knowledge complementarities and patenting: do new universities...
TRANSCRIPT
UniversitätZürichIBW–InstitutfürBetriebswirtschaftslehre
Working Paper No. 164 Knowledge Complementarities and Patenting: Do New Universities of Applied Sciences Foster Regional Innovation? Patrick Lehnert, Curdin Pfister, Dietmar Harhoff and Uschi Backes-Gellner
January 2020 First Draft. Preliminary Results.
Die Discussion Papers dienen einer möglichst schnellen Verbreitung von neueren Forschungsarbeiten des Leading Houses und seiner Konferenzen und Workshops. Die Beiträge liegen in alleiniger Verantwortung der
Autoren und stellen nicht notwendigerweise die Meinung des Leading House dar.
Disussion Papers are intended to make results of the Leading House research or its conferences and workshops promptly available to other economists in order to encourage discussion and suggestions for revisions. The
authors are solely responsible for the contents which do not necessarily represent the opinion of the Leading House.
The Swiss Leading House on Economics of Education, Firm Behavior and Training Policies is a Research Program of the Swiss State Secretariat for Education, Research, and Innovation (SERI).
www.economics-of-education.ch
Working Paper No. 164 Knowledge Complementarities and Patenting: Do New Universities of Applied Sciences Foster Regional Innovation? Patrick Lehnert, Curdin Pfister, Dietmar Harhoff and Uschi Backes-Gellner
Knowledge Complementarities andPatenting: Do New Universities of Applied
Sciences Foster Regional Innovation?∗
Patrick Lehnert†1, Curdin Pfister1, Dietmar Harho↵2, and UschiBackes-Gellner1
1University of Zurich2Max Planck Institute for Innovation and Competition, Munich
January 16, 2020
First Draft. Preliminary Results.
∗ The authors would like to thank Simone Balestra, David Deming, Simon Janssen, Edward Lazear, JensMohrenweiser, Samuel Muhlemann, Laura Rosendahl Huber, Paul Ryan, Guido Schwerdt, seminarparticipants at the University of Zurich, participants of the “Higher Education in Modern Ecosystems:E�ciency, Society and Policies” conference in Augsburg, participants of the 8th ZEW/MaCCI Confe-rence on the Economics of Innovation and Patenting in Mannheim, participants of the Munich SummerInstitute 2019, participants of DRUID19 in Copenhagen, participants of the 30th SASE conference inNew York City, participants of the 14th EPIP conference in Zurich, and participants of the RISE2workshop in Munich for helpful comments, as well as Felix Poge for help with data preparation.Marc Egli, Michael Niederberger, Marie-Lena Pfa↵, and Vanessa Sticher provided excellent researchassistance. This study is partly funded by the Swiss State Secretariat for Education, Research, andInnovation (SERI) through its Leading House on the Economics of Education, Firm Behavior andTraining Policies as well as by the German Commission of Experts for Research and Innovation.
† Corresponding author. Address: Patrick Lehnert, University of Zurich, Plattenstrasse 14, CH-8032Zurich, Switzerland. E-Mail: [email protected]
Abstract
This study analyzes how universities of applied sciences (UASs)—bachelor-granting three-year colleges teaching and conducting applied research—a↵ectregional innovation. We particularly focus on regional complementaritiesbetween such applied research institutions and basic ones. We exploit variationin the location and timing of UAS establishments in Germany. To accountfor endogeneity, we apply fixed e↵ects estimations and control for regionaleconomic activity by implementing a proxy based on 30 years of satellite data.We find that UASs increase innovation with a substantially larger e↵ect inregions where other (basic and applied) public research organizations coexist.This result indicates strong knowledge complementarities. In an additionalanalysis, we also find that UASs accelerate innovation in the regions belongingto the former German Democratic Republic.
JEL Classification Numbers: I23, O31, O38, R11
2
1. Introduction
While a great number of studies have identified the positive innovation e↵ects of basic
research performed at traditional academic universities (e.g., Anselin et al., 1997; Autant-
Bernard, 2001; Ja↵e, 1989; Toivanen and Vaananen, 2016), only a few studies have
investigated the innovation e↵ects of other types of research institutions, particularly those
with a more application-oriented research focus. For example, Adams et al. (2003) find
positive innovation e↵ects for federal laboratories cooperating with industrial laboratories;
Popp (2017) identifies application-oriented Public Research Organizations (called PROs in
the following) as more important for private firms in the energy sector than traditional
Academic Universities (UNIs in the following); and Pfister et al. (2018) examine whether
Universities of Applied Sciences (UASs in the following) in Switzerland lead to an increase
in regional patenting activities, finding strong positive e↵ects.
On top of innovation e↵ects that stem from a single institution itself, the question of
whether additional e↵ects arise due to the existence of an ecosystem of research institutions
in close proximity remains open. We expect di↵erent types of knowledge creation and
knowledge that come together in regions where research institutions coexist to lead to strong
complementarity e↵ects. Such complementarities enhance the capacity for innovation in
regions with bundles of di↵erent research institutions. In such a context, we expect UASs
to foster the transfer into applied research of the scientific knowledge that basic research
institutions produce into applied research, leading to a higher degree of innovation.
This paper analyzes the single and combined regional innovation e↵ects of UASs and
neighboring research institutions. We are able to (a) identify the innovation e↵ect of the
establishment of UASs in Germany and (b) to separate which part of the e↵ect goes back
to UASs themselves and which part goes back to complementarities with other research
institutions.
UASs o↵er three-year study programs awarding bachelor-level degrees and that thereby
focus on applied research. To analyze the UASs’ contribution to regional innovation, we
examine the evolution of patents in Germany from 1980 through 2012. We investigate at
the regional level whether this contribution depends on the existence of other research
3
institutions—UNIs and PROs—in the regions where UASs are located. A larger UAS e↵ect
on innovation in regions with a basic research institution would indicate complementarities
between basic and applied research.
For our analysis, we draw on two main data sources. First, we use patent data from
the European Patent O�ce’s (EPO) Worldwide Patent Statistical Database (April 2017
version) to measure innovation. Second, we collect information on the exact locations and
establishment years of research institutions. For this purpose, we initially start with the
directory of Higher Education Institutions (HEIs in the following), two types of which
exist in Germany (UNIs and UASs), from the German Rectors’ Conference and a list
of institutes belonging to PROs from BMBF (2018b). We augment these data with
extensively researched information on the particular campus locations of each HEI, the
locations of institutes belonging to PROs, and the establishment years of all identified
locations. Combining the two data sources provides us with a rich dataset highly suitable
for analyzing the role of knowledge complementarities between di↵erent types of research
institutions for innovation outcomes.
As an identification strategy, we follow a growing literature analyzing the establishments
of HEIs (e.g., Eyles and Machin, 2019; Kamhofer et al., 2019; Pfister et al., 2018; Toivanen
and Vaananen, 2016). In so doing, we exploit variation in the location and the timing of
UAS establishments to investigate if patenting activities increase after these establishments.
As the spatial and temporal distribution of UASs is not random in Germany, we
need to account for potential endogeneity in the location of UASs to identify their e↵ect
on patenting activities. That is, we need to control for time-invariant and time-variant
regional economic factors that may determine both the existence of a UAS in a region and
the development of patents. To control for any such regional factors that are time-invariant,
we apply fixed e↵ects (FE) estimation.
In addition, we need to control for time-variant regional economic factors. However,
such data are not readily available at a detailed regional level and for the necessary
time period. Therefore, we use a proxy that Lehnert et al. (2020) derive from satellite
imagery with machine-learning methods. Dating back until the 1980s, this proxy covers a
4
su�ciently long time series for our analysis of German UASs. Furthermore, the proxy is
available at a very high level of regional detail. The proxy contains the composition of a
region’s terrestrial surface (i.e., the types of land that cover the region). This composition
comprises six surface groups indicating di↵erent types of land, for example built-up land,
cropland, or forests. As Lehnert et al. (2020) demonstrate, these surface groups validly
predict economic activity even at very small regional levels. Therefore, we use the surface
groups as control variables in our estimations to account for (at least a part of the)
time-variant regional factors determining UAS establishments.
Our results show that UASs have a positive and significant e↵ect on both patent
quantity and patent quality. However, this e↵ect is substantially larger in regions where
other research institutions exist at the time of the UAS establishment. In regions where
UASs can draw upon the scientific knowledge of preexisting research institutions, strong
knowledge complementarities between UASs and these preexisting institutions exist and
these complementarities contribute significantly to regional patent quantity and quality.
An additional analysis of UAS regions in the new federal states suggests that UASs help
accelerate the technological development in these regions as compared to UAS regions in
the old federal states.
2. Literature Review
Many studies have emphasized the positive influence of research institutions performing
basic research on regional innovation activities. In an important study, Ja↵e (1989) finds
evidence of spillovers from UNI research to private-sector R&D, in science, technology,
engineering, and mathematics (STEM)-related industries in particular, thereby contributing
to local patenting activities. Other studies attesting to the positive relationship between
UNI research and the evolution of patents include Autant-Bernard (2001), Cowan and
Zinovyeva (2013), Leten et al. (2014), and Toivanen and Vaananen (2016).
In addition, other types of (public) research institutions, in particular those with a
more applied research focus, positively a↵ect innovation. In their review of the traditional
linear model of innovation, Leyden and Menter (2018) argue that basic public research
5
alone—without accompanying applied research—is insu�cient for generating knowledge
spillovers from the public to the private sector. A variety of PROs can serve this applied
function within an innovation system, with empirical studies providing evidence for the
positive e↵ect of these PROs (e.g., Comin et al., 2019; Intarakumnerd and Goto, 2018;
Popp, 2017). Case studies of regional innovation systems in Germany also show that PROs
can play a key role in fostering innovation (e.g., Broekel and Graf, 2012; Graf, 2011).
A particular type of applied research institution belonging to the higher education
sector is the UAS, which Pfister et al. (2018) show to have a large positive e↵ect on
patenting in Switzerland. For Germany, Fritsch and Slavtchev (2007) provide evidence
for a positive innovation e↵ect of HEIs (i.e., UNIs and UASs taken together without
di↵erentiating between the two) using six years of patent data from the German Patent
O�ce. Furthermore, using two waves of Community Innovation Survey data, Robin and
Schubert (2013) find that collaboration with public research institutions (i.e., UNIs, UASs,
and PROs taken together without di↵erentiating between the three) increases innovation
in firms. Other studies use cross-sectional surveys to investigate the role of public research
institutions (di↵erentiating between UNIs, UASs, and PROs) for innovation in firms,
providing mixed results (Beise and Stahl, 1999; Fritsch and Schwirten, 1999).
Few studies explicitly address cooperation between research institutions. Fritsch and
Schwirten (1999) find that in the regions surveyed for their analysis, cooperation with other
research institutions is more common for UNIs and PROs than for UASs, which in turn
cooperate more often with firms. Investigating how the number of cooperation partners
a↵ects innovation in firms (but without explicitly di↵erentiating between di↵erent types of
cooperation partners), Becker and Dietz (2004) argue that a mix of heterogeneous coope-
ration partners creates synergies that further increase research and development (R&D)
activities. However, the innovation e↵ect resulting from knowledge complementarities
between di↵erent types of research institutions remains an open question.
In sum, the empirical evidence suggests that both basic and applied public research
institutions increase innovation and that cooperation between the two potentially enhances
this e↵ect. In this paper, we contribute to the literature by studying the establishment of
6
UASs at a detailed regional level, allowing us to disentangle their e↵ect on patenting from
the e↵ect of other research institutions. While either research institutions focusing on basic
research or those focusing on applied research constitute valuable knowledge resources
for innovation activities, a region’s capacity for innovation might even further increase if
innovators can draw on both basic and applied research knowledge. The interaction of
basic and applied research could thus speed up and improve the innovation process.
Assuming that knowledge spillovers are geographically concentrated, as many empirical
studies argue (e.g., Audretsch and Feldman, 1996; Cabrer-Borras and Serrano-Domingo,
2007; Ponds et al., 2007), we expect that opening a UAS in a region where it can cooperate
with a basic research institution yields a larger innovation e↵ect than opening a UAS
elsewhere. Moreover, if the two types of institutions produce complementary applied
research knowledge, opening a UAS in a region where another type of applied research
institution already exists might also lead to positive knowledge spillovers. In regions where
no other institution exists, we expect UASs to increase innovation, but to a lesser extent.
