Environmental Engineering Reference
In-Depth Information
emissions
as a control variable representing the environmental system which can
capture any possible inducement effect due to the stringency of different national
environmental regulations. By doing this, we seek to test the potential effects of
environmental regulations operating as a mechanism through which cleaner tech-
nological change can be induced, which has positive impacts on the countries in
which regulations are in force [
14
]. More precisely, we rely on the hypothesis that
the lower the level of CO
2
per capita, the higher the level of technological capa-
bilities is, measured by the patent count. Since this variable measures the
—
nal goal
of environmental regulations
lower carbon intensity, that is the impact of the
overall level of emissions from all sectors in a country weighted by its population
—
—
we capture any environmental induced-effect policies. Moreover, due to the gen-
erality of this variable, the analysis is also effective in countries where the frame-
work of green regulation is weak but other implicit mechanisms are at work, as for
instance in the case of Italy [
23
]. We use emissions data from IEA CO
2
Emissions
from Fuel Combustion Statistics [
38
], measured in Mt of CO
2
.
4 Econometric Strategy and Empirical Results
The use of patent data as a proxy of innovation-related activity means that we have
to deal with count variables, i.e. variables with non-negative integer values. In our
analysis, the variable under scrutiny is the patent count. Patent data on the EE
residential sector are divided into three sub-sectors according to policy data:
buildings, lighting and four electrical appliances (refrigerators, freezers, washing
machines and dishwashers), respectively. As con
rmed by Hausman et al. [
28
] and
Baltagi [
5
], these data usually show a high degree of skewness with upper tails over
dispersion (relatively low medians and high means) and a large proportion of zeros.
Such features can re
ect observed factors such as the size of
rms (larger
rms
usually
rm
may patent less than another but produce breakthrough technologies). Empirical
literature suggests speci
le more patents than smaller ones) and unobserved heterogeneity (one
c modelling strategies for dealing with patents which can
be reduced to two main options: the Poisson Regression Model (PRM) and the
Negative Binomial Regression Model (NBRM). When the dependent variable is
affected by the presence of many zeros the Zero-In
ated Negative Binomial Model
(ZINB) may also be a good modelling strategy (for a comprehensive explanation
see [
11
,
78
]. In our dataset, the presence of zeros in the dependent variable is
negligible and Vuong
s test does not justify the use of the ZINB.
6
In the light of
this, we decided to use the NBRM, in which the variance is modelled as a quadratic
term (NB-2). Equation (
4
) represents the general expression of the models esti-
mated, taking into account the
'
ve groups of variables for the speci
c drivers of
innovation described above:
6
Test results are available upon request.
Search WWH ::
Custom Search