Do competitively acquired funds induce universities to increase productivity?

Research Policy 40 (2011) 136–147

Contents lists available at ScienceDirect

Research Policy journal homepage: www.elsevier.com/locate/respol

Do competitively acquired funds induce universities to increase productivity? Thomas Bolli, Frank Somogyi ∗ ETH Zürich, KOF Swiss Economic Institute, Weinbergstraße 35, 8092 Zürich, Switzerland

a r t i c l e

i n f o

Article history: Available online 26 November 2010 Keywords: Productivity Research University Technology transfer Third-party funding Endogeneity

a b s t r a c t This paper analyzes the impact of private and public third-party funds on the productivity of Swiss university departments and public research institutions. Estimating a production function assuming that labor inputs produce master students and scientific publications reveals a positive effect of public third-party funding on productivity but not for private funds. However, once we include technology transfer as an additional output, the coefficient for public third-party funding turns insignificant while private funding becomes significant, indicating that the disciplining effect of public donors focuses on publications while private donors foster technology transfer. We employ three alternative approaches to tackle endogeneity and find qualitatively robust results. © 2010 Elsevier B.V. All rights reserved.

1. Introduction Due to budgetary pressure and the perception of universities and public research institutions as a central piece of the national innovation system, political pressure to use university funding productively has increased in recent years. Consequently, the use of quasi-market instruments, e.g. competitive funding distribution has improved prominence (see e.g. de Boer and File, 2009). However, there exists little evidence about the impact of third-party funding on productivity. Furthermore, both the theoretical and empirical findings are ambiguous, indicating that further research in this area is required to allow policymakers to make evidence-based decisions (van der Ploeg and Veugelers, 2008). We use a survey among natural science, mathematics, physics, medicine, and economics departments of Swiss universities and public research institutions (henceforth denoted university departments) in 2005 to estimate an output-oriented multi-output production function that uses master degrees,1 scientific publications and the intensity of technology transfer as outputs. Including the share of private and public third-party funding in the regression reveals the impact of these funding sources on productivity.

∗ Corresponding author. Tel.: +41 44 632 42 38; fax: +41 44 632 10 42. E-mail addresses: [email protected] (T. Bolli), [email protected], [email protected] (F. Somogyi). 1 In 2005, Swiss Universities awarded all students with a degree (licentiat) equivalent to a M.Sc. in the Anglo-saxon system. Bachelor’s degrees did not exist at the time and are hence not included in this study. 0048-7333/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.respol.2010.10.001

We add to the existing literature in a number of ways. First, we include a measure for technology transfer in our production function. Since we further differentiate between public and private third-party funding, our approach identifies differences between the two funding sources across the innovation stages, i.e. whether private and public third-party funding have a different impact on basic and applied research. Second, we address the problem of endogeneity between the ability to acquire third-party funding and productivity using three approaches, namely analyzing the impact of third-party funding for universities where reverse causality is not present separately, exploiting the variance within universities to account for unobserved heterogeneity, and implement an IV estimation. Third, using data at the department level ensures the proximity to the decision making unit while comprising the whole university sector as opposed to a single university. And fourth, we present the first estimates for the relationship between the funding structure and productivity in Switzerland. We find that public third-party funds exert a positive effect on productivity in production functions that exclude technology transfer, while private funds have no significant influence. However, including technology transfer, the universities’ “third mission”, renders private third-party funds positive significant, while public funds have no impact anymore. Hence, controlling for technology transfer in such an estimation setting is relevant. We further analyze the output dimensions individually, which reveals that public third-party funds increase publication productivity but have no effect on technology transfer. Private funding on the other hand increases both publication and technology transfer productivity. Teaching productivity is not affected by either funding source.

T. Bolli, F. Somogyi / Research Policy 40 (2011) 136–147 Table 1 Summary of channels of third-party funding on productivity.

Table 2 Population, data sample and response rates by department.

Direction of adm., mis. and disc. effects on productivity

Effect of monitoring intensity on adm., mis. and disc. effects

Institutions

∂productivity ∂adm.eff.

<0

∂adm.eff. ∂monit.

∂productivity ∂mis.eff.

<0

∂mis.eff. ∂monit.

<0

∂productivity ∂disc.eff.

>0

∂disc.eff. ∂monit.

<0

ETH-Domain Swiss Federal Inst. of Technology Zurich Swiss Federal Inst. of Technology Lausanne Federal Research Institutes

>0

The paper is structured as follows: Section 2 discusses the theoretical framework and section 3 summarizes the existing empirical evidence. Sections 4 and 5 describe the applied methodology and the data. We discuss the estimation results in section 6. Section 7 concludes the paper. 2. Theoretical framework This section provides the theoretical background of the relationship between competitive funding acquisition and productivity in a principal-agent framework (see e.g Kivistö, 2005). The donor is the principal and the researcher is the agent. If the two have different utility functions, the principal faces a trade-off between monitoring the agent and accepting diverging outcomes caused by information asymmetry and uncertainty. Hence, third-party funding and the corresponding monitoring influence productivity through three main channels: The administration effect and the misallocation effect decrease productivity, while the discipline effect increases it. The first channel concerns the administrative work that comes along with third-party funding. The acquisition of external funds and the enclosure of details about the work requires the investment of time and money. We call this channel the administration effect. The second channel, which we call the misallocation effect, arises due to the difference in the utility functions. Therefore, the principal has an interest to control the agent’s behavior by restricting the use of funds. This may cause behavioral distortions and a suboptimal outcome, as these restrictions are based on incomplete information of the principal and will therefore lead to misallocated resources (see e.g. Schiller and Liefner, 2006). The third channel, which we call the discipline effect, causes research productivity to increase. This channel is based on the same information asymmetry as the misallocation effect. However, its effect on productivity is positive, as restrictions on the use of funds also limit the possibilities of the researcher to pursue his own goals and to use funds in an inefficient way (see e.g. Niskanen, 1975). These effects further depend on the monitoring intensity of the donor, which mirrors the autonomy of universities. The administration effect increases with monitoring intensity, i.e. with the amount of time the researcher devotes to reporting and justifying his behavior. The misallocation and discipline effects on the other hand decrease if monitoring intensity increases. As the researcher reports his behavior more accurately, the information asymmetry diminishes. Therefore resources can be allocated in a more optimal way, and the researcher’s opportunities to pursue his goals decrease. This implies that the administration effect, which lowers research productivity, is increasing in monitoring intensity, while the misallocation effect, which also lowers research productivity, and the discipline effect, which increases research productivity, both decrease the more intense monitoring is conducted. As shown in Table 1, the three channels through which thirdparty funding affects productivity have opposite signs and depend on monitoring intensity. Therefore theoretical predictions about the direction of the net effect are ambiguous and remain an empirical issue. A further aspect of the principal-agent relationship is that monitoring intensity of private donors is expected to increase in the

137

Population

Sample

Response rate (%)

87

45

51.7

31

12

38.7

11

11

100.0

Cantonal University of Basle Berne Fribourg Geneva Italian Switzerland Lausanne Neuchâtel St. Gallen Zurich

32 84 17 46 9 69 22 21 74

11 33 5 15 2 12 6 8 22

34.4 39.3 29.4 32.6 22.2 17.4 27.3 38.1 29.7

University of Applied Sciences of Berne Central Switzerland Eastern Switzerland Italian Switzerland Northwestern Switzerland Western Switzerland Zurich

13 10 36 7 27 12 22

9 5 14 2 17 4 8

69.2 50.0 38.9 28.6 63.0 33.3 36.4

630

241

38.3

Total

innovation stage due to two reasons. First, the relevance of uncertainty and hence of creative control decreases in the innovation stage, i.e. the misallocation effect is stronger for basic research than for applied research (Aghion et al., 2008b). Second, the absorptive capacity of firms is geared towards applied research, implying that monitoring costs of the firm decrease in the innovation stage. Therefore we expect the monitoring intensity and consequently the impact of private third-party funding on productivity to increase in the innovation stage. It is even possible that monitoring is not profitable for early stages of the innovation process, in which case no monitoring by private donors takes place at all. Presumably, the absorptive capacity of public donors is better for basic research and worse for applied research. Therefore the expected degree of monitoring of public donors increases less with the innovation stage than the monitoring intensity of private donors. 3. Literature review A number of recent papers study the effect of third-party funding on research productivity and the different channels described in the previous section. Many of these papers are outcomes of the discussion on New Public Management (NPM) (see e.g. Leisyte and Kizniene, 2006; de Boer et al., 2007 and the references cited therein). Schubert (2009) finds a significant relationship between the introduction of NPM government schemes and productivity. Concretely, the negative impact of accounting schemes indicates the relevance of the administration effect while the positive sign of introducing goal agreements and evaluations suggests the presence of a discipline effect. In line with the NPM literature, Aghion et al. (2008a) present macroeconomic evidence for the misallocation effect by showing that autonomy and productivity are positively related. Similarly, Aghion et al. (2009) find a significant positive impact of autonomy on the Shanghai university ranking (see Liu and Cheng, 2005). They further show that competition for public funding increases the Shanghai university ranking of the research institution, thereby providing evidence for the discipline effect. Kempkes and Pohl (2008) find evidence for the misallocation effect in their exam-

138

T. Bolli, F. Somogyi / Research Policy 40 (2011) 136–147

Table 3 Summary statistics of variables. Variable

Variable description

Obs

Mean

S.D.

