X hits on this document

PDF document

Poverty, Income Inequality and Economic Growth in U.S. Counties: - page 16 / 33





16 / 33

fixed effects could reduce this problem considerably. Second, I use income inequality,

not asset inequality. While Deininger and Squire (1998) argue that asset inequality is a

more robust determinant of inequality than income inequality, Aghion et al. (1999), claim

that both income and wealth inequality vary together in cross-sectional data. Data

limitations preclude me from using asset inequality and compare it with income

inequality measures.

Spatial Model

Several studies show that geographical location and location parameters

significantly affect productivity, inequality and growth (Quah 1996, Redding and

Venables 2002, Rupasingha et al. 2002, Rupasingha and Goetz 2007 among many

others). I therefore do not employ OLS, which assumes that errors from different counties

are independent. The presence of spatial dependence can thus yield misleading results

from employing OLS (LeSage 1999). For instance, growth in a specific county can have

spill over effects to the neighboring county, in which case the errors are dependent. Thus,

I estimate equation (3) by three alternative spatial specification models, namely the

Spatial Autoregressive (SAR) model, the Spatial Error Model (SEM) and the general

spatial model (SAC) that accounts for dependence in the error terms arising out of spatial

dependence. These models employ the Maximum Likelihood Estimation method for

estimation and I use Matlab to estimate the three spatial models.

The SAR model accounts for the spatial dependence in the dependent variable and

the SEM incorporates spatial dependence in the error term. The SAR model takes the

following form:


Document info
Document views126
Page views126
Page last viewedSun Jan 22 18:28:57 UTC 2017