Uddannelses- og Forskningsudvalget 2015-16
UFU Alm.del
Offentligt
1660560_0001.png
Analysis of the Danish Research
and Innovation System
– A compendium of excellent systemic
and econometric impact assessments
 UFU, Alm.del - 2015-16 - Supplerende svar på spørgsmål 168: Spm. om ministeren vil fremsende relevante evalueringer af Globaliseringspuljen 2007-12, til Uddannelses- og forskningsministeren
Analysis of the Danish Research and Innovation System
Analysis of the Danish Research and Innovation System
– A compendium of excellent systemic and econometric
impact assessments
Published by
Thomas Alslev Christensen, PhD and Head of
Department for Innovation Policy and Research Analyses,
Hanne Frosch, Special Advisor
David Boysen Jensen, Head of Section
Danish Ministry of Higher Education and Science
Danish Agency for Science, Technology and Innovation
Bredgade 40,
1260 Copenhagen K,
Denmark
www.ufm.dk
Layout
Formidabel ApS
The publication can be downloaded from the website
http://ufm.dk/en/publications
ISBN Print: 978-87-93151-26-0
Danish Agency for Science, Technology and Innovation
 UFU, Alm.del - 2015-16 - Supplerende svar på spørgsmål 168: Spm. om ministeren vil fremsende relevante evalueringer af Globaliseringspuljen 2007-12, til Uddannelses- og forskningsministeren
Analysis of the Danish Research and Innovation System
Foreword
When doing research and innovation analysis it
is important to use methods that are based on
proper scientific principles. It is, however also
important to constantly challenge and develop
the knowledge behind the methodology used
to assess the research and innovation system.
A sound knowledge of what works and what
do not work is a condition for progress and we
therefore need state-of-the-art econometric
analysis for making qualified choices.
The compendium is an example of systemic
analyses of the Danish innovation system and
Nordic business investments in R&D. The two
impact assessments are systemic rather than
program assessments and they show the effect
of investing in either corporate R&D activi-
ties or in the research and innovation system.
The compendium consists of two new impact
assessments and a new manual for carrying out
high-quality analysis:
• The Short-run Impact on Total Factor Pro-
ductivity Growth of the Danish Innovation
and Research Support System
– is a short-
run impact assessment of participating in
the Danish innovation and research support
system.
• Economic Impact of Business Investments
in R&D in the Nordic Countries – A microe-
conomic analysis
– is an impact assessment
of Nordic companies’ investment decisions.
• Central Innovation Manual on Excellent
Econometric Evaluation of the Impact of
Public R&D Investments (CIM 2.0)
– is a
manual on how to carry out high-quality
analyses.
Because of the constant need for qualified
and state-of-the-art econometric analyses
and evidence-based policy making there is an
ongoing demand for updating the guidelines
and procedures for evaluations and impact
assessments. Therefore, the Danish Agency for
Science, Technology and Innovation publish an
updated version of the 2012-CIM manual for
making excellent econometric evaluation.
The two system impact assessments make use
of the guidelines in the CIM 2.0 manual and
are examples of impact assessments which
make use of excellent econometric methods.
The assessments are possible because of the
establishment of impressive and comprehen-
sive national register data bases on firm R&D
and innovation projects and activities as well as
on firm panel data.
I hope the reader will find the new systemic
analyses interesting and relevant and the com-
pendium as exciting as we have. We encour-
age the reader to disseminate our work and to
make active use of the CIM 2.0 manual in order
to produce excellent impact assessment studies
of their own and find inspiration to improve
analytical methods and of attaining sound
knowledge of econometric assessments.
Dr. Thomas Alslev Christensen
Head of Department, Danish Agency for
Science, Technology and Innovation
Danish Agency for Science, Technology and Innovation
3
 UFU, Alm.del - 2015-16 - Supplerende svar på spørgsmål 168: Spm. om ministeren vil fremsende relevante evalueringer af Globaliseringspuljen 2007-12, til Uddannelses- og forskningsministeren
Analysis of the Danish Research and Innovation System
Contents
SECTION A
The Short-run Impact on Total Factor Productivity Growth of the Danish
Innovation and Research Support System
Foreword
1. Introduction
2. Description of innovation support programmes
3. Data
4. Method
4.1 Estimation
5. Results
5.1 Main results
5.2 Robustness
5.3 Discussion
6. Conclusion
7. References
8. About the project
A-4
A-5
A-7
A - 10
A - 14
A - 15
A - 19
A - 19
A - 22
A - 27
A - 30
A - 32
A - 33
Economic Impacts of Business Investments in R&D in the Nordic Countries
SECTION B
1. Executive Summary and Conclusions
1.1 Main results
1.2 The effect of increased investment in private R&D
1.3 Overall business R&D investments in the Nordic countries
1.4 Analytical framework and data in the study
2. Impact and efficiency of R&D and innovation in the private sector:
How do we measure the effectiveness of the R&D and innovation system?
2.1 Introduction and motivation
2.2 Lack of economic impact evidence in scoreboards and rankings
2.3 Evidence from the literature - empirical studies of business R&D
2.4 Which questions are addressed in the study?
2.5 Contributors to the study
2.6 Structure and novelty of the study
B-6
B-7
B-8
B - 10
B - 11
B - 11
B - 11
B - 11
B - 13
B - 13
B - 13
B - 14
Danish Agency for Science, Technology and Innovation
4
 UFU, Alm.del - 2015-16 - Supplerende svar på spørgsmål 168: Spm. om ministeren vil fremsende relevante evalueringer af Globaliseringspuljen 2007-12, til Uddannelses- og forskningsministeren
