To these are added XINF, which refers to the level
of cross-cutting resource commitments and policy
choices that constitute the common innovation infrastructure,
YCLUS, which refers to the particular
environments for innovation in a countries’ industrial
clusters, and ZLINK, which captures the strength
of linkages between the common infrastructure and
the nation’s industrial clusters. Under Eq. (2), we
assume that the various elements of national innovative
capacity are complementary in the sense that the
marginal boost to ideas production from increasing
one factor is increasing in the level of all of the other
factors.
Deriving an empirical model from Eq. (2) requires
addressing three issues: the source of statistical
identification, the precise specification of the innovation
output production function, and the source
of the econometric error. Our choices with respect
to each of these issues reflects our overarching
goal of letting the data speak for itself as much as
possible.
First, the parameters associated with Eq. (2) are
estimated using a panel dataset of 17 OECD countries
over 20 years. These estimates can therefore depend
on cross-sectional variation, time-series variation, or
both. Choosing among the two potential sources of
identification depends on the production relationships
to be highlighted in our analysis. While comparisons
across countries can easily lead to problems of
unobserved heterogeneity, cross-sectional variation
provides the direct inter-country comparisons that
can reveal the importance of specific determinants
of national innovative capacity. Time-series variation
may be subject to its own sources of endogeneity
(e.g. shifts in a country’s fundamentals may reflect
idiosyncratic circumstances in its environment), yet
time-series variation provides insight into how a
country’s choices manifest themselves in terms of
observed innovative output