elsewhere (see Rovolis and Spence 1997a and 1997b), the following analysis focuses only on the
infrastructure effects on the other categories of regional GDP.
It has to be noted that all the quasi-production functions of this section, as well as the analysis
regarding the returns of variety for the manufacturing sector (next section), have been formulated
with regional specific effects and a time trend. These are least squares dummy variable models
(LSDV), (see Greene 1997 for an analysis of this models). This means that the error term of equation
35 should be written as:
uit = mi + vit∣ (36)
where μi is the unobservable regional specific effect, and vit is the remainder disturbance. However,
the results for these dummy variables are not reported here due to space limitations. Thus, the panel
of data used in this analysis has an N regional dimension, and a T time dimension.
There are three tables of results regarding the effects of public capital on the non-
manufacturing sector. Table 1 presents the findings for the total of infrastructure taken as a whole,
table 2 examines only the productive part of public capital, and table 3 the social component.
In all these tables a Hausman specification test has been performed. This test can demonstrate
if the hypothesis of the exogeneity public capital is valid or not. Usually the question of whether
infrastructure investment in a specific geographical area is exogenous or not, is circumvented in
empirical work. A potential theoretical justification for this could be that public investment has been
decided on by a central state mechanism, without any relation to the regional output. Another reason
could be the fact that when the effects of infrastructure on the output of the secondary sector of the
economy are examined, the latter can constitute only a rather small part of the regional economy
(when the primary and/or the tertiary sectors are the most significant). However, in the case where the
whole regional GDP is the dependent variable, there is a strong possibility that GDP disparities are the
source of public capital disparities.
The way in which the Hausman specification test is conducted can be found in most
econometrics texts18. In a nutshell, the test principle is that the regressor (of an original regression),
which is to be tested for exogeneity, should be used in an auxiliary regression as the dependent
variable, in which the other regressors (of the original regression) are the explanatory variables. The
residuals of this auxiliary regression are then used as another regressor in the original regression. If
the coefficient of the residuals in this augmented regression is statistically significant, then there is a
simultaneity problem.
18 See for instance Berndt 1991, pp. 379-380, or the original paper of Hausman (1976), and Nakamura and Nakamura
paper (1981)). For the application of the test in a panel data analysis context, see, for instance, Baltagi 1995, pp. 68-
73).
14