3. The German Landscape of Research Institutions
3.1. Higher Education Institutions (HEIs)
In Germany, two types of public1 HEIs perform research in the STEM fields: UNIs
and UASs.2 Like all education institutions in Germany, UNIs and UASs fall under the
jurisdiction of the Lander3 governments (BMBF, 2018a). Although both institutions
award equivalent bachelor’s and master’s degrees (International Standard Classification of
Education (ISCED) 2011 levels 6 and 7), UNIs hold the exclusive right to award doctoral
degrees (BMBF, 2004; Enders, 2010).4 Moreover, UNIs focus on basic research and impart
1 We restrict our analysis to public HEIs. We exclude private ones primarily because relatively fewer ofthem are active in STEM-related research (see Buschle and Haider, 2016).
2 In addition to UNIs and UASs, the German higher education sector also comprises colleges ofeducation (Padagogische Hochschulen), of theology (Theologische Hochschulen), of art and music(Kunsthochschulen), and of public administration (Verwaltungsfachhochschulen) (BMBF, 2018a).However, as these institutions specialize in subjects unrelated to STEM and thus produce innovationsthat do not usually result in patents, we do not consider them in our analysis.
3 Germany is divided into 16 federal states called Lander.4 Recently, the federal state of Hesse granted research-intensive UASs a limited right to award doctoral
degrees (Meurer, 2018).
7
basic research skills to their students, whereas UASs emphasize vocational practice and
applied research (BMBF, 2004).
While some UNIs have a centuries-long tradition, the first UASs in Germany were
established in 1968 as teaching institutions (BMBF, 2004; Enders, 2010). UASs target
students who have completed an apprenticeship (a dual vocational education and training
program). In comparison to UNIs, teaching at UASs aims at knowledge relevant for
vocational practice and problem-solving (BMBF, 2004). This focus also manifests at
UASs in, for example, bachelor degree programs usually including a mandatory practical
semester in the form of an internship at a firm (BMBF, 2004).
Only in 1985 did an amendment of the German Higher Education Framework Act
add applied research to the purpose of UASs (Enders, 2010; Kulicke and Stahlecker, 2004;
Wissenschaftsrat, 2002). By engaging in applied research projects jointly conducted with
firms, faculty members at UASs maintain their vocational practice, which they can then
pass on to their students (Hinz et al., 2016; Kulicke and Stahlecker, 2004). A large part of
the research projects that UASs undertake is in STEM fields such as IT, materials science,
or mechanical engineering (Kulicke and Stahlecker, 2004).
UASs have become an important research partner for local firms. Small and medium-
sized firms in particular profit from the knowledge that UASs generate, because these
firms value the application-oriented knowledge and the vocational practice and because
they often do not possess the capacities for carrying out research projects by themselves
(Hachmeister et al., 2015; Kulicke and Stahlecker, 2004). Moreover, participation in joint
research projects with larger firms or in publicly funded research projects is also important
for UASs (Hachmeister et al., 2015).
The UASs’ focus on applied—in comparison to basic—research and on combining
vocational practice with applied research in their teaching distinguishes the UASs from
academic universities. Thus, by providing applied research knowledge to local players
in the R&D ecosystem, UASs can positively contribute to regional innovation activities.
When cooperating with research institutions that have a stronger focus on basic research,
UASs can contribute to the transfer of basic scientific knowledge, thereby increasing
8
regional innovation. Through this cooperation, a local combination of basic and applied
research institutions can be an important driver of (regional) innovation outputs.
3.2. Public Research Organizations (PROs)
In addition to UNIs and UASs, four PROs—the Max Planck Society, the Leibniz Associa-
tion, the Helmholtz Association, and the Fraunhofer Society—constitute another unique
pillar of the German research and innovation system. These PROs are publicly funded and
act independently, serving functions that are complementary to those of HEIs (see, e.g.,
EFI, 2010). Each organization maintains several research institutes throughout Germany,
often employing hundreds and sometimes even thousands of researchers (BMBF, 2018b).
The four PROs serve di↵erent functions and have di↵erent ranges of activity:
• The Max Planck Society conducts basic research in the natural sciences, the life
sciences, the social sciences, and the humanities (BMBF, 2018a), that is, in both
STEM and non-STEM fields. The Max Planck Society autonomously chooses
its research subjects and aspires to scientific excellence, a goal also reflected in
the organization’s high degree of internationalization (BMBF, 2018a; Hohn, 2010).
Examples include the Max Planck Institute for Plasmaphysics or the Max Planck
Institute of Biochemistry (see BMBF, 2018b).
• The Leibniz Association originally put those institutes that did not fit into the
structures of the other three organizations under one umbrella (Hohn, 2010). Conse-
quently, the Leibniz Association has the broadest research scope of the four, ranging
from basic to applied research in both STEM and non-STEM fields, and having
a decentralized organizational structure (BMBF, 2018a; Hohn, 2010). Examples
include the Leibniz Institute of Polymer Research or the Leibniz Institute of Plant
Genetics and Crop Plant Research (see BMBF, 2018b). In addition to research
institutes, the Leibniz Association also comprises non-research institutes such as
museums and further education institutes (e.g., training centers) (BMBF, 2018a;
Hohn, 2010).
9
• The Helmholtz Association comprises research centers that are active in technology-
intensive (i.e., STEM-related) fields with a long-term perspective, such as aeronautics
and materials science (BMBF, 2018a). Therefore, the organization is involved in
the transfer of basic research knowledge to technological products (Hohn, 2010).
In comparison to the Max Planck Society and the Fraunhofer Society, the public
funding that the Helmholtz Association receives is tied to specific subjects or projects
(Hohn, 2010). Examples include the German Aerospace Center or the Helmholtz
Centre for Infection Research (see BMBF, 2018b).
• The Fraunhofer Society engages in applied research projects on a range of topics
(e.g., health, environment, mobility, energy), with the majority being STEM-related
(BMBF, 2018a). As the Fraunhofer Society receives government funding that calls
for research collaboration, both private and public demand drive the organization’s
choice of research projects (Hohn, 2010). Examples include the Fraunhofer Institute
for Solar Energy Systems or the Fraunhofer Institute for Laser Technology (see
BMBF, 2018b).
Figure 1a graphically summarizes the results of a study on publication and patenting
activities of research institutions (EFI, 2010), depicting the institutions’ research profiles.
The Max Planck Society focuses on basic research and the Fraunhofer Society focuses on
applied research. The Helmholtz and Leibniz Associations range somewhere in between,
with the Leibniz Association tending towards basic research. With respect to HEIs, the
study did not di↵erentiate between UNIs and UASs, but the average over all HEIs locates
in the middle of the spectrum. In Figure 1b, the two dots denoting UASs and UNIs,
respectively, represent our assessment of their profiles as Section 3.1 describes.
Given their di↵ering profiles, PROs strategically engage in research cooperation with
HEIs (BMBF, 2018a). Again, such cooperation can play a key role in fostering the transfer
of basic scientific knowledge to actual applications. For PROs with a basic research focus,
such as the Max Planck Society, cooperation with UASs can provide an important source
of applied research knowledge, while UASs, in turn, might value such cooperation as a
source of basic research knowledge. Furthermore, PROs with an applied research focus,
10
Figure 1: Profiles of research institutions in Germany
10 20 30 40 50 60 70 80
0.2
0.4
0.6
0.8
1.0
1.2
1.4
Applied research (patent applications per 1,000 researchers)
Basic
research
(scientificpublication
sper
researcher)
Average over all
HEIs
Max Planck
Fraunhofer
Helmholtz
Leibniz
(a) EFI (2010): Positioning ofinstitutions according to
empirical analyses
10 20 30 40 50 60 70 80
0.2
0.4
0.6
0.8
1.0
1.2
1.4
Applied research
Basic
research
UASs
UNIs
Max Planck
Fraunhofer
Helmholtz
Leibniz
(b) Authors’ extension: Approximatepositioning of UNIs and UASs according
to their legal mandates
Source: Illustration based on EFI (2010, p. 40) and authors’ assessment.Notes: The original analysis in EFI (2010) is based on an analysis of publications in the Science CitationIndex and patent applications per researcher (in full-time equivalents) for three periods, 1994–1996,1999–2001, and 2004–2006 (dots show averages over these three periods). The original analysis does notdi↵erentiate between UASs and UNIs. The dots for UASs and UNIs in panel (b) represent our assessmentof their research activity as described in Section 3.1.
such as the Fraunhofer Society, might also profit from knowledge complementarities with
UASs.
4. Data and Methodology
To analyze the innovation e↵ect of UASs, we use data from the EPOs Worldwide Patent
Statistical Database (April 2017 version) from 1980 through 2012. Among other items, this
data contains complete information on patents, including the exact geographic locations
of inventors, the application date, and the number of patent citations three years after
publication.5 We use geodata provided by the Federal Agency for Cartography and Geodesy
(BKG)6 to assign every inventor to a German municipality. Using these assignments,
5 The EPO data include citation information also for five and ten years after publication, respectively.We analyze the citation lag of three years to get a conservative estimate and to be able to use a longertime series.
6 Available from http://www.geodatenzentrum.de/geodaten/gdz rahmen.gdz div?gdz spr=deu&gdz aktzeile=5&gdz anz zeile=1&gdz unt zeile=15&gdz user id=0 (last retrieved on May 5, 2017).
11
we follow Pfister et al. (2018) and compute the fractionated7 number of patents per
municipality i and year t (patent quantity, PQUAN), as well as a patent’s average
number of citations in municipality i and year t (patent quality, PQUALcit) as innovation
outcomes. To investigate change rates instead of changes in absolute numbers, we take the
variables’ natural logarithms after adding the value 1, i.e., we observe ln(PQUAN+1) and
ln(PQUALcit + 1), respectively. Appendix A shows the descriptive statistics for patent
quantity and patent quality.
To determine which types of HEIs exist in a municipality at a given time, we had to
collect detailed data on the location and timing of the establishment of all their campuses.
We performed extensive online searches in which we studied for every single HEI the
available information on the web to nail down the HEI’s individual history, campus
locations, and fields. As a starting point for these searches, we used data from the German
Rectors’ Conference.8 This data contains, among other items, the name, type, main
address, and establishment year of every HEI that existed in Germany on January 1,
2017,9 but it does not contain their individual campus locations or fields. Furthermore, to
identify HEIs that closed before January 1, 2017 (and are thus not part of the German
Rectors’ Conference data),10 we use data on student numbers ranging as far back as 1998
from the German Federal Statistical O�ce’s GENESIS database.11 To obtain more details
on single campuses, we researched online the campus addresses, campus establishment
years,12 and campus profiles (i.e., the departments or study programs that each houses),13
7 For example, if a patent lists one inventor from municipality A, one inventor from municipality B, andone inventor from abroad, the patent counts as 1
3 of a patent for municipality A and 13 for municipality
B. The remaining 13 does not enter our estimations.
8 Available from http://www.hs-kompass2.de/kompass/xml/download/hs liste.txt (last retrieved onJanuary 20, 2017).
9 We exclude HEIs that o↵er only distance learning, because we do not expect their innovation e↵ects, ifany, to be locally concentrated.
10 Nonetheless, we might miss some HEIs that closed before 1998. However, drawing on historicaldescriptions of HEIs (see also section 3.1), we find this number to be very small, therefore not biasingour estimations.
11 Available from ht tp s : //www- gene s i s . d e s t a t i s . d e/gene s i s /on l i n e/da ta ; j s e s s i on id=7CF9926E22E6DDE18B248F7571AD3949.tomcat GO 2 3?operation=abruftabelleBearbeiten&levelindex=1&levelid=1500539721416&auswahloperation=abruftabelleAuspraegungAuswaehlen&auswahlverzeichnis=ordnungsstruktur&auswahlziel=werteabruf&selectionname=21311-0002&auswahltext=&werteabruf=Werteabruf (last retrieved on July 20, 2017).
12 If the websites or other sources do not indicate an establishment year for a campus, we assume it to bethe establishment year of the respective institution as indicated in the original dataset.
13 For example, the main campus of the Weihenstephan-Triesdorf University of Applied Sciences with the
12
gathering this information primarily from the institutions’ websites. Drawing on the
campus profile, we categorize HEI campuses according to whether they are active in
STEM or not. If a campus o↵ers at least one study program in STEM or houses a STEM
department, we categorize it as a campus in STEM.
