or in working form:
ln ML = (a - 1)ln nu + ln Glt +1 + uit
nit
(39)
where Mit is the Gross Production Value of manufacturing (as a proxy of the finished manufactured
goods) in a specific region or sector in time t, n the number of manufacturing establishments (as an
index of the range of economic activity), a the degree of homogeneity in equation 37, Git the
infrastructure variable, t is a time variable, and eit the error term of the form eit = μi + vit, [where μi is
the unobservable regional or sectoral specific effect, and vit is the remainder disturbance]
A point that begs clarification is the fact that public capital in the above equation appears to be
a pure public good. However, as Holtz-Eakin and Lovely (1996, p. 119, footnote 17) argue “entering
public capital in per-firm units would not affect its coefficient. Instead, only the coefficient on the
growth of firms (and our estimate of a) would be affected”.
A similar formulation has been employed for the Greek case. However, here four different
datasets were used. The first comprises information on manufacturing (large industry, employing
more than 20 persons) for the 49 prefectures. (For a more detailed analysis see Rovolis and Spence
1997a, 1997b, and 1998.) It has to be remembered that these manufacturing data refer to the total of
all industrial sectors, no sectoral breakdown being available. The analysis by Holtz-Eakin and Lovely
has had the added luxury of regional data together with a sectoral breakdown.
There are, however, three other datasets at hand, which do have a sectoral dimension. The
first provides a sectoral breakdown for Greece as a whole. The second has a similar breakdown for
the metropolitan area of Athens, and the last refers to the Rest of Greece. The last mentioned is a
derivative set of data, as it is the difference between that for the whole of Greece and the Athens panel
(again for more details on these datasets see Rovolis and Spence 1997a, 1997b, and 1998). These
data, along with the regional panel, allow analysis of the different channels by which public capital can
affect manufacturing sector at four different spatial levels.
Gross Production Value (GPV) has been used as a measure for manufacturing output in tables
4 to 7, in which are presented the results for the different datasets. Public capital has been introduced
again either as total infrastructure (but excluding Miscellaneous and Administrative expenditures as
before) or as a breakdown into productive and social infrastructure (again as defined previously). As
in all previous cases where panel data analysis has been used, regional dummies were introduced into
the regressions to capture the regional specific effects. This constitutes the Least Squares Dummy
Variable model. This model and the organisation of the dummy variables designed to capture the
regional effects (and similarly the sectoral effects for the sectoral panels) are described in section 4.4.
19