This article was downloaded by: [University of Arizona]

On: 18 November 2012, At: 22:43

Publisher: Routledge
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer
House, 37-41 Mortimer Street, London W1T 3JH, UK

Economics of Innovation and New Technology

Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/gein20

An Empirical Evaluation of The Effects of R&D

Subsidies
a
Isabel Busom
a
Departament d'Economia Aplicada, Universitat Autonoma de Barcelona, Bellaterra,
08193, Spain
Version of record first published: 28 Jul 2006.

To cite this article: Isabel Busom (2000): An Empirical Evaluation of The Effects of R&D Subsidies, Economics of
Innovation and New Technology, 9:2, 111-148

To link to this article: http://dx.doi.org/10.1080/10438590000000006

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions

This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to
anyone is expressly forbidden.

The publisher does not give any warranty express or implied or make any representation that the contents
will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses
should be independently verified with primary sources. The publisher shall not be liable for any loss, actions,
claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or
indirectly in connection with or arising out of the use of this material.
 Emm hnm Nmv Teahn.20aO. W.9. pp. 111-148 Q OPA (- PubUshrs Assodaion) N.V.
Repias available directly Born tkpublkhu PubWbyti~au~llllaa
P b a o c o d n e ~ ~ b y ~ & f k Hmvoml Acaacmie Pub- hpiut,
pan of the Gordon snd Breach Pub- Gmup.
Rimed in MEhyaia

AN EMPIRICAL EVALUATION
OF THE EFFECTS OF R&D SUBSIDIES
Downloaded by [University of Arizona] at 22:43 18 November 2012

Departament d'Economia Aplicada, UniuersitatAutbnoma de Barcelona, 08193

Bellatema, Spain

(ReceivedNovember15. 1998 ;1n finalfonn April 15.1999)

Rgd> subsidies are a common tool of technology policy, but little is known about the effects
they have on the behavior of firms. This paper presents evidence on the effects that R&D sub-
sidies have on the R&D effort of recipients. and on the probability that a firm will participate
in a propun granting R&D subsidies. The empirical model consists of a system of equations:
a participation equation; and an R&D effort equation. Endogeneity of public funding is con-
trolled for. Estimates are obtained with a cross-saction sample of Spanish firms. The main
findings are that: 1) small firms are more likely to obtain a subsidy than large firms, probably
reflecting one of the public agency's gods; 2) overall, public funding induces more private
effort, but for some firms (30% of participants) full crowding out effects cannot be ruled out,
and 3) firm size remains related to effort. whether or not a firm gets public funding.

Kcywonls: Technology policy; RBtD,subsidies;policy evaluation

JEL Chsification: 031; H32; L52

1. INTRODUCTION

Promoting the generation and diffusion of innovations is becoming an

important goal of public intervention in most OECD countries, as techno-
logical change is acknowledged to be one of the main determinants of eco-
nomic growth. Within the European Union the increasing budget allocated
to successive Framework Programs, complementing national level science
* The author is also a research affihte at the Institut d'Analisi Econbmica, Barcelona.
Spain. This paper has kencompleted during a stay at the Department of Economics, Univer-
sity of California, Berkeley. Financial support provided by the NATO Fellowships Program is
gratefully acknowledged, as is additional support from the Ministerio de Educacib of Spain.
WICYT PB93-0679 and ClCYT SEC96-2300.
Conesponding Author: E-mail: Isabeldusom@uab.es
 112 ISABEL BUSOM

and technology programs, illustrates this point. Yet, evidence on the

effects of technology policy tools on the behavior and productivity of
firms is rather limited.* This paper develops an empirical model to ana-
lyze an R&D subsidy program in Spain, focusing on both the determinants
of h s ' participation in the program and on the effects that participation
has on the firms' R&D e f f m t
Three of the several issues that can be raised in evaluating this specific
Downloaded by [University of Arizona] at 22:43 18 November 2012

