What we learn in school: Cognitive and non-cognitive skills in the educational production function

This dissertation revisits the traditional educational production function, offering alternative strategies to model how achievement and socio-emotional skills enter the relationship and how they are affected during the schooling period. The proposed analyses use a combination of estimation methodologies (longitudinal, multilevel and simultaneous equations models) to empirically assess the importance of the different inputs in the educational process. These estimates can be compared to those obtained using traditional estimation methods to complement our understanding of what educational outcomes are generated in school and which school inputs are most important in producing certain outcomes. The analyses try to provide a broader understanding –both conceptually and statistically- of how education is produced and unbiased estimates of the relative importance of the determinants of academic and behavioral performance.;Study design and methods. This dissertation is composed of three empirical questions about the conceptual and statistical structure of the educational production function, aimed at identifying what educational outcomes are generated and what determinants affect them. The empirical analyses use the Early Childhood Longitudinal Study, Kindergarten Class of 1998–99. 1. Estimation of cognitive achievement: an overview of the traditional educational production function; 2. Estimation of non-cognitive achievement: educational production function for non-cognitive skills; 3. A simultaneous equations model of the determinants of educational outcomes: achievement and behavioral skills.;Question 1 involves the estimation of the production of cognitive skills, and educational achievement in reading and mathematics; Question 2 involves the estimation of the production of non-cognitive skills in school, and particular behavioral skills such as internalizing and externalizing behavioral problems, and self-control (reported by the teacher). I use three different estimation methods for both questions: ordinary least squares; students’ fixed effects; and multilevel students’ fixed-effects.;In Question 3 I model the production of simultaneous outcomes, using a cross-sectional and a dynamic simultaneous equation model of the production of education. This framework is an attempt to account for simultaneity and interdependence between outcomes and several educational inputs, leading to a more realistic formulation of how different educational ingredients can be interrelated over time, and acknowledging that educational components can be both inputs and outputs of the process, at different points in time. The estimation methods are three-stage least squares for the cross-sectional estimates; and within-three stages least squares for the longitudinal model.;Findings. The findings obtained from the estimation of the three research questions indicate, in accordance with the existing literature, that the associations between teacher and schools characteristics and the production of cognitive skills and non-cognitive skills are small and mainly statistically insignificant.;First, the results using students’ fixed effects estimation suggest that the effects of teacher’s educational attainment on the cognitive skills index; and experience on the non-cognitive skills index. Some effects of class size are also detected for the production of both skills.;Secondly, the estimates using the multilevel students’ fixed effects estimation, which controls for the clustered structure of the ECLS-K dataset, indicate that some the effects of certain school level characteristics are statistically significant for the production of reading achievement. These variables are type of school (Catholic school versus public) or class size (medium size versus small). Similarly, the effects of certain teacher characteristics, such as higher educational attainment are statistically significant for the production of mathematics achievement. Regarding the non-cognitive skills, teachers with more experience lead to better non-cognitive skills, while students in private schools, versus students in public schools, have lower non-cognitive skills, as reported by their teachers.;Finally, the results using the cross-sectional simultaneous equations model confer a statistically significance importance to the associations between cognitive and non-cognitive skills in all the grade-levels. Compared to the teacher and school characteristics associations, the coefficients associated with the simultaneous relationships are educationally important.;Policy implications. The design of a comprehensive model of educational outcomes and the study of their associations with the different school inputs are expected to uncover interesting features of the educational production process. Consequently, and building on all the existing knowledge on the production of education, these analyses can help to shed some light on fundamental knowledge for educational research: to better understand the educational process. The empirical findings arising from the study can be useful for informing policymakers and school practitioners and guiding decision making, by offering complementary frameworks that more accurately represent the educational process. Finally, the results may be useful for designing and evaluating educational interventions that are efficient and effective in producing higher quality and quantity of educational outcomes, by incorporating the assessment of non-cognitive skills into the interventions’ expected outcomes. Indirectly, this could also stimulate the creation of newer theoretical frameworks, statistical methods and more comprehensive empirical sources for the study of education.