Does boarding benefit the mathematics achievement of primary and middle school students? Evidence from China

ABSTRACT

There is an increasing interest of policies and projects to adapt boarding schools as a means to provide quality education for educationally disadvantaged students. However, little evidence is available for the effectiveness of boarding schools. The current study examined the boarding effect on primary and middle school student mathematics achievement using stratified analysis and propensity score matching approach. The boarding on campus showed a negative and insignificant effect on fourth-grade student mathematics achievement, but a significantly positive effect on eighth-grade student achievement. The mechanisms through which boarding on campus might affect boarders’ achievement were also explored.

KEYWORDS:

• Boarding effect
• boarding school
• mathematics achievement
• contact with families
• school boarding environment

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. Another recoding approach was also tried. For the “boarders” in day schools, if they didn’t answer any specific question developed for boarders, their boarding statuses were recoded as non-boarder; else, their boarding statuses were recoded as missing data. For the “non-boarders” in the 100-percent-boarder schools, if they answered any specific question developed for boarders, their boarding statuses were recoded as boarder; else, their boarding status were recoded as missing data. Analyses were conducted separately with original boarding information from students, and the two different recoded boarding information. The results based on the three sets of boarding information were similar. To be conservative, the results with these inconsistent boarding information recoded as missing data were presented in this paper.

2. With available data on parents’ education level, there were 1784 fourth-grade students and 277 eighth-grade students with missing data on the family structure. These missing data was recoded. When a student indicated both of his/her parents’ education levels, his/her family structure was recoded as “complete family”. When a student indicated only one of his/her parents’ education level, his/her family structure was recoded as “single-parent family”. Else, his/her family structure was kept as missing data.

3. Stratified analysis, PSM, and OLS were conducted to examine the boarding effects in the current study. Each of them had pros and cons, and they could be considered as robustness checks to each other. In the stratified analysis, the comparison was made between boarders and non-boarders who shared the same parental education level in the same school. It made the strictest comparison in terms of unobserved school effects. However, it could not consider any other student-level covariates, and kept the least number of cases. In PSM, the matched treated and control groups shared similar propensities of being a boarder. The school fixed effects and effects of student-level covariates were considered in PSM, and the number of cases it used was larger than that in stratified analysis. In the OLS, all cases were kept while it didn’t balance the treated and control groups.

4. Student gender and student left-behind status were also used as stratified variables to estimate the boarding effects in the additional stratified analyses. They provided similar boarding effect estimates to those using SES stratifications. No heterogeneous boarding effect by either student gender or student left-behind status was found. Their results were not presented.

5. Additional models with interactions of gender and boarding status, and left-behind status and boarding status were built for all samples to examine the heterogeneous boarding effects. No significant heterogeneous boarding effect by either student gender or student left-behind status was found, and the results from these additional models were not presented.

6. The generalized coefficient of determination is calculated as $R^2=1-\{L(\mathbf{0})/L(\hat{\beta }){\}}^{2/n}$, where $L(\mathbf{0})$ is the likelihood of the intercept-only model, and $L(\hat{\beta })$ is the likelihood of the specified model, and n is the sample size. A different set of logistic models were also built with student characteristics, school means of student characteristics, number of students in each school, and the school-level percentage of boarders as the predictors in the model. The coefficient estimates for student characteristics were similar in the two sets of models. As the generalized coefficients of determination were larger in the models with school dummies, the propensity scores based on these models were used for following analyses.

Additional information

Funding

This work was supported by the College of Education and Graduate School, Michigan State University [Summer Research Fellowship].

Notes on contributors

Siwen Guo

Siwen Guo, PhD, is an assistant professor in the Department of Psychology at Renmin University of China. Her research interests focus on the application of quantitative methods to large-scale data and policy research, and the effect of mathematics curriculum on student performance.

Lingyan Li

Lingyan Li, PhD, is a professor of education evaluation at Beijing Normal University. Her research interests focus on the school development policies, looking specifically at how principals’ leadership affect the quality of school managements, and how reforms and school-based self-evaluations affect students’ living situation in schools, and how reforms affect principals’ decisions and effectiveness. She also studies the relationship between schools and national, provincial and local policies.

Yan Sun

Yan Sun, is currently a PhD candidate at Graduate School of Education, Rutgers University, New Jersey, US. His research interests include quasi-experiment designs and advanced psychometric models that can facilitate classroom teaching and learning

Richard Houang

Richard Houang, PhD, is a senior research specialist in the Center for the Study of Curriculum and an adjunct faculty member in measurement and quantitative methods at Michigan State University. His current research focuses on curriculum assessment, the relationship between mathematics and science curriculum and student achievement, item characteristics and student performance, domain-referenced and classroom assessment.

William H. Schmidt

William H. Schmidt, PhD, is a University Distinguished Professor in measurement and quantitative methods and the Department of Statistics, founder and director of the Center for the Study of Curriculum, and director of the Education Policy Center at Michigan State University. His current writing and research focuses on issues of academic content in K-12 schooling, the effects of curriculum on academic achievement, assessment, and educational policy related to mathematics, science, and testing in general.