A Meta-Analysis of Selected Studies on the Effectiveness of Bilingual Education

Abstract

A meta-analysis of selected studies on the efficacy of bilingual education was conducted and the results were compared with a traditional review of the same literature. When statistical controls for methodological inadequacies were employed, participation in bilingual education programs consistently produced small to moderate differences favoring bilingual education for tests of reading, language skills, mathematics, and total achievement when the tests were in English, and for reading, language, mathematics, writing, social studies, listening comprehension, and attitudes toward school or self when tests were in other languages. The magnitude of effect sizes was influenced by the types of programs compared, language of the criterion instruments, academic domain of the criterion instruments, random versus nonrandom assignment of students to programs, formula used to calculate effect sizes, and types of scores reported in the studies. Programs characterized by instability and/or hostile environments were associated with lower effect sizes. The synthesized studies contained a variety of methodological weaknesses which affected the magnitude of the effect sizes. Initial group differences—in language dominance, in environmental language exposure, in need for the bilingual program—were not uncommon. In some cases, comparison groups contained bilingual program “graduates.” In others, experimental groups changed in composition during the study through the exiting of successful students and their replacement with newcomers subsequent to pretesting and prior to posttesting. Although the technique of meta-analysis allows for statistical control of methodological inadequacies, the methodological inadequacies in the synthesized studies render the results less than definitive and highlight the need for quality research in the area of bilingual education. Problems inherent in conducting research on bilingual programs are discussed in relation to the outcomes of this synthesis, and guidelines for future research are proposed.