Min

Output Publ * Mas ** TTinformal ** TTfacility ** TTtraining ** TTresearch ** TTconsultancy ** TTtotal 1 ** TTtotal 2 **

Scientific publications Number of master degrees Informal technology transfer Facility technology transfer Training technology transfer Research technology transfer Consultancy technology transfer Unweighted technology transfer Form intensity weighted technology transfer

187 187 187 187 187 187 187 187 187

82.88 47.40 2.89 1.82 2.59 2.98 2.82 13.10 38.70

119.34 84.14 1.01 0.96 0.95 1.23 1.23 4.26 13.18

Input Acad * Other * Budget per Emp *

Full-time academic staff Full-time administrative staff Budget per employee

187 187 187

3995.22 2801.46 11.40

6530.26 7741.11 0.88

Scientific field Engineering Nat Sci Med Econ/Bus Math and Phy

Engineering (dummy) Natural sciences (dummy) Medicine (dummy) Economics and business (dummy) Mathematics and physics (dummy)

187 187 187 187 187

0.35 0.14 0.23 0.18 0.10

0.48 0.35 0.42 0.39 0.30

0 0 0 0 0

1 1 1 1 1

0 0 1 1 1 1 1 5 15 150.10 0.10 7.32

Max 999 600 5 5 4.56 5 5 22.56 68.89 50000.10 75000.10 12.84

University types Fed Inst Tech Fed Res Inst UAS University Budget shares Budgetsha Pub * Budgetsha Priv *

Federal Institute of Technology (dummy) Federal Research Institute (dummy) University of Applied Sciences (dummy) Cantonal University (dummy)

187 187 187 187

0.26 0.05 0.24 0.44

0.44 0.23 0.43 0.50

0 0 0 0

1 1 1 1

Public third-party funds/total funds in % Private third-party funds/total funds in %

187 187

24.37 18.01

19.63 21.85

0 0

100 93.12

Instruments Winner Instrument 1 Instrument 2

Dummy indicating whether TT has increased funding Financial motive for TT: basic research funds Financial motive for TT: commercialization

161 164 164

0.81 3.00 2.60

0.39 1.44 1.34

0 1 1

1.00 5.00 5.00

Work shares Wsha Teach Wsha BRes Wsha ARes

Share of work devoted to teaching Share of work devoted to basic research Share of work devoted to applied research

180 180 180

24.58 23.82 33.15

17.47 22.44 23.43

0 0 0

90 90 90

* **

Enters estimation in logs, normalized by the median. Enters estimation in logs, normalized by the median and Publ.

ination of German universities by showing that independence increases university efficiency. Mensah and Werner (2003) as well as Kuo and Ho (2007) study funding restrictions directly. Mensah and Werner (2003) evaluate the impact of financial autonomy on the efficiency of universities, using the share of unrestricted assets as a measure for financial flexibility. Applying a stochastic frontier methodology, they find a positive correlation between restriction and efficiency. This finding is supported by Kuo and Ho (2007) who examine a change in the budget regime in Taiwan that has led to more flexibility of the utilization of private funds. They find a negative impact of the policy adoption on the efficiency of universities. This negative correlation between (financial) autonomy and efficiency can be interpreted as evidence for the misallocation effect. Butler (2003) finds that introducing a competitive funding distribution scheme based on output counts has increased the share of Australia’s ISI publications despite declining resources, indicating the presence of a disciplining effect. However, she further presents evidence of the misallocation effect, as the quality of publications measured by the share of citations has stagnated within the period. Agasisti (2009) analyzes the relationship between competition and tertiary education in Italy, finding a positive effect on productivity, and Abbott and Doucouliagos (2009) find that competition for overseas students increases university efficiency in Australia. Cherchye and Abeele (2005) find a positive correlation between the share of scientific research grants and efficiency, but a negative effect of contract research funds. Bonaccorsi et al. (2006) analyze

the impact of private funding on efficiency and find an inverse Ushape for Italian universities. Using data on individual researchers at Louis Pasteur University, Carayol and Matt (2006) find a small effect of public third-party funds but none for private funds. Jansen et al. (2007) and Schmoch and Schubert (2009) present evidence on the relation between third-party funding and research productivity in Germany. Both papers find an inverted U-shaped relation between third-party funding and the total number of academic publications in a research group. They further show considerable heterogeneity across scientific fields. Robst (2001) shows that the share of tuition has no effect on productivity. Hence, the existing evidence indicates a positive net effect of public third-party funding, while the results are ambiguous for private funding. However, with the exception of Aghion et al. (2009) who use a political instrument to evaluate increases in public third-party funding, these articles do not address the problem of endogeneity. 4. Data In Switzerland four types of public research institutes exist: cantonal universities, federal institutes of technology, universities of applied sciences and federal research institutions. Of the 10 cantonal universities, only those in Lucerne (only social sciences and theology), Lugano (only social sciences and architecture) and St. Gallen (only economics, law and management) limit the range of covered disciplines, while the others offer a broad spectrum. The

T. Bolli, F. Somogyi / Research Policy 40 (2011) 136–147

139

Table 4 Cross-correlations among the variables. Publ Publ Mas TTinformal TTfacility TTtraining TTresearch TTconsultancy TTTotal 1 TTTotal 2 Acad Other Budget per Emp Budgetsha Pub Budgetsha Priv Instrument 1 Instrument 2

1 0.0192 −0.0346 0.2071 * −0.0058 0.0688 0.0518 0.0718 0.1033 0.7659 * 0.6462 * 0.0727 −0.0438 −0.1686 * 0.1658 * −0.057 TTTotal 1

TTTotal 1 TTTotal 2 Acad Other Budget per Emp Budgetsha Pub Budgetsha Priv Instrument 1 Instrument 2

Instrument 1 Instrument 2 *

1 0.9877 * 0.1610 * 0.1867 * 0.0959 −0.1392 0.3214 * 0.3376 * 0.3863 *

Mas 1 0.1294 −0.0283 0.1739 * 0.0823 0.2568 * 0.1605 * 0.1368 0.1927 * 0.0075 0.1628 * −0.1935 * 0.0751 −0.0663 0.0661 TTTotal 2 1 0.1768 * 0.2028 * 0.0741 −0.0941 0.3117 * 0.3461 * 0.3597 *

TTinformal

TTfacility

1 0.4507 * 0.6682 * 0.6659 * 0.5308 * 0.8307 * 0.7630 * 0.0934 0.1196 0.1281 −0.1971 * 0.2217 * 0.1858 * 0.2779 *

1 0.4556 * 0.4489 * 0.3272 * 0.6559 * 0.6916 * 0.2232 * 0.2672 * −0.0865 −0.0007 −0.0126 0.2672 * 0.1505

Acad

Other

1 0.7788 * 0.0634 −0.1003 −0.1103 0.0906 0.025

1 0.1067 −0.2097 * −0.1265 0.0395 0.0057

Instrument 1

Instrument 2

1 0.0064

1

TTtraining

TTresearch

TTconsultancy

1 0.6736 * 0.6268 * 0.8570 * 0.8170 * 0.0838 0.0991 0.1633 * −0.1374 0.3547 * 0.2045 * 0.3748 *

1 0.5112 * 0.8438 * 0.8887 * 0.1318 0.1471 * 0.0992 0.0057 0.3206 * 0.2598 * 0.2890 *

1 0.7734 * 0.7439 * 0.1115 0.1182 0.0702 −0.2214 * 0.3492 * 0.2238 * 0.2553 *

Budget per Emp

Budgetsha Pub

Budgetsha Priv

1 −0.3119 * −0.0406 −0.1016 0.1031

1 −0.1105 −0.0312 0.027

1 0.0952 0.2525 *

Significant correlation among variables at the 5% level.