Analysis of the Danish Research and Innovation System
3. Methodology and data
3.1 Methodology
3.2 Data
4. The microeconomic impacts of investments in business R&D in the Nordic countries
4.1 The marginal return of business investments in R&D
4 .2 Estimating the R&D capital elasticity
5. Business R&D in the Nordic countries
5.1 Business R&D investment by country
5.2 Business R&D investment by industry
5.3 Business R&D investment by company size
5.4 R&D collaboration
Bibliography
Appendix A - Company innovation in the Nordic countries
Appendix B - Educational level in the Nordic countries
Appendix C - Descriptive statistics
B - 14
B - 14
B - 19
B -21
B -21
B-25
B -29
B -30
B -32
B -37
B -39
B -41
B -43
B -51
B -56
SECTION C
Central Innovation Manual on Excellent Econometric Evaluation
of the Impact of Public R&D Investments
1 Preface
2 PART I – What is CIM (2.0)?
3 Overview of important standards and minimum requirements
3.1 Principle 1 – Data quality and harmonisation of data collection
3.2 Principle 2 – Selection of comparable companies and/or individuals to
control groups
3.3 Principle 3 – Use of the difference-in-difference method and balanced
panel data
3.4 Principle 4 – Treatment of outliers.
3.5 Principle 5 – Long-time series
3.6 Principle 6 – Robustness test
3.7 Principle 7 – Impact indicators should be made relative
3.8 Principle 8 – Peer review of results
3.9 Principle 9 – Failures and stress tests
4 PART II - Standard for performance objectives: Key performance indicators
4.1 Ex ante evaluation
4.2 Baseline measurement at ex post evaluation
5 PART III – Overview of the most important key performance indicators,
impact assessments and results in Denmark
5.1 Results of impact assessments in Denmark
C-4
C-6
C - 11
C - 12
C - 12
C - 15
C - 17
C - 18
C - 18
C - 18
C - 18
C - 18
C - 20
C - 21
C - 21
C - 24
C - 24
SECTION D
Publications
D-1
Danish Agency for Science, Technology and Innovation
5
 UFU, Alm.del - 2015-16 - Supplerende svar på spørgsmål 168: Spm. om ministeren vil fremsende relevante evalueringer af Globaliseringspuljen 2007-12, til Uddannelses- og forskningsministeren
Analysis of the Danish Research and Innovation System
Danish Agency for Science, Technology and Innovation
6
 UFU, Alm.del - 2015-16 - Supplerende svar på spørgsmål 168: Spm. om ministeren vil fremsende relevante evalueringer af Globaliseringspuljen 2007-12, til Uddannelses- og forskningsministeren
1660560_0007.png
The Short-run Impact on Total Factor Productivity Growth
The Short-run Impact on Total
Factor Productivity Growth
of the Danish Innovation and
Research Support System
Danish Agency for Science, Technology and Innovation
A -
1
 UFU, Alm.del - 2015-16 - Supplerende svar på spørgsmål 168: Spm. om ministeren vil fremsende relevante evalueringer af Globaliseringspuljen 2007-12, til Uddannelses- og forskningsministeren
The Short-run Impact on Total Factor Productivity Growth
The Short-run Impact on Total Factor Productivity Growth
of the Danish Innovation and Research Support System
Published by
Ministry of Higher Education and Science
Danish Agency for Science, Technology and Innovation
Bredgade 40
1260 København K
Telefon: +45 3544 6200
E-mail: [email protected]
www.ufm.dk
Editor
Head of Department
Thomas Alslev Christensen
Danish Agency for Science, Technology and Innovation
A -
2
 UFU, Alm.del - 2015-16 - Supplerende svar på spørgsmål 168: Spm. om ministeren vil fremsende relevante evalueringer af Globaliseringspuljen 2007-12, til Uddannelses- og forskningsministeren
The Short-run Impact on Total Factor Productivity Growth
Contents
Foreword
1. Introduction
2. Description of innovation support programmes
3. Data
4. Method
4.1 Estimation
5. Results
5.1 Main results
5.2 Robustness
5.3 Discussion
6. Conclusion
7. References
8. About the project
A- 4
A- 5
A- 7
A - 10
A -14
A -
15
A - 19
A -
19
A - 22
A - 27
A - 30
A - 32
A - 33
Danish Agency for Science, Technology and Innovation
A -
3
 UFU, Alm.del - 2015-16 - Supplerende svar på spørgsmål 168: Spm. om ministeren vil fremsende relevante evalueringer af Globaliseringspuljen 2007-12, til Uddannelses- og forskningsministeren
The Short-run Impact on Total Factor Productivity Growth
Foreword
As a new initiative, the Danish Agency for Science, Technology and Innovation has
initiated a comprehensive impact study of the Danish system of innovation and sup-
port systems. This is possible because of the Innovation Danmark database which
has a comprehensive amount of information about the innovation and support
programs. With this new information available, we have an obligation to make use
of the new opportunities that is provided to us for creating new knowledge; not only
about the innovation system itself, but about the way we assess the system.
The comprehensive information from the Innovation Danmark database makes
it possible to assess the innovation system, which is a rare opportunity. During the
work with this report I have received very positive feedback from colleagues regard-
ing the collection of information and the opportunities that this presents. Also when
presenting drafts of this report I have received positive and impressed comments
regarding the level at which we assess the Danish system of innovation and support
systems.
This report is first and foremost a methodology report on the edge of the re-
search frontier of impact assessments. We have accepted the new possibilities of
assessing the system, by trying to clear the impact effect from other sources. There-
fore I advise the reader to be careful when interpreting the results of the report and
for a deeper analysis of the individual innovation programs; I refer to the individual
impact assessments of the innovation programs.
I hope the reader of this report will find it as enlightening and inspiring as we
have and will use this as an inspiration for further studies of impact assessments.
Thomas Alslev Christensen
Head of Department
Danish Agency for Science, Technology and Innovation
Danish Agency for Science, Technology and Innovation
A -
4
 UFU, Alm.del - 2015-16 - Supplerende svar på spørgsmål 168: Spm. om ministeren vil fremsende relevante evalueringer af Globaliseringspuljen 2007-12, til Uddannelses- og forskningsministeren
The Short-run Impact on Total Factor Productivity Growth
1. Introduction
This study performs the first joint estima-
tion of the economic impact of innovation
and research support programs. We focus
solely on firms with less than 500 employ-
ees, and later restrict our analysis to firms
with less than 100 employees.
This report features three important types of
findings:
1)
We quantify relative impact on
productivity
We are the first who attempt to per
form a causal study of multiple and
simultaneous support programs
1
We use the cleanest sample of partici-
pants and non-participants, to date,
because for the first time we have
access to extensive information about
multiple program participation
participants and non-participants. We find
that these criteria are necessary, as we wish to
make causal inference on our estimates.
When estimating impact, we take into account
the historical productivity performance of
firms to rule out that firms participating were
growing fast in the first place, and that we are
simply picking a select group of firms that are
growing faster.
Using our sample, we find that firms establish-
ing contact with the support system, sub-
sequently, on average, grow 2.5 percentage
points faster annually the first two years, com-
pared to non-participating firms. Behind this
average estimate lies highly varying estimates
for the individual programs.