policy tool are the following. The first issue concerns the analysis of who
are the participants in a program: What types of firms end up getting pub-
lic subsidies? To what extent does the observed r d t agree with the pub-
lic agency's intended goals? The second issue refers to how receiving a
subsidy affects a firm's behavior with respect 0(9 R&D decisions: does the
firm increase its R&D effort or does public financing crowd out private
financing, partially or completely? Are there interaction effects between a
firm's participation in a public R&D program and its R8cD effort? Does
the type of product market competition matter equally, whether participat-
ing or not? Finally, the third question is in a way the final test of the effec-
tiveness of the policy: how does the productivity of firmsreceiving R&D
subsidies change (a direct effect) and what is the impact on the productiv-
ity of other firms (the i n k t or spillover effect) and on consumers?
This paper addresses the first NO, not having enough information in the
data used here to address the third. However, because they are related
issues it is worthwhile to comment briefly om the evidence available so far.
The relationship between publicly funded R&D and productivity at the
firm or industry level has been addressed in studies by Griliches (1986),
Griliches and Lichtenberg (1984). Cunh (1984), and Hall and Mairesse
(1995). These authors use production function models to estimate and test
the effects of publicly provided R&D funding on the productivity of firms
receiving it. In general, very small or non-significant effects were found.
The alternative dual approach is followed by Nadhi and Marnuneas (1994,
* There are a number of recent empirical studies, closely related to the endogenous growth
literatme, that focus on the effect of public and private R&D on growth at the aggregate level.
Lichtenberg (1992) finds, using aggregate data for several countries, that government funded
research has a much lower marginal ptoduct than privately funded r e m h . Park (1995) pro-
vides additional alferences to this literatwe. However, to umderstartdhow and why public and
private RBtl) have an effect on growth requires an analysis at the decision-maker level, that
is, firms and public agencies. Kauko (1996), Mowery (1995) and Capron (1992) provide a
review of studies at a disaggregate level.
t Work evaluating similar R&D policy tools is found in Hail (1993). on tax credits in the
USA, or Folster (1992) on subsidies to encourage cooperative RBrH) in Sweden.
 R&D SUBSIDES 113

1995). who estimate cost and factor demand functions that depend on pub-
licly financed M D capital and infrastructures. Using two-digit indus-
try-level data, they do find significant positive effects of public R&D
capital on the cost structure of industries but also find that these effects
vary over time and across industries. As pointed out in these and other
studies, more attention has to be paid to the composition of R&D, to the
dynamic structure of effects, to measurement of spillovers, and to the
Downloaded by [University of Arizona] at 22:43 18 November 2012

interaction between public and private R&D effort.

As a first step in evaluating public funding for R&D it may be of interest
to investigate the firms immediate response to subsidies. The question is
whether public funding induces or replaces private R&D spending. If it
replaces private R&D spending, it must be concluded that a subsidy policy
may not be appropriate. The evidence here is again controversial. Scon
(1984). Levin and Reiss (1984) for U.S. firms, Holemans and Sleuwaegen
(1988) for Belgian firms, Antonelli (1989) for Italian h s , and Klette and
Moen (1997) for Norwegian firms, address this issue by estimating regres-
sion models where private R&D expenditure is the dependent variable and
publicly financed R&D is one of the explanatory variables. Sample sizes
and empirical methods vary,* but in all of these studies authors find evi-
dence of a statistically significant and positive elasticity of privately
funded R&D with respect to the subsidy. Using a similar empirical fi-ame-
work Lichtenberg (1984, 1987, 1988), however, finds evidence of crowd-
ing-out. Lichtenberg (1984), controlling for the possible endogeneity of
being a recipient of public funding through firm fixed effects, finds that
public funding may substitute for company funding.+ In Lichtenberg
(1987), he finds further evidence of overestimation of the positive effects
of public funding of R&D, when sales are decomposed into sales-to-Gov-
ernment and other sales. When controlling for the buyer, Lichtenberg finds
that federally funded R&D has no effects on private R&D investment.
Finally Lichtenberg (1988) concludes that government R&D contracts
obtained through competition have a positive effect on company funded

* Scon uses a sample of 3388 line of business observations corresponding to 437 U.S.
firms; Holemans and Sleuwaegen a sample of 236 0b~ervation~ of Belgian firms and
Antonelli a sample of 83 Italian firms. Levin and Reiss use industry-level data (20 indusuies.
3 years). Klene and Moen use 816 observations of line of business data for high tech sectors
in Norway.
t Lichtenberg (1984) used NSF firm-level panel data with 991 observations. His 1987 and
1988 work is based on a panel of 187 finns observed for 6 years (1979-1984).
 114 ISABEL BUSOM

