What lies behind South Africa’s improvements in PIRLS? An Oaxaca-Blinder
27 June 2022
The Oaxaca-Blinder decomposition provides a sense of how much of the change in the outcome is due to, say, the presence of more books and how much is due to better utilisation of the books. [...] Column totals, and row totals, each produce the same 28.93 value, which is the change in the country’s PIRLS score, using just the schools available for the analysis (hence the 28.93 differs from the 25 of Table 1). [...] tests, which in the ‘demo’ model of Table 5 emerged as statistically insignificant, reveals that only in 13 of the 269 analyses containing this variable, did the change linked to the coefficient emerge as statistically significant, in the sense that p was less than or equal to 0.100 (here the reference is to the significance levels which would appear in the ‘Coefficients’ column of Table 5). [...] As in the case of the Indonesia and Iran studies mentioned in section 3, most of the change is unexplained in the sense that changes in the regression coefficients could be driven by various unknown factors. [...] The closer each marker is to the intersection of the two axes, the smaller the difference between the mean using all the data, and the mean using just observations from the analysis.