Appendix
A couple of potential issues arise with our use of innovation module questions. Firstly, since most innovation activities and all sources of information questions sit within a routed part of the innovation module, a tautological "no innovation intent"-"no innovation activity" relationship is imposed for routed respondents. While logically appropriate, imposing an exact relationship for roughly half the population strengthens the apparent statistical association between the activities and outcomes (and, therefore, risks imposing the result we seek to test).28 Routing may also provide a bad incentive to some respondents – almost 93% of firms that have not successfully innovated claim they are also making no ongoing effort towards innovation (table 1). Perhaps a better format for the survey would be to ask questions about activities not specifically tied to outcomes and to not have routing – the trade-off being that of compliance cost against potentially better power to discern practices that drive innovation.29
Secondly, we have multiple measures of R&D in the dataset (chart 5). In the first module of the survey, respondents are asked whether they undertook or funded R&D in the last financial year and, if so, how much was spent on R&D and what proportion was conducted in-house. Of those firms that reported R&D in this module, their mean expenditure on R&D was 3.4% of total operating expenditure, with the average firm performing 65% of that R&D in-house.30 In the innovation module of the survey, firms that have not been routed out of the question are asked whether they have conducted in-house or external R&D in the last two years "while trying to innovate". These indicator questions are hard to reconcile with the quantitative data in the first module of the survey. Specifically, of the 27% of firms that report some R&D in the innovation module, four fifths report no R&D expenditure in the prior year. It seems implausible that 21% of firms did R&D in 2004, but didn't do it in 2005, especially since only 7% of firms were doing R&D in 2005 overall. We use the one-year (quantitative) R&D variables on the grounds that these questions: define R&D for the respondent; provide continuous, rather than binary, measures (ie, they have potentially greater explanatory power); and do not suffer from the routing issues described above. This choice may exacerbate any effects of lags on the relationship between innovation outcomes & R&D spending, since our preferred variables only measure the latter over the last financial year. We test this choice in the regressions presented.
Similar issues arise with the two employee training questions – one innovation-related & the other general. However, in this case there are two mitigating factors that lead us to leave both variables in our regressions: there is a plausible case that innovation-specific training is different from general human capital raising activities; and general training is so pervasive that it may have little discriminatory power anyway.31
Back to Top