Our main results (annual effects in percentage
points) are that firms participating in Innova-
tion Network (3.6), Innovation Voucher
(3.6),
and
Innovation Assistant (2.9) tend to grow
faster the first two years.2 The qualitative
results are robust to alternative specifications,
however, when we limit our analysis to firms
with less than 100 employees and control for
firm individual productivity growth trends
(depending on firms size), we find that effects
are larger for some programs. While Innova-
tion Assistant effects are robust to alternative
specifications, Innovation
Networks (4.3)
and
Innovation Voucher (4.1) effects are
amplified, and Innovation
Consortia
(4.6)
now enters significantly in the analysis. All of
these programs are designed spur an increase
of knowledge via the channels collaboration,
counseling or within-firm skill upgrading.
We find no enhanced productivity growth fol-
lowing participation in
Industrial PhD
(nega-
tive but insignificant impact), which is in line
2)
3)
1
Impact of several of the
programs have been studied
individually or grouped as for
example “research projects”.
See e.g. CEBR (2009, 2011b,
2013a), DASTI (2011), DASTI
& DAMVAD (2013), Kaiser &
Kuhn (2012), and Chai & Shih
(2013).
2
Results are from the instru-
mental variable approach in
TABLE 5.2. Consult the table
for significance levels.
We follow firms two years after participation,
which is a short period. However, we have to
make a compromise when aiming to cover
as many programs as possible. This short
window has two important downsides: 1) In
programs, where we find higher productivity
growth for participants, we cannot conclude
on whether the effect on growth is a perma-
nent effect, or 2) whether productivity growth
rises in the short run because the participation
effect induces a one-time lift to the productiv-
ity level.
Because we add strict criteria to avoid con-
taminated estimates, we perform our analyses
on a sample of firms that most notably did
not receive support two years before observed
participation or two years following observed
participation. These criteria apply to both
Danish Agency for Science, Technology and Innovation
A -
5
 UFU, Alm.del - 2015-16 - Supplerende svar på spørgsmål 168: Spm. om ministeren vil fremsende relevante evalueringer af Globaliseringspuljen 2007-12, til Uddannelses- og forskningsministeren
The Short-run Impact on Total Factor Productivity Growth
with previous studies, and
Innovations Agents
(zero impact).3 The finding that Innovation
Agents participation does not return dif-
ferential growth is not surprising, but rather
comforting. The Innovation
Agents
program
is designed to give firms a “checkup” and then
forward them to relevant private consult-
ing or to other programs such as
Innovation
Voucher. One possible conclusion is that
Innovation Agents check up on Danish firms
with exhibiting productivity growth rates that
are not different from that of the typical non-
participating firm.
In the report we suggest other explanations
for missing effects. One important circum-
stance is that this study does not look at
productivity
levels,
only productivity
growth.
Thus, programs with no documented pro-
ductivity enhancing effects may still play an
important role by, for example, helping highly
productive firms to expand product markets
(possibly export markets) and thereby grow.
This is, however, not within the scope of this
study, but we encourage further studies into
other performance measures.
The report proceeds as follows: Section 2
describes the different innovation support
programmes. Section 3 presents the data and
how we construct the sample, while section
4 explains the estimation method. In section
5 we present the main results (section 5.1) of
our analysis as well as results using alternative
specifications for robustness check (section
5.2), before finally discussing of our results
(5.3). We conclude in section 6.
3
We have somewhat few
observations on Industrial
PhD to firmly conclude. We
have enough observations
to conclude on Innovation
Agents. Consult sections 3
and 5.1 for further informa-
tion on which programs we
have too few observations
to conclude upon.
Danish Agency for Science, Technology and Innovation
A -
6
 UFU, Alm.del - 2015-16 - Supplerende svar på spørgsmål 168: Spm. om ministeren vil fremsende relevante evalueringer af Globaliseringspuljen 2007-12, til Uddannelses- og forskningsministeren
The Short-run Impact on Total Factor Productivity Growth
2. Description of innova-
tion support programmes
The description of the programmes contained
in this section was written by The Danish
Agency for Science, Technology and Innova-
tion (DASTI).
Danish Council for Strategic Research
The primary focus of the Danish Council
for Strategic Research (CSR) is to promote
excellent and relevant research that will be of
benefit to future development and economic
growth in Denmark. Hence, the research must
be of high standard and lie within areas of
research that is related to societal challenges.
CSR offers a number of different support
programmes (including SPIR) aimed at both
private firms and research institutions.
EUopStart
Danish firms and research institutions may
apply the EUopStart programme for a grant
(up to 20,000 euros) when applying for par-
ticipation in selected European and interna-
tional research programmes. The grants cover
different activities related to the application
process such as salary, travel, conference and
consultancy. The receiving firm or research
institution has to put down 50 percent of the
grant in self-financing.
Industrial PhD
The Industrial PhD programme aims at
increasing knowledge sharing between uni-
versities and private sector firms, promoting
research with commercial perspectives, and
taking advantage of competences and research
facilities in private firms to increase the num-
ber of PhDs with knowledge about industrially
focused research and innovation. For this pur-
pose, the Industrial PhD student is employed
in a firm and enrolled at a university at the
same time. The student spends all his or her
time on the project both places and shares
his or her time equally between the firm and
the university while taking the degree. The
Danish Agency for Science, Technology and
Innovation subsidises the Industrial PhD’s
salary with a fixed monthly amount and the
expenses at the university with a fixed amount
over the three years. A grant is approximately
134,000 euro divided between the firm and
the university.
Eurostars
The Eurostars programme offers grants to
small and medium sized firms (SME) and re-
search institutions who participate in research
and development programmes under the
Eurostars programme. Hence, the Eurostars
programme supports business-to-business
cross border collaboration projects between
enterprises from minimum two countries,
and promotes market oriented R&D activi-
ties among research intensive SMEs. Grants
amount to a maximum of 310,000 euros.
FP7
The Seventh Framework Programme is the
European Union’s chief instrument for public
funding of research and for increasing private
R&D. The Seventh Framework Programme is
based on four principal programmes (Coop-
eration, Ideas, People and Capacities), with
public sector bodies eligible to participate
across all four. The major fields of research
supported by the themes of the Cooperation
programme are industry led and bring to-
gether public and private sector stakeholders
to define research and development priorities,
timeframes and action plans on a number
of issues that are strategically important to
achieving Europe’s future growth, competi-
tiveness and sustainability. The Marie-Curie
actions funded under the People programme
aims to increase mobility between public and
Danish Agency for Science, Technology and Innovation
A -
7
 UFU, Alm.del - 2015-16 - Supplerende svar på spørgsmål 168: Spm. om ministeren vil fremsende relevante evalueringer af Globaliseringspuljen 2007-12, til Uddannelses- og forskningsministeren
The Short-run Impact on Total Factor Productivity Growth
private sectors, as well as between countries.
