Abstract:

In this study we estimate a truncated 2nd order translog production function to explain variation in State of Rhode Island, USA high school achievement indexes. The state-wide three year average mathematics achievement index (MAI) and English language arts index (ELAI) are indices that capture overall performance in their respective domains. For each index, two alternative models are estimated based on a double-logarithmic specification. The motivation for expenditures on education lies in the belief that the education of children is fundamental to the future economic growth and a lasting democracy. Contemporary research offers many examples of including education factors in growth models in order to better understand the implications of education policy on macroeconomic performance as well as local and regional socio-economic economics. For example, Barro and Sala-i-Martin (1995) have found that years of secondary and higher education exposure contribute positively toward economic growth. A number of studies involving small growing economies have examined the relationship between the production embedded in human capital and sources of productivity (Haouas and Yagoubi (2005)). In a slightly different approach, Park (2006) empirically investigated the growth implications of a dispersed population in terms of educational attainment levels. The commonality among these studies and others, is an agreement that there is a statistically significant link between human productivity and economic growth.

This paper extends prior efforts directed at modeling the determinants of predictive factors that explain variability in educational achievement at the high school level. Because achievement models are difficult to estimate efficiently and tend to lack temporal stability this paper introduces an augmented version of the K4 radial basis function artificial neural network (RANN). The K4 RANN has proven robust in modeling financial data across time and against competing ANN topologies (see, Dash et. al. (2003) and Dash and Kajiji (2003, 2008). The new multivariate version, dubbed the K7 RANN, is derived to permit simultaneous estimation of two or more target (dependent) variables in a single equation double-log model specification. This joint and simultaneous approach to the estimation of production elasticity metrics is more representative of the way education administrators actually view student performance – jointly, and not as sequential independent domains.

The paper proceeds as follows. Section 2 presents the economic reasoning supporting the use of a RBF ANN to estimate a double-log production function. Section 3 presents the data and identifies data reduction and transformation related issues. Econometric modeling results are presented in section 4. Among some of the more interesting policy findings we report significant evidence that wealth is correlated with social and economic standing - a finding that suggests inter-generational family wealth may influence current generation literacy rates. Additionally, all solved econometric models find that increasing PPE will not lead to positive changes in MAI scoring. Section 5 provides a summary and conclusion.

Keywords:

Neural Networks, Quantitative Policy Modeling, Analysis of Education.