Complex interplay effects of multiple classroom instructional and context factors on student growth in EFL in Vietnam: The case of nonlinear, regularized regression and the relevance of the scaling model (Dissertation)

"The primary aims of the study are (1) to identify classroom instructional factors which have a crucial effect on the academic growth of ninth-graders in EFL in Vietnam, and (2) to gain insight into their interplay with each other and with context factors. Besides, this study has a strong focus on methodological approaches: (a) using multiple methods in order to deal with the “large p, small n” problem, (b) to understand the relevance of the scaling model used for the results. Data from a research project carried out in Vietnam during the school year 2006–2007 were used in this study. Besides a longitudinal design with two measurement points (MPs) using adapted English tests and questionnaires from the DESI-study in Germany, a video study was conducted in the middle of the school year between two MPs. The recorded video data were transcribed, micro-analytically coded, and lessons were rated to gain indicators of classroom instruction. Different IRT scaling models were chosen to estimate student ability in the pretest and posttest. For the C-test, the unidimensional 1PL and 2PL models, the Rasch testlet model, and testlet 2PL model were selected to model student ability. To estimate student ability via the listening comprehension test (LC-test), the Rasch model, the unidimensional 2PL, and 3PL models were applied. The student ability estimates at the two MPs were linked to one common scale using the concurrent calibration approach with different a priori ability distributions. The plausible values (PVs) were generated and treated as student ability estimates for all analyses. To understand the relationship between the instructional variables and student growth, we explored the hypothesized linear and nonlinear, additive and interactive effects of classroom instructional factors. To examine these hypothetical effects, OLS and regularized regression models using lasso (least absolute shrinkage and selection operators) were applied, including main effects as well as quadratic and interaction terms of instructional variables. Initial student ability and the socioeconomic status of students were treated as context variables. The results show, on the one hand, a positive view of important general instructional quality dimensions of teaching effectiveness and, on the other hand, a strongly teacher-centered and textbook-driven instruction and poor instructional quality from the point of view of EFL didactics. The most important instructional factors of student growth in the C-test were quality aspects of motivation in instruction as well as aspects related to the teaching language. Regarding the LC-test results, language-related aspects together with the relative frequency of repeated questions were the most important predictors of student growth. While the findings confirmed all the hypothesized instructional effects on student growth, aptitude treatment interaction effects of instruction were only confirmed with regard to student growth in the C-test. The different scaling models produced significant differences in the results regarding instructional effects on student growth. Online publication: https://kola.opus.hbz-nrw.de/frontdoor/index/index/docId/1617 "