SES-Achievement Gaps in East Asia: Evidence from PISA 2003–2018

Abstract

Widening achievement gaps driven by socioeconomic disparity have become a global concern. Yet, few studies have been able to track the changes and developments of socioeconomic-achievement gaps (SES-achievement gaps henceforward) across time. Using PISA data from 2003 to 2018, we estimated family SES-achievement gaps in seven Asian education systems. The findings suggest that in mainland China the gaps are most pronounced, whereas in Macao, they remain consistently modest. Other systems fall in between these two extremes. Based on the new synchronic and diachronic pieces of evidence, we extrapolate possible explanations of varying SES-achievement gaps across education systems. We found that the gaps tend to be smaller when the low-socioeconomic students perform better in academic environments. Based on our analysis, we argue that if the policy goal is to reduce the learning gap associated with social and economic division, public attention in education should be directed towards students from disadvantaged backgrounds.

Appendices

Appendix

Appendix A. SES-Achievement Gap Trends in Reading and Science

See Figs. 

Fig. 3

SES-achievement gap trends in reading

3,

Fig. 4

SES-achievement gap trends in science

4.

Appendix B. Computing SES-Achievement Gaps and Respective Standard Errors

Given PISA uses an assessment design known as the multiple-matrix sampling (Shoemaker, 1973), none of the students completed the full-length test but rather were allocated a portion of the test (Rutkowski et al., 2010). As a result, although the traditional “average out” approach still works for the point estimate, it does not work for the standard errors. Therefore, secondary analysis with plausible values follows Rubin (1987)’s rules.

Let $\overline{Q}$ denote the final point estimate and U its estimated variance. Here are equally plausible estimates $\hat{{Q}_{1}}$,$\hat{{Q}_{2}}$,…,$\hat{{Q}_{m}}$ ( equals to 5 or 10 in PISA) and their corresponding variances ${U}_{1}$,${U}_{2}$,…,${U}_{m}$.

We first compute the overall estimate, namely an average of all point estimates, which is given by

$$\overline{Q}=\frac{1}{m}\sum_{i=1}^{m}\hat{{Q}_{i}}.$$

The within-imputation variance is simply the average of the estimated variances:

$$\overline{U}=\frac{1}{m}\sum_{i=1}^{m}{U}_{i}.$$

The between-imputation variance is the sample variance of the estimates themselves:

$$B=\frac{1}{m-1}\sum_{i=1}^{m}(\hat{Q}-\overline{Q}{)}^{2}.$$

The total variance, is the sum of variance within and between imputation with an additional correction factor to account for the simulation error in $\overline{Q}$,

$$ T = \overline{U} + \left( {1 + \frac{1}{m}} \right)B $$

The square root of T is the overall standard error associated with $\overline{Q}$. The 95% confidence interval can be obtained using the approximation

$$\overline{Q}\pm {t}_{df}\sqrt{T},$$

where ${t}_{df}$ denotes a quantile of Student’s $t$-distribution with degrees of freedom

$$df=(m-1)(1+\frac{m\overline{U}}{(m+1)B}{)}^{2}.$$

P-values can be obtained by comparing the ratio $\overline{Q}$/$\sqrt{T}$ to the same $t$-distribution (Schafer & Olsen, 1998).

Appendix C. Mean Performance of PISA Scores at Specific SES Percentiles

See Tables 3, 4, 5.

Table 3 Mean performance of PISA score at specific SES percentiles (Math)

Table 4 Mean Performance of PISA score at specific SES percentiles (Reading)

Table 5 Mean Performance of PISA score at specific SES percentiles (Science)