Year: 2010
Type: article
Study design: Observational Study
Abstract: New models are commonly designed to solve certain limitations of other ones. Quantile regression is introduced in this paper because it can provide information that a regular mean regression misses. This research aims to demonstrate its utility in the educational research and measurement field for questions that may not be detected otherwise. Quantile regression is appropriate when the assumption of a normal distribution of the error term is violated. It is most useful when the interest is at various locations along the complete distribution rather than just the central tendency. The first part of this research used quantile regression to explore a changing relationship between language proficiency and math achievement. Results reveal that language proficiency affects math achievement differently at different math ability levels. Other commonly used covariates such as social economic status and gender are also related to math achievement differently at different locations on the math score distribution. It is shown that regular mean regression analyses fail to capture this information. The second part of the research models math growth longitudinally adjusting for language proficiency. Four rounds of data for a cohort of students are used to detect the long term math achievement gap between English Language Learners (ELLs), Former ELLs and NonELLs. Model-building process suggests that language demand in tests may have contributed to the big achievement gap between ELL and Non-ELLs. Long term and differential effects of other background variables are also detected. Implication of the results and limitations of the technique are discussed.
Author: Fang Chen
Language: English
Subfield: Education
Field: Social Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W267390009