Advances in Spatial Science - Editorial Board Manfred M. Fischer Geoffrey J.D. Hewings Phần 6 pps

where VA is value added and Emp is employment. In contrast to the subsequent

results, here we do not discriminate between product and process innovation and

consider any form of determinant of productivity growth.

Contrary to our expectations, no significant positive effects of innovation on

labor productivity growth are revealed in the top panel of Table 8.6. Moreover,

small manufacturing firms (between 10 and 50 employees) even experienced a

significant negative “treatment” effect of innovation on labor productivity growth

(significant at 10% only). It remains to be seen in the later specification whether this

result is robust.

One possible explanation for failure to find more conclusive results may be that

we are not capturing the relevant growth period. It may take longer than 2 years

after the initial innovation for firms to internalize all the benefits of it. To control for

this we redefined productivity growth so that we explore the growth in labor

productivity between the second and fourth year after the innovation:

growth½ ¼ ðt þ 4Þðt þ 2Þ ln VA

Emp

tþ4

ln VA

Emp

tþ2

(8.10)

The bottom panel of Table 8.6 presents estimates of the average treatment effect

of innovation on labor productivity growth between the second and fourth years

after the innovation was initially made. By changing the period of observation we

hope to capture the effects of innovation on productivity that were not apparent in

the first 2 years after the time of innovation. As before, we find that innovating firms

did not grow significantly faster (in terms of productivity) than comparable non￾innovating firms. We no longer find negative impacts of innovation on productivity

growth in small manufacturing firms. Interestingly, while a non-significant impact

of innovation on productivity growth of manufacturing firms has been expected

with respect to our previous OLS results, finding non-significant results for services

firms is a little more surprising. Matching innovating and non-innovating services

firms and comparing their relative performance fails to uncover significant differ￾ences in post-treatment (i.e. post-innovation) performance between both groups.

To further disentangle the cause of this lack of evidence on the effects of

innovation on productivity growth, we opt for a more specific definition of innova￾tion by explicitly discriminating between product and process innovations in

Table 8.7. This is based on the findings that process innovations have labor

displacement effects and are expected to result in significant productivity growth,

while, due to the demand effect, product innovations may likely cause employment

growth and, thus, may not result in significant productivity growth (Harrison et al

2005; Parisi et al 2006; Hall et al 2007).

Evidence on changes in employment after a firm has conducted some innova￾tion, however, do not confirm these differentiated expectations (see Table B1 in

Appendix B). Notwithstanding what kind of innovation a firm has conducted, both

process and product innovating firms seem on average to decrease their employ￾ment levels. This is true for virtually all size classes with only a few exceptions.

8 Innovation and Firms’ Productivity Growth in Slovenia 187

Decreases in employment levels should therefore result in positive changes in

productivity growth in both groups of innovating firms.

Table 8.7 presents estimates of the average treatment effect separately for

process and product innovation on labor productivity growth.12 In line with the

evidence on employment changes, results for separate sets of process and product

innovating firms do not differ substantially from those presented for aggregate

innovations. Again, little evidence is found in favor of innovations positively

affecting productivity growth. As was the case before, most of the estimates are

not significantly different than zero, whereby small manufacturing firms (between

10 and 50 employees) in the case of process innovations and medium sized services

firms (between 50 and 250 employees) in the case of product innovations, are found

to experience a significant negative “treatment” effect of innovation on labor

productivity growth. These negative effects disappear when taking into account

productivity growth between the second and fourth years after the innovation

(see Tables A1 and A2 in the Appendix A).

Possibly, the reason for the insignificance of the results may be that the effects of

innovation are not adequately captured by labor productivity and that total factor

productivity should have been used instead. Additionally, our productivity proxy

may fail to control for contemporaneous growth in inputs, which may conceal the

actual productivity dynamics. In order to control for this we use a TFP measure of

productivity estimated by the Levinsohn and Petrin (2003) method. For obvious

reasons this is done for manufacturing firms only. The results shown in Table 8.8

again indicate that there is no significant relationship between innovation activity

and subsequent increases in productivity after 2 or 4 years. The only exception are

micro firms (less than 10 employees) in the period of 4 years after innovation,

where a negative relationship is found, but this result is not repeated in any other

alternative specification.

Conclusions

The paper examines the implications of endogenous growth theory on the relation￾ship between firm productivity, innovation and productivity growth using firm￾level innovation (CIS) and accounting data for a large sample of Slovenian firms in

the period 1996–2002. Two different approaches – simple OLS after the Cre´pon–

Duguet–Mairesse (CDM) approach, and matching techniques – are used to check

the robustness of the results. We also distinguish between product and process

innovations.