federal institutes of technology in Zurich and Lausanne focus on engineering, natural sciences, mathematics and physics. The universities of applied sciences have been pure teaching institutions until the mid nineties. Their mandate was broadened to include applied research in the disciplines they cover: engineering, management, social work, pedagogy, health professions and fine arts. Besides these three types of higher education institutions, there are four governmental research institutions: the Swiss Federal Institute of Aquatic Science and Technology (EAWAG), the Research Institute for Material Sciences and Technology (EMPA), the Swiss Federal Institute for Forest, Snow and Landscape Research (WSL) and the Paul Scherrer Institute (PSI) which conducts research on energy technologies and elementary particles physics. In 2005, KOF Swiss Economic Institute conducted a survey among Swiss universities and public research institutions. The questionnaire was sent to the directors of the research institutions and the heads of university departments.2 The sample covers only fields related to technology and science: engineering, natural sciences, mathematics, physics, medicine, economics and business administration. 241 of the 630 questionnaires were returned, implying a response rate of 38.3%. The response rate varies substantially between the types of institutions (see Table 2 for more details). As a consequence, the cantonal universities are underrepresented, while federal institutes and universities of applied sciences are overrepresented.3 The data entails the number of master degrees measuring teaching output. Research output is quantified as the number of papers published in scientific journals. Since these reflect the realized, or

2 The questionnaire is available in German, French and English http://www.kof.ethz.ch/surveys/structural/panel/wissensaustausch 2005. 3 For more information concerning the data, see Arvanitis et al. (2008).

at

ex post output, our estimates implicitly assume no uncertainty on the output side (Cherchye and Abeele, 2005). The survey further contains questions capturing the relevance of 19 technology transfer channels. The respondents attribute a value between one and five to each channel, where one refers to ‘not important’ and five to ‘extremely important’. The channels are divided into five channel groups: ‘Informal contacts’, ‘Technical facilities’, ‘Training’, ‘Research collaboration’ and ‘Consulting’. The number of questions within each channel group ranges from 2 to 9. If the answers to the questions in the different channel groups were simply averaged to arrive at a single number measuring technology transfer, the different number of questions in each channel group would imply different weights across groups. Therefore we construct our measure for technology transfer as the sum of the average relevance in each group. In order to account for the informality character of each channel group, we follow Arvanitis et al. (2007) and assign weights from one to five to the channel groups ‘Informal contacts’, ‘Training’, ‘Consulting’, ‘Technical facilities’ to ‘Research collaboration’. As the choice of weights is somewhat arbitrary, we also estimate a version of our model that uses no weighting scheme. Concretely, the robustness check that includes all five channel groups separately allows the weights to be datadriven. Labor input is available by the categories ‘Professor’, ‘Post doc’, ‘Graduate’, ‘Technical Staff’ and ‘Administrative Staff’. Due to multicollinearity problems, we aggregate professors, postdocs and graduates to ‘Academic Staff’ and technical staff and administrative staff to ‘Administrative Staff’ (Filippini and Lepori, 2007, report similar problems). Furthermore, the data entails information about the share of labor devoted to the activities teaching, basic research and applied research. We combine the information about labor inputs and labor activity shares to construct activity adjusted labor inputs.

140

T. Bolli, F. Somogyi / Research Policy 40 (2011) 136–147

Thereby, we assume that labor activity shares are the same across input classes, i.e. that all labor input categories devote the same share of time for each activity. Finally, the data includes information about financial resources of each department. Besides the overall budgets, the respondents disclose the share of third-party funds from governmental sources and from the business sector. We use this information to construct the two variables at the heart of this analysis. We denote third-party funds from the business sector divided by total funds as ‘Bsha Priv and for third-party funds from governmental sources ‘Bsha Pub . The latter mainly contains funds from the two Swiss research promotion agencies ‘Swiss National Science Foundation’ (SNSF) and ‘Innovation Promotion Agency’ (CTI). The reference funding category consists of the global budget of universities from various sources. Summary statistics and cross-correlations of the variables are presented in Tables 3 and 4. We add 0.1 to all observations to eliminate observations taking the value 0, thereby allowing universities to specialize in corner solutions despite the logarithmic production function. 187 of the total 241 observations have no missing values in the utilized variables.

Table 5 Main results. Dep Var: Publ

1

2

3

4

Mas

−0.330*** (0.038) −0.013 (0.011)

−0.348*** (0.034) −0.019 (0.011)

0.520*** (0.062) 0.043 (0.031) 0.177*** (0.027) 0.030*** (0.008) −0.079*** (0.027) −0.023 (0.026) 0.031* (0.015)

0.515*** (0.063) 0.037 (0.028) 0.190*** (0.027) 0.031*** (0.008) −0.083*** (0.026) −0.019 (0.029) 0.031** (0.014)

0.229*** (0.046) 0.083 (0.107) −0.254** (0.100) 0.241* (0.121) 0.278** (0.130) −0.769*** (0.212) −0.160 (0.125) 0.005 (0.109) 0.149*** (0.029)

−0.081*** (0.022) −0.004 (0.011) −0.738*** (0.039) −0.044** (0.016) 0.010 (0.025) 0.173*** (0.034) −0.010 (0.024) 0.088*** (0.030) 0.012** (0.004) −0.023 (0.021) −0.002 (0.026) −0.001 (0.014) −0.051 (0.034) 0.003 (0.021) 0.051 (0.034) −0.108 (0.080) −0.258** (0.095) 0.010 (0.056) −0.249*** (0.077) −0.097 (0.123) −0.034 (0.066) −0.137* (0.067) 0.032 (0.024)

−0.522 (0.400)

−0.085*** (0.022) −0.007 (0.011) −0.733*** (0.041) −0.046** (0.016) 0.013 (0.026) 0.170*** (0.030) −0.010 (0.023) 0.092*** (0.028) 0.013*** (0.004) −0.024 (0.019) −0.001 (0.026) −0.001 (0.014) −0.046 (0.033) 0.002 (0.020) 0.050 (0.035) −0.129 (0.082) −0.258** (0.094) 0.024 (0.054) −0.231*** (0.079) −0.092 (0.125) −0.017 (0.062) −0.119* (0.067) 0.016 (0.032) −0.006 (0.008) 0.055** (0.022) −0.016 (0.013) −0.473 (0.419)

187 0.961

187 0.962

Mas squ TTTotal TTTotal squ Mas TTTotal Acad Acad squ Other Other squ Acad Other Mas Acad Mas Other

5. Methodology TTTotal Acad

We estimate an output-oriented multi-output production function in translog form using OLS, assuming that labor produces scientific publications and master students and compare the results to a model that introduces technology transfer as a third output. Including budget shares in the production function yields estimates of their impact on productivity. We address the issue of endogeneity between the ability to acquire third-party funding and productivity brought forth in the literature by three approaches. The first approach is based on a control question in the survey capturing the presence of reverse causality, the second accounts for unobserved heterogeneity by including university fixed effects and the third approach consists of instrumenting third-party funding shares. All three approaches are explained in detail below. Since price data is difficult to obtain, prices rarely reflect free market forces and the cost minimization assumption appears questionable for universities, we follow Coelli and Perelman (2000) and Coelli (2000) and estimate an output-oriented multi-output production function in translog form4 : lny1,i =

M 

˛0 +

∗ ˛m lnym,i

 1  ∗ ∗ + ˛m,n lnym,i ∗ lnyn,i + ˇk lnxk,i 2 K

m=2 n=2 K K

k=1 K

 1  ∗ ˇk,l lnxk,i ∗ lnxl,i + ˇm,k lnym,i ∗ lnxk,i (1) 2 k=1 l=1

M

 

  2

m=2 k=1

+1 ∗ Bsha Pubi + 2 ∗ Bsha Pub2i +ı1 ∗ Bsha Privi + ı2 ∗ Bsha Privi C 

+

Budget per Emp

Constant

−2.520*** (0.493)

0.202*** (0.046) 0.026 (0.098) −0.230** (0.096) 0.272** (0.122) 0.265* (0.135) −0.664*** (0.215) −0.123 (0.105) 0.021 (0.108) −0.003 (0.061) −0.071** (0.026) 0.014 (0.041) −0.029 (0.021) −2.116*** (0.514)

N R2

187 0.846

187 0.856

Natural Sciences Medicine Economics and Business Mathematics and Physics UAS Cantonal Universities Fed Res Inst Budgetshare Pub Budgetshare Pub squ Budgetshare Priv

0.053 (0.036)

Budgetshare Priv squ

m=2 M M

+

TTTotal Other

c controlci + εi

0.069*** (0.020)

OLS estimates of production functions assuming a translog form. Table 3 provides variable definitions. Budgetshare Pub and Priv show the effect of funding shares on university productivity. The table shows coefficients and robust standard errors, which are clustered at the university level. *, ** and *** denote significance levels 10%, 5% and 1%. Columns 1 and 2: No technology transfer (TT) measure; Columns 3 and 4: Informality weighted TT measure Column 1 and 3: Linear budget shares; Columns 2 and 4: Quadratic budget shares.

c=1

The dimension i refers to the analyzed department. The dependent variable, the number of scientific publications, captures the

4

MThis m=1

M

M

approach assumes linear homogeneity of degree +1 in outputs, i.e.

