This paper empirically assesses the effect of the determinants of the share of educational expenditures in the GDP in low-income and lower-middle income countries while taking into consideration the issue of potential simultaneity bias by introducing interaction variables. 
Based on data from the World Bank and using a sample of twenty-five low-income economies in 2014, we find that the share of educational expenditures in the GDP growth is dependent upon both the log of per capita  gross national income and its square, the square of the log of the share of public health expenditures in the GDP, and the Polity IV index of democracy/autocracy.  Using a sample of thirty-eight lower middle-income countries we find that the ratio of education expenditures to the GDP is dependent upon the square of the log of per capita  gross national income, both the Polity IV index of democracy/autocracy and its square, and the log of share of public health expenditures in the GDP. 
When interaction variables are added to the model, regression results show that the share of educational expenditures in the GDP is influenced by the Polity IV index of democracy/autocracy, the square of the log of per capita  gross national income, the log of the share of public health expenditures in the GDP, the interaction term between the Polity IV index of democracy/autocracy and its square, and that between the Polity IV  index of democracy/autocracy and the log of the share of public health expenditures in the GDP.  Since education is an example of human capital, which has been shown to be a driver of economic growth in developing countries (see, for instance, Dao, 2014), assessing empirically the effect of the determinants of the relative importance of education in the economy will enable developing countries to devise strategies aimed at fostering education, and hence at economic growth.  Data for all variables are from the 2016 World Development Indicators and the 2015 Polity Series. We specify and estimate a semi log and quadratic model and observe that some coefficient estimates do not have the expected sign due to possible collinearity among some independent variables.