5.2. Variables
The dependent variable is the innovative intensity (innovation per employee)
in the LPS, expressed as the annual average of patents per employee between
2001 and 2006 and using 2001 as the base year for employment.
R&D data comes from two sources. First, as in Boix and Galletto
(2008a), R&D by LPS in the year 2001 was assigned from regional data
departing from regional R&D intensity per employee in each institutional
sector (business sector, universities and public administrations) and
multiplied by the jobs by institutional sector in each LPS. Since university
R&D and jobs are concentrated in few LPS, which cause problems with the
logarithms, the data was grouped into two categories: business and public
R&D. Second, business R&D expenditures have been directly collected from
microdata (SABI database by Bureau van Dijk). The average expenditures
between 1998 and 2001 have been used in order to reduce the variability of
microdata by year. This approach to business R&D is considered to be more
precise.
Since there are 206 LPS without innovations for which logarithms
cannot be computed, the problem is treated as a censured sample by means of
a Heckman estimate of the fixed-effects model.
5.3. Results
The results of the estimates are divided in two tables. The first table (Table 4)
contains the input coefficients (R&D) and the basic tests. For the detailed
interpretation of the fixed effects, a table of results is proposed where the
combined effects are in the central part, and the separated territorial and
specialization effects are in the margins (Table 5). The main findings can be
summarized in four points:
1. The results for input variables show that both business and public
R&D are economic and statistically significant for the three estimated
models. The coefficients for business R&D range between 0.26 for imputed
data and 0.09 for microdata. Public R&D ranges between 0.19 and 0.24
(Table 4).
2. Similar to Boix and Galletto (2008a), the estimates of the fixed
effects by territory (Table 5, lower row), provide robust evidence of the
existence of an I-district effect of 0.49 in unitary deviations from the
averaged group effect, and close to the 47% deduced from Table 19. The
manufacturing LPS of large firms have a fixed effect of 0.11 although it is
not statistically significant. The other manufacturing LPS also show a high
9 The reported estimates refer to business R&D expenditures from microdata, which
are considered to be more precise. Fixed effects reported using R&D assigned from
regional data are very similar.
15