To obtain information on research institutes belonging to one of the four PROs (Max
Planck Society, Leibniz Association,14 Helmholtz Association,15 and Fraunhofer Society),
we proceed similarly. Drawing on BMBF (2018b) and the respective PRO’s websites, we
collect the addresses, establishment years, and profiles of these institutes and categorize
them according to whether they are active in STEM or not. Geocoding these addresses and
combining the information that we collected on the campuses belonging to HEIs with that
on the institutes belonging to PROs provides us with a rich longitudinal dataset allowing
us to identify the spatial distribution of public research institutions at any geographical
level.
Throughout the observation period we analyze in this paper, 134 public UASs with
campuses in 347 locations exist in Germany. Of these campus locations, we categorize
212 as campus locations in STEM.16 These STEM campus locations are geographically
distributed across 142 (of a total of 11,266) municipalities.17 Furthermore, 99 UNIs have
campuses in 588 locations (310 in STEM distributed across 74 municipalities), the Max
Planck Society has 137 institutes in 174 locations (145 in STEM distributed across 50
municipalities), the Leibniz Association has 171 institutes in 174 locations (101 in STEM
distributed across 49 municipalities), the Helmholtz Association has 38 institutes in 114
locations (114 in STEM distributed across 52 municipalities), and the Fraunhofer Society
department of bioengineering sciences (among others) is in Freising. However, the Weihenstephan-Triesdorf University of Applied Sciences has a second campus in Weidenbach (more than 100 kilometersfrom Freising), which contains the departments of agriculture, food, and nutrition and of environmentalengineering.
14 The Leibniz Assocation, formerly named Blue List Partnership and Blue List Science Assocation,emerged in the 1990 from the Blue List institutes (see, e.g., Brill, 2017). Former Blue List institutesalso enter our analysis as Leibniz institutes.
15 In 1995, the Association of National Research Centres was transfered into the Helmholtz Association(see, e.g., Ho↵mann and Trischler, 2015). Institutes belonging to the former Association of NationalResearch Centres also enter our analysis as Helmholtz institutes.
16 Here, a location comprises all departments or branches with the same postal address. Thus one locationcan consist of multiple departments or branches.
17 Territorial status of January 1, 2017.
13
has 105 institutes in 162 locations (161 in STEM distributed across 84 municipalities).
Appendix B shows the distribution of campus and institute locations in STEM across
municipalities for the year 2012.
To estimate the e↵ect of UASs on innovation, we assign each German municipality
either to the control group or to the treatment group. To the control group, we assign
all municipalities not treated by a UAS campus in STEM. To the treatment group, we
assign all municipalities that are treated by a UAS campus in STEM. We consider a
municipality treated by a UAS campus if the municipality is located within a 25-kilometer18
travel-distance19 radius of the campus. The year 1985, when UASs began to conduct
applied research, constitutes the first possible UAS treatment period for a municipality,
even if the UAS the municipality is treated by was established before 1985 as a teaching
institution. Comparing the development of patent quantity and patent quality in treated
and untreated municipalities allows us to identify the e↵ect of UASs on innovation.
To investigate knowledge complementarities between UASs and other research insti-
tutions, we further di↵erentiate municipalities according to whether they lie within the
25-kilometer radii of other types of research institutions. To disentangle the e↵ect of
a establishing a UAS from that of establishing another type of research institution, we
generate indicators for the existence of each other type of research institution at the time of
the UAS campus establishment and control for establishment of other research institutions
after the UAS establishment in our estimations. We then interact the treatment variable
with the indicators for the existence of other types of research institutions. These inte-
raction terms show the surplus in the UASs’ e↵ect on innovation resulting from knowledge
complementarities with other research institutions.
18 We chose the threshold of 25 kilometers because in Germany, the majority of the working population(79.2 percent in 2016) commute 25 kilometers or less. See https://www.destatis.de/DE/ZahlenFakten/GesamtwirtschaftUmwelt/Arbeitsmarkt/Erwerbstaetigkeit/TabellenPendler/Pendler1.html (lastretrieved on February 19, 2019). Empirical evidence on the innovation e↵ects of research institutionsalso suggests a concentration of these e↵ects within a 25-kilometer radius of an institution (Helmersand Overman, 2017; Pfister et al., 2018). We provide estimation results with varying treatment radiias a robustness check in Appendices C and D to demonstrate that the choice of treatment radius doesnot alter our main results.
19 We compute the travel distance between the geographical center of a municipality and a campus orinstitute location using the HERE application programming interface. See https://developer.here.com/(last retrieved on July 10, 2019)
14
Figure 2 shows the distribution of treatment regions in Germany for 2012, the last
year of observation in our analysis. Of the 11,266 municipalities, 3,720 (33.02 percent) lie
within the 25-kilometer treatment radii of UAS campuses in STEM, 2,186 (19.40 percent)
within the radii of UNI campuses in STEM, 1,837 (16.31 percent) within the radii of Max
Planck institutes in STEM, 1,188 (10.55 percent) within the radii of Leibniz institutes in
STEM, 1,302 (11.56 percent) within the radii of Helmholtz institutes in STEM, and 2,078
(18.44 percent) within the radii of Fraunhofer institutes in STEM.
To identify the UAS e↵ect innovation, we need to control for potential endogeneity
in the locations of UAS campuses. To do so, we use FE estimation to account for time-
invariant characteristics of municipalities that determine both UAS locations and patenting
activities. Furthermore, we add year FE to capture time trends that are common to all
municipalities. However, municipality FE and year FE still do not account for time-variant
regional factors potentially determining both UAS locations and patenting, such as regional
economic activity. Neglecting such factors biases our results if these factors are correlated
with both the treatment, i.e., the timing and location of UAS campus establishments, and
patenting. To be able to control for these factors in our estimations, we need appropriate
data that reach as far back in time as 1985, when applied research was added to the
purpose of UASs.
Unfortunately, no direct measure of regional economic activity or of other factors
related to UAS establishments and patenting exists at a su�ciently disaggregated regional
level for a time series dating as far back as 1985. Therefore, we use satellite-based data
from Lehnert et al. (2020) to proxy for regional economic activity. These data are available
from 1984 onwards and indicate the extent to which each of six surface groups covers
a region’s area: built-up land (i.e., artificial materials such as buildings), grass, forests,
cropland, land without vegetation (e.g., bare rock, pure soil), and water. As Lehnert et al.
(2020) demonstrate, changes in these surface groups are a valid predictor of changes in
regional economic activity even at detailed regional levels such as the municipality level.
We include the surface groups as control variables in our estimations to account for (at
least a part of the) time-variant regional heterogeneity determining both UAS locations
15
Figure 2: Treated municipalities in 2012 (25-kilometer travel-distance radius)
Source: Authors’ illustration with geodata from the BKG, data on HEIs from the German Rectors’Conference, data on PROs from BMBF (2018b), and self-collected data on HEIs and PROs.
16
and innovation.20
Combining all data sources, we thus estimate the following model:
Yi,t =�0 + �1UASi,t�3 + �2UASi,t�3 ⇥ UNIi+
�3UASi,t�3 ⇥MaxPlancki + �4UASi,t�3 ⇥ Leibnizi+
�5UASi,t�3 ⇥Helmholtzi + �6UASi,t�3 ⇥ Fraunhoferi+
i+ t+Xi,t�3 + SGi,t + µi,t
(1)
with i indicating the municipality, t the year ranging from 1984 (the first year in the surface
groups data)21 through 2012 (the last year in the patent data containing complete citation
information), Y the dependent variable (patent quantity or patent quality), UAS the
treatment status by a UAS campus in STEM, UNI, MaxPlanck, Leibniz, Helmholtz,
and Fraunhofer the existence of a campus or institute belonging to the respective
institution or organization at the time of the UAS campus establishment, and µ the error
term. The vector X includes controls for the establishment of other research institutions in
STEM during the observation period (i.e., the vector includes UNIi,t�3, MaxPlancki,t�3,
Leibnizi,t�3, Helmholtzi,t�3, and Fraunhoferi,t�3). The vector SG contains the surface
groups. We allow for a time lag of three years in the treatment variable UAS, because the
regional research structures of UASs need some time to establish after the announcement
of applied research becoming a goal of UASs.22
20 As the six surface groups are collinear, we follow Lehnert et al. (2020) and exclude the forests group asreference. Furthermore, we add a variable indicating the percentage of unidentified surfaces to accountfor potential measurement error in the surface groups variables (see Lehnert et al., 2020).
21 Appendix G provides the estimation results without using surface groups as control variables for thetime period 1980 through 2012. In these estimations, the pattern of the results is identical to thatin our main results, but the coe�cients are larger magnitude. Therefore, surface groups do indeedaccount for potential endogeneity in UAS locations.
22 For example, graduates are one important channel of knowledge transfer (see, e.g., Andrews, 2019;Lehnert et al., 2018) and the minimum number of years a UAS student needs to graduate is threeyears.
17
5. Results
5.1. Main Results
Table 1 shows the FE estimation results for the dependent variable patent quantity,
including the satellite-based surface groups to proxy for regional economic activity. In
essence, these estimations show that UASs in STEM have a positive and significant e↵ect
on patent quantity, with this e↵ect being substantially larger in regions where other
research institutions exist at the time of the UAS campus establishment. According to
our preferred specification in column 7, the establishment of a UAS campus in STEM
significantly increases patent quantity in regions without another research institution and
in regions where a UAS can draw upon the knowledge of Max Planck insitutes, Helmholtz
institutes, or Fraunhofer institutes.
When we do not take into account interaction e↵ects of UASs with other research
institutions, we find a positive and significant UAS e↵ect of 6.5 percent (table 1, column 1).
However, as this e↵ect becomes smaller when including the interaction terms indicating the
presence of other research institutions (columns 2 through 6), the presence of other research
institutions explains this e↵ect at least partly, with positive and significant interaction
e↵ects for all other types of research institutions. Finally, when including the full set of
interaction terms, the UAS e↵ect on innovation decreases to 2.7 percent and the interaction
terms with UNIs and Leibniz institutes become insignificant (column 7).
A look at the estimation results in Table 2 reveals that to increase patent quality,
UASs rely on the knowledge of other research institutions within a region. The pattern
of the e↵ects on patent quality is similar to that on patent quantity, although the e↵ects
are smaller in size. However, in the specification including the full set of interactions
(column 7), the UAS e↵ect becomes insignificant (column 1). Furthermore, only the
interaction term with Max Planck institutes remains statistically significant. Consequently,
the establishment of UASs only increases patent quality as measured by patent citations if
UASs can draw upon the knowledge of Max Planck institutes. This result is in line with
the Max Planck Society’s mission of aspiring to scientific excellence.
18
Tab
le1:
FEestimationresultson
patentqu
antity
(25-km
travel-distance
radius,3-y.
lag)
Dep.var.:
ln(P
QUAN
+1)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
UASi,t�
30.065⇤
⇤⇤0.043⇤
⇤⇤0.041⇤
⇤⇤0.060⇤
⇤⇤0.055⇤
⇤⇤0.044⇤
⇤⇤0.027⇤
⇤⇤
(0.006)
(0.007)
(0.006)
(0.006)
(0.006)
(0.006)
(0.007)
UASi,t�
3⇥UNI i=
10.067⇤
⇤⇤0.010
(0.012)
(0.013)
UASi,t�
3⇥M
axPlanck
i=
10.130⇤
⇤⇤0.094⇤
⇤⇤
(0.015)
(0.016)
UASi,t�
3⇥Leibn
izi=
10.068⇤
⇤⇤-0.021
(0.022)
(0.023)
UASi,t�
3⇥Helmholtz
i=
10.143⇤
⇤⇤0.069⇤
⇤⇤
(0.022)
(0.023)
UASi,t�
3⇥Fraunhof
eri=
10.126⇤
⇤⇤0.082⇤
⇤⇤
(0.016)
(0.017)
Municipalities
11,266
11,266
11,266
11,266
11,266
11,266
11,266
N308,087
308,087
308,087
308,087
308,087
308,087
308,087
R2(w
ithin)
0.097
0.097
0.098
0.097
0.097
0.098
0.099
R2(between)
0.210
0.225
0.247
0.215
0.197
0.233
0.249
R2(overall)
0.107
0.114
0.132
0.111
0.113
0.122
0.138
Sou
rce:
Authors’
calculation
sbased
onpatentdatafrom
theEPOs,
dataon
surfacegrou
psfrom
Lehnertet
al.(202
0),dataon
HEIs
from
the
German
Rectors’Con
ference,dataon
PROsfrom
BMBF(2018b
),an
dself-collected
dataon
HEIs
andPROs.