To this end they will support industry train-
ing, joint research partnerships and staff
secondments between the two sectors. As well
as specific actions to help SMEs, the Capaci-
ties programme aims to develop European
research infrastructures, optimise their use
and improve access for researchers, including
from industry. It will also support regional
research-driven clusters, involving enterprises
as well as universities and local authorities.
Research Voucher
The Research Voucher scheme was offered
in the period 2008-2009. It provided sup-
port for research based collaboration between
SMEs and knowledge institutions (Universi-
ties, RTOs etc.). The purpose of the Research
Voucher scheme was to enhance innovation in
SMEs as well as to make public research more
application-orientated. The financial support
was solely for the activities in the knowledge
institutions, and could be up to a maximum
of 200,000 euros for projects with duration
of up to 2 years. The financial support could
not surpass 25 pct. of the total budget for the
project. Support was granted at a first come,
first served basis. A total of 17 projects were
initiated under the Research Voucher scheme.
Gazelle Growth
The Gazelle Growth programme helped small
firms achieving their growth potential on
foreign markets – especially the US-market.
Due to the size of the home market, especially
small gazelle firms from small economies have
to look at foreign markets sooner than small
gazelle firms from big economies, if they want
to grow. That can be at a time, where their net-
work and knowledge of foreign market can be
limited. With the Gazelle Growth programme
small gazelle firms was advised and trained,
so the entry on a foreign market can go faster
and succeed then if they tried themselves. The
Danish Gazelle Growth programme was termi-
nated by the end of 2010.
The Danish National Advanced Tech-
nology Foundation
The Danish National Advanced Technology
Foundation offers private firms and universi-
ties the funds and the framework for devel-
oping new and important technologies. The
general objectives of the Danish National
Advanced Technology Foundation is to en-
hance growth and strengthen employment by
supporting strategic and advanced technologi-
cal priorities within the fields of research and
innovation. Up to this day the Foundation has
invested in 273 advanced technology projects
with a total budget exceeding 700 million
euros. Half of the finance comes from firms
and research institutions themselves. Average
support per project is approximately 1.5 mil-
lion euros with a support range of each project
from 0.5 to 12 million euros.
Innovation Agents
The aim of the Innovation Agents is to create
innovation in small and medium-sized firms.
Innovation Agents are public funded consult-
ants that help firms identify barriers to inno-
vation by performing an “innovation check”.
The consultants identify the most important
development opportunities for the firms and
work closely together with regional growth
houses and business advice offices to provide
firms with one access point to the public in-
novation system.
Innovation Consortia
Innovation Consortia subsidies and facilitate
collaboration projects between firms, research
institutions and non-profit advisory and
knowledge dissemination parties. The purpose
of the programme is that the parties jointly de-
velop knowledge or technologies that benefit
not only individual firms but entire industries
within the Danish business community. The
joint projects should result in the completion
of high-quality research relevant to Danish
firms. Furthermore, the project should ensure
that the new knowledge is converted into
competences and services specifically aimed
at firms, and that the acquired knowledge
is subsequently spread widely to the Danish
business community – including in particular
Danish Agency for Science, Technology and Innovation
A -
8
 UFU, Alm.del - 2015-16 - Supplerende svar på spørgsmål 168: Spm. om ministeren vil fremsende relevante evalueringer af Globaliseringspuljen 2007-12, til Uddannelses- og forskningsministeren
The Short-run Impact on Total Factor Productivity Growth
SMEs. A consortium can apply for financial
grants at the Danish Agency for Science,
Technology and Innovation, and the grants
subsequently finance the expenses incurred by
the research and knowledge institutions whilst
undertaking the cooperative project. Typically
grants amount to approximately 1-2 million
euros.
Innovation Incubators
The objective of the innovation incubator pro-
gramme is to promote commercialisation of
new innovative ideas, inventions and research
in particular through the creation of new
knowledge based start-ups. The innovation in-
cubators provide professional counselling and
early stage gap funding (pre-seed and seed
capital) for entrepreneurs and new innovative
enterprises. The innovation incubators operate
at the very early stage of the investment chain,
where venture capitalists and other private
investors are reluctant to engage. The innova-
tion incubators funds 50 – 60 new knowledge
based firms per year, and has a total budget of
approximately 30 million euros.
Innovation Network Denmark (The Na-
tional Danish Cluster Programme)
The Innovation Network Denmark pro-
gramme supports the establishment of
network and cluster organizations. An In-
novation Network is a cluster organization
with participation of all relevant Danish
universities and technology institutes within
a specific technological area, a business sector
or a cross-disciplinary theme. Today a total of
22 innovation networks are scattered all over
Denmark. Each network has pools for inno-
vation projects where firms and researchers
work together to solve concrete challenges.
The innovation networks also carry out idea
generation processes and matchmaking
activities, and they hold theme meetings and
specialist events. Hence, the overall objective
for the innovation networks is to facilitate and
encourage knowledge exchange between SMEs
and knowledge institutions.
SPIR – Strategic Platforms for Innova-
tion and Research
SPIR funds initiatives which seek to
strengthen the link between strategic re-
search and innovation and thereby pro-
moting efficient knowledge dissemination
and possibilities for fast application of new
knowledge in connection with innovation
in the private and public sectors. Typically
grants amount to approximately 8 - 10 mil-
lion euros.
Innovation Voucher
The Innovation Voucher scheme supports
collaborative projects between a small or
medium sized firm and a knowledge in-
stitution. The objective of the Innovation
Voucher scheme is to encourage more SMEs
to collaborate with universities, research
and technology institutes and education
institutions. The maximum amount of public
support is 13,500 euro. The public support
must not exceed 40 pct. of the total innova-
tion project.
Innovation Assistant
The Innovation Assistant program provides
an incentive for small and medium-sized
firms to hire a highly educated person. The
rationale is that highly educated people
working on an innovative project promotes
growth in the SMEs. The firm must have
between 2 and 100 employees in order to re-
ceive subsidy (up to one year) to employ the
highly educated person. Also the firm must
pay at least half of the Innovation Assistants
wages. Each grant is approximately 20,100
euro.