R&D spending, but that non-competitive R W contracts have a negative

effect*
These simple regression models have two shortcomings. Fit, there is
the problem of the endogeneity of public funding, only implicitly
addressed in fichtenberg and Klette and Moen, and not in the remaining
studies. In order for a firm to receive public funding it must apply for
funding, and the public agency may or may not a w d it, given firm and
Downloaded by [University of Arizona] at 22:43 18 November 2012

project characteristics. This makes public funding an endogenous variable,

and its inclusion in a linear regression will cause inconsistent estimates if
it happens to be correlated with the error term. It is very likely that this
correlation will be non zero, because unobservable factors may determine
both receiving public funding and private RBd4 effort decisions. On the
other hand, if the sample of firms used to evaluate the effect of public
funding includes only recipient firms, then there will be truncation or self
selection effects, and estimation methods must account for this in order to
obtain consistent estimates.
Second, the conclusion that should be drawn from finding a negative
relationship between public and private R&D expenditure in former stud-
ies is not clear, as the public agency may choose precisely to finance more
heavily R&D projects with higher spillover potential (for instance, those
involving basic research or higher appropriability difficulties), where
incentives for private funding may be small. In that case, there is no reason
to expect positive or higher private spending by the recipient firm, unless
the project triggers additional applied and development research generat-
ing private ~etums.So higher public funding may be associated with very
small private funding, producing a non significant or even a negative coef-
ficient in this type of regression, yet not implying crowding-out effects. In
addition, if the public subsidy is granted on the basis of some matching
condition, a positive relationship is likely to be found in the regression
equation, but the firm may have reduced its private effort relative to what
it would have done without the subsidy, implying partial crowding out
Therefore, to make inferences about crowding out effects, what is needed
is a consistent estimator of the firm's hypothetical effort in R&D in
absence of a subsidy.
The study presented here intends to address these difficulties to some
extent. Fht, it investigates which factors associated with receiving
* This suggests that the way public funding is provided (with firms competing or not com-
peting for it) may matter a lot in changing firms incentives.
 R&D SUBSIDIES 115

R&D subsidies, an issue not addressed before. An attempt will be made to

identify the decision of the firm to apply for funding and the decision of
the public agency to give it. Second, it estimates how much R&D effort
firms that received subsidies would have made had they not received the
subsidies, using the sub-sample of non-participants and controlling for
selection. It will be concluded that public funding has had crowding out
effects if the private spending of a subsidy recipient is found to be smaller
than the effort it would have made had it not received a subsidy. The
Downloaded by [University of Arizona] at 22:43 18 November 2012

remainder of the paper is organized as follows. Section XI presents the

econometric model, section III contains a description of the data, section
IV explains the behavioral hypothesis, and sections V and VI contain a
discussion of the results.

2. A STRUCTURAL MODEL OF PARTICIPATION IN A PUBLIC

R&D PROGRAM AND OF R&D PRIVATE EFFORT

In most public R&D promotion programs a iirm has to apply for a subsidy
in order to get one. Then the public agency has to decide whether to give it
or not. There are two decisions involved, one by the firm and one by the
public agency. Together these decisions give rise to the participation status
of the firm:participant or non-participant in the program. Conditional on
obtaining or not obtaining public funding, a iirm then decides how much
R&D effort to make, in terms of expenditure, personnel or type of project
(i-e., process or product, basic research content). If this is the process gen-
erating the observations, an empirical model to represent it will include
four structural equations, the first two b e i i a sort of demand and supply
equations for subsidies, and the third and fourth being the R&D effort
decision if participating and if not participating respectively. These equa-
tions are:
A* = f,(Z. U) (1)
G* = fg(W, V ) (2)

Yl* = l~l(Xl?~l) (3)