˛m = 1,

n=1

˛m = 0 for m = 1, 2, . . ., M and

m=1

ˇm,k = 0 for k = 1, 2, . . .,

K, as well as symmetry, i.e. ˛m,n = ˛n,m for m, n = 1, 2, . . ., M and ˇk,l = ˇl,k for k, l = 1, 2, . . ., K (see e.g. Coelli and Perelman, 2000).

amount of basic research and serves as the normalizing output. The remaining outputs, y, appear as explanatory variables normalized ∗ = y /y . These consist by scientific publications, meaning that ym m 1 of the number of master degrees (Mas) approximating teaching output and technology transfer intensity (TT), measuring applied research. The input vector, x, contains the amount of labor input differentiated by ‘Academics’ (Acad), and ‘Technical and Administrative Staff’ (Other).

Table 6 Simultaneous production function estimation for each output. Dep Var: Output

Acad Teach Admin Teach Acad Teach squ Admin Teach squ Acad Other Teach

Natural Science Medicine Econ and Business Math and Physics UAS University Fed Res. Inst.

(3a)

(4a)

0.591*** (0.094) −0.051 (0.078) 0.110*** (0.040) −0.009 (0.008) −0.077* (0.043) 0.165 (0.101)

0.568*** (0.094) −0.035 (0.077) 0.109*** (0.039) −0.008 (0.008) −0.077* (0.043) 0.127 (0.100)

0.591*** (0.094) −0.052 (0.078) 0.111*** (0.040) −0.009 (0.008) −0.078* (0.043) 0.166 (0.101)

0.568*** (0.094) −0.039 (0.077) 0.112*** (0.039) −0.008 (0.008) −0.080* (0.043) 0.128 (0.100)

−0.329 (0.294) −1.236*** (0.316) 0.701*** (0.252) −0.555* (0.336)

−0.348 (0.291) −1.142*** (0.313) 0.699*** (0.249) −0.545 (0.333)

−0.328 (0.294) −1.235*** (0.316) 0.700*** (0.252) −0.555* (0.336)

−0.346 (0.291) −1.140*** (0.313) 0.697*** (0.249) −0.545 (0.333)

Natural Science

0.787*** (0.251) 0.066 (0.253) 0.205 (0.404)

0.792*** (0.247) 0.065 (0.250) 0.162 (0.397)

0.786*** (0.251) 0.066 (0.253) 0.206 (0.404)

0.792*** (0.247) 0.067 (0.250) 0.164 (0.397)

UAS

0.066 (0.076)

0.066 (0.076)

−1.990* (1.154)

−0.155 (0.112) −0.105** (0.041) 0.001 (0.070) −0.025 (0.041) −1.457 (1.152)

180 0.51

180 0.52

Constant

−1.985* (1.154)

−0.154 (0.112) −0.105** (0.041) 0.002 (0.070) −0.025 (0.041) −1.445 (1.152)

N R2

180 0.51

180 0.52

Budgetsha Pub Budgetsha Pub squ Budgetsha Priv

Publications

(2a)

0.043 (0.063)

Budgetsha Priv squ

0.043 (0.063)

Acad BRes Admin BRes Acad BRes squ Admin BRes squ Acad Other BRes Budget per Emp

Medicine Econ and Business Math and Physics

University Fed Res. Inst. Budgetsha Pub

Technology transfer

(1b)

(2b)

(3b)

(4b)

0.244*** (0.078) 0.209*** (0.061) 0.057 (0.040) 0.028*** (0.007) −0.054 (0.042) 0.302*** (0.075)

0.242*** (0.080) 0.216*** (0.061) 0.062 (0.040) 0.029*** (0.007) −0.060 (0.043) 0.289*** (0.076)

0.244*** (0.078) 0.208*** (0.061) 0.058 (0.040) 0.028*** (0.007) −0.055 (0.042) 0.303*** (0.075)

0.241*** (0.079) 0.213*** (0.061) 0.063 (0.040) 0.028*** (0.007) −0.061 (0.043) 0.290*** (0.076)

0.265 (0.216) 0.555** (0.238) −0.229 (0.182) 0.587** (0.260)

0.260 (0.216) 0.585** (0.238) −0.237 (0.182) 0.572** (0.263)

0.265 (0.216) 0.556** (0.238) −0.230 (0.182) 0.588** (0.260)

0.260 (0.216) 0.587** (0.238) −0.238 (0.182) 0.573** (0.263)

−1.651*** (0.241) −0.399** (0.186) 0.095 (0.301)

−1.617*** (0.244) −0.403** (0.187) 0.084 (0.300)

−1.652*** (0.241) −0.399** (0.186) 0.098 (0.301)

−1.617*** (0.244) −0.402** (0.187) 0.089 (0.300)

UAS

0.116** (0.054)

0.116** (0.054)

Budgetsha Pub

−3.336*** (0.858)

0.043 (0.082) −0.037 (0.031) 0.128** (0.052) 0.000 (0.031) −3.166*** (0.870)

180 0.72

180 0.72

Constant

−3.330*** (0.858)

0.043 (0.082) −0.036 (0.031) 0.128** (0.052) 0.001 (0.031) −3.159*** (0.870)

N R2

180 0.72

180 0.72

Budgetsha Pub squ Budgetsha Priv

0.135*** (0.046)

Budgetsha Priv squ

0.135*** (0.046)

Acad ARes Admin ARes Acad ARes squ Admin ARes squ Acad Other ARes Budget per Emp Natural Science Medicine Econ and Business Math and Physics

University Fed Res. Inst.

(3c)

(4c)

0.046 (0.029) 0.035 (0.024) 0.022** (0.010) 0.005* (0.003) −0.025** (0.012) 0.014 (0.031)

0.045 (0.029) 0.032 (0.024) 0.021** (0.010) 0.005 (0.003) −0.022* (0.012) 0.021 (0.031)

−0.068 (0.091) −0.077 (0.096) −0.052 (0.077) −0.295*** (0.104)

−0.084 (0.091) −0.096 (0.096) −0.043 (0.077) −0.279*** (0.104)

0.019 (0.075) 0.035 (0.077) −0.208 (0.128)

0.011 (0.075) 0.049 (0.077) −0.185 (0.127)

−0.013 (0.022)

Constant

−0.203 (0.354)

0.023 (0.034) 0.020 (0.013) 0.090*** (0.022) −0.012 (0.012) −0.261 (0.356)

N R2

180 0.42

180 0.43

Budgetsha Pub squ Budgetsha Priv

0.096*** (0.019)

Budgetsha Priv squ

T. Bolli, F. Somogyi / Research Policy 40 (2011) 136–147

Budget per Emp

Master degrees (1a)

Columns 1a-b/2a-b: Regressions excluding Technology Transfer. Columns 3a-c/4a-c: Regressions including technology transfer. Colums 1a-b/3a-c: Linear terms of budget shares. Columns 2a-b/4a-c: Linear and quadratic terms of budget shares. ***, ** and * indicate significance at the 1%, 5% and 10% levels. Standard errors in parentheses.

141

142

T. Bolli, F. Somogyi / Research Policy 40 (2011) 136–147 Table 8 IV methodology.

Table 7 Reverse causality and unobserved heterogeneity. Dep Var: Publ

1

2

3

4

Dep Var: Publ

IV 1

IV 2

First 1

First 2

Mas

−0.299*** (0.044) −0.017 (0.012)

−0.021 (0.019) −0.002 (0.010) −0.895*** (0.040) −0.009 (0.022) −0.005 (0.023) 0.105*** (0.033) 0.000 (0.018) 0.025 (0.036) 0.004 (0.004) 0.000 (0.020) −0.014 (0.023) −0.007 (0.010) 0.003 (0.028) 0.008 (0.013) 0.021 (0.026) 0.024 (0.026) 0.099* (0.048) 0.038*** (0.012)

−0.311*** (0.043) −0.006 (0.012)

−0.064** (0.030) −0.002 (0.012) −0.756*** (0.042) −0.047** (0.021) 0.013 (0.033) 0.159*** (0.045) 0.007 (0.026) 0.102*** (0.032) 0.013** (0.005) −0.022 (0.023) 0.005 (0.031) −0.002 (0.016) −0.039 (0.034) 0.007 (0.026) 0.045 (0.037) 0.034 (0.026)

Mas

−0.308*** (0.042) −0.017 (0.013)

0.017 (0.028) −0.006 (0.015) −0.953*** (0.056) 0.030 (0.046) 0.013 (0.058) −0.036 (0.069) 0.005 (0.044) 0.052 (0.041) 0.007 (0.006) 0.066* (0.039) 0.023 (0.057) −0.023 (0.021) 0.046 (0.079) 0.047 (0.031) 0.129*** (0.048) 0.021 (0.044) 0.345*** (0.075)

