sampling error in the aggregation of individual characteristics to form group level constructs; extensions include latent non-linear effects as well as single-level (L1xL1) and cross-level (L1xL2) latent interactions.