Notes:Rob
ust
stan
darderrors
inparentheses.Allmod
elsincludeintercep
t,municipalityFE,year
FE,surfacegrou
ps,
andcontrols
forUNI i
,t�3,
MaxPlanck
i,t�
3,Leibn
izi,t�
3,Helmholtz
i,t�
3,an
dFra
unhof
eri,t�
3.⇤p<
0.10,⇤⇤p<
0.05,⇤⇤
⇤p<
0.01.
19
Tab
le2:
FEestimationresultson
patentqu
ality(citations)
(25-km
travel-distance
radius,3-y.
lag)
Dep.var.:
ln(P
QUALcit+1)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
UASi,t�
30.015⇤
⇤⇤0.010⇤
⇤⇤0.008⇤
⇤0.014⇤
⇤⇤0.014⇤
⇤⇤0.011⇤
⇤⇤0.006
(0.003)
(0.004)
(0.003)
(0.003)
(0.003)
(0.003)
(0.004)
UASi,t�
3⇥UNI i=
10.018⇤
⇤⇤0.005
(0.006)
(0.007)
UASi,t�
3⇥M
axPlanck
i=
10.039⇤
⇤⇤0.035⇤
⇤⇤
(0.007)
(0.008)
UASi,t�
3⇥Leibn
izi=
10.019⇤
⇤-0.001
(0.010)
(0.010)
UASi,t�
3⇥Helmholtz
i=
10.016
-0.007
(0.010)
(0.011)
UASi,t�
3⇥Fraunhof
eri=
10.025⇤
⇤⇤0.014
(0.008)
(0.009)
Municipalities
11,266
11,266
11,266
11,266
11,266
11,266
11,266
N308,087
308,087
308,087
308,087
308,087
308,087
308,087
R2(w
ithin)
0.028
0.028
0.028
0.028
0.028
0.028
0.028
R2(between)
0.080
0.097
0.118
0.082
0.084
0.095
0.126
R2(overall)
0.037
0.039
0.044
0.037
0.038
0.039
0.045
Sou
rce:
Authors’
calculation
sbased
onpatentdatafrom
theEPOs,
dataon
surfacegrou
psfrom
Lehnertet
al.(202
0),dataon
HEIs
from
the
German
Rectors’Con
ference,dataon
PROsfrom
BMBF(2018b
),an
dself-collected
dataon
HEIs
andPROs.
Notes:Rob
ust
stan
darderrors
inparentheses.Allmod
elsincludeintercep
t,municipalityFE,year
FE,surfacegrou
ps,
andcontrols
forUNI i
,t�3,
MaxPlanck
i,t�
3,Leibn
izi,t�
3,Helmholtz
i,t�
3,an
dFra
unhof
eri,t�
3.⇤p<
0.10,⇤⇤p<
0.05,⇤⇤
⇤p<
0.01.
20
In sum, the FE estimations suggest that UASs can tap their full potential as a driver
of regional innovation when other research institutions coexist within UAS regions. For
patent quantity, we find positive interaction e↵ects of UASs in regions where they can
draw upon basic research knowledge (Max Planck institutes), applied research knowledge
(Fraunhofer institutes), or a combination of both (Helmholtz institutes). For patent quality,
we find a positive interaction e↵ect of UASs in regions where they can draw upon basic
research knowledge (Max Planck institutes). Therefore, knowledge complementarities
between UASs and other research institutions foster regional innovation.
5.2. Robustness Checks
To assess the role of the radius of 25 kilometers we chose for the assignment of municipalities
to the treatment and the control groups, we estimate Equation 1 with radii of 20 kilometers
and 30 kilometers. Appendices C and D show the results of these estimations. We find
the same overall pattern of e↵ects as with the 25-kilometer radius. Interestingly, in the
model specification with the whole set of interaction terms, the UAS e↵ect on patent
quantity (without any coexisting research institutions) turns insignificant when we use the
30-kilometer radius (column 7 in table D1). On the contrary, when we use the 20-kilometer
radius, the UAS e↵ect on patent quality remains highly significant (column 7 in table D2).
This finding suggests that the impact of a UAS campus establishment concentrates in
regions closer to the UAS campus.
To investigate how the timing of the treatment lag a↵ects our results, we repeat our
estimations with a lag of five years in Appendix E. Again, the overall pattern of e↵ects
remains unchanged, with two notable exceptions. First, although rather small in magnitude,
the UAS e↵ect on patent quality (without any coexisting research institutions) remains
significant at the five percent level (column 7 in table E2). This finding indicates that
UASs might need more time than three years to increase the quality of regional innovation.
Second, the interaction with Leibniz institutes yields a significantly negative e↵ect on
patent quantity (column 7 of table E1). One potential reason for this finding might be
that Leibniz sta↵ who produced innovations move to UASs after a certain time since the
21
UAS establishment and at the UAS, they do not engage in innovation activities anymore.
Another potential explanation might be that UASs and Leibniz institutes compete for
cooperation partners, thereby impeding innovation processes.
Furthermore, to test if the inclusion of the new federal states from 1991 onwards biases
our results, we repeat our main analysis for the old federal states only. Appendix F
shows the results of these regressions. Again, the pattern of the results is similar to our
main analysis that includes the new federal states. The only notable exceptions are a
negative e↵ect (significant at the ten-percent level) on patent quantity for UASs in regions
with Leibniz institutes (column 7 in table F1) and, somewhat surprisingly, a significant
negative e↵ect (significant at the five-percent level) on patent quality for UASs in regions
with Helmholtz institutes (column 7 in table F2). Again, movements of research sta↵ or
competition for cooperation partners could explain these e↵ects. Furthermore, in light of
this finding, the insignificance of the two interactions in the full sample also suggests that
regions in the new federal states yield a greater potential for knowledge complementarities
between UASs and Leibniz or Helmholtz institutes to arise.
As patent citations might not capture all aspects of a patent’s quality, we consider
two additional patent quality measures (for an overview of patent quality indicators see
Squicciarini et al., 2013). Following Pfister et al. (2018), we use the number of claims
(PQUALclaims, an indicator for a patent’s technological scope) and family size (PQUALsize,
the number of countries where the invention has legal protection) as additional outcome
variables. Appendix H shows the estimation results for these two variables.
The analysis of the additional patent quality measures yields two important insights.
First, for both number of claims and family size, the UAS e↵ect (without coexisting
research institutions) remains positive and highly significant (columns 7 in tables H1
and H2). Second, none of the interaction terms remains positive and significant, but the
interaction with Leibniz institutes even becomes negative and highly significant for both
additional patent quality measures (columns 7 in tables H1 and H2) and the interaction
with UNIs becomes negative and significant for number of claims (column 7 in tables H1).
These findings indicate that for patent quality as measured by number of claims or family
22
size, UASs might be substitutes for other types of research institutions with which they
compete in case of coexistence. Without this competition, UASs do increase patent quality
in the two additional dimensions.
6. The Role of UASs for Innovation in the New Fe-
deral States
To highlight the role of UASs for innovation and technological development, we perform
an additional preliminary analysis comparing treated regions in the new federal states
(former East Germany) to treated regions in the old federal states (former West Germany).
Upon reunification in 1991, the levels of innovation and technological development were
substantially lower in the regions belonging to the new federal states due to their di↵erent
history behind the iron curtain. After reunification, UASs—due to their focus on the local
economy—are probably one instrument for catching up to the old federal states. In this
subsection, we therefore analyze whether treated regions in the new federal states yield a
larger increase in innovation and technological development than regions in the old federal
states, thereby extending comparative case studies such as Fritsch and Graf (2011).
To measure innovation and technological development, we consider three outcome
variables. As indicators of innovation, we use (1) patent quantity and (2) patent quality
analogous to our preceding analyses. As an indicator of technological development, we
use (3) patent lag (PLAG). At the level of patent families, the patent lag indicates the
time distance (in years) between the patent application year and the average publication
year of the sources that the patent references. The smaller the patent lag, the closer is
the patented technology to the technological frontier. Analogous to the procedure for the
patent quantity and patent quality measures, we aggregate patent lag at the municipality
level (see section 4).23
Figure 3 descriptively shows the trends in the three outcome variables and their 95
23 For municipalities without any patent applications within a given year, we set patent lag to the largestvalue in the data (i.e., to the longest observed time distance between a patent application and itsreferenced sources). Furthermore, as the patent lag variable contains a few negative values due to thereporting level of patent families, we set these negative values to zero.
23
Figure 3: Trends of outcome variables in old and new federal states
1980
1985
1990
1995
2000
2005
2010
0.5
1.0
1.5
2.0
2.5
3.0
Year
PQUAN
Old federal states
New federal states
(a) Patent quantity
1980
1985
1990
1995
2000
2005
2010
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
Year
PQUALcit
Old federal states
New federal states
(b) Patent quality (citations)
1980
1985
1990
1995
2000
2005
2010
55
60
65
70
75
80
85
90
Year
PLAG
Old federal states
New federal states
(c) Patent lag
Source: Authors’ illustrations based on patent data from the EPO.Note: Graphs show variable means and their 95 percent confidence intervals.
24
percent confidence intervals separately for the old and the new federal states.24 In 1991—so
just after the fall of the iron curtain and before any catch-up process—municipalities in
the new federal states exhibit lower levels of patent quantity (figure 3a) and of patent
quality (figure 3b), and a higher level of patent lag (figure 3c). Afterwards and on average
(independent of UAS treatment status), we observe for patent quality and patent lag that
the trends run parallel over time. Given the lower initial level, this parallel trend implies
that the change rate is larger in the new federal states than in the old ones. For all three
outcome variables, the new federal states never reach the same level as the old ones.
To investigate whether UASs accelerate innovation and technological development in
the new federal states, we divide the post-reunification period (beginning in 1991) into
three-year periods p (i.e., p = 1993 denotes the first period from 1991 through 1993,
p = 1996 the second period from 1994 through 1996, etc.). We use the first three-year
period as the baseline period (representing the levels of innovation and technological
development immediately after the reunification) to compare the subsequent periods to
this baseline. For every three-year period, we calculate the means of the three outcome
variables at the municipality level (denoted as PQUANavg, PQUALavgcit , and PLAGavg).
To analyze whether UASs in the new federal states yield di↵erent e↵ects than UASs
in the old ones, we perform propensity-score matching we the goal of identifying causal
e↵ects. In so doing, we restrict the sample of municipalities to those that are treated by a
UAS campus in STEM in the first year of a three-year period. To ensure a comparison of
regions with similar levels of economic activity before the reunification and with similar
research infrastructures in the observed three-year period, we match municipalities in
the new federal states and those in the old ones based on the following characteristics:
average growth in the surface groups from Lehnert et al. (2020) as a proxy for the pre-
reunification trend in economic activity and current treatment by other types of research
institutions (UNIs, Max Planck institutes, Leibniz institutes, Helmholtz institutes, and
Fraunhofer institutes). Thus the treatment in these analyses is whether a UAS region
is located in the new federal states, denoted as newFS. We conduct separate analyses
24 Due to its historical situation, we exclude the city of Berlin from the analyses in this subsection.
25
for all six three-year periods following the baseline period. As outcome variables, we use
the di↵erences in patent quantity, patent quality, and patent lag between the observed
three-year period and the baseline period, denoted as PQUANdiff , PQUALdiffcit , and
PLAGdiff (e.g., PQUANdiff2011 = PQUANavg
2011 � PQUANavg1993).
Table 3: Propensity-score matching results
PQUANdiff PQUALdiffcit PLAGdiff
p N (1) (2) (3)
newFSi 1996 2,961 -0.475⇤⇤⇤ -0.006 -1.649
(0.096) (0.030) (1.348)
1999 3,494 -1.369⇤⇤⇤ -0.016 -3.806⇤⇤⇤
(0.196) (0.022) (1.254)
2002 3,494 -1.839⇤⇤⇤ 0.125 -3.418⇤
(0.258) (0.128) (1.820)
2005 3,526 -1.924⇤⇤⇤ -0.057⇤⇤ -4.425⇤⇤⇤
(0.328) (0.023) (1.523)
2008 3,526 -1.866⇤⇤⇤ 0.069⇤ -7.212⇤⇤⇤
(0.277) (0.040) (1.783)
2011 3,707 -1.583⇤⇤⇤ 0.063 -8.492⇤⇤⇤
(0.256) (0.050) (1.789)
Source: Authors’ calculations based on patent data from the EPO, data on surfacegroups from Lehnert et al. (2020), data on HEIs from the German Rectors’ Conference,data on PROs from BMBF (2018b), and self-collected data on HEIs and PROs.Notes: Robust standard in parentheses. ⇤p < 0.10, ⇤⇤p < 0.05, ⇤⇤⇤p < 0.01.
