Achievement-related within-school socioeconomic gaps in science subjects in China: evidence on existence, consistency, and compensation

ABSTRACT

Using data from the 2011 (Chinese) Student Academic Achievement Evaluation, we examined whether within-school socioeconomic gaps in science achievement exist across science subjects, how consistent they are, and whether there are relationships between school average science achievement and within-school socioeconomic gaps in science achievement. Results of multivariate multilevel analyses indicate that for both fathers and mothers within-school socioeconomic gaps in science achievement existed among schools but did not vary much across schools. School mean socioeconomic status and teacher experience were related to these gaps. Schools were strongly consistent in within-school socioeconomic gaps in science achievement across science subjects, and this consistency was independent of (robust to) student and school characteristics. The relationships between school average science achievement and within-school socioeconomic gaps in science achievement were rather weak among schools across science subjects, and the addition of school characteristics to student characteristics effectively demolished the relationships.

KEYWORDS:

• Within-school socioeconomic gaps in science achievement
• science education
• multivariate multilevel model

Acknowledgements

The authors are grateful to principal investigators of Student Academic Achievement Evaluation (SAAE) for providing data for the present analysis and to Yaya Yuan, Yue Xie, Maohua Wang, and Shiji Feng for their useful comments on an earlier draft.

Notes

1. We did not use the overall science test scores with a higher Cronbach’s alpha. As mentioned earlier, like in many countries, Chinese education teaches science as separate school subjects. A general measure of science achievement is often considered less specific or less meaningful to, say, earth science teachers. So we pursued separate analyses by content areas in science (also to examine any differences in the predictor variables on the different learning outcomes in science).

2. SES was estimated for fathers and mothers separately, but the procedure was the same. We adopted the methodology that O.D. Duncan (1961) employed to produce the Socioeconomic Index (SEI). This approach produces a SES measure descriptive of occupational prestige. Specifically in our case, occupation categories in SAAE (see Appendix 1) were rank ordered within the Chinese socioeconomic context and then normalized. The resulting SES measure was a standardized index appropriate to the Chinese socioeconomic context. For example, the SES score for mothers in the worker sector was –1.283, and the SES score for fathers in the business management sector was 1.558. This SES measure was created based on the entire sample, not the sample that took part in data analysis (with the deletion of some cases necessary for data analysis).

3. To calculate effect size units (as effect size), for each science subject, a statistically significant coefficient is divided by the SD of the outcome (dependent) variable.

4. One way to understand the meanings of these variance components is to take the square root of a variance component to obtain its standard deviation (SD) to accompany the average within-school socioeconomic gap in academic achievement (the intercept). After all, when SES is made random, fixed effect (average) and random effect (SD) behave very much like descriptive statistics (mean and SD) for a regular variable. For example, for scientific inquiry in the case of fathers, mean (average within-school socioeconomic gap in academic achievement) = .020 (Table 3) and SD (square root of variance component) = .026 (Appendix 2).

5. This conclusion from Table 3 is seemingly a contradiction to the results from Table 2 that showed very small effect sizes of SES. Table 2 pertains to the conventional school effects analysis in which SES is fixed at the school level just as any other student-level variables (so that we can estimate the average effect of SES assuming no variation in the effect of SES across schools). In Table 3, SES is random at the school level while other student-level variables remain fixed (so that we can estimate the average effect of SES and the variation in the effect of SES across schools). Furthermore, in Table 2, because SES is fixed at the school level, school-level variables are not used to adjust the effect of SES across schools. In Table 3, where SES is random at the school level, school-level variables are used to adjust the effect of SES across schools. To address our three research questions, the model specification in Table 3 is necessary. Both tables are valid for different research purposes. To avoid confusion and fulfill the research purposes of the present analysis, we would base our discussion on the existence of the within-school socioeconomic gap in academic achievement in Table 3 for the rest of our presentation.

6. Apart from analytical results reported in Table 3, the multivariate multilevel model also estimated the effects of school characteristics on school average academic achievement across science subjects. Later on, apart from analytical results presented in Table 5, the model also estimated correlation coefficients among school average academic achievement across science subjects. Because these results were not the primary concerns of the present analysis, they were not reported but are available from the authors.

Additional information

Notes on contributors

Xin Ma

Xin Ma is Professor of Education Statistics and Professor of Mathematics Education in the Department of Educational, School, and Counseling Psychology, University of Kentucky, Lexington, KY, USA. His areas of specialization are advanced statistical (quantitative) methods, advanced analysis of large-scale (national and international) surveys, psychology of mathematics education, program evaluation and policy analysis, and school effectiveness and improvement.

Jing Yuan

Jing Yuan is a doctoral candidate in the Department of Educational, School, and Counseling Psychology, University of Kentucky, Lexington, KY, USA. Her areas of specialization are advanced statistical (quantitative) methods and science education.

Xingkai Luo

Xingkai Luo is Professor of Science Education at Guangxi Normal University, Guilin, Guangxi, PRC. His area of specialization is science education.