# counties as ones with per capita income of less than $16,707.^{13 }Since the counties are

ordered based on their real per capita income, they are not necessarily contiguous to each

other. Thus, the spatial models, which employ the contiguity matrix, are not appropriate

to test this hypothesis. Columns (1) and (2) in table 4 therefore gives the results of OLS

regression for poor counties and rich counties, respectively. The standard errors are

clustered by state to account for unobserved correlation of errors within a given state.

# Results from table 4 show that the coefficient estimate on initial level of real per

capita income is the same for rich and poor counties, and is equal to the one obtained for

all counties taken together (Table 2). This result implies that irrespective of whether

counties are rich or poor, the poorer ones catch up with the relatively richer ones at a

similar rate. Initial level of income inequality and poverty negatively affects economic

growth in poor counties.

For rich counties, initial level of inequality reduces subsequent economic growth

and the coefficient estimate is statistically significant at less than 1% level. However,

initial level of poverty increases economic growth in rich counties although the

coefficient estimate is not statistically significant. The high correlation between initial

level of poverty and income inequality of 0.76 in rich counties (as against 0.49 for poor

counties) could possibly cause the coefficient estimates on poverty to be positive in rich

counties. On excluding initial income inequality from the model, I get the coefficient

estimate on initial level of poverty to be negative, which is statistically significant at the

5% level.^{14 }The models explain 79% and 64% of variation in growth rates in poor and

13 The cut-off’s are selected to have roughly equal number of observation under each category (1450 in each). I once again use equation (3); however, estimate it separately for rich and poor counties.

14 Including income inequality and leaving out poverty also gives a negative sign on the coefficient estimate of income inequality, which is statistically significant at the 1% level.

24