YO* = hz(Xo,wo) (4)
where Z,W, XIand Xg are vectors of explanatory variables, and u, v, w0
and w are error terms that may contain unobservable characteristics of the
 116 ISABEL BUSOM

firm.The covariance matrix will be non-diagonal if common unobservable

variables underlie u, v, wl and wo.
A* in equation (1) represents the firm's expected profitability of apply-
ing for an R&D subsidy relative to not applying. Equation (2) captures the
decision rule of the public agency granting the subsidies. The dependent
variable, G*,is the value to the agency of funding a particular project, and
it is not observed. The agency's decision may k based on a number of fac-
Downloaded by [University of Arizona] at 22:43 18 November 2012

tors, some related to the social interest of the R&D project, and some to
the estimated ability of the firm to carry it out. Equations (3) and (4) cap
ture the total R&D effort made by a firm, usually measured by R&D
expenditure or R&D personnel.
To the researcher, A* and G* are unobservable. yl' is observed only for
participating firms; y< is observed only for non-participants.* These two
equations will allow for differences in the coefficients associated to
explanatory variables under each pdcipation regime. These differences
may arise as a result of the policy. For instance, fmn size is often found to
be positively related to R&D effort, but for firms receiving an R&D sub-
sidy, size may not be a relevant factor to explain effort. These type of
effects can be tested within this framework.
Because of the unobservable nature of the profitability of a decision for
each agent, the model above is in fact a latent variable model, where only the
qualitative outcome of the decisions is observed. gf we know whether a firm
applies or not, we can define a binary variable If that equals 1 if the firm
applies. We may observe whether the agency awards the subsidy or not,
which can in turn be translated into a binary variable Ig. The observation
mechanism gives rise to a discrete choice model for participation in the R&D
program and to an effort model with endogenous switching. The complete
suuctural model includes in addition to equations (1)-(4), the following:
> 0 and I = 0 otherwise
If = 1 if A* (5)
Ig = 1if G* > 0 and I = 0 otllerwise (6)
If we only observe who is a participant in the program (a case of limited
observability) and who is not, then(5) and (6) collapse into
I = If*&= g (Z,W , u. v ) (7)
where 1=1implies that a firm has applied for and obtained a subsidy.
* This frameworkis very similar to that used in evaluating the impact of training programs
on wages; see for example Heckman and Robb (1985).
 R&D SUBSIDIES 117

Equations (1) and (2). and their observable counterparts (5) and (6) or
(7),are a discrete choice model of participation or selection equation and
may be estimated with probit or logit multivariate models. Equations (3)
and (4) may be estimated for each subsample (of participants and non-par-
ticipants) correcting for endogenous self-selection.
To sum up, this model allows for testing the public agency's allocation
rule, inferring additional effects of the policy and estimating the hypothet-
Downloaded by [University of Arizona] at 22:43 18 November 2012

ical RBrD expenditure of participants, using non-participants as the com-

parison group after controlling for selection.

3. THE DATA

The data used consists of a sample of 154 Spanish firms that were con-
ducting R&D activities in 1988.75 (45%) of which received public fund-
ing for their R&D projects through the Centro para el Desarrollo
Tecnol6gico e Industrial (CDTI), an agency of the Spanish Ministry of
Industry: That year, the CDTI granted subsidized loans to 213 projects of
a total of 541 applications (a 39% approval rate). Some firms had more
than one project approved. According to CI)TI's annual report, on average
public funding amounted to 39 % of total R&D investment for approved
projects (CDTI,1988).
Firms in the sample were asked three sets of questions. First, standard
questions related to firm characteristics such as size, export volume, indus-
try,firm's age, ownership.
Second, some questions concerning R&D activities such as R&D
expenditure (total expenditure, including the subsidy if received), R&D
personnel, availability of public funding, patents obtained in the previous
10 years, type of research, sources of ideas for R&D projects. And finally,
questions related to strategic attitudes or behavior of firms in the product
market or with respect to R&D. Most of the answers to the last type of
questions were finally coded as binary variables. Table I provides the defi-
nition of the variables constructed from the survey.
Firms could apply for two possible sources of public funding: from
national programs or from European-level programs. Eighty three firms in
* For a description of data sources, see Appendix I.
 118 ISABEL BUSOM

the sample (about 54 %) declared having applied for a national R&D sub-
sidy, and 75 obtained it* As for European-level programs, 47 firms
applied for and 40 obtained funding. Only 15 % of firms declared having
both types of public funding. However, we do not have information for
firms in our sample about how many projects did a firm submit, nor of the
magnitude of the subsidy received.
Some information was obtained on how is R & D done. Main sources of
Downloaded by [University of Arizona] at 22:43 18 November 2012