Table 3 shows the results of the propensity-score matching analyses. In line with the
descriptive results, UASs yield a smaller e↵ect on patent quantity in the new federal states
than in the old ones (column 1). This di↵erence persists throughout the entire observation
period. For patent quality, most coe�cients are rather small in magnitude and statistically
insignificant (column 2). Thus the UASs’ e↵ect on patent quality does not di↵er between
the new and the old federal states.
Furthermore, we find that in the new federal states, UASs accelerate technological
development. There, UASs decrease the patent lag to a significantly larger extent than
26
in the old federal states (column 3 in table 3). This finding implies that UASs help the
regions in the new federal states in catching up to the technological development of the
old federal states. This e↵ect constantly increases in magnitude over time, suggesting that
UAS influence technological development in the long term.
7. Conclusion
This paper analyzes the innovation e↵ect of UASs in Germany and focuses in particular
on the role of knowledge complementarities between UASs and other basic and applied
research institutions. To do so, we exploit variation in the location and timing of the
establishments of UASs in Germany to compare the development of patenting activities
in municipalities with a UAS and those without one. We analyze whether the e↵ect of
UASs varies within regions in which other types of research institutions coexist to assess
complementarities between these institutions and UASs. To account for endogeneity in
the location and timing decisions of the UAS establishments, we estimate FE models and
adjust for regional economic activity by including a proxy for regional economic activity
derived from satellite data by Lehnert et al. (2020) in our estimations.
Our estimations yield the result that UASs have a positive and significant e↵ect on
patent quantity (i.e., the number of patent applications in a municipality) and patent
quality (i.e, the average number of patent citations three years after publication) and that
this e↵ect is substantially larger when the scientific knowledge of other research institutions
is available in a region. More specifically, knowledge complementarities between UASs and
three other types of research institutions—Max Planck institutes (which perform basic
research), Fraunhofer institutes (which perform applied research), and Helmholtz institutes
(which perform a mixture of both)—lead to an increase in patent quantity. Regarding
patent quality, complementarities arise only between UASs and Max Planck institutes.
An additional analysis yields the preliminary finding that UASs help regions in the new
federal states to catch up with regard to technological development.
Nevertheless, our robustness checks indicate that when we consider additional dimensi-
ons of patent quality (number of priority claims and patent family size), the establishment
27
of UASs can also lead to a decrease in these outcomes when UASs are established in regions
with UNIs (which perform basic research) or in regions with Leibniz institutes (which
have a tendency towards basic research). Although we cannot conclusively determine the
reasons for these decreases, one potential explanation could be competition between UASs
and other research institutions, e.g., competition for cooperation partners from the private
sector. If the institution that existed before the establishment of a UAS was already able
to provide the necessary knowledge resources to the local economy, establishing a new
institution could create a type of competition that prohibits knowledge complementarities.
Therefore, when designing reforms of the education and innovation system, policyma-
kers should consider the regional preconditions in decisions on the locations of research
institutions. Establishing a research institution in regions without other institutions can
increase regional innovation. However, if other research institutions already exist and if
the research knowledge of these existing institutions is complementary to that of new
institutions, the interaction of existing and new institutions yields great potential to boost
regional innovation outcomes. However, in some cases the opposite e↵ect might occur if
the research knowledge of two types of institutions is not complementary. If policymakers
aim at increasing regional innovation, they should exercise caution in the decision on new
locations of research institutions.
28
References
Adams, J.D., Chiang, E.P., Jensen, J.L., 2003. The influence of federal laboratoryR&D on industrial research. The Review of Economics and Statistics 85, 1003–1020.doi:10.1162/003465303772815899.
Andrews, M., 2019. How do institutions of higher education a↵ect local invention? Evidencefrom the establishment of U.S. colleges. doi:10.2139/ssrn.3072565.
Anselin, L., Varga, A., Acs, Z., 1997. Local geographic spillovers between universityresearch and high technology innovations. Journal of Urban Economics 42, 422–448.doi:10.1006/juec.1997.2032.
Audretsch, D.B., Feldman, M.P., 1996. R&D spillovers and the geography of innovationand production. The American Economic Review 86, 630–640.
Autant-Bernard, C., 2001. Science and knowledge flows: Evidence from the French case.Research Policy 30, 1069–1078. doi:10.1016/S0048-7333(00)00131-1.
Becker, W., Dietz, J., 2004. R&D cooperation and innovation activities of firms: Evidencefor the German manufacturing industry. Research Policy 33, 209–223. doi:10.1016/j.respol.2003.07.003.
Beise, M., Stahl, H., 1999. Public research and industrial innovations in Germany. ResearchPolicy 28, 397–422. doi:10.1016/S0048-7333(98)00126-7.
BMBF [Federal Ministry of Education and Research], 2004. Die Fachhochschulen inDeutschland. 4 ed., Federal Ministry of Education and Research, Bonn, Berlin.
BMBF [Federal Ministry of Education and Research], 2018a. Bundesbericht Forschungund Innovation 2018: Forschungs- und innovationspolitische Ziele und Maßnahmen.BMBF, Berlin.
BMBF [Federal Ministry of Education and Research], 2018b. Organisationen undEinrichtungen in Forschung und Wissenschaft: Organisationenband BundesberichtForschung und Innovation 2018. Federal Ministry of Education and Research, Berlin.
Brill, A., 2017. Von der “Blauen Liste” zur gesamtdeutschen Wissenschaftsorganisation.Dokserver des Zentrums fur Zeithistorische Forschung Potsdam, Potsdam. doi:10.14765/zzf.dok.1.1115.
Broekel, T., Graf, H., 2012. Public research intensity and the structure of GermanR&D networks: A comparison of 10 technologies. Economics of Innovation and NewTechnology 21, 345–372. doi:10.1080/10438599.2011.582704.
Buschle, N., Haider, C., 2016. Private Hochschulen in Deutschland. WISTA 1. StatistischesBundesamt. Wiesbaden.
Cabrer-Borras, B., Serrano-Domingo, G., 2007. Innovation and R&D spillover e↵ects inSpanish regions: A spatial approch. Research Policy 36, 1357–1371. doi:10.1016/j.respol.2007.04.012.
29
Comin, D., Licht, G., Pellens, M., Schubert, T., 2019. Do companies benefit frompublic research organizations? The impact of the Fraunhofer Society in Germany.ZEW Discussion Paper No. 19-006. Leibniz Centre for European Economic Research.Mannheim.
Cowan, R., Zinovyeva, N., 2013. University e↵ects on regional innovation. Research Policy42, 788–800. doi:10.1016/j.respol.2012.10.001.
EFI [Commission of Experts for Research and Innovation], 2010. Research, innovationand technological performance in Germany – EFI report 2010. Commission of Expertsfor Research and Innovation, Berlin.
Enders, J., 2010. Hochschulen und Fachhochschulen, in: Simon, D., Knie, A., Hornbostel, S.(Eds.), Handbuch Wissenschaftspolitik. VS Verlag fur Sozialwissenschaften, Wiesbaden,pp. 443–456.
Eyles, A., Machin, S., 2019. The introduction of academy schools to England’s education.Journal of the European Economic Association 17, 1107–1146. doi:10.1093/jeea/jvy021.
Fritsch, M., Graf, H., 2011. How sub-national conditions a↵ect regional innovationsystems: The case of the two Germanys. Papers in Regional Science 90, 331–354.doi:10.1111/j.1435-5957.2011.00364.x.
Fritsch, M., Schwirten, C., 1999. Enterprise-university co-operation and the role of publicresearch institutions in regional innovation systems. Industry and Innovation 6, 69–83.doi:10.1080/13662719900000005.
Fritsch, M., Slavtchev, V., 2007. Universities and innovation in space. Industry andInnovation 14, 201–218. doi:10.1080/13662710701253466.
Graf, H., 2011. Gatekeepers in regional networks of innovators. Cambridge Journal ofEconomics 35, 173–198. doi:10.1093/cje/beq001.
Hachmeister, C.D., Duong, S., Roessler, I., 2015. Forschung an Fachhochschulen aus derInnen- und Außenperspektive: Rolle der Forschung, Art und Umfang. CHE ArbeitspapierNr. 181. Centrum fur Hochschulentwicklung (CHE). Gutersloh.
Helmers, C., Overman, H.G., 2017. My precious! The location and di↵usion of scientificresearch: Evidence from the synchrotron diamond light source. The Economic Journal127, 2006–2040. doi:10.1111/ecoj.12387.
Hinz, M., Gothe, K., von Polier, X., Wagner, B., Wimmer, M., Falke, U., 2016. Forschung anFachhochschulen: Wie aus praxisorientierter Forschung Produkte und Dienstleistungenwerden. BMBF, Bonn.
Ho↵mann, D., Trischler, H., 2015. Die Helmholtz-Gemeinschaft in historischer Perspektive,in: Mlynek, J., Bittner, A. (Eds.), 20 Jahre Helmholtz-Gemeinschaft. Hermann vonHelmholtz-Gemeinschaft Deutscher Forschungszentren e.V., Bonn, pp. 9–47.
Hohn, H.W., 2010. Außeruniversitare Forschungseinrichtungen, in: Simon, D., Knie, A.,Hornbostel, S. (Eds.), Handbuch Wissenschaftspolitik. VS Verlag fur Sozialwissenschaf-ten, Wiesbaden, pp. 457–477.
30
Intarakumnerd, P., Goto, A., 2018. Role of public research institutes in national innovationsystems in industrialized countries: The cases of Fraunhofer, NIST, CSIRO, AIST, andITRI. Research Policy 47, 1309–1320. doi:10.1016/j.respol.2018.04.011.
Ja↵e, A.B., 1989. Real e↵ects of academic research. The American Economic Review 79,957–970.
Kamhofer, D.A., Schmitz, H., Westphal, M., 2019. Heterogeneity in marginal non-monetaryreturns to higher education. Journal of the European Economic Association 17, 205–244.doi:10.1093/jeea/jvx058.
Kulicke, M., Stahlecker, T., 2004. Forschungslandkarte Fachhochschulen: Potenzialstudie.BMBF, Bonn, Berlin.
Lehnert, P., Niederberger, M., Backes-Gellner, U., 2020. Measuring economic activity withsatellite imagery: Small regions and long time series. Leading House Working Paper(forthcoming). Swiss Leading House on Economics of Education, Firm Behavior andTraining Policies. Zurich.
Lehnert, P., Pfister, C., Backes-Gellner, U., 2018. Firms’ Changes in R&D Personnel Afterthe Introduction of Universities of Applied Sciences in Switzerland. Leading HouseWorking Paper No. 141. Swiss Leading House on Economics of Education, Firm Behaviorand Training Policies.
Leten, B., Landoni, P., Van Looy, B., 2014. Science or graduates: How do firms benefitfrom the proximity of universities? Research Policy 43, 1398–1412. doi:10.1016/j.respol.2014.03.005.
Leyden, D.P., Menter, M., 2018. The legacy and promise of Vannevar Bush: Rethinkingthe model of innovation and the role of public policy. Economics of Innovation and NewTechnology 27, 225–242. doi:10.1080/10438599.2017.1329189.
Meurer, P., 2018. Zugang von FH-/HAW-Absolventinnen und -Absolventen zur Promotion,kooperative Promotionen und Promotionsrecht. Studien zum deutschen Innovationssy-stem No. 16-2018. Commission of Experts for Research and Innovation. Berlin.
Pfister, C., Rinawi, M., Harho↵, D., Backes-Gellner, U., 2018. Regional innovation e↵ectsof applied research institutions. BFI Working Paper No. 2019-50. Swiss Leading Houseon Economics of Education, Firm Behavior and Training Policies. Chicago.
Ponds, R., van Oort, F., Frenken, K., 2007. The geographical and institutional proximityof research collaboration. Papers in Regional Science 86, 423–444. doi:10.1111/j.1435-5957.2007.00126.x.
Popp, D., 2017. From science to technology: The value of knowledge from di↵erent energyresearch institutions. Research Policy 46, 1580–1594. doi:10.1016/j.respol.2017.07.011.
Robin, S., Schubert, T., 2013. Cooperation with public research institutions and successin innovation: Evidence from France and Germany. Research Policy 42, 149–166.doi:10.1016/j.respol.2012.06.002.
31
Squicciarini, M., Dernis, H., Criscuolo, C., 2013. Measuring patent quality: Indicators oftechnological and economic value. OECD Science, Technology and Industry WorkingPapers 2013/03. Organisation for Economic Co-operation and Development. Paris.doi:10.1787/5k4522wkw1r8-en.
