Estimating UK investment in intangible assets and Intellectual Property Rights
27
23
Where on the left hand side we use R&D as a % of sales. On the right hand side, the vector
represents firm level IP use intensity and contains standardized responses for the individual IP
mechanisms. The firm level mean and standard deviation of these responses is captured in the vector
. The IP use vector can be expanded out as:
0 1 2 3
4 5 6 7
ijt
X Design Trademarks Pat ents Copyright
Confidentiality Secrecy Complexity Lead Time
αβ β β β
β β β β
= + + + +
+ + + +
captures the R&D/Sales intensity of those firms that do not rely on any IP mechanism,
captures the elasticity of R&D intensity to patent use holding all other forms of IP use constant, and
the elasticity of R&D intensity to the use of design registration holding all other forms constant.
The
coefficients are not structural parameters and therefore do not represent a causal change in
demand for R&D intensity i.e. we are not saying that firms first decide on IP use and then how much
to spend on R&D. Rather, they are the reduced form correlations and can be regarded as a summary
of the various elasticities. As long as the use/importance of each IP mechanism is measured in the
same units, we can calculate the ratio
which would give the share of spending that would be
accounted for by firms using IP type i as protection, as a share of total spending by those firms using
any form of protection.
As implied by (7) we also include a number of control variables. As noted we include the firm level
mean and standard deviation of IP responses. We also include the % of employees with a degree in
science or engineering, the log of employment and industry dummies.
The R&D relationship is analysed using standard OLS. For Design, we note that questions on design
expenditure are less well answered on the CIS (Awano, Franklin, Haskel et al (2010a)). There are
also many instances where firms respond positively to the question of whether they undertake activity
but the expenditure estimate is zero or missing. Therefore we use the binary question on design
activity and so employ a probit procedure. Since the questions on R&D expenditure appear to be
better answered, we have more confidence in using the expenditure data in the case of R&D.
The results are presented below in Table 4. Column 1 presents a probit regression for design, where
the left-hand side variable is a binary yes/no response and the right hand side includes demeaned
responses for each IP protection mechanism as described above. Also included in the regression, but
Where on the left hand side we use R&D as a % of sales. On the right hand side, the vector
represents firm level IP use intensity and contains standardized responses for the individual IP
mechanisms. The firm level mean and standard deviation of these responses is captured in the
vector
. The IP use vector can be expanded out as:
X
ijt
αβ
Design +
β
Trademarks +
β
Patents +
β
Copyright
(8)
=+
0 1 2 3
+
β
Confidentiality +
β
Secrecy +
β
Complexity +
β
Lead Time
4 5 6 7
Where
captures the R&D/Sales intensity of those firms that do not rely on any IP mechanism,
2
captures the elasticity of R&D intensity to patent use holding all other forms of IP use
constant, and
the elasticity of R&D intensity to the use of design registration holding all other
0
forms constant. The
coefficients are not structural parameters and therefore do not represent
a causal change in demand for R&D intensity i.e. we are not saying that firms first decide on IP
use and then how much to spend on R&D. Rather, they are the reduced form correlations and
can be regarded as a summary of the various elasticities. As long as the use/importance of
each IP mechanism is measured in the same units, we can calculate the ratio
which would
give the share of spending that would be accounted for by firms using IP type i as protection,
as a share of total spending by those firms using any form of protection.
As implied by (7) we also include a number of control variables. As noted we include the firm
level mean and standard deviation of IP responses. We also include the % of employees with a
degree in science or engineering, the log of employment and industry dummies.
The R&D relationship is analysed using standard OLS. For Design, we note that questions on
design expenditure are less well answered on the CIS (Awano, Franklin, Haskel et al (2010a)).
There are also many instances where firms respond positively to the question of whether they
undertake activity but the expenditure estimate is zero or missing. Therefore we use the binary
question on design activity and so employ a probit procedure. Since the questions on R&D
expenditure appear to be better answered, we have more confidence in using the expenditure
data in the case of R&D.
The results are presented below in Table 4. Column 1 presents a probit regression for design,
where the left-hand side variable is a binary yes/no response and the right hand side includes
demeaned responses for each IP protection mechanism as described above. Also included in
the regression, but not shown, are the control variables described above. Column 2 reports the
marginal effects from this regression. Column 3 presents the results of an OLS estimation,
where the left-hand side variable is R&D/Sales and the right hand side includes demeaned
responses on the importance of IP mechanisms and the control variables. In each regression,
in order to identify the idiosyncratic impact of the different mechanisms we assign one mechanism
as the reference category and constrain its impact to zero.