ideas for their projects and reaction to a rival's W&D activities were issues
of particular interest, the first as an indicator of possible knowledge spillo-
vers, the second as one of strategic behavior. A Binn's own patents were an
important source of ideas for 56 % of firms, but even more (63 %) consid-
ered rivals' patents or licenses or their rivals' products to be an important
source of ideas. Almost all firms considered general sources of informa-
tion (Universities or research centers, scientific and technical publications)
quite or very important. The existence of knowledge spillovers in research
ideas will be proxied by binary variables constructed on the basis of these
questions?
In an attempt to obtain information on firms' strategic behavior in the
product market, firms were asked about their pricing strategy.* Most firms
(44%) declared to choose prices and then adjust production to demand,
and about 29 % declared to set production plans first, and determine the
price afterward. Some firms were regulated, and a specific binary variable
was constructed for this case. As for R&D specific strategies, 58 % of
firms would accelerate their own R&D effort if they found out a rival firm
was doing similar R&D rather than cooperate or stop. Finally, all firms
said RBrD played a very importaut role in the firm's strategy in the long
run, but only half of them thought R&D was important in the short run.

The rejection rate in the sample is very small relative to that in the population; CDTI
reports rejection rates of about 60 % of the submitted proposals for that year. It is possible
that some firms that were rejected declared not having applied This will have implications
for estimation.
t These are knowledge spillovers a firm benefits from. We do not have a measure for
*
extent of appropriabilityof the firm's own R&D.
Firms were asked in particular whether they usually set prices and then production was
adjusted to demand @rice competition), or vice versa Firms following the first rule may be
interpreted to be Bertrand competitors, while those following the second rule would be
Cournot cornpetiton.
 R&D SUBSIDIES

TABLE I Variable Definition

Subsidies
Cdti =1 if a firm received a subsidy from CDTI,0 otherwise.

Empean -1 if a firm was a partner in EUREKA or any EC R&D program.

Ownership
Downloaded by [University of Arizona] at 22:43 18 November 2012

Public =1 if firm was partly publicly owned

Foreign =1 if firm was participated by foreign capital
Strategic Variables
Rice =1 if firm declared to set prices and then adjust production to sales

Qmtity =1 if firm declared to make production plans and then adjust prices ,

Regulated =1 if firm declared prices to be regulated

Other =1 if none of the above
Monopoly -1 if firm declared behaving as such
Frival =1 if f m declared it would increase own R&D in response to a rival's
Shortrun =1 f firm declared R&D to be important in the short nm
R&D Pmcess
Ideariv =1 if firm looked into competitor's products for ideas for own R&D
Ideapt =1 if firm used own patents as sources of ideas
Ideaext =1 if firm declared scientific and technical publications to be important
=1 if firm cooperated with others in R&D activities
Basic =l if firm does basic or applied m h
Development =1 if firm does development
Rocess =1 if R&D activities an oriented towards process innovation
Product =1 if R&D activities are oriented towards product innovation
Industry

Dchemical =1 if firm is in chemical or pharmaceutical industry

Detronics =I if f m is in electrical o electronics industry
Dequipmt =1 if firm is in machinery or transportation equipment industry
Denergy =1 if firm is in the energy sector
 120 ISABEL BUSOM

Binary Variables Definition

Dtraditional =1 if firm is in textile, food, metal industries

Dservices =1 if f m provides services to other industries
DOrherI =1 if firm is in other industries
Continuous Vm'ables
Patents Number of patenu obtained by f m during the previous 10 years
Downloaded by [University of Arizona] at 22:43 18 November 2012

R&D expenditure Total R&D expenditure in 1987,in Million pesetas

R&D personnel Number of employees involved in R&D activities
Age Number of years since f m was created
Employment Total number of employees
Exportshare Exportdtotal Sales

Tables 11 and KII show descriptive statistics for the sample, which con-
tains 147 firms after those with missing values were deleted. The sample is
biased towards large firms, thereby producing high means for some varia-
bles (see Appendix I).