−0.094 (0.069) −0.021 (0.028)

−0.293** (0.137) 0.077 (0.115)

−0.132* (0.074) 0.030 (0.056) 0.157 (0.138) −0.139 (0.135) −0.074 (0.164) 0.375* (0.193) −0.038 (0.155) −0.085 (0.148) −0.011 (0.021) −0.189* (0.097) −0.092 (0.153) 0.054 (0.057) −0.199 (0.262) −0.108 (0.092) −0.338** (0.131) 0.001 (0.117)

0.219** (0.089) 0.128** (0.058) 2.332 (1.578)

0.219** (0.080) 0.117* (0.062) 2.889* (1.544)

Yes Yes 164 0.363

Yes Yes 164 0.403

Mas squ TTTotal TTTotal squ Mas TTTotal Acad Acad squ Other Other squ Acad Other Mas Acad Mas Other

0.539*** (0.071) 0.045 (0.037) 0.181*** (0.036) 0.030*** (0.009) −0.094** (0.034) −0.031 (0.029) 0.027* (0.014)

TTTotal Acad TTTotal Other Budget per Emp Budgetsha Pub Budgetsha Priv Non-Winners Budgetsha Priv Winners

0.215*** (0.053) 0.143*** (0.040) 0.127 (0.080) 0.034 (0.045)

Budgetsha Priv Total

0.529*** (0.069) 0.032 (0.038) 0.175*** (0.034) 0.029*** (0.009) −0.081** (0.031) −0.028 (0.034) 0.031 (0.018)

0.231*** (0.048) 0.167*** (0.030)

Mas squ TTTotal TTTotal squ Mas TTTotal Acad Acad squ Other Other squ Acad Other Mas Acad Mas Other

0.522*** (0.074) 0.048 (0.032) 0.191*** (0.027) 0.032*** (0.009) −0.094*** (0.034) −0.025 (0.030) 0.028** (0.014)

TTTotal Acad TTTotal Other Budget per Emp Budgetsha Pub Budgetsha Priv

0.236*** (0.054) 0.149*** (0.035) 0.083 (0.127)

Instrument 1

Constant

−2.317*** (0.582)

−0.163 (0.300)

0.045 (0.041) −2.498*** (0.528)

Field FE Institution Type FE University FE N R2

Yes Yes No 161 0.843

Yes Yes No 161 0.978

Yes No Yes 187 0.861

0.067** (0.024) −0.417 (0.428)

Instrument 2 Constant

−2.572*** (0.599)

−1.445*** (0.530)

Yes No Yes 187 0.966

Field FE Institution Type FE N R2 p (Overidentification)

Yes Yes 164 0.844 0.5663

Yes Yes 164 0.925 0.7356

OLS estimates of production functions assuming a translog form. Table 3 provides variable definitions. Budgetshare Pub and Priv show the effect of funding shares on university productivity. The table shows coefficients and robust standard errors, which are clustered at the university level. *, ** and *** denote significance levels 10%, 5% and 1%. Columns 1 and 2: “Budgetsha Priv” with and without reverse causality; Columns 3 and 4: Inclusion of university fixed effects; Columns 1 and 3: No Technology Transfer (TT) measure; Columns 2 and 4: Informality weighted TT measure.

In order to assess for the potential non-linearity of the relationship between third-party funding and productivity, budget shares enter as quadratic polynomials (‘Budgetsha Pub’, ‘Budgetsha Priv’, ‘Budgetsha Pub squ’ and ‘Budgetsha Priv squ’). The vector of control variables (‘Control’) contains dummy variables for the following scientific fields: Natural Sciences (‘Nat Sci’), Medicine (‘Med’), Economics and Business (Econ and Bus), Mathematics and Physics (‘Math and Phy’). The base category is engineering. Furthermore, dummy variables for the institution types ‘Cantonal University’ (‘University’), ‘University of Applied Sciences’ (‘UAS’) and ‘Federal Research Institution’ (‘Fed Res Inst’) capture the differences to the base category ‘ETH Swiss Federal Institute of Technology’. Finally, the vector of control variables includes the log of the total budget per employee (‘Budget per Emp’) to account for differences in the available budget.  denotes a normally distributed error term, meaning that ∼N(0, 2 ).

0.193 (0.166) 0.011 (0.093) −0.057 (0.145) −0.017 (0.023) −0.091 (0.075) −0.060 (0.061) 0.003 (0.026)

OLS estimates of production functions assuming a translog form. Table 3 provides variable definitions.Budgetshare Pub and Priv show the effect of funding shares on university productivity. The table shows coefficients and robust standard errors, which are clustered at the university level. *, ** and *** denote significance levels 10%, 5% and 1%. p (Overidentification) captures the p-value of a Sargan test, which tests the Nullhypothesis, that the instruments are valid. Instrument 1 is the TT motive intensity of “Acquisition basic research funds”. Instrument 2 is the TT motive intensity of “Commercialization”. Columns 1 and 2: Distance functions, “Budgetsha Priv” instrumented; Columns 3 and 4: First stage equations; Columns 1 and 3: No TT measure; Columns 2 and 4: Informality weighted TT measure.

In order to provide a credible identification strategy, we use three alternative approaches to tackle the econometric problems of endogeneity due to omitted variable bias (e.g. research quality) and reverse causality. However, as the survey was designed to capture technology transfer, only one of these approaches allows tackling endogeneity of public third-party funding, implying that a causal interpretation of our findings is more convincing for private funding. The first approach separates departments based on a survey question whether technology transfer has increased research funding resources.5 If this question is answered with yes, reverse

5 See question 5.1 in the questionnaire http://www.kof.ethz.ch/surveys/structural/ panel/wissensaustausch 2005.

T. Bolli, F. Somogyi / Research Policy 40 (2011) 136–147

causality and consequently endogeneity exists. We distinguish non-winners from winners, where we define winners as departments that have gained additional financial resources from technology transfer. We include two variables for private funding in our estimations, which capture the impact for non-winners (‘Budgetsha Priv Non-Winners’) and winners (‘Budgetsha Priv Winners’) separately. Concretely, we multiply private funding shares by a dummy for non-winners and winners, respectively.6 This approach allows us to asses whether our results are solely driven by departments for which reverse causality is present. Our second approach to tackle endogeneity in the estimation consists of exploiting within-university variation by including university fixed effects, thereby controlling for unobserved heterogeneity, e.g. quality differences. This approach has the advantage that it accounts for unobserved heterogeneity for both public and private funding, while the other approaches only address problems related to private funding. These two approaches are corroborated by a traditional instrumental variable (IV) approach, which solves the problems of measurement error in research output due to omitted quality, reverse causality and any remaining endogeneity bias (see e.g. Angrist and Krueger, 2001). We instrument the share of private funding by the answers to two questions concerning financial motives for engaging in technology transfer with private companies.7 Besides of the output-oriented multi-output production function described above, we also employ a 3SLS methodology, which estimates separate production functions for each output simultaneously (see e.g. Wenger, 2000). This method allows us to identify the productivity effect of third-party funding on each output separately. We adjust total labor input by the share of work devoted to teaching, basic research and applied research in the equations for master degrees, publications and technology transfer, respectively. Thereby we assume that the share of labor devoted to outputs is the same for academic and non-academic staff.8 In order to test whether heterogeneity in the production technology across scientific fields affects the robustness of our results (see e.g. Schmoch and Schubert, 2009), we drop individual disciplines from our sample in order to check if our results are driven by single disciplines. In addition, we drop medical departments and economics departments at the same time, keeping a sample of only science fields.9 We find that our results are robust to these variations in the sample (Tables 10 and 11).

6 That is, the first variable, ‘Budgetsha Priv Non-Winners’, gives the value of the share of third-party funding if the department has not gained additional financial means from technology transfer (non-winners) and zero, if the department has gained (winners), as well as a corresponding variable, ‘Budgetsha Priv Winners’, for winners, that gives the value of the share of third-party funding if the department has gained additional financial means (winner) and zero, if the department has not gained additional funds from technology transfer (see e.g. Arvanitis and Hollenstein, 2002). 7 The research departments were asked to rank several financial motives to engage in technology transfer with private companies on a 1 (not important) to 5 (extremely important) scale. The options used as instruments here are ‘resources for expanding basic research’ and ‘commercial success’. See question 4.1 in the questionnaire. 8 We also estimated model variations that entail alternative input specifications and find similar results. Notably, we test the sensitivity to including total labor input as well as to introducing the full vector of adjusted labor inputs in each equation. Furthermore, the results are stable across the robustness checks shown in the paper concerning the output distance function. 9 Due to the small number of observations by field, regressions for single fields result in production function estimates that are not well-behaving. Therefore we do not report them here.