Open Funds
Open Funds where earmarked for innova-
tive collaboration projects between firm and
public knowledge institutions. The objective
was to ensure that innovation projects that
would benefit entire industries did not fall
flat because they did not fit into the innova-
tion system. Open Funds could finance up to
50 percent of a project. The programme was
terminated in 2012.
Danish Agency for Science, Technology and Innovation
A -
9
 UFU, Alm.del - 2015-16 - Supplerende svar på spørgsmål 168: Spm. om ministeren vil fremsende relevante evalueringer af Globaliseringspuljen 2007-12, til Uddannelses- og forskningsministeren
The Short-run Impact on Total Factor Productivity Growth
3 Data
We use data from two different sources:
-
The
Innovation Danmark
database
created by the Danish Agency for
Science, Technology and Innovation
(DASTI) containing a list of firms that
have received support (hereafter
participants)
Worker-firm matched registry data
from Statistics Denmark
One advantage of Experian data over Statistics
Denmark data is that it has one more year of
observations (2012 over 2011). Some programs
were introduced in later years, whereby adding
one more year of observations would be very
important to the analysis. However, due to the
poor mechanical data match result, it does not
add crucial information to the analysis.
For this analysis, we generally prefer data from
Statistics Denmark to Experian data, because
we can control for the skill of employees and
use the effective size (full-time employment) of
the firm level workforce instead of the number
of employees. The skill level at participating
firms is, on average, different from that of
non-participants. Not controlling for the skill
level introduces an upward bias on the impact
assessment of productivity growth. Using the
number of employees (the only available op-
tion in Experian data) instead of the fulltime
equivalent number of employees (available in
Statistics Denmark registry data) also creates a
possible bias, because participating firms may
differ from other firms in terms of the share of
full time workers. Thus, we must compare firms
using effective unit input of labor.
The Estimation Sample
Measuring productivity growth impact is not
straightforward, because several circumstances
affect firm performance. For instance, a natural
bias of this sample is that we observe only
firms that are neither bankrupt, bought up,
nor reconstructed. We enforce strict criteria to
isolate potential effects, implying that our sam-
ple shrinks from information of about 3,000
participation activities to about 1,100.
In this section we describe the process of creating
the estimation sample(s). We illustrate the pro-
cess in FIGURE 3.1 and TABLE 3.1, respectively.
-
The databases have a common firm identi-
fier that allows us to match the list of program
participants with firm information. Firm
information is crucial to performing impact
assessment. We utilize information on value
added, capital, number of employees, full-time
employment, skills of employees, and industry
(using the NACE3-classification)
4
.
4.The
NACE-classification
(Nomenclature statistique
des activités économiques
dans la Communauté euro-
péenne) is the EU stan-dard
industry classification.
5.CEBR
(2011b) focused on
the Industrial PhD program
and were able to recover
a substantive number of
missing observations.
We have tried to combine the Innovation
Danmark database with a different firm panel
of annual reports data (Experian data, formerly
also known as KOB-data). However, we are
effectively able to match fewer participants
using Experian data than through Statistics
Denmark. Searching for missing matches after
matching on firm identifier and year, is a much
too comprehensive and ad hoc task for this
project, as it involves searching through firm
names in the panel data, or parts of names,
from an extensive list of firm names that
were not matched (either due to missing firm
identifier (cvr-number) or, likely, mistyping in
the Innovation Danmark database).
5
Why we
find more mechanical matches using Statistics
Denmark registry data, we cannot tell, because
we do not control the data matching process
(restricted for regulatory reasons to enforce
anonymity of the firms in the registry data).
Danish Agency for Science, Technology and Innovation
A -
10
 UFU, Alm.del - 2015-16 - Supplerende svar på spørgsmål 168: Spm. om ministeren vil fremsende relevante evalueringer af Globaliseringspuljen 2007-12, til Uddannelses- og forskningsministeren
1660560_0017.png
The Short-run Impact on Total Factor Productivity Growth
We measure the impact of a particular pro-
gram on firm performance relative to non-par-
ticipating firms. We adjust the raw sample of
firms from a set of criteria that are intended to
center on capturing
participation effects. Our
point of reference is the
Raw Sample,
which
is simply the result of matching the complete
worker-firm panel of private Danish firms
with the Innovation Danmark database. The
raw sample spans from 2000-2011.
Using the full sample to measure these
par-
ticipation effects delivers an average firm per-
formance difference between non-participants
and participants. We control for a range of
differences between firms based on statistical
facts about the firms, and we leave out firms in
industries where no participants are found.
For an observation to be included we need a
full set of information on each observation.
The observations that fulfill the requirement
of a full set of information make up the Esti-
mation Sample.
We foremost use Estimation
Sample 1,
includ-
ing all firms that have less than 500 employees
and can be observed in a four year window.
The estimation samples are not just the result
of mechanical changes to the data butalso the
result of the chosen estimation strategy. The
strategy imposes certain requirements to the
data. We formally walk through the estimation
strategy in section 4, but some of the criteria
mentioned in this section are the result of the
estimation strategy.
Using the same criteria as for
Estimation Sam-
ple 1,
we create
Estimation sample 2,
where
the only altered criteria is that firm employ-
ment must be less than 100. We want to rule
out as many biases as possible, i.e. in this case
that firm size band is too wide. With so many
programs and also repeated firm appearances in
the support system we have to drop firm obser-
vations associated with participation before and
after observed participation status in a given
year.
TABLE 3.1 demonstrates how almost 11,000
observations of contact with the system in the
Innovation Danmark database become about
1,100 observed participations in
Estimation
Sample 1.
6
We begin with the full Innova-
tion Danmark database spanning from 2002
to 2012, imposing no criteria.
7
Here we have
almost 11,000 observed participation activities
from 8,300 firms. When we matched this data
with the firm panel spanning from 2002 to 2011
(step 1 in TABLE 3.1), we drop more than 4,000
observations, most of which are from 2012.
We observe productivity growth development
for two years. Thus, given that the last year of
the sample is 2011, we can only measure impact
on participation initiated no later than 2009.
Therefore, we cut the number of observations in
half to 3,100 by excluding information on sup-
port in 2010 and 2011 (step 2).
We limit our main analysis to firms with less
than 500 employees, dropping more than 300
observations (step 3).
To measure productivity growth impact, we
must observe productivity two years ahead
and also other participation activity, dropping
800 observations (step 4).
