The Influence of Using Plausible Values and Survey Weights on Multiple Regression and Hierarchical Linear Model Parameters


Creative Commons License

TAT O., KOYUNCU İ., GELBAL S.

JOURNAL OF MEASUREMENT AND EVALUATION IN EDUCATION AND PSYCHOLOGY-EPOD, vol.10, no.3, pp.235-248, 2019 (ESCI) identifier identifier

Abstract

In large-scale assessments like Programme for International Students Assessment (PISA) and the Trends in International Mathematics and Science Study (TIMSS), plausible values are often used as students' ability estimations. In those studies, stratified sampling method is employed in order to draw participants, and hence, the data gathered has a hierarchical structure. In the context of large-scale assessments, plausible values refer to randomly drawn values from posterior ability distribution. It is reported that using one of plausible values or mean of those values as independent variable in linear models may lead to some estimation errors. Moreover, it is observed that sampling weights sometimes are not used during analysis of large-scale assessment data. This study aims to investigate the influence of three approaches on the parameters of linear and hierarchical linear regression models: 1) using only one plausible value, 2) using all plausible values, 3) incorporating sampling weights or not. Data used in the present study is obtained from school and student questionnaires in PISA (2015) Turkey database. Results revealed that the use of sampling weights and number of plausible values has significant effects on regression coefficients, standard errors and explained variance for both regression models. Findings of the study were discussed in details and some conclusions were drawn for practice and further research.