firms. Epidemic effects relate the dynamic pattern of diffusion with the spread
of information about the features of the technology and, therefore, with the
cumulative number of previous adopters. Stock and order effects originate from
game-theoretic models: early adopters have a first-mover advantage over the
others because of the possibility of preemption on resources critical for the use
of the technology.
The model presented in the following is simplified and involves a reduced
number of regressors: it should be considered only as an example of the meth-
ods introduced in the previous sections. The only time-invariant covariate con-
sidered is the size of the plant expressed in thousands of employees at June
1989: previous studies show the evidence of a positive and highly significant
effect of size on the probability of adoption. This could be due to the fact that
the profitability of a technological innovation requires a critical mass to become
effective; moreover, great plants enjoy advantages connected with the reduc-
tion of the risk, because of preferential accesses to the capital markets and to
managerial and organizational resources.
In addition we consider different time scales. We conform to Colombo and
Mosconi (1995) in using both the calendar time t, and the duration of non
adoption ∏ιt = t — max (θ, tf). For plants that entered the sector before 1970,
the two time scales coincide. We expect calendar time to reflect phenomena
which do not depend on the existence of the firm, notably changes in the prices
and performance improvements of the technologies over time, epidemic effects,
and other time varying factors. Instead, the duration of non adoption captures
effects related to the existence of the firm: insofar as is negatively related to the
age of plant i, we may expect a negative duration dependence as younger firms
and plants exhibit on average higher growth opportunities than older ones and
thus more frequently face investment decisions.
As a whole, the covariates introduced in the model are the following:
Xi,t = (sizei,t,τitt) ;
in our application, for expositional purposes, p is assumed not to depend of Xi,t.
We have estimated two different classes of models:16 the first group is formed
by the saturated ones; the second is composed of the unsaturated ones. In each
class, the baseline model, which will be referred to as Hjj, is estimated under
no restriction on the parameters. Within this model it is possible to test for
поп-causality through Wald or likelihood ratio tests.
Table 5 reports the estimates of the unrestricted unsaturated model, which
allows for a more parsimonious representation of the economic aspects.17These
estimates might suffer some bias due to the omission of several variables that
have proved significant in previous studies: however, all of our hypotheses have
been confirmed. The size of the firm is highly significant for both CADCAM
16Estimation has been performed as in the previous Section.
17The saturated models give no significant improvement of fit in this case, as can be seen
comparing through a Likelihood Ratio test the value of the Ioglikelihoods reported in Table
6.
23