143

Table 9 TT measure. Dep Var: Publ

1

2

3

4

Mas

−0.330*** (0.038) −0.013 (0.011)

−0.076*** (0.021) −0.007 (0.011) −0.744*** (0.036) −0.054*** (0.016) 0.024 0.024 0.169*** (0.032) −0.010 (0.025) 0.085*** (0.029) 0.012** (0.005) −0.025 (0.020) 0.004 (0.024) −0.001 (0.014) −0.059* (0.032) 0.001 0.021 0.050 (0.031) 0.022 (0.023) 0.062*** (0.021)

−0.045 (0.027) −0.012 (0.016)

0.229*** (0.046) 0.149*** (0.029) 0.053 (0.036)

−0.081*** (0.022) −0.004 (0.011) −0.738*** (0.039) −0.044** (0.016) 0.010 (0.025) 0.173*** (0.034) −0.010 (0.024) 0.088*** (0.030) 0.012** (0.004) −0.023 (0.021) −0.002 (0.026) −0.001 (0.014) −0.051 (0.034) 0.003 (0.021) 0.051 (0.034) 0.032 (0.024) 0.069*** (0.020)

Constant

−2.520*** (0.493)

−0.522 (0.400)

−0.529 (0.378)

0.051* (0.025) −0.013 (0.032) 0.060* (0.029) −0.244* (0.121) −0.220** (0.092) −0.156* (0.088) −0.086 (0.057) −0.146* (0.074) −0.493 (0.296)

Field FE Institution Type FE N R2

Yes Yes 187 0.846

Yes Yes 187 0.962

Yes Yes 187 0.964

Yes Yes 187 0.977

Mas squ TTTotal TTTotal squ Mas TTTotal Acad Acad squ Other Other squ Acad Other Mas Acad Mas Other

0.520*** (0.062) 0.043 (0.031) 0.177*** (0.027) 0.030*** (0.008) −0.079*** (0.027) −0.023 (0.026) 0.031* (0.015)

TTTotal Acad TTTotal Other Budget per Emp Budgetshare Pub Budgetshare Priv TTinformal TTfacility TTeducation TTresearch TTconsulting

0.103*** (0.032) 0.014 (0.020) 0.048 (0.028) 0.003 (0.006) −0.026 (0.026) −0.017 (0.022) 0.029** (0.013)

OLS estimates of production functions assuming a translog form. Table 3 provides variable definitions. Budgetshare Pub and Priv show the effect of funding shares on university productivity. The table shows coefficients and robust standard errors, which are clustered at the university level. *, ** and *** denote significance levels 10%, 5% and 1%. Column 1: No Technology Transfer (TT) measure; Column 2: Informality weighted TT measure; Column 3: Unweighted TT measure; Column 4: Data-driven weights of TT measure.

6. Results 6.1. Main results Table 5 displays the results of OLS production function estimations in translog form. The specifications shown in columns 1 and 2 exclude technology transfer intensity, while columns 3 and 4 include it. Columns 1 and 3 include budget shares linearly (‘Budgetsha Pub’ and ‘Budgetsha Priv’), while the regressions shown in columns 2 and 4 also include the corresponding quadratic terms (‘Budgetsha Pub squ’ and ‘Budgetsha Priv squ’). We further test whether our results are robust to a change in the method of estimation, i.e. estimating a production function for each

144

T. Bolli, F. Somogyi / Research Policy 40 (2011) 136–147

Table 10 Dropping individual scientific fields excluding TT. Dep Var: Publ

mp

en

ec

ns

md

ec&md

Mas

−0.342*** (0.049) −0.000 (0.011) 0.491*** (0.061) 0.186*** (0.041) 0.035 (0.031) 0.017 (0.012) −0.044 (0.040) −0.006 (0.036) 0.035 (0.029) 0.205*** (0.033) 0.160*** (0.046) 0.077* (0.042) −2.250*** (0.370)

−0.308*** (0.064) −0.004 (0.015) 0.544*** (0.087) 0.145*** (0.043) 0.034 (0.049) 0.026** (0.011) −0.059 (0.036) −0.020 (0.036) 0.027 (0.019) 0.196*** (0.048) 0.126*** (0.036) 0.065 (0.040) −1.925*** (0.503)

−0.320*** (0.038) −0.014 (0.012) 0.521*** (0.064) 0.170*** (0.028) 0.048 (0.033) 0.029*** (0.009) −0.079** (0.028) −0.019 (0.027) 0.030* (0.016) 0.230*** (0.048) 0.147*** (0.032) 0.055 (0.039) −2.531*** (0.516)

−0.338*** (0.029) −0.046*** (0.013) 0.531*** (0.067) 0.221*** (0.042) 0.021 (0.034) 0.035*** (0.009) −0.082** (0.031) −0.055* (0.031) 0.032** (0.014) 0.291*** (0.081) 0.073 (0.057) 0.002 (0.044) −3.192*** (0.891)

−0.358*** (0.049) 0.000 (0.012) 0.488*** (0.071) 0.169*** (0.037) 0.018 (0.029) 0.039*** (0.010) −0.069** (0.025) −0.036 (0.031) 0.061** (0.023) 0.247*** (0.055) 0.162*** (0.030) 0.052 (0.040) −2.749*** (0.594)

−0.360*** (0.033) −0.034*** (0.011) 0.491*** (0.058) 0.212*** (0.057) 0.02 (0.031) 0.016 (0.016) −0.031 (0.033) −0.051 (0.049) 0.048 (0.039) 0.227*** (0.067) 0.105 (0.067) 0.02 (0.051) −2.469*** (0.747)

Yes Yes 153 0.843

Yes Yes 122 0.843

Yes Yes 169 0.836

Yes Yes 144 0.872

Yes Yes 160 0.845

Yes Yes 110 0.873

Mas squ Acad Other Acad squ Other squ Acad Other Mas Acad Mas Other Budget per Emp Budgetsha Pub Budgetsha Priv Constant Field FE Institution Type FE N R2

OLS estimates of production functions assuming a translog form. Table 3 provides variable definitions. Budgetshare Pub and Priv show the effect of funding shares on university productivity. The table shows coefficients and robust standard errors, which are clustered at the university level. *, ** and *** denote significance levels 10%, 5% and 1%. Columns 1–5 each exclude one of the fields: “Mathematics and Physics”, “Engineering”, “Economics and Business”, “Natural Sciences” and “Medicine”. Column 6 excludes the fields: “Economics and Business” and “Medicine”.

output separately by employing a 3SLS estimator (see e.g. Wenger, 2000). Table 6 portrays the results. Columns 1a-b and 2a-b show the results for the estimations including teaching and publications as output variables, while columns 3a-c and 4a-c also include technology transfer as a third output. The odd numbered columns contain linear estimations, while the even numbered columns also feature squared terms of the budget shares. The production function estimations behave well in the sense that the first order coefficients of labor inputs (‘Acad’ and ‘Other’) enter positive and significant in all regressions. Furthermore, outputs (‘Mas’ and ‘TTTotal’) have the expected negative sign in all setups. The budget available per employee enters the production functions significant and positive, which might capture the fact that the price of labor inputs depends on their quality. Alternatively it might reflect productivity gains from improved capital endowment. The dummy variables for scientific fields reveal that medical departments are significantly less productive than the base category, engineering departments. This finding might reflect difficulties to attribute resources correctly to university departments and hospitals, respectively. Since economics and business as well as math and physics departments are typically less technology transfer oriented than engineering and natural sciences departments, regressions excluding technology transfer result in a significantly positive coefficient of these departments. However, once we control for technology transfer, the coefficient for economics and business departments becomes insignificant (Table 5, columns 3 and 4), while the coefficients for math and physics departments become significantly negative, implying that controlling for technology transfer is essential to compare productivity across fields. Natural science departments are not statistically different from engineering, independent of whether technology transfer activities are accounted for or not.