6
“Contact with the system”
can refer to multiple participa-
tion in different programs
within a year. However, this
is rare.
7
After initializing this
project, the database now
contains information on
some firms before 2002,
and also current (not full)
status for 2013 (constantly
updated). The full sample
refers to the sample of firms
that Statistics Denmark was
able to identify. CEBR has
no control over this process
due to data regulatory rea-
sons. Firms are anonymous
in the registry data and
must remain so.
Notes:
Figure 3.1
Procedure to narrow the sample
Sample adjustment
process
Raw sample
Adjustments made
Full sample
All firms with less than 500
employees and only from
industries with program
participants
Firms from the Full Sample
that neither received
support in the two years
preceding the observation
year nor in the two
subsequent years
Same as Estimation Sampe
1 but for firms with less
than 100 employees.
Estimation sample 1
Estimation sample 2
The figure shows the narrowing of the full sample of firms to
comprise only relevant firms under stricter criteria.
Danish Agency for Science, Technology and Innovation
A -
11
 UFU, Alm.del - 2015-16 - Supplerende svar på spørgsmål 168: Spm. om ministeren vil fremsende relevante evalueringer af Globaliseringspuljen 2007-12, til Uddannelses- og forskningsministeren
1660560_0018.png
The Short-run Impact on Total Factor Productivity Growth
Table 3.1
The effective number of participation observations in
estimation sample 1
Steps
Criteria
None
1
2
3
4
5
6
Matched with registry data
Effective event window
Firms with less than 500
Observations required
(forward-looking)
Observations about the firm
historical growth
Only firms not participating in
two years before nor after
observed participation status
First year
2002
2002
2002
2002
2002
2002
2002
Last year
2012
2011
2009
2009
2009
2009
2009
# obs.
10887
6409
3152
2815
2022
1665
1096
# firms
8307
4840
2488
2357
1720
1424
1071
# obs.
/#firms
1.310581
1.324174
1.266881
1.194315
1.175581
1.169242
1.023343
Revenue
-
449.0
495.0
92.7
109.0
117.0
94.5
Value
added
-
149.0
177.0
29.5
36.8
40.0
32.0
Full-Time
empl.
-
249.4
323.6
48.3
55.0
59.2
46.2
Notes:
The table step by step demonstrates each of the added criteria resulting in the final Estimation Sample 1.
Source:
CEBR calculations using Innovation Danmark Database and Statistics Denmark registry data.
To control for historical productivity growth
and participation activity adds further re-
strictions to the information criteria, drop-
ping about 350 observations (step 5).
Finally, we restrict observations of par-
ticipation to include only firm observations
in those years where they did not receive
support in the preceding two years and the
following two years (step 6).
From TABLE 3.1 we observe that the number
of observed participations across the 2002-
2009 period is 1,096 split on 1,071 unique
firms. Some few firms appear twice in the
sample period. The 1,096 are indicative of
activity.
Behind that aggregate number we find 1,140
individual program participation indications.
These are shown in TABLE 3.2. Vertically the
rows indicate the individual program. Hori-
zontally, the columns indicate which types of
programs fit into which group. We have seven
groups but we include group 3 in group 2. Ef-
fectively, we can measure average group im-
pact on group 2, 4, 6 and 7. Note that group 7
only comprises
Innovation Assistants.
From TABLE 3.2, we see that the number
of observed participations that fulfill all the
necessary criteria to be included in
Esti-
mation Sample 1
varies greatly from one
program to another. For example, we have
one observation of the
Danish Council for
Strategic Research (DCSR), but 327 on
Innovation Networks. We are not able to
make inference from the estimates of impact
concerning participation in initiatives under
DCSR; SPIR, EUOpSTART and Eurostars
(all started recently); FP7 (started in 2007
and many applications made by large firms);
Research Voucher and Gazelle Growth
(few
applicants, fewer observations); The
Danish
National Advanced Technology Foundation
(effectively few observations).
Danish Agency for Science, Technology and Innovation
A -
12
 UFU, Alm.del - 2015-16 - Supplerende svar på spørgsmål 168: Spm. om ministeren vil fremsende relevante evalueringer af Globaliseringspuljen 2007-12, til Uddannelses- og forskningsministeren
1660560_0019.png
The Short-run Impact on Total Factor Productivity Growth
Table 3.2
The effective number of observed program participations in
estimation sample 1
PROGRAMS GROUPS
3. Intl. 4. Counceling
collaboration
and support
Program
Danish Council for
Strategic Research
EUopSTART
Industrial PhD
Eurostars
FP7
Research Voucher
Gazelle Growth
The Danish National
Advanced Technology
Foundation
Innovation Agents
Innovation Consortia
Innovation Incubators
Innovation Networks
SPIR
Innovation Voucher
Scheme
Innovation Assistant
Open funds
GROUP TOTALS
Notes:
1. Strategic
research
1
2. Colla-
boration
5. Financing 6. Industrial
PhD
7. Skill enhancing
employment
0
51
0
14
2
10
11
252
91
2
327
0
180
167
32
1
136
14
589
2
51
167
The table shows the effective number of observations found in Estimation Sample 1 and used for the main analysis (see construction procedure above).
The horizontal grouping of the 16 individual programs has been determined in collaboration with the Danish Agency for Science, Technology and
Innovation.
Source:
CEBR work on Innovation Danmark Database and Statistics Denmark registry data.
Next, in section 4, we present the estimation
strategy.
Danish Agency for Science, Technology and Innovation
A -
13
 UFU, Alm.del - 2015-16 - Supplerende svar på spørgsmål 168: Spm. om ministeren vil fremsende relevante evalueringer af Globaliseringspuljen 2007-12, til Uddannelses- og forskningsministeren
1660560_0020.png
The Short-run Impact on Total Factor Productivity Growth
4 Method
In this chapter we discuss in general terms
the estimation methods used. The estimation
design must suit the impact measure, in our
case: Productivity growth differences between
participating firms and non-participants,
ruling out as many other factors as possible
that may also have an impact, but founded
on a well-formulated production function.
Productivity is directly related to the avail-
ability of technology to a firm and the firm’s
ability to utilize the available technology. This
is referred to as total factor productivity (TFP).
To measure TFP we must specify a production
function. However, by the estimation method
that we choose, we obtain productivity growth
directly from a transformation of the produc-
tion function.
