• Home
• About MED
• Publications
• Occasional Papers
• Archive
• 2007
• 07/04
• 3. Innovation Results 2005

• Publications

• Occasional Papers

• Archive

• 2007

Firstly, we compare and contrast different innovation measures (across both outcomes & practices) within the BOS 2005 cross-section. Since this paper is partly concerned with reporting the results of the expanded innovation collection format (ie, the introduction of two "new" innovation types), we segment firms that have successfully innovated into three distinct groups: product and/or operational process only (PP) innovators; organisational/managerial process and/or marketing only (OM) innovators; and innovators that have succeeded in producing a combination (COMBO) of PP & OM innovations.¹⁴ Given our prior discussion, we should note that the breakdown into innovation groups is inconsistent with our advocated holistic view of the firm. It is done purely for ease of exposition and not for conceptual reasons.

Chart 4: Headline innovation outcome comparison 2003-2005 (consistent population)

Headline rates for innovation outcomes & our innovation groups are presented in table 1. The top panel of the table shows overall rates for the four innovation outcomes and, to the right, rates of successful (and ongoing attempts at) innovation conditional on having innovated on another dimension.¹⁵ For example, of those firms that successfully introduced new (or significantly improved) operational processes, 63.5% also introduced new organisational/managerial processes. These results support the motivating hypothesis of the paper – technological progress does not operate independently of wider practices within the firm. Applying the innovation group definitions above emphasises this point. In the bottom panel of the table, we see that our COMBO group (i.e., firms with innovations spanning "technological" & "non-technological" dimensions) has by far the largest population (of the innovator groups). Chart 4 puts these results in the context of economy-wide innovation by comparing innovation in 2003 to 2005. Not too much attention should be paid to overall rates as both populations have had to be adjusted to achieve comparability, and because the innovation reference periods are different. The important point to draw from this graph is that the 2003 results miss the complexity of the innovation story for a significant proportion of product & operational process innovators.

The distinction between PP & OM innovator groups is itself murky, since over a fifth of each of these groups considers they are ongoing innovators on the other dimension (bottom right of table 1). This last data suggests a potential mislabelling of the innovation groups as "distinct". We choose to continue with the groups as defined on the grounds that: we prefer to specify our subsequent regressions with outcomes on the left-hand side and activities on the right-hand side (despite the obvious problem of whether causality can be asserted in this contemporaneous relationship); and we have already admitted our groupings violate the holistic approach advocated – firms crossing our artificially imposed boundaries does nothing but emphasise this point further.¹⁶

Chart 5: Innovation activities 2005 (BPS-consistent industry coverage)

→ Full size version of Chart 5 [117 kB JPEG]

Chart 6: Sources of ideas/info for innovation 2005 (BPS-consistent industry coverage)

→ Full size version of Chart 6 [102 kB JPEG]

We now turn to measuring the underlying innovation activities in the population. Chart 5 shows overall participation rates in innovation activities measured in BOS, while chart 6 shows sources of innovation ideas. There is significant variation in the rates of business uptake of these practices with general training of staff being conducted by 85% of firms, while M&A activity affects less than 3% of the population. The question we wish to ask is: How are these activities related to innovation group outcomes (contemporaneously)? To answer this question we conduct a series of multinomial probit regressions of our innovation groups on firm characteristics and various combinations of innovation activities (Table 2).¹⁷ We draw two interpretations from the table coefficients. First, they indicate whether the characteristic, practice or source of information is more or less likely to be associated with the innovation group that heads the column the coefficient is in (relative to being in the NON group). A p-value under each coefficient indicates the statistical significance of this interpretation. Second, looking across a row (within a panel), coefficients can be compared to see whether an independent variable is more likely to be associated with some successful innovation groups over others. A supplementary test of the equivalence of the OM and PP coefficients tells us whether the characteristic, practice or source of information is significantly more likely to be related to one of these outcome (p-values for these tests are not reported in the tables, but significant differences – at the 5% level – of this type are denoted in the tables by bolded coefficients).

Note: All panels are multinomial probits with innovation group as the dependent variable (NON is the base outcome). Regressions contain ANZSIC division dummies (coefficients not shown). Stars denote significance at 5% (*) & 1% (**) level (two-sided test – robust p-values in square brackets below coefficients). Bold coefficients indicate significant (5% level) difference between PP & OM innovator coefficients (larger of the two highlighted). Proportions of each innovation group accurately predicted are shown below the table.

Firm size, export performance and ODI all have significantly positive coefficients in panel (1), but these results are not robust to the introduction of firm practices. FDI is related to innovation outcomes across all specifications and is more likely to be associated with PP innovation than OM innovation. Subsidiary firms are significantly less likely to be innovative, perhaps explained by a division of responsibilities within the business group (the panel (3) effect of the business group as a source of ideas would support this hypothesis). Firm age is not significant in any of our regressions. It may be that the simple model specified is not appropriate, or that other variables, particularly firm size, are picking up any life-cycle effect.¹⁸

At the bottom of each panel we report the proportions of accurately predicted innovation outcomes. Panel (1) is poor at identifying innovators of any type.¹⁹ Our model becomes better at discerning successful innovators once we introduce innovation activities and/or sources of information. Part of the increase in the overall prediction rate from the first panel (54%) to the last panel (75%) is likely to be due to the routing in the innovation module of the survey (see the Appendix for more on this). However, the way the model allocates innovators to innovation groups has also improved, suggesting that the practice-inclusive models are adding explanatory power over and above the routing effect.

Focussing on panel (4), most sources of innovation ideas are not individually significantly related to innovation outcomes, the strongest positive effect coming from existing staff (whereas, new staff are negatively related to PP innovation outcomes). In contrast, most innovation practices are significantly related to innovation outcomes (bearing in mind our caveat around these significance tests). We highlight just a few points: innovation-specific employee training dominates general employee training (general training being pervasive and, therefore, a commonly held characteristic of NON-innovators); the newly measured innovation activities of changed marketing strategy, new strategy/management techniques, and organisational restructuring have a significantly higher relationship with OM than PP innovation;²⁰ marketing of new products is a PP innovation property (perhaps suggesting that existing marketing methods are more commonly used to introduce new products); higher shares of in-house R&D are important to PP innovators; and the contemporaneous relationship of R&D intensity to innovation outcomes is, if anything, negative. Given the existing literature on the effect of R&D, this last point should raise concerns about causality and the importance of considering lags between practices and outcomes.²¹ These concerns lead us to turn to the BPS-BOS panel.