12Note that we only show results for the first two years after the innovation has been introduced,

while the results for productivity growth between the second and fourth years after the innovation

was initially introduced are shown in the Appendix (Tables B1 and B2).

188 J.P. Damijan et al.

OLS estimates seem to provide some empirical support to theoretical proposition

of a positive impact of innovation on productivity growth. Both the actual innova￾tion variables from CIS as well as probabilities to innovate estimated using the

system of the research capital equation and innovation equation indicate that

innovating firms increase their productivity at a faster rate than non-innovating

firms. Refinements of the empirical tests allowing for sectoral differences and within

sector heterogeneity, however, reveal that the above results rely mainly on the

exceptional performance of a specific group of services firms. It is shown that it is

medium sized, more (but not the most) productive firms and firms with high (but not

the highest) R&D expenditures to sales in the services sectors that are the frontrun￾ners in innovation. They demonstrate the highest potential to increase productivity

and are capable of using innovations the most efficiently. Separate estimation results

for product and process innovations show no significant differences.

As a robustness check we use nearest neighbor matching approach in order to

match innovating and non-innovating firms with similar characteristics and then

perform average treatment tests of the impact of innovation on performance of

innovating firms as compared to the performance of non-innovating firms. Esti￾mates arrived at by the matching techniques do not reveal any significant positive

effects of innovation on labor productivity growth, regardless of the length of the

period after the innovation was made. Results do not differ for the samples of

manufacturing versus services firms or the samples of firms classified by their size.

The results also do not show any different effects for product and process innova￾tions. Both types of innovations bring about a reduction in employment, however,

little evidence is found in favor of innovations – be it product or process – positively

affecting productivity growth. The result is not sensitive to the use of a TFP or of a

VA/emp as a measure of productivity.

The overall conclusion is that the results of the exercise are not robust to

different econometric approaches. There are several possible reasons why our

analysis has not yielded the expected positive relationship between innovative

activity and productivity growth. In our opinion, the primary reason for these results

lies in the quality of the survey data, primarily with regard to the definition of

innovation. A simple indicator of conducting at least one (product or process)

innovation in the past 2 years may not indicate firm’s true innovativeness in a

satisfactory way. An indicator pointing out the number of innovations conducted

would be more informative. Similarly, a longer series of information about the

share of sales obtained through innovated products and services would be of

extreme importance. Secondly, we do not have the information on the exact time

of innovation, as innovative activity could happen in either of the 2 years between

surveys. Finally, it may be the case that a longer time series is required to capture

the full effects of innovation.

8 Innovation and Firms’ Productivity Growth in Slovenia 189

Appendix A

Table A1 Average treatment effects estimates of innovation on growth in VA/Emp (difference in

logs) between two and four periods after innovation (t + 4) (t + 2) [Process innovation]

Firm size Manufacturing (NACE 15–37) Services (NACE 45–90)

ATT SE No. of obs.

treat. (control)

ATT SE No. of obs.

treat. (control)

Emp  10 0.084 0.140 52 (43) 0.019 0.103 65 (47)

10 < Emp  50 0.003 0.083 114 (70) 0.062 0.133 39 (28)

50 < Emp  250 0.044 0.040 404 (194) 0.027 0.096 22 (16)

Emp > 250 0.042 0.066 318 (106) 0.027 0.136 13 (9)

Note: ***,**,* denote statistical significance at 10%, 5% and 1% level. The number of observa￾tions is given in terms of both the number of treatment and control observations (the latter is in

parentheses). SE- bootstrapped standard errors

Table A2 Average treatment effects estimates of innovation on growth in VA/Emp (difference in

logs) between two and four periods after innovation (t + 4) (t + 2) [Product innovation]

Firm size Manufacturing (NACE 15–37) Services (NACE 45–90)

ATT SE No. of obs.

treatm. (control)

ATT SE No. of obs.

treatm. (control)

Emp  10 0.084 0.140 52 (43) 0.019 0.103 65 (47)

10 < Emp  50 0.003 0.083 114 (70) 0.062 0.133 39 (28)

50 < Emp  250 0.044 0.040 404 (194) 0.027 0.096 22 (16)

Emp > 250 0.042 0.066 318 (106) 0.027 0.136 13 (9)

Note: ***,**,* denote statistical significance at 10%, 5% and 1% level. The number of observa￾tions is given in terms of both the number of treatment and control observations (the latter is in

parentheses). SE- bootstrapped standard errors

190 J.P. Damijan et al.