A widely used method for estimating partici-
pation effects of a single program is a twin
study using a matching estimator. In this type
of study, we match participating firms with,
statistically speaking, twin firms that do not
participate. This estimation procedure has
some advantages over, for example, linear
regression models. Communicating the analy-
sis is reasonably straightforward: 1) A clear-
cut control group of non-participating firms
similar to participants is constructed. Thus, we
can argue that any found effects are likely the
true isolated effects of participation. 2) Given
certain assumptions, we can conclude that the
effect found is causal.
Given these clearly attractive properties of
matching methods, we still cannot rule out a
well-specified regression model, which is
more flexible. One important downside of
matching is that we match on level variables,
which are “snapshot” characteristics, because
matching on growth patterns preceding par-
ticipation is very complicated. Thus we may be
BOX 4.1 PRODUCTIVITY GROWTH
When a firm uses inputs of production it incurs production costs. We can measure the total
extra value created by the firm by subtracting production costs other than remuneration of
capital and labor from revenue obtained from the sale of its production of goods or services.
Economists refer to this extra value as value added. A firm can create more value
added
if it
grows in size, for example by increasing capital use and/or hiring more labor. However, that
does not per se imply increased production efficiency.
Often the public debate focuses on
labor productivity,
which is simply valued added per
employee. It is easy to calculate for descriptive purposes. However, labor productivity is
indicative for comparing productivity differences across firms, industries (to some extent)
etc. but does not take into account intensive use of capital. Thus, the productivity measure
that we are interested in is one that takes into account the use of both labor and capital in
production. Economists refer to this as total
factor productivity.
We measure total
factor productivity growth as the growth in firm value added that
cannot be attributed to increased use of capital or labor
Danish Agency for Science, Technology and Innovation
A -
14
 UFU, Alm.del - 2015-16 - Supplerende svar på spørgsmål 168: Spm. om ministeren vil fremsende relevante evalueringer af Globaliseringspuljen 2007-12, til Uddannelses- og forskningsministeren
1660560_0021.png
The Short-run Impact on Total Factor Productivity Growth
matching firms that at a snapshot in time have
identical revenue, capital intensity, produc-
tivity level, workforce skill level, but actually
follow two different dynamic paths. In such a
case the firms are not suitable twin pairs to be
compared.
The linear regression method (estimated us-
ing ordinary least squares, OLS) is still the
best linear unbiased estimator available, and
often we can justify that linearity of effects is
a fair assumption. Measurement of growth
differences is definitely such a case, and
controlling for historical growth is uncompli-
cated, broadly used and well-described in the
literature. Furthermore, we can specify our
regression model and select our estimation
sample such that any differences between a
regression model and a matching procedure to
assess impact of participation are, for practical
purposes, eliminated.
4.1
Estimation
We rely on OLS estimation with fixed effects
to estimate firm productivity growth from
the firm level production function. Using this
method, we can directly obtain a measure of
participation effects from the estimates of pro-
ductivity growth differences between partici-
pants and non-participants without having to
estimate productivity separately for partici-
pants and non-participants in the first place.
We derive our estimating equation from a
standard production function for firm i in year
t:
function to include firm i’s
individual produc-
tivity level component,
c
i
:’
(2)
Hence, firm level total factor productivity,
, is the scale product of cross-firm com-
mon technology
and firm individual abil-
ity to take advantage of common technology, c
i
(i.e. the firm fixed effect).
Under the assumption that the above speci-
fication holds, each firm has an intrinsic
productivity growth potential, because the
individual component acts as a scale factor
on firm productivity growth from changes
in
. This intrinsic ability of a firm to
utilize available technology is unobservable.
For shorter time periods we assume that
this unobservable characteristic of the firm
remains constant. Consequently, we focus
on fixed effects estimation, which deals with
time-constant unobservable characteristics.
We therefore do not worry about the firm indi-
vidual component
c
i
.
Taking logs of the production function (rep-
resented below by small letters) we can write
up a basic estimating equation (leaving out
potential control variables) for the production
function:
(3)
Note the unobserved fixed effect of firm (i).
We can remove the unobserved individual
fixed effect by taking first differences (Δ), and
when we then add some control variables and
a participation indicator variable we arrive at
our core estimating equation:
(1)
Firm level value added, Y,
is produced using
capital
(K)
and labor
(L)
inputs, but also
depends on firm level total factor productiv-
ity
(A). The total factor productivity level of a
specific firm can be perceived as the result of
available technology and its capabilities (e.g.
strong management) to utilize labor and capi-
tal inputs. To see this, rewrite the production
8
Unless we specify another
forward year, we always
consider two-year forward
differences.
(4)
We estimate the linear regression model
above using pooled OLS.
Danish Agency for Science, Technology and Innovation
A -
15
 UFU, Alm.del - 2015-16 - Supplerende svar på spørgsmål 168: Spm. om ministeren vil fremsende relevante evalueringer af Globaliseringspuljen 2007-12, til Uddannelses- og forskningsministeren
The Short-run Impact on Total Factor Productivity Growth
Our dependent variable is
Δy
t
measured in log
points between time
t
and
t+2.
8
This gives us
the percentage point growth in firm total value
added. We account for the growth contribution
to value added from increasing use of capital
and labor resources.
We choose a two-year lead period for two rea-
sons. First, we find one year to be too short,
and second, we lose too many observations if
we use longer lead periods.
An observable variable, which is an indicator
for a firm’s ability to absorb new technology,
is whether the firm a priori is skill intensive.
Our fulltime equivalent labor stock variable
cannot be divided into different skill types of
labor. Thus, to account for the fact that labor is
a heterogeneous input, we introduce a variable
accounting for the initial share of workers that
hold at least a bachelor degree. Furthermore,
we account for industry specific trends in pro-
ductivity growth (δ
j
), and time varying trends
in productivity affecting all firms (η
t
).
Apart from accounting for the initial relative
skill level of firm labor stock, we do not add
further level variables (such as size or pro-
ductivity level) to our estimating equation,
because we stick to our model specification,
i.e. the production function. Adding further
variables on an ad hoc basis distorts the
theoretically motivated estimation strategy.
As explained above, the share of high skill
workers is justified from the criteria of act-
ing as a proxy for labor quality. In section 5.2
we perform robustness checks, adding level
control variables.
We measure whether an average trend dif-
ference in
Δy
t
exists between firms receiving
support and firms not participating. Thus, we
obtain an estimate of potential participation
effects from the coefficient(s) γ
s
on the partici-
pation indicator variable(s) (participation
i,s,t
).
