(1992), and Barro and Sala-i-Martin (1991) for several reasons. The GDP data are from the
most recent and updated estimate series from the BEA. According to Beemiller and
Woodruff (2000) the state GDP data are revised and updated twice annually, with
benchmark revisions occurring approximately every five years. Moreover, our study covers a
longer time period than the previous two studies; 1963—1997 against 1963—1986 and 1963—
1989 for Sala-i-Martin (1991) and Bernard and Jones (1996), respectively.
Educational data were obtained from the Economic Research Service (ERS) for the
years 1970, 1980, 1990 and 2000. Human capital is defined as the average proportion over
the 4 years for the population 25 years and older with at least a 4-year college degree.
There are no capital stock series available for US states by industries. Garofalo and
Yamarik (2002) attempted to construct state-by-state capital stock and gross investment
estimates using data on the service life and amount of capital equipment, and apportioning
the national capital stock among the states. Due to data limitations for sectors, we follow a
somewhat different approach to construct the state capital stock. For each sector, we
constructed the series on the basis of the national capital stock data in constant 1997 prices
(i.e., the stock of privately-owned and government-owned durable equipment and
structures), which were allocated across states using wage and salary disbursements at the
state level.
In order to account for the spatial typology of states, a weight matrix is used. The
weight matrix defines the spatial connection between regions. Due to the typology of the US
states, we consider that a distance-based weight matrix is most appropriate to capture
potential spatial effects. We therefore define the spatial weight matrix on the basis of arc
distances between the geographical midpoints of the states considered. It is a Boolean