Toivanen, O., Vaananen, L., 2016. Education and invention. The Review of Economicsand Statistics 98, 382–396. doi:10.1162/REST_a_00520.
Wissenschaftsrat, 2002. Empfehlungen zur Entwicklung der Fachhochschulen. Drs. 5102/02,Wissenschaftsrat.
32
Appendices
Appendix A. Descriptive Statistics
Table A1: Descriptive statistics for patent quantity (PQUAN)
Treatment group Control group
Year N Mean SD Min. Max. N Mean SD Min. Max.
1980 3,111 1.43 9.21 0.00 373.03 5,494 0.27 1.97 0.00 109.82
1981 3,111 1.51 8.92 0.00 319.33 5,494 0.30 1.90 0.00 97.62
1982 3,111 1.57 9.20 0.00 378.96 5,494 0.30 1.86 0.00 102.90
1983 3,111 1.80 9.87 0.00 393.11 5,494 0.37 2.12 0.00 114.90
1984 3,111 1.97 10.10 0.00 387.92 5,494 0.40 2.06 0.00 109.42
1985 3,111 2.14 11.01 0.00 425.76 5,494 0.42 2.25 0.00 119.70
1986 3,111 2.24 11.02 0.00 416.78 5,494 0.45 2.22 0.00 111.72
1987 3,111 2.50 11.77 0.00 425.19 5,494 0.53 2.61 0.00 129.78
1988 3,111 2.72 12.94 0.00 463.36 5,494 0.57 2.55 0.00 111.46
1989 3,111 2.76 12.89 0.00 476.14 5,494 0.57 2.75 0.00 138.49
1990 3,111 2.59 11.78 0.00 438.52 5,494 0.53 2.31 0.00 101.23
1991 3,720 2.20 11.17 0.00 391.32 7,546 0.40 2.08 0.00 116.36
1992 3,720 2.23 11.43 0.00 435.26 7,546 0.41 2.10 0.00 123.70
1993 3,720 2.26 11.05 0.00 380.13 7,546 0.43 1.98 0.00 104.56
1994 3,720 2.42 11.99 0.00 399.79 7,546 0.45 2.19 0.00 118.22
1995 3,720 2.52 12.69 0.00 465.81 7,546 0.48 2.13 0.00 91.48
1996 3,720 3.00 15.31 0.00 591.50 7,546 0.58 2.54 0.00 119.88
1997 3,720 3.41 17.66 0.00 711.12 7,546 0.64 2.60 0.00 114.72
1998 3,720 3.79 20.65 0.00 895.03 7,546 0.72 2.77 0.00 114.95
1999 3,720 4.13 22.05 0.00 899.42 7,546 0.74 2.92 0.00 131.53
2000 3,720 4.36 24.73 0.00 1,009.36 7,546 0.78 2.99 0.00 106.95
2001 3,720 4.34 24.04 0.00 940.04 7,546 0.76 2.91 0.00 113.15
2002 3,720 4.31 23.41 0.00 840.96 7,546 0.77 2.91 0.00 112.89
2003 3,720 4.35 23.33 0.00 802.77 7,546 0.80 3.08 0.00 121.79
2004 3,720 4.51 23.56 0.00 797.36 7,546 0.84 3.14 0.00 112.81
2005 3,720 4.66 24.59 0.00 802.33 7,546 0.88 3.38 0.00 135.32
2006 3,720 4.64 25.10 0.00 902.28 7,546 0.91 3.29 0.00 112.21
2007 3,720 4.69 25.08 0.00 870.89 7,546 0.91 3.22 0.00 108.08
2008 3,720 4.43 24.73 0.00 860.18 7,546 0.88 3.12 0.00 92.67
2009 3,720 4.49 23.89 0.00 795.56 7,546 0.89 3.24 0.00 97.64
2010 3,720 4.52 24.54 0.00 859.75 7,546 0.86 3.14 0.00 106.30
2011 3,720 4.41 24.20 0.00 819.56 7,546 0.86 3.17 0.00 91.38
2012 3,720 4.19 23.73 0.00 844.43 7,546 0.83 3.23 0.00 114.08
Total 116,061 3.31 18.69 0.00 1,009.36 226,446 0.64 2.72 0.00 138.49
Source: Authors’ calculations based on patent data from the EPO, data on HEIs from the German Rectors’Conference, data on PROs from BMBF (2018b), and self-collected data on HEIs and PROs.
33
Table A2: Descriptive statistics for patent quality (citations) (PQUALcit)
Treatment group Control group
Year N Mean SD Min. Max. N Mean SD Min. Max.
1980 3,111 0.11 0.33 0.00 4.00 5,494 0.04 0.23 0.00 4.00
1981 3,111 0.13 0.42 0.00 7.00 5,494 0.05 0.29 0.00 10.00
1982 3,111 0.13 0.37 0.00 5.00 5,494 0.05 0.27 0.00 6.22
1983 3,111 0.15 0.41 0.00 5.00 5,494 0.06 0.27 0.00 4.40
1984 3,111 0.19 0.50 0.00 10.00 5,494 0.08 0.33 0.00 6.00
1985 3,111 0.19 0.53 0.00 15.00 5,494 0.09 0.40 0.00 11.00
1986 3,111 0.21 0.46 0.00 4.00 5,494 0.10 0.42 0.00 7.00
1987 3,111 0.23 0.52 0.00 7.00 5,494 0.10 0.40 0.00 7.00
1988 3,111 0.22 0.57 0.00 17.00 5,494 0.10 0.37 0.00 8.00
1989 3,111 0.23 0.49 0.00 6.48 5,494 0.11 0.40 0.00 6.00
1990 3,111 0.24 0.60 0.00 13.50 5,494 0.12 0.48 0.00 18.00
1991 3,720 0.23 0.52 0.00 6.00 7,546 0.10 0.46 0.00 15.00
1992 3,720 0.24 0.59 0.00 10.85 7,546 0.11 0.48 0.00 19.00
1993 3,720 0.25 0.60 0.00 7.22 7,546 0.12 0.58 0.00 24.00
1994 3,720 0.29 0.63 0.00 10.00 7,546 0.12 0.50 0.00 11.00
1995 3,720 0.30 0.63 0.00 8.00 7,546 0.13 0.51 0.00 8.00
1996 3,720 0.34 0.76 0.00 17.00 7,546 0.16 0.57 0.00 14.47
1997 3,720 0.35 0.67 0.00 9.00 7,546 0.17 0.60 0.00 24.00
1998 3,720 0.37 0.74 0.00 13.00 7,546 0.17 0.65 0.00 26.00
1999 3,720 0.39 0.76 0.00 17.00 7,546 0.19 0.75 0.00 28.86
2000 3,720 0.38 1.14 0.00 54.00 7,546 0.19 0.66 0.00 18.00
2001 3,720 0.43 1.17 0.00 38.00 7,546 0.20 0.73 0.00 21.00
2002 3,720 0.43 0.92 0.00 13.63 7,546 0.21 0.67 0.00 16.00
2003 3,720 0.45 0.93 0.00 19.73 7,546 0.23 0.70 0.00 12.00
2004 3,720 0.42 0.80 0.00 10.00 7,546 0.24 0.82 0.00 24.00
2005 3,720 0.44 0.89 0.00 19.79 7,546 0.24 0.68 0.00 12.00
2006 3,720 0.41 0.79 0.00 11.93 7,546 0.23 0.66 0.00 12.00
2007 3,720 0.42 0.81 0.00 19.00 7,546 0.24 0.81 0.00 33.00
2008 3,720 0.46 1.03 0.00 32.93 7,546 0.26 0.80 0.00 19.00
2009 3,720 0.42 0.86 0.00 16.40 7,546 0.27 0.93 0.00 30.00
2010 3,720 0.39 0.88 0.00 28.00 7,546 0.21 0.92 0.00 50.00
2011 3,720 0.34 0.79 0.00 30.00 7,546 0.19 0.63 0.00 13.00
2012 3,720 0.27 0.54 0.00 9.00 7,546 0.16 0.54 0.00 13.00
Total 116,061 0.31 0.74 0.00 54.00 226,446 0.16 0.61 0.00 50.00
Source: Authors’ calculations based on patent data from the EPO, data on HEIs from the German Rectors’Conference, data on PROs from BMBF (2018b), and self-collected data on HEIs and PROs.
34
Appendix B. Distribution of Campus and Institute Locations
Figure B1: Distribution of campus and institute locations in 2012
Source: Authors’ illustration with geodata from the BKG, data on HEIs from the German Rectors’Conference, data on higher education institutions from BMBF (2018b), and self-collected data on HEIsand PROs.
35
Appendix C. Results with 20-kilometer Radius
Tab
leC1:
FEestimationresultson
patentqu
antity
(20-km
travel-distance
radius,3-y.
lag)
Dep.var.:
ln(P
QUAN
+1)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
UASi,t�
30.085⇤
⇤⇤0.060⇤
⇤⇤0.059⇤
⇤⇤0.079⇤
⇤⇤0.076⇤
⇤⇤0.063⇤
⇤⇤0.045⇤
⇤⇤
(0.007)
(0.008)
(0.008)
(0.007)
(0.007)
(0.008)
(0.008)
UASi,t�
3⇥UNI i=
10.088⇤
⇤⇤0.021
(0.016)
(0.017)
UASi,t�
3⇥M
axPlanck
i=
10.163⇤
⇤⇤0.115⇤
⇤⇤
(0.020)
(0.022)
UASi,t�
3⇥Leibn
izi=
10.110⇤
⇤⇤-0.002
(0.031)
(0.031)
UASi,t�
3⇥Helmholtz
i=
10.156⇤
⇤⇤0.053⇤
(0.030)
(0.032)
UASi,t�
3⇥Fraunhof
eri=
10.150⇤
⇤⇤0.088⇤
⇤⇤
(0.021)
(0.023)
Municipalities
11,266
11,266
11,266
11,266
11,266
11,266
11,266
N308,087
308,087
308,087
308,087
308,087
308,087
308,087
R2(w
ithin)
0.097
0.098
0.098
0.097
0.098
0.098
0.099
R2(between)
0.208
0.216
0.234
0.211
0.203
0.225
0.240
R2(overall)
0.109
0.115
0.129
0.113
0.114
0.121
0.133
Sou
rce:
Authors’
calculation
sbased
onpatentdatafrom
theEPOs,
dataon
surfacegrou
psfrom
Leh
nertet
al.(202
0),dataon
HEIs
from
the
German
Rectors’Con
ference,dataon
PROsfrom
BMBF(2018b
),an
dself-collected
dataon
HEIs
andPROs.
Notes:Rob
ust
stan
darderrors
inparentheses.Allmod
elsincludeintercept,
municipalityFE,year
FE,surfacegrou
ps,
andcontrols
forUNI i
,t�3,
MaxPlanck
i,t�
3,Leibn
izi,t�
3,Helmholtz
i,t�
3,an
dFra
unhof
eri,t�
3.⇤p<
0.10,⇤⇤p<
0.05,⇤⇤
⇤p<
0.01.
36
Tab
leC2:
FEestimationresultson
patentqu
ality(citations)
(20-km
travel-distance
radius,3-y.
lag)
Dep.var.:
ln(P
QUALcit+1)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
UASi,t�
30.021⇤
⇤⇤0.018⇤
⇤⇤0.017⇤
⇤⇤0.021⇤
⇤⇤0.020⇤
⇤⇤0.017⇤
⇤⇤0.015⇤
⇤⇤
(0.004)
(0.004)
(0.004)
(0.004)
(0.004)
(0.004)
(0.004)
UASi,t�
3⇥UNI i=
10.010
-0.004
(0.008)
(0.008)
UASi,t�
3⇥M
axPlanck
i=
10.029⇤
⇤⇤0.024⇤
⇤
(0.010)
(0.011)
UASi,t�
3⇥Leibn
izi=
10.012
-0.009
(0.013)
(0.014)
UASi,t�
3⇥Helmholtz
i=
10.024⇤
0.007
(0.014)
(0.016)
UASi,t�
3⇥Fraunhof
eri=
10.028⇤
⇤⇤0.021⇤
(0.010)
(0.011)
Municipalities
11,266
11,266
11,266
11,266
11,266
11,266
11,266
N308,087
308,087
308,087
308,087
308,087
308,087
308,087
R2(w
ithin)
0.028
0.028
0.028
0.028
0.028
0.028
0.028
R2(between)
0.075
0.081
0.096
0.076
0.077
0.084
0.097
R2(overall)
0.036
0.037
0.040
0.037
0.037
0.038
0.040
Sou
rce:
Authors’
calculation
sbased
onpatentdatafrom
theEPOs,
dataon
surfacegrou
psfrom
Leh
nertet
al.(202
0),dataon
HEIs
from
the
German
Rectors’Con
ference,dataon
PROsfrom
BMBF(2018b
),an
dself-collected
dataon
HEIs
andPROs.
