Department of Statistics, UCLA

This paper uses new business micro data from the Business Research and

Development and Innovation Survey (BRDIS) for the years 2008-2011 to relate

the discrete innovation choices made by U.S. companies to features of the com-

pany that have long been considered to be important correlates of innovation.

We use multinomial logit to model those choices. Bloch and Lopez-Bassols

(2009) used the Community Innovation Surveys (CIS) to classify companies

according dual, technological or output-based innovation constructs. We found

that for each of those constructs of innovation combinations considered, man-

ufacturing and engaging in intellectual property transfer increase the odds

of choosing innovation strategies that involve more than one type of cate-

gories (for example, both goods and services, or both tech and non-tech) and

radical innovations, controlling for rm size, productivity, time and type of

R&D. Company size and company productivity as well as time do not lean

the choices in any particular direction. These associations are robust across

the three multinomial choice models that we have considered. In contrast with

other studies, we have been able to use companies that do and companies that

do not innovate, and this has allowed to rule out to some extent selectivity

bias.