# where is the scalar spatial error coefficient. If spatial dependence operates via both the

dependent variable and the error term, in other words, if both and are statistically

significant, then one has to use the general spatial model (SAC) (LeSage, 1999).

# The SAC, which incorporates spatial dependence in both the dependent variable

and shocks to omitted variables in the model, takes the following form: (The weighting

matrix W is the same for all the three models).

y Wy X u u Wu

N ~ ( , ) 0 2

# Data Source and Summary statistics

# The two primary sources of data in this analysis come from a) The Census of

# Population and Housing Summary Tape File 1 and 3 - U.S. Bureau of the Census 1980,

1990 and 2000, and b) the U.S. County and City Data Book 1984. I obtain the data for

local government general expenditure from the latter source, and the rest from the Census

Bureau.

I include all the states and counties in the United States except Alaska because

data for most counties in Alaska are not strictly comparable across 1980 and 2000 U.S.

# Census.^{9 }Table 1 gives the summary statistics, which reveal huge variations in the

variables. While some counties experienced a negative growth between 1979 and 1999,

some counties experienced more than 100% growth rate during the period. The standard

deviation for this variable is very high revealing varied pattern of growth rates across the

counties. San Miguel County in Colorado experienced the highest growth rate of 234%

9

This was mainly because of newly formed counties after 1990.

18