Notes:Rob
ust
stan
darderrors
inparentheses.Allmod
elsincludeintercept,
municipalityFE,year
FE,surfacegrou
ps,
andcontrols
forUNI i
,t�3,
MaxPlanck
i,t�
3,Leibn
izi,t�
3,Helmholtz
i,t�
3,an
dFra
unhof
eri,t�
3.⇤p<
0.10,⇤⇤p<
0.05,⇤⇤
⇤p<
0.01.
37
Appendix D. Results with 30-kilometer Radius
Tab
leD1:
FEestimationresultson
patentqu
antity
(30-km
travel-distance
radius,3-y.
lag)
Dep.var.:
ln(P
QUAN
+1)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
UASi,t�
30.048⇤
⇤⇤0.023⇤
⇤⇤0.024⇤
⇤⇤0.043⇤
⇤⇤0.036⇤
⇤⇤0.027⇤
⇤⇤0.006
(0.005)
(0.006)
(0.006)
(0.005)
(0.005)
(0.005)
(0.006)
UASi,t�
3⇥UNI i=
10.068⇤
⇤⇤0.024⇤
⇤
(0.010)
(0.010)
UASi,t�
3⇥M
axPlanck
i=
10.115⇤
⇤⇤0.081⇤
⇤⇤
(0.012)
(0.013)
UASi,t�
3⇥Leibn
izi=
10.052⇤
⇤⇤-0.031
⇤
(0.017)
(0.018)
UASi,t�
3⇥Helmholtz
i=
10.143⇤
⇤⇤0.084⇤
⇤⇤
(0.017)
(0.018)
UASi,t�
3⇥Fraunhof
eri=
10.108⇤
⇤⇤0.060⇤
⇤⇤
(0.013)
(0.014)
Municipalities
11,266
11,266
11,266
11,266
11,266
11,266
11,266
N308,087
308,087
308,087
308,087
308,087
308,087
308,087
R2(w
ithin)
0.096
0.097
0.097
0.096
0.097
0.097
0.098
R2(between)
0.154
0.186
0.224
0.161
0.154
0.197
0.234
R2(overall)
0.082
0.092
0.114
0.086
0.094
0.101
0.124
Sou
rce:
Authors’
calculation
sbased
onpatentdatafrom
theEPOs,
dataon
surfacegrou
psfrom
Leh
nertet
al.(202
0),dataon
HEIs
from
the
German
Rectors’Con
ference,dataon
PROsfrom
BMBF(2018b
),an
dself-collected
dataon
HEIs
andPROs.
Notes:Rob
ust
stan
darderrors
inparentheses.Allmod
elsincludeintercept,
municipalityFE,year
FE,surfacegrou
ps,
andcontrols
forUNI i
,t�3,
MaxPlanck
i,t�
3,Leibn
izi,t�
3,Helmholtz
i,t�
3,an
dFra
unhof
eri,t�
3.⇤p<
0.10,⇤⇤p<
0.05,⇤⇤
⇤p<
0.01.
38
Tab
leD2:
FEestimationresultson
patentqu
ality(citations)
(30-km
travel-distance
radius,3-y.
lag)
Dep.var.:
ln(P
QUALcit+1)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
UASi,t�
30.013⇤
⇤⇤0.007⇤
⇤0.007⇤
⇤0.011⇤
⇤⇤0.011⇤
⇤⇤0.008⇤
⇤⇤0.003
(0.003)
(0.003)
(0.003)
(0.003)
(0.003)
(0.003)
(0.003)
UASi,t�
3⇥UNI i=
10.017⇤
⇤⇤0.006
(0.005)
(0.005)
UASi,t�
3⇥M
axPlanck
i=
10.031⇤
⇤⇤0.022⇤
⇤⇤
(0.006)
(0.006)
UASi,t�
3⇥Leibn
izi=
10.024⇤
⇤⇤0.007
(0.008)
(0.008)
UASi,t�
3⇥Helmholtz
i=
10.024⇤
⇤⇤0.005
(0.008)
(0.008)
UASi,t�
3⇥Fraunhof
eri=
10.024⇤
⇤⇤0.013⇤
⇤
(0.006)
(0.007)
Municipalities
11,266
11,266
11,266
11,266
11,266
11,266
11,266
N308,087
308,087
308,087
308,087
308,087
308,087
308,087
R2(w
ithin)
0.028
0.028
0.028
0.028
0.028
0.028
0.029
R2(between)
0.054
0.074
0.105
0.059
0.063
0.075
0.117
R2(overall)
0.034
0.036
0.042
0.035
0.035
0.037
0.043
Sou
rce:
Authors’
calculation
sbased
onpatentdatafrom
theEPOs,
dataon
surfacegrou
psfrom
Leh
nertet
al.(202
0),dataon
HEIs
from
the
German
Rectors’Con
ference,dataon
PROsfrom
BMBF(2018b
),an
dself-collected
dataon
HEIs
andPROs.
Notes:Rob
ust
stan
darderrors
inparentheses.Allmod
elsincludeintercept,
municipalityFE,year
FE,surfacegrou
ps,
andcontrols
forUNI i
,t�3,
MaxPlanck
i,t�
3,Leibn
izi,t�
3,Helmholtz
i,t�
3,an
dFra
unhof
eri,t�
3.⇤p<
0.10,⇤⇤p<
0.05,⇤⇤
⇤p<
0.01.
39
Appendix E. Results with Five-year Lag
Tab
leE1:
FEestimationresultson
patentqu
antity
(25-km
travel-distance
radius,5-y.
lag)
Dep.var.:
ln(P
QUAN
+1)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
UASi,t�
50.071⇤
⇤⇤0.048⇤
⇤⇤0.047⇤
⇤⇤0.068⇤
⇤⇤0.061⇤
⇤⇤0.051⇤
⇤⇤0.033⇤
⇤⇤
(0.006)
(0.007)
(0.006)
(0.006)
(0.006)
(0.006)
(0.007)
UASi,t�
5⇥UNI i=
10.068⇤
⇤⇤0.017
(0.012)
(0.013)
UASi,t�
5⇥M
axPlanck
i=
10.117⇤
⇤⇤0.088⇤
⇤⇤
(0.014)
(0.016)
UASi,t�
5⇥Leibn
izi=
10.039⇤
-0.047
⇤⇤
(0.020)
(0.021)
UASi,t�
5⇥Helmholtz
i=
10.132⇤
⇤⇤0.072⇤
⇤⇤
(0.021)
(0.022)
UASi,t�
5⇥Fraunhof
eri=
10.117⇤
⇤⇤0.071⇤
⇤⇤
(0.015)
(0.017)
Municipalities
11,266
11,266
11,266
11,266
11,266
11,266
11,266
N308,087
308,087
308,087
308,087
308,087
308,087
308,087
R2(w
ithin)
0.097
0.098
0.098
0.097
0.098
0.098
0.099
R2(between)
0.202
0.219
0.242
0.206
0.188
0.226
0.239
R2(overall)
0.101
0.106
0.122
0.103
0.104
0.114
0.126
Sou
rce:
Authors’
calculation
sbased
onpatentdatafrom
theEPOs,
dataon
surfacegrou
psfrom
Leh
nertet
al.(202
0),dataon
HEIs
from
the
German
Rectors’Con
ference,dataon
PROsfrom
BMBF(2018b
),an
dself-collected
dataon
HEIs
andPROs.
Notes:Rob
ust
stan
darderrors
inparentheses.Allmod
elsincludeintercept,
municipalityFE,year
FE,surfacegrou
ps,
andcontrols
forUNI i
,t�5,
MaxPlanck
i,t�
5,Leibn
izi,t�
5,Helmholtz
i,t�
5,an
dFra
unhof
eri,t�
5.⇤p<
0.10,⇤⇤p<
0.05,⇤⇤
⇤p<
0.01.
40
Tab
leE2:
FEestimationresultson
patentqu
ality(citations)
(25-km
travel-distance
radius,5-y.
lag)
Dep.var.:
ln(P
QUALcit+1)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
UASi,t�
50.020⇤
⇤⇤0.013⇤
⇤⇤0.012⇤
⇤⇤0.018⇤
⇤⇤0.018⇤
⇤⇤0.015⇤
⇤⇤0.009⇤
⇤
(0.003)
(0.003)
(0.003)
(0.003)
(0.003)
(0.003)
(0.004)
UASi,t�
5⇥UNI i=
10.020⇤
⇤⇤0.008
(0.006)
(0.006)
UASi,t�
5⇥M
axPlanck
i=
10.037⇤
⇤⇤0.030⇤
⇤⇤
(0.007)
(0.008)
UASi,t�
5⇥Leibn
izi=
10.018⇤
⇤-0.003
(0.009)
(0.010)
UASi,t�
5⇥Helmholtz
i=
10.023⇤
⇤0.002
(0.009)
(0.010)
UASi,t�
5⇥Fraunhof
eri=
10.027⇤
⇤⇤0.013
(0.007)
(0.008)
Municipalities
11,266
11,266
11,266
11,266
11,266
11,266
11,266
N308,087
308,087
308,087
308,087
308,087
308,087
308,087
R2(w
ithin)
0.028
0.029
0.029
0.028
0.028
0.029
0.029
R2(between)
0.053
0.068
0.090
0.053
0.056
0.067
0.099
R2(overall)
0.033
0.035
0.039
0.033
0.034
0.036
0.040
Sou
rce:
Authors’
calculation
sbased
onpatentdatafrom
theEPOs,
dataon
surfacegrou
psfrom
Leh
nertet
al.(202
0),dataon
HEIs
from
the
German
Rectors’Con
ference,dataon
PROsfrom
BMBF(2018b
),an
dself-collected
dataon
HEIs
andPROs.
Notes:Rob
ust
stan
darderrors
inparentheses.Allmod
elsincludeintercept,
municipalityFE,year
FE,surfacegrou
ps,
andcontrols
forUNI i
,t�5,
MaxPlanck
i,t�
5,Leibn
izi,t�
5,Helmholtz
i,t�
5,an
dFra
unhof
eri,t�
5.⇤p<
0.10,⇤⇤p<
0.05,⇤⇤
⇤p<
0.01.
41
Appendix F. Results for Old Federal States Only
Tab
leF1:
FEestimationresultson
patentqu
antity
(25-km
travel-distance
radius,3-y.
lag,
oldfederal
states
only)
Dep.var.:
ln(P
QUAN
+1)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
UASi,t�
30.065⇤
⇤⇤0.041⇤
⇤⇤0.043⇤
⇤⇤0.063⇤
⇤⇤0.058⇤
⇤⇤0.045⇤
⇤⇤0.029⇤
⇤⇤
(0.006)
(0.008)
(0.007)
(0.007)
(0.007)
(0.007)
(0.008)
UASi,t�
3⇥UNI i=
10.069⇤
⇤⇤0.020
(0.013)
(0.015)
UASi,t�
3⇥M
axPlanck
i=
10.112⇤
⇤⇤0.090⇤
⇤⇤
(0.015)
(0.017)
UASi,t�
3⇥Leibn
izi=
10.022
-0.046
⇤
(0.024)
(0.025)
UASi,t�
3⇥Helmholtz
i=
10.096⇤
⇤⇤0.029
(0.025)
(0.026)
UASi,t�
3⇥Fraunhof
eri=
10.110⇤
⇤⇤0.074⇤
⇤⇤
(0.017)
(0.019)
Municipalities
8,605
8,605
8,605
8,605
8,605
8,605
8,605
N249,545
249,545
249,545
249,545
249,545
249,545
249,545
R2(w
ithin)
0.103
0.103
0.104
0.103
0.103
0.104
0.104
R2(between)
0.368
0.398
0.371
0.368
0.345
0.374
0.382
R2(overall)
0.121
0.130
0.145
0.123
0.128
0.133
0.150
Sou
rce:
Authors’
calculation
sbased
onpatentdatafrom
theEPOs,
dataon
surfacegrou
psfrom
Leh
nertet
al.(202
0),dataon
HEIs
from
the
German
Rectors’Con
ference,dataon
PROsfrom
BMBF(2018b
),an
dself-collected
dataon
HEIs
andPROs.
