Affluence, Inequality, and Educational Achievement: A Structural Analysis of 97 Jurisdictions across the Globe

Abstract

How does economic inequality affect societies' average levels of student achievement, and how might the answer to this question inform our understanding of U.S. students' achievement in cross-national perspective? Analyses of country-level data (N = 97) from three recent international studies reveal that while more affluent countries exhibit higher average achievement than less affluent countries, more unequal countries have lower average achievement than less unequal countries—even net of affluence and other factors. This helps explain why U.S. students score relatively low compared to their counterparts in other developed nations. I discuss the implications of these findings for understanding the role of stratification in societies generally and for raising achievement in the U.S. specifically.

Acknowledgments

I thank Jim Ainsworth, Claudia Buchmann, Tomeka Davis, Doug Downey, Kendra Freeman, Linda Grant, Anne McDaniel, Linda Renzulli, and Regina Werum for helpful comments and suggestions.

Notes

¹This study involves both countries and jurisdictions. The term “jurisdiction” refers to political entities or economies that are not countries but contribute a case to my data set along with the actual countries. Throughout the study, I use the terms “countries” and “jurisdictions” interchangeably.

²Basque Country (Spain) and Malta were missing socioeconomic background data while Denmark, France, and Portugal were missing school resource data. Given the need to maximize the N, and since I am only missing data on one variable for a few jurisdictions, I retain these cases for the regressions by imputing values. The imputed affluence measures are based on school resources while the imputed school resource measures are based on affluence. The pattern of findings is the same using listwise deletion.

*p < .05; **p < .01; ***p < .001 (two-tailed tests).

⁺ p < .10; *p < .05; **p < .01; ***p < .001 (two-tailed tests based on robust HC3 standard errors).

³Heteroskedasticity (variation in error variances across observations in OLS regression) presents a potential problem due to the aggregate nature of the data and the relatively small number of cases. When heteroskedasticity is present, OLS regression estimates are not biased but tests of statistical significance often are. It is therefore beneficial to conduct a test based on a heteroskedasticity consistent covariance matrix (HCCM). Long and Ervin (2000) explored the impact of four types of HCCM tests and concluded that whenever heteroskedasticity is suspected, a significance test known as HC3 should be used for samples sizes under 250. Therefore, I use HC3 tests of significance in all regression analyses to ensure that standard errors and thus levels of statistical significance are estimated properly.

⁴Multicollinearity could present a problem in the regression analyses, as Table 3 reveals several moderate correlations between independent variables. None of these correlations, however, eclipses .7. In addition, the relationships estimated in the regressions are consistently in the same direction as those estimated in the correlations. Finally, collinearity diagnostics run on each model show that variance inflation factors (VIFs) typically fall in the range of 1 to 3 with a mean around 2, with the highest VIF being 3.65 for socioeconomic affluence in Model 5. Therefore, multicollinearity is probably not causing problems in the regressions.

⁵Cross-national variation in the length of school years constitutes one potentially important factor. PISA, however, does not contain this information. In supplemental analyses, I limited the sample to the 65 cases that contain the number of days in the school year (from the TIMSS and PIRLS data) and assessed the impact of this measure on achievement. It did not exhibit a significant effect and did not alter the conclusions drawn from Table 4.

^a Coded as an affluent jurisdiction.

^b Coded as an other (non-country) jurisdiction.