evaluating the variables for their statistical significance is not a big concern. Therefore, I
include both poverty and income inequality in the regressions.10
Columns (1) through (3) show the results from Spatial Auto Regressive, Spatial
Error and General Spatial models respectively. Since the coefficient estimate on both the
autoregressive parameter, (from SAR) and spatial error (from SEM) are statistically
significant, inferring the results from SAC model (Column 3) is preferred over the other
two models (LeSage 1999).
Column (3) shows that the coefficient estimate on initial level of per capita
income is negative indicating conditional convergence. In other words, counties with a
lower initial level of income grow faster and catch-up with rich counties over time.
Hundred dollars increase in the initial level of real per capita income leads to 0.05
percentage point decrease in the subsequent growth rate, which is statistically significant
at the 1% level. To explain, if a county has $ 14,615 (mean in the sample) as its’ initial
level of income and grows at 4% per annum (on average), then ceteris paribus, another
county having 14,715 dollars as its’ initial level of income would grow only at 3.95% per
annum. This result is indeed consistent with the theory that poor economies on average
grow faster than rich ones, which is supported by many studies (Barro 1991, Sala-I-
Martin 1996, Barro 1999).
The coefficient estimate on initial level of income inequality is also negative
indicating that counties with higher income inequality than others in 1979 experienced
slower growth between 1979 and 1999, which is consistent with the credit-market
imperfection theory, explained earlier. Column (3) shows that a ten-percentage point
10 As a robustness check, I ran the regression with only one of them; the results were still similar to the ones obtained in Table 2.