Notes:Rob
ust
stan
darderrors
inparentheses.Allmod
elsincludeintercept,
municipalityFE,year
FE,surfacegrou
ps,
andcontrols
forUNI i
,t�3,
MaxPlanck
i,t�
3,Leibn
izi,t�
3,Helmholtz
i,t�
3,an
dFra
unhof
eri,t�
3.⇤p<
0.10,⇤⇤p<
0.05,⇤⇤
⇤p<
0.01.
42
Tab
leF2:
FEestimationresultson
patentqu
ality(citations)
(25-km
travel-distance
radius,3-y.
lag,
oldfederal
states
only)
Dep.var.:
ln(P
QUALcit+1)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
UASi,t�
30.015⇤
⇤⇤0.008⇤
⇤0.008⇤
⇤0.014⇤
⇤⇤0.015⇤
⇤⇤0.011⇤
⇤⇤0.005
(0.003)
(0.004)
(0.004)
(0.003)
(0.003)
(0.004)
(0.004)
UASi,t�
3⇥UNI i=
10.020⇤
⇤⇤0.010
(0.006)
(0.007)
UASi,t�
3⇥M
axPlanck
i=
10.036⇤
⇤⇤0.035⇤
⇤⇤
(0.007)
(0.008)
UASi,t�
3⇥Leibn
izi=
10.007
-0.010
(0.010)
(0.011)
UASi,t�
3⇥Helmholtz
i=
1-0.003
-0.024
⇤⇤
(0.011)
(0.012)
UASi,t�
3⇥Fraunhof
eri=
10.021⇤
⇤⇤0.012
(0.008)
(0.009)
Municipalities
8,605
8,605
8,605
8,605
8,605
8,605
8,605
N249,545
249,545
249,545
249,545
249,545
249,545
249,545
R2(w
ithin)
0.031
0.031
0.031
0.031
0.031
0.031
0.031
R2(between)
0.105
0.158
0.162
0.107
0.101
0.125
0.159
R2(overall)
0.036
0.039
0.043
0.036
0.036
0.038
0.043
Sou
rce:
Authors’
calculation
sbased
onpatentdatafrom
theEPOs,
dataon
surfacegrou
psfrom
Leh
nertet
al.(202
0),dataon
HEIs
from
the
German
Rectors’Con
ference,dataon
PROsfrom
BMBF(2018b
),an
dself-collected
dataon
HEIs
andPROs.
Notes:Rob
ust
stan
darderrors
inparentheses.Allmod
elsincludeintercept,
municipalityFE,year
FE,surfacegrou
ps,
andcontrols
forUNI i
,t�3,
MaxPlanck
i,t�
3,Leibn
izi,t�
3,Helmholtz
i,t�
3,an
dFra
unhof
eri,t�
3.⇤p<
0.10,⇤⇤p<
0.05,⇤⇤
⇤p<
0.01.
43
Appendix G. Results without Surface Groups
Tab
leG1:
FEestimationresultson
patentqu
antity
(25-km
travel-distance
radius,3-y.
lag,
withou
tsurfacegrou
ps)
Dep.var.:
ln(P
QUAN
+1)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
UASi,t�
30.114⇤
⇤⇤0.080⇤
⇤⇤0.081⇤
⇤⇤0.110⇤
⇤⇤0.101⇤
⇤⇤0.083⇤
⇤⇤0.058⇤
⇤⇤
(0.007)
(0.008)
(0.007)
(0.007)
(0.007)
(0.007)
(0.008)
UASi,t�
3⇥UNI i=
10.099⇤
⇤⇤0.026⇤
(0.014)
(0.015)
UASi,t�
3⇥M
axPlanck
i=
10.155⇤
⇤⇤0.115⇤
⇤⇤
(0.016)
(0.017)
UASi,t�
3⇥Leibn
izi=
10.051⇤
⇤-0.054
⇤⇤
(0.022)
(0.023)
UASi,t�
3⇥Helmholtz
i=
10.161⇤
⇤⇤0.069⇤
⇤⇤
(0.024)
(0.025)
UASi,t�
3⇥Fraunhof
eri=
10.172⇤
⇤⇤0.115⇤
⇤⇤
(0.017)
(0.019)
Municipalities
11,266
11,266
11,266
11,266
11,266
11,266
11,266
N342,507
342,507
342,507
342,507
342,507
342,507
342,507
R2(w
ithin)
0.128
0.129
0.130
0.128
0.129
0.130
0.132
R2(between)
0.036
0.046
0.076
0.040
0.045
0.063
0.088
R2(overall)
0.056
0.061
0.076
0.058
0.062
0.069
0.082
Sou
rce:
Authors’
calculation
sbased
onpatentdatafrom
theEPOs,
dataon
HEIs
from
theGerman
Rectors’Con
ference,dataon
PROsfrom
BMBF(2018b
),an
dself-collected
dataon
HEIs
andPROs.
Notes:Rob
ust
stan
darderrors
inparentheses.Allmod
elsincludeintercep
t,municipalityFE,year
FE,an
dcontrolsforUNI i
,t�3,M
axPlanck
i,t�
3,Leibn
izi,t�
3,Helmholtz
i,t�
3,an
dFra
unhof
eri,t�
3.⇤p<
0.10,⇤⇤p<
0.05,⇤⇤
⇤p<
0.01.
44
Tab
leG2:
FEestimationresultson
patentqu
ality(citations)
(25-km
travel-distance
radius,3-y.
lag,
withou
tsurfacegrou
ps)
Dep.var.:
ln(P
QUALcit+1)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
UASi,t�
30.030⇤
⇤⇤0.021⇤
⇤⇤0.019⇤
⇤⇤0.029⇤
⇤⇤0.028⇤
⇤⇤0.023⇤
⇤⇤0.015⇤
⇤⇤
(0.003)
(0.003)
(0.003)
(0.003)
(0.003)
(0.003)
(0.003)
UASi,t�
3⇥UNI i=
10.025⇤
⇤⇤0.007
(0.005)
(0.006)
UASi,t�
3⇥M
axPlanck
i=
10.049⇤
⇤⇤0.044⇤
⇤⇤
(0.006)
(0.007)
UASi,t�
3⇥Leibn
izi=
10.015⇤
-0.013
(0.008)
(0.009)
UASi,t�
3⇥Helmholtz
i=
10.024⇤
⇤⇤-0.001
(0.009)
(0.010)
UASi,t�
3⇥Fraunhof
eri=
10.036⇤
⇤⇤0.021⇤
⇤⇤
(0.007)
(0.008)
Municipalities
11,266
11,266
11,266
11,266
11,266
11,266
11,266
N342,507
342,507
342,507
342,507
342,507
342,507
342,507
R2(w
ithin)
0.041
0.041
0.041
0.041
0.041
0.041
0.041
R2(between)
0.000
0.000
0.009
0.000
0.000
0.001
0.012
R2(overall)
0.029
0.030
0.035
0.029
0.030
0.031
0.036
Sou
rce:
Authors’
calculation
sbased
onpatentdatafrom
theEPOs,
dataon
HEIs
from
theGerman
Rectors’Con
ference,dataon
PROsfrom
BMBF(2018b
),an
dself-collected
dataon
HEIs
andPROs.
Notes:Rob
ust
stan
darderrors
inparentheses.Allmod
elsincludeintercep
t,municipalityFE,year
FE,an
dcontrolsforUNI i
,t�3,M
axPlanck
i,t�
3,Leibn
izi,t�
3,Helmholtz
i,t�
3,an
dFra
unhof
eri,t�
3.⇤p<
0.10,⇤⇤p<
0.05,⇤⇤
⇤p<
0.01.
45
Appendix H. Results for Additional Patent Quality Measures
Tab
leH1:
FEestimationresultson
patentqu
ality(priorityclaims)
(25-km
travel-distance
radius,3-y.
lag)
Dep.var.:
ln(P
QUALclaim
s+1)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
UASi,t�
30.012⇤
⇤⇤0.019⇤
⇤⇤0.015⇤
⇤⇤0.016⇤
⇤⇤0.014⇤
⇤⇤0.015⇤
⇤⇤0.021⇤
⇤⇤
(0.003)
(0.004)
(0.004)
(0.004)
(0.004)
(0.004)
(0.004)
UASi,t�
3⇥UNI i=
1-0.020
⇤⇤⇤
-0.017
⇤⇤
(0.007)
(0.007)
UASi,t�
3⇥M
axPlanck
i=
1-0.017
⇤⇤0.002
(0.008)
(0.009)
UASi,t�
3⇥Leibn
izi=
1-0.052
⇤⇤⇤
-0.049
⇤⇤⇤
(0.010)
(0.011)
UASi,t�
3⇥Helmholtz
i=
1-0.020
⇤-0.002
(0.010)
(0.011)
UASi,t�
3⇥Fraunhof
eri=
1-0.014
⇤-0.000
(0.008)
(0.009)
Municipalities
11,266
11,266
11,266
11,266
11,266
11,266
11,266
N308,087
308,087
308,087
308,087
308,087
308,087
308,087
R2(w
ithin)
0.041
0.041
0.041
0.041
0.041
0.041
0.041
R2(between)
0.046
0.038
0.034
0.039
0.042
0.040
0.034
R2(overall)
0.039
0.037
0.036
0.038
0.038
0.038
0.036
Sou
rce:
Authors’
calculation
sbased
onpatentdatafrom
theEPOs,
dataon
surfacegrou
psfrom
Leh
nertet
al.(202
0),dataon
HEIs
from
the
German
Rectors’Con
ference,dataon
PROsfrom
BMBF(2018b
),an
dself-collected
dataon
HEIs
andPROs.
Notes:Rob
ust
stan
darderrors
inparentheses.Allmod
elsincludeintercept,
municipalityFE,year
FE,surfacegrou
ps,
andcontrols
forUNI i
,t�3,
MaxPlanck
i,t�
3,Leibn
izi,t�
3,Helmholtz
i,t�
3,an
dFra
unhof
eri,t�
3.⇤p<
0.10,⇤⇤p<
0.05,⇤⇤
⇤p<
0.01.
46
Tab
leH2:
FEestimationresultson
patentqu
ality(fam
ilysize)(25-km
travel-distance
radius,3-y.
lag)
Dep.var.:
ln(P
QUALsize+1)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
UASi,t�
30.043⇤
⇤⇤0.044⇤
⇤⇤0.040⇤
⇤⇤0.048⇤
⇤⇤0.043⇤
⇤⇤0.038⇤
⇤⇤0.042⇤
⇤⇤
(0.008)
(0.010)
(0.009)
(0.009)
(0.009)
(0.009)
(0.010)
UASi,t�
3⇥UNI i=
1-0.002
-0.019
(0.015)
(0.017)
UASi,t�
3⇥M
axPlanck
i=
10.017
0.031
(0.018)
(0.022)
UASi,t�
3⇥Leibn
izi=
1-0.059
⇤⇤-0.084
⇤⇤⇤
(0.024)
(0.027)
UASi,t�
3⇥Helmholtz
i=
10.007
0.004
(0.023)
(0.026)
UASi,t�
3⇥Fraunhof
eri=
10.032⇤
0.042⇤
(0.019)
(0.022)
Municipalities
11,266
11,266
11,266
11,266
11,266
11,266
11,266
N308,087
308,087
308,087
308,087
308,087
308,087
308,087
R2(w
ithin)
0.066
0.066
0.066
0.066
0.066
0.066
0.066
R2(between)
0.096
0.096
0.101
0.094
0.096
0.100
0.103
R2(overall)
0.066
0.066
0.068
0.066
0.066
0.068
0.069
Sou
rce:
Authors’
calculation
sbased
onpatentdatafrom
theEPOs,
dataon
surfacegrou
psfrom
Leh
nertet
al.(202
0),dataon
HEIs
from
the
German
Rectors’Con
ference,dataon
PROsfrom
BMBF(2018b
),an
dself-collected
dataon
HEIs
andPROs.
Notes:Rob
ust
stan
darderrors
inparentheses.Allmod
elsincludeintercept,
municipalityFE,year
FE,surfacegrou
ps,
andcontrols
forUNI i
,t�3,
MaxPlanck
i,t�
3,Leibn
izi,t�
3,Helmholtz
i,t�
3,an
dFra
unhof
eri,t�
3.⇤p<
0.10,⇤⇤p<
0.05,⇤⇤
⇤p<
0.01.
47