9
The subscript s indexes the number of up to N
different programs (or groups of programs) in
question.
10
By using first differences estimation, we
eliminate unobserved time-invariant firm fixed
effects that may drive firm-specific productiv-
ity growth effects. In the longer run, this may
turn out to be a strict assumption. If firms
enter an innovation support program that
initiates a new firm specific growth trend, then
we are dealing with time-varying firm effects.
However, in the short event windows that we
measure impact, we do not consider this to be
a likely source of inconsistency.
We effectively measure annual productivity
growth rates over two years for all firms that
received support in any given year from 2002
to 2009 and compare them with non-partici-
pating firms.
FIGURE 4.1 illustrates the principle of meas-
uring participation. Participation can happen
in any year, but we only include an observation
if a firm has no participation activity before
nor after the observation year – in this case
the observation year is 2005. From 2003 to
2005 neither firm participates. In 2005 some
firms participate and some do not. We effec-
tively compare firm productivity growth rates
between 2005 and 2007, taking into account a
range of other sources of productivity growth.
Thus we can isolate the potential participation
effect.
What happens after two years? We do not
know. Will the firm remain on a higher pro-
ductivity growth path? Intuitively that seems
unlikely that entering a program suddenly
transforms how a firm runs its business in
any situation. We find it reasonable to assume
that a firm temporarily grows faster than it
would have and that the observed increased
productivity growth rate is a combination of
the normal, underlying growth rate and a one-
time increase in productivity.
9
We do not consider
dynamic additive effects
between programs, e.g. that
firms join one program in
2003 and another program
in 2007. We showed in sec-
tion 3, that very few firms are
represented multiple times,
across time, in our sample.
10 We also measure the
overall impact of partici-
pating in any program. In
this case we have just
one indicator variable,
participation
i,s,t
,and the fol-
lowing estimating equation:
Danish Agency for Science, Technology and Innovation
A -
16
 UFU, Alm.del - 2015-16 - Supplerende svar på spørgsmål 168: Spm. om ministeren vil fremsende relevante evalueringer af Globaliseringspuljen 2007-12, til Uddannelses- og forskningsministeren
1660560_0023.png
The Short-run Impact on Total Factor Productivity Growth
Figure 4.1
Assessment of impact on firm productivity
growth from participating in a program
PRODUCTIVITY GROWTH RATE
Participator
PARTICIPATION
EFFECT
Non-participator
2003
2004
2005
2006
2007
PARTICIPATION
No participation
No other participation
Notes:
The figure shows the narrowing of the full sample of firms to comprise only relevant firms under stricter criteria.
11
The underlying motiva-
tion for assuming produc-
tivity potential from highly
educated workers comes
from numerous correlation
studies that document the
relationship
Selection
A concern when performing impact assess-
ment of programs that are designed to spur
innovation and R&D activities is that the firms
receiving support irrespective of participation
or not have the potential to innovate and in-
crease productivity growth, or plainly grow at
a faster pace. One descriptive fact is that firms
that innovate tend to employ more intensively
highly educated workers (see CEBR 2013b).
Our inclusion of the share of highly educated
workers at the time program participation is
initiated can account for this possible con-
founding effect. The inclusion of this informa-
tion accounts for trend differences stemming
from unleashed productivity potential of a
highly educated workforce in participating
firms that initially deliver relatively low pro-
ductivity levels.
11
However, participating firms could already
be growing at a faster pace than non-partici-
pants. Clearly we must address this issue.
One way is to specify a lagged dependent
variable model by adding lagged productiv-
ity to equation (3). This gives us the fol-
lowing fixed effect specification of a lagged
dependent variable model (LDP) as an
alternative to equation (4):
(5)
We estimate the above equation using
pooled OLS.
If the decision to participate in a program
at time t is correlated with growth in pro-
ductivity leading up to time
t, Δy
i,t-2
, then
leaving out
Δy
i,t-2
(as in equation 4) will bias
the estimated coefficient of participation,
γ
s
.
If θ<0, the estimate will be biased down-
ward if we leave out
Δy
i,t-2
, and if θ>0, the
Danish Agency for Science, Technology and Innovation
A -
17
 UFU, Alm.del - 2015-16 - Supplerende svar på spørgsmål 168: Spm. om ministeren vil fremsende relevante evalueringer af Globaliseringspuljen 2007-12, til Uddannelses- og forskningsministeren
1660560_0024.png
The Short-run Impact on Total Factor Productivity Growth
estimate will be biased upwards if we leave
out
Δy
i,t-2
,. Note that, in general, we do not
need
Δy
i,t-2
but only
Δy
i,t-1
(i.e. a one period
difference from t-1 to t). We use two periods
because 1) it is more stable to use annual-
ized growth rate over two periods, and 2)
we are looking back two periods anyway to
observe prior participation activity.
The fixed effects specification of the LDP
model suffers from
Δy
i,t-2
and
Δ
i,t
being
correlated by construction, making the OLS
estimator never fully consistent.
Instead of accounting for the omitted vari-
able bias using a fixed effects LDP model
we can use a two-stage least squares (2SLS)
approach, instrumenting lagged productivity
growth with further lags of the productiv-
ity level.
12
This instrumental variable (IV)
approach will account for selection of firms
that were already growing at faster pace
before participating in a program.
As we described in section 3, the estimation
samples only include participating and non-
participating firms that did neither receive
support two years before the starting year
of the observed difference or during the two
subsequent years we observe firm perfor-
mance.
Thus, using a clean sample of participation
activity, accounting for lagged productiv-
ity growth both using the LDP approach
and performing an IV estimation taking
into account historical productivity growth,
delivers a sound foundation for estimating
participation effects.
In the next section we present the results
of performing the simple pooled OLS fixed
effects estimation not account for historical
growth (equation 4), pooled OLS fixed ef-
fects estimation of the LDP model (equation
5), and the 2SLS IV approach.
12
Anderson & Hsiao (1981)
suggested the idea of using
productivity levels lagged
two periods as an instru-
ment for productivity growth
lagged one period. See Ver-
beek (2008) for a discussion
of the method and alterna-
tive specifications. See also
Nickell (1981) and Angrist &
Pischke (2009). Griffith, Red-
ding & Van Reenen (2004)
argue to use IV approach for
robustness if TFP measure-
ment error is a concern.
Danish Agency for Science, Technology and Innovation
A -
18