In the regressions excluding technology transfer, Universities of Applied Sciences are found to be significantly less productive than the base category, Federal Institutes of Technology (ETH). Controlling for technology transfer, the coefficient turns insignificant. The opposite effect is observed for Federal Research Institutions, which are as productive as ETH departments if we exclude technology transfer. However, Federal Research Institutions become significantly less productive once technology transfer is accounted for. Cantonal Universities are statistically indistinguishable from the ETH in all setups. Turning to our variables of main interest, the shares of public and private third-party funding (‘Budgetsha Pub’ and ‘Budgetsha Priv’), we find a significantly positive impact of public third-party funds and no significant effect of private funds on productivity in column 1 of Table 5, which includes budget shares linearly and excludes technology transfer. The positive impact of public third-party funding is plausible since the Swiss research policy is geared towards the promotion of basic research, and hence the share of public funds mainly impacts basic research production, which is more likely to result in publications than technology transfer. The 3SLS results in Table 6 support the hypothesis by showing a positive impact on research productivity but not for technology transfer. A potential explanation for the insignificant coefficient of the private funding share is that positive and negative effects cancel each other out. Alternatively, private donors decide to conduct no or ineffective monitoring in respect to basic research. Potential explanations include the large costs that private donors incur in monitoring scientific research. While public authorities delegate monitoring to the SNSF and CTI that use a low-cost peer-review process, private companies have to build up absorptive capacity or buy in expertise in the respective field. Additionally, the nature of basic research renders contracts typically incomplete and diffi-

T. Bolli, F. Somogyi / Research Policy 40 (2011) 136–147

145

Table 11 Dropping individual scientific fields including TT. Dep Var: Publ

mp

en

ec

ns

md

ec&md

Mas

−0.104*** (0.026) 0.014 (0.014) −0.710*** (0.062) −0.041** (0.016) 0.174*** (0.032) −0.002 (0.030) 0.113*** (0.031) 0.008 (0.005) −0.018 (0.030) −0.035 (0.040) 0.040 (0.028) −0.013 (0.046) −0.074* (0.039) −0.052** (0.024) 0.037 (0.030) 0.029 (0.031) 0.073** (0.027) −0.378 (0.353)

−0.082* (0.044) 0.004 (0.011) −0.700*** (0.053) −0.048 (0.032) 0.180*** (0.049) −0.002 (0.046) 0.090* (0.042) 0.013** (0.006) 0.011 (0.030) 0.000 (0.035) −0.002 (0.016) −0.065 (0.072) 0.009 (0.025) −0.021 (0.030) 0.060 (0.041) 0.039 (0.038) 0.076*** (0.025) −0.653 (0.437)

−0.067*** (0.022) −0.002 (0.011) −0.754*** (0.044) −0.040** (0.015) 0.171*** (0.032) −0.004 (0.022) 0.069** (0.030) 0.010** (0.004) 0.001 (0.026) −0.004 (0.026) −0.005 (0.013) −0.042 (0.030) 0.007 (0.017) −0.021 (0.019) 0.043 (0.037) 0.026 (0.027) 0.064*** (0.022) −0.437 (0.432)

−0.077*** (0.019) −0.035*** (0.009) −0.774*** (0.024) −0.071** (0.026) 0.171*** (0.034) −0.013 (0.020) 0.109*** (0.026) 0.015*** (0.004) 0.071* (0.036) 0.009 (0.035) 0.000 (0.009) −0.062 (0.039) 0.006 (0.014) −0.030* (0.015) 0.099* (0.056) 0.018 (0.029) 0.060*** (0.019) −1.039 (0.659)

−0.085** (0.030) 0.003 (0.013) −0.751*** (0.045) −0.040** (0.016) 0.142*** (0.038) −0.031 (0.026) 0.071** (0.032) 0.014** (0.007) −0.002 (0.026) −0.009 (0.029) 0.010 (0.025) −0.046 (0.030) −0.003 (0.025) 0.003 (0.023) 0.041 (0.036) 0.017 (0.032) 0.067*** (0.021) −0.405 (0.433)

−0.097*** (0.022) −0.029* (0.015) −0.762*** (0.056) −0.093*** (0.03) 0.174*** (0.031) 0.003 (0.03) 0.117*** (0.019) 0.014 (0.009) 0.086 (0.053) 0.002 (0.065) 0.042 (0.032) −0.049 (0.064) −0.090** (0.036) −0.085*** (0.025) 0.074 (0.045) 0.025 (0.032) 0.074** (0.027) −0.77 (0.524)

Yes Yes 153 0.960

Yes Yes 122 0.951

Yes Yes 169 0.962

Yes Yes 144 0.971

Yes Yes 160 0.964

Yes Yes 110 0.973

Mas squ TTTotal TTTotal squ Acad Acad squ Other Other squ Mas TTTotal Mas Acad Mas Other TTTotal Acad TTTotal Other Acad Other Budget per Emp Budgetsha Pub Budgetsha Priv Constant Field FE Institution Type FE N R2

OLS estimates of production functions assuming a translog form. Table 3 provides variable definitions. Budgetshare Pub and Priv show the effect of funding shares on university productivity. The table shows coefficients and robust standard errors, which are clustered at the university level. *, ** and *** denote significance levels 10%, 5% and 1%. Columns 1 to 5 each exclude one of the fields: “Mathematics and Physics”, “Engineering”, “Economics and Business”, “Natural Sciences” and “Medicine”. Column 6 excludes the fields: “Economics and Business” and “Medicine”.

cult to enforce accurately, limiting the utility of monitoring further. These problems might induce private donors to abstain from monitoring activities when they finance basic research. Specification two excludes technology transfer as well, but third-party funding shares enter as quadratic polynomials instead of linearly. The share of private funding remains insignificant. However, the coefficients of public third-party funding become negative, where only the quadratic term is negative. Table 6 reveals that a negative correlation between public third-party funding and teaching productivity drives the negative impact. The picture changes for regressions that include technology transfer as an additional output. The coefficient of the public third-party funding share turns insignificant, while private funds enhance productivity significantly. Hence, ignoring technology transfer as an output biases the estimation. This result can be explained by the fact that private funding is more likely to be directed towards technology transfer than towards publications, and that while private donors do not monitor basic research, they do take care that the results are transferred and monitor the process of applied research in a productivityenhancing manner. In addition, the absorptive capacity of private donors geared towards applied research, implying relatively low monitoring costs. However, Table 6 indicates a positive relationship

between private funding and productivity in respect to research and technology transfer. These results hold for a quadratic specification of budget shares as well. As shown in column 4 of Table 5, the coefficients of the quadratic terms are insignificant. Therefore, we specify budget shares linearly in the following. 6.2. Reverse causality, unobserved heterogeneity and endogeneity To tackle the endogeneity problems discussed above, we employ three different approaches. First, we separate departments that have gained additional financial means through technology transfer from those that have not gained additional funds and include an interaction term between the private funding share and a dummy variable capturing whether conducting technology transfer has increased the available research funding. Comparing the coefficients of these two groups reveals the presence of reverse causality. Assuming that our estimates are unbiased, the two coefficients for winners and non-winners are the same. Therefore, a higher coefficient for winners indicates reverse causality. Second, we introduce institution fixed effects to control for unobserved heterogeneity at the university level, e.g. differences in research quality. Finally, we instrument the share of private funds using an IV approach.

146

T. Bolli, F. Somogyi / Research Policy 40 (2011) 136–147

Table 7 displays the estimation results, where column 1 reveals that a specification excluding technology transfer yields the same results as the base specification, i.e. the public third-party funding share has a significant positive impact while the private funding share remains insignificant for both non-winners and winners. Column 2 introduces technology transfer. We find that the share of public third-party funding is insignificant. The private funding share has a significant positive impact for both non-winners and winners, indicating that the finding is not driven by reverse causality. In fact, the magnitude of the coefficient is larger for nonwinners than for winners, though the level of significance is smaller. The results of our third approach, the IV estimations are shown in Table 8. Columns 1 and 2 show the results of the second stage and columns 3 and 4 display those of the first stage. The results support the base estimation. In the setup without technology transfer variables shown in column 1, we find a positive and significant effect of the public third-party funding share on productivity and no significant effect of private funding. Column 2 reveals that if we account for technology transfer, the influence of the public third-party funding share becomes insignificant, while the private funding share turns out to be productivity enhancing. 6.3. Robustness In order to evaluate the robustness of our results, Table 9 reports two additional specifications in columns 3 and 4, which use variations of our technology transfer measure. Columns 1 (no technology transfer) and 2 (informality weighted technology transfer) repeat the results from our baseline specification in Table 5 to facilitate the comparison of the results. In column 3, total technology transfer enters the equation without any weighting of the single technology transfer channel groups available from the survey, while in column 4, we include all technology transfer channel groups (informal contacts, technical facilities, training, research collaboration, consulting) separately,10 implying that weights are data driven. Varying the definition of the technology transfer measure barely affects the coefficients for inputs, outputs and third-party funding shares. An additional robustness check is presented in Tables 10 and 11, where we drop one scientific field at a time from our sample (columns 1–5 in both tables) in order to check if our results are driven by individual scientific fields. In addition, we drop both medical departments and economics departments together in column 6, implying that the remaining sample consists only of natural science departments. Table 10 reports the results for the regressions without controls for technology transfer, while it is controlled for in Table 11. We find that our results are qualitatively robust to these variations in the sample.

