When data from the NCES (National Center for Education Statistics) national study, Early Childhood Longitudinal Study: Kindergarten 2011 (ECLS-K), first became available, Ya Mo and Nell Sedransk began an examination of students’ characteristics and demographics and their influence on learning in reading and math during the kindergarten year. ECLS-K is particularly suited for this because it involves paired assessments at the beginning and at the end of the school year in both reading and math. So it is possible to study competence upon entry into kindergarten, the attainment by the end of kindergarten, and the growth during the kindergarten year.
It quickly became apparent from a standard (hierarchical) modeling approach that not only were demographic and other background factors interrelated in complicated ways or sometimes redundant, but the salient factors were not the same for all students. So what began as a “straightforward analysis” of factors important to kindergarten learning expanded into a more detailed analysis using classification/regression tree models (CRT). These models allow the data analysis to be driven by the data’s internal structure so that complex relationships can emerge. As the work progressed, the value of a nonlinear modeling structure for complex education data like these became clear: for these ECLS-K data, tree-based methods of analysis revealed important regional, ethnic, and classroom context factors that were jointly affecting kindergartners’ learning, but in different ways for different subsets of students.
At the same time, these data provide an excellent case study that demonstrates the advantage of using a model that allows for non-uniform impacts of factors across the population. Thus the manuscript was expanded to introduce tree-based methods more generally for the study of student achievement and other education metrics and to offer the fully worked case study as an exemplar. The manuscript has passed through NCES disclosure review and been submitted to a journal for peer-review.
Ya Mo began this study as a NISS Senior Postdoctoral Fellow; she is now on the faculty of the College of Education at Boise State University, where she continues to teach and carry out research. Ya also continues as a NISS research partner; with Brian and Alexi now joining the team, new research directions will focus on applied psychometric research, math performance metrics, and statistical modeling approaches, both nonlinear as well as linear.