We find no impact of third-party funding on teaching productivity. However, the share of both private and public third-party funding improves publication productivity. Technology transfer productivity is independent of public third-party funding, but is increased by private funding. The external validity of our results for private funding beyond the borders of Switzerland is straightforward. For public third-party funding on the other hand, external validity depends on the similarity of the grant system, as a more bureaucratic approach might promote productivity less effectively. However, the use of peerreview processes is wide-spread, indicating that our results hold for other countries as well. In particular, the German and Austrian systems are very similarly structured as in Switzerland. A limitation of our paper is that while the categories public and private funds might capture quite heterogeneous funding sources, our data only allows the separation of these two broad categories. It is left to future research to delve deeper into the issue and analyze the impact of more accurately specified funding sources on the behavior of researchers. The estimation of dynamic effects provides an additional direction of future research, as this paper, due to the cross-sectional nature of the data, studies only static effects. Dynamic effects might arise, because if external funds flow to the most efficient researchers, the less productive researchers will acquire less funding resources. This opens the possibility of selection, either through self-selection based on income or through promotion decision by supervisors. This might have an impact on the average researcher quality and consequently on research productivity. A further politically relevant topic not addressed in this paper is that third-party funds might have effects on the behavior of researchers beyond the impact on productivity. Of particular relevance for politicians is the possibility that private third-party funding induces the researcher to devote more time and effort to applied projects, thereby reducing the work devoted to basic research (see e.g. Florida and Cohen, 1999; Geuna, 2001; Schiller ˜ and Macho-Stadler, 2008). Finally, and Liefner, 2006; Banal-Estanol the present state of research does not identify the relevance of individual influence channels but only the aggregate impact. Disentangling these effects might be a further interesting path of research. Acknowledgments The authors would like to thank the participants of the KOF Doctoral Seminar, the Young Swiss Economist Meeting, the European Public Choice Society Meeting and the COST Workshop at the European Workshop for Efficiency and Productivity Analysis as well as Spyros Arvanitis, Mehdi Farsi, Marius Ley, Tobias Stucki, Jan-Egbert Sturm, Martin Woerter and two anonymous referees for helpful comments and discussions.

7. Conclusions This paper analyzes the impact of private and public thirdparty funding on the productivity of Swiss university departments and public research institutions. By accounting for potential endogeneity, we enhance the understanding of the causal relationship between funding mechanisms and productivity, hence improving politicians’ ability to make evidence-based decisions.

10 The full regression in Table 9 contains also squared terms of the TT measures, interactions of the TT measures with the output ‘master students’, interactions between all TT measures and interactions between the TT measures and the inputs (Acad and Other), which gives a list of 35 different variables. We only report the five linear terms in Table 9 to increase the clarity of exposition. Full estimations are available upon request from the authors.

References Abbott, M., Doucouliagos, C., 2009. Competition and efficiency: overseas students and technical efficiency in Australian and New Zealand universities. Education Economics 17 (1), 31–57. Agasisti, T., 2009. Market forces and competition in university systems: theoretical reflections and empirical evidence from Italy. International Review of Applied Economics 23, 463–483. Aghion, P., Dewatripont, M., Hoxby, C.M., Mas-Colell, A., Sapir, A., 2008a. Higher Aspirations: An Agenda for Reforming European Universities. Bruegel Blueprints. Aghion, P., Dewatripont, M., Hoxby, C.M., Mas-Colell, A., Sapir, A., 2009. The governance and performance of research universities: evidence from Europe and the U.S. NBER Working Paper 14851. Aghion, P., Dewatripont, M., Stein, J.C., 2008b. Academic freedom, private-sector focus, and the process of innovation. Rand Journal of Economics 39, 617–635. Angrist, J.D., Krueger, A.B., 2001. Instrumental variables and the search for identification: from supply and demand to natural experiments. Journal of Economic Perspectives 15 (4), 69–85.

T. Bolli, F. Somogyi / Research Policy 40 (2011) 136–147 Arvanitis, S., Hollenstein, H., 2002. The impact of spillovers and knowledge heterogeneity of firm performance: evidence from Swiss manufacturing. In: Kleinknecht, A., Mohnen, P. (Eds.), Innovation and Firm Performance: Econometric Explorations of Survey Data. Palgrave, pp. 225–252. Arvanitis, S., Kubli, U., Sydow, N., Woerter, M., 2007. Knowledge and technology transfer (KTT) activities between universities and firms in Switzerland – the main facts: an empirical analysis based on firm-level data. ICFAI Journal of Knowledge Management 17, 17–75. Arvanitis, S., Kubli, U., Woerter, M., 2008. University-industry knowledge and technology transfer in Switzerland: what university scientists think about cooperation with private enterprises. Research Policy 37 (10), 1865–1883. ˜ Banal-Estanol, A., Macho-Stadler, I., 2008. Commercial incentives in academia. City University Economics Discussion Papers. Bonaccorsi, A., Daraio, C., Simar, L., 2006. Advanced indicators of productivity of universities: an application of robust nonparametric methods to Italian data. Scientometrics 66 (2), 389–410. Butler, L., 2003. Explaining Australia’s increased share of ISI publications—the effects of a funding formula based on publication counts. Research Policy 32 (1), 143–155. Carayol, N., Matt, M., 2006. Individual and collective determinants of academic scientists’ productivity. Information Economics and Policy 18 (March (1)), 55–72. Cherchye, L., Abeele, P.V., 2005. On research efficiency: a micro-analysis of Dutch university research in economics and business management. Research Policy 34 (4), 495–516. Coelli, T., 2000. On the econometric estimation of the distance function representation of a production technology. Core Discussion Papers, Université Catholique de Louvain, Center for Operations Research and Econometrics (CORE). Coelli, T., Perelman, S., 2000. Technical efficiency of European railways: a distance function approach. Applied Economics 32 (15), 1967–1976. de Boer, H., File, J., 2009. Higher education governance reforms across Europe. Modern Project Report, European Center for Strategic Management of Universities. de Boer, H.F., Enders, J., Schimank, U., 2007. On the way towards new public management? The governance of university systems in England, the Netherlands, Austria and Germany. In: Jansen, D. (Ed.), New Forms of Governance in Research Organizations – Disciplinary Approaches, Interfaces and Integration. Springer, Dordrecht, pp. 135–152. Filippini, M., Lepori, B., 2007. Cost structure economies of capacity utilization and scope in Swiss higher education institutions. In: Bonaccorsi, A., Daraia, C. (Eds.), Universities and Strategic Knowledge Creation. Specialisation and Performance in Europe. Edward Elgar.

147

Florida, R., Cohen, W., 1999. Engine or infrastructure? The university role in economic development. In: Branscomb, L., Kodama, F., Florida, R. (Eds.), Industrializing Knowledge: University–Industry Linkages in Japan and the United States. MIT-Press, Cambridge, pp. 589–610. Geuna, A., 2001. The changing rationale for European university research funding: are there negative unintended consequences. Journal of Economic Issues 35, 607–632. Jansen, D., Wald, A., Franke, K., Schmoch, U., Schubert, T., 2007. Third party research funding and performance in research: on the effects of institutional conditions on research performance of teams. Kölner Zeitschrift für Soziologie und Sozialpsychologie 59, 125–149. Kempkes, G., Pohl, C., 2008. The efficiency of German universities – some evidence from non-parametric and parametric methods. Applied Economics Online First. Kivistö, J., 2005. The government-higher education institution relationship: theoretical considerations from the perspective of agency theory. Tertiary Education and Management 11, 1–17. Kuo, H., Ho, Y., 2007. The cost efficiency impact of the university operation fund on public universities in Taiwan. Economics of Education Review 27, 603–612. Leisyte, L., Kizniene, D., 2006. New public management in Lithuania’s higher education. Higher Education Policy 19, 377–396. Liu, N., Cheng, Y., 2005. The academic ranking of world universities. Higher Education in Europe 30, 127–136. Mensah, Y.M., Werner, R., 2003. Cost efficiency and financial flexibility in institutions of higher education. Journal of Accounting and Public Policy 22 (4), 293– 323. Niskanen, W., 1975. Bureaucrats and politicians. Journal of Law and Economics 18 (3), 617–643. Robst, J., 2001. Cost efficiency in public higher education institutions. The Journal of Higher Education 72, 730–750. Schiller, D., Liefner, I., 2006. Higher education funding reform and university–industry links in developing countries: the case of Thailand. Higher Education 54, 543–556. Schmoch, U., Schubert, T., 2009. Sustainability of incentives for excellent research – the German case. Scientometrics 81 (1), 195–218. Schubert, T., 2009. Empirical observations on new public management to increase efficiency in public research-boon or bane? Research Policy 38, 1225–1234. van der Ploeg, F., Veugelers, R., 2008. Towards evidence-based reform of European universities. CESifo Economic Studies 54 (2), 99–120. Wenger, J., 2000. What do schools produce? Implications of multiple outputs in education. Contemporary Economic Policy 18 (1), 27–36, 01.