TABLE 11Descriptive Statistics. Continuous Variables

Variable N Mean Stand Dev. Mini- Maxiwuun

R&D Expenditure (M. Pts) 147 330 721

R&D P e r s o ~ e l 147 51 105
Num.Employees 147 1622 3310
Age of the finn (years) 147 31 24
Patents (Number) 147 6 18
R&D expendlemployee 147 0.61 1.1
R&D perslemployee (9%) 147 13 19
Exports share 147 16 18.6
 R&D SUBSlDIES 121

TABLE ILI Descriptive Statistics. B i Variables

Variable Nwn oflfm

with X=l
Variable
(X)
strategy
Rice
Q-ti9
Downloaded by [University of Arizona] at 22:43 18 November 2012

Regulat
R&D Roeess Monopoly
Patent Other

Royalty Frival
Ideaext Shortrun
Ideariv
Cooperate IndustrJt
Rasic Chemical
Development Electronics
Rocess Ekpipment
Roduct Energy
Senices
Ownemhip Traditional'
Public Other1
Foreign
a It includes several industries: food textile, steel and metallic products.

4. SELECTION OF EXPLANATORY VARIABLES:

BEHAVIORAL HYPOTBESES

The data described above may have been generated by the process out-
lined in section 2. Given the available information and the model, the
dependent variables will be the following: application for an R&D subsidy
(equation (S)), granting a subsidy (equation (6)) and R&D effort (equa-
tions (3) and (4)). Some hypothesis may now be formulated as to which
variables enter the vectors of explanatory variables Z, W, and X,.
 122 ISABEL BUSOM

Firms' decisions: R&D effort and subsidy application

Both the theoretical and empirical literature point to a number of factors

associated with private R&D effort in general. Research has focused on
the role that firm size and market power or strategic behavior have on
incentives to undertake R&D activities. Ease of imitation or a low degree
of appropriability of R&D results are also a crucial trait which creates a
Downloaded by [University of Arizona] at 22:43 18 November 2012

wedge between private and social returns. The interaction with market
structure is complex, but theoretical models of patent races do predict it
will affect R&D activity. One example is found in Delbono and Denicolo
(1990), who develop a model predicting that firms competing in quantities
in the product market will spend less in R&D than firms competing in
prices. Other factors that may affect R&D investment decisions are liquid-
ity constraints (Hall et al, 1998), type of ownership (Veugelers and
Vanden-Houten (1990) provide a model and empirical evidence that multi-
national firms may lead to reduced domestic R&D), and the extent to
which a firm sells its products in foreign markets. Finally, differences in
technological opportunity and in expected demand growth may explain
interindustxy differences in R&D effort.
Consequently, total employment, strategy and ownership indicators,
share of exports over total sales and industry d d e s will be included in
the effort decision. In addition, the indicator of spillovers the firm receives
from others and the previous patents, measuring experience in RBrD will
be added. The sign to be expected of the spillover indicator is ambiguous:
on one hand, benefiting from others' ideas may lower the h ' s costs of
innovating, but on the other it may trigger more projects, and thus more
effort.
Tuming now to the application decision, it would seem that since all
fums in the sample conduct R&D activities, a d the eligibility conditions
set by the public agency do not appear too restrictive ex ante in Spain, all
firms would be expected to apply for a subsidy, provided there are no fixed
costs associated with applying.* The benefits from obtaining a subsidy
need not be equal for all firms, however. If there are fixed costs associated
with research, small firms will presumably be less able to finance them
* This may be true for firms already conducting R&D. but firms not doing R&D may have
to incur expenses just to present a good proposal, and the expected subsidy may be too low
for them to make applying worthwhile.
 R&D SUBSIDIES 123

and more eager to participate in the subsidy program. Also firms that per-
ceive R&D to be an important strategic variable for competing with rivals
even in the short run* and iirms that compete in international markets
(exporters) may also profit relatively more from RBd) subsidies than other
firms. Finally, firms with fully domestic capital are more likely to apply,
because those with foreign capital participation may benefit from R&D
conducted in another location, and their R8cD activities in Spain may be
Downloaded by [University of Arizona] at 22:43 18 November 2012

limited to adapting products or technology. Included explanatory variables

in the application equation will be firm size (expecting a negative relation-
ship), capital ownership, importance given to R&D in the short run,pric-
ing strategy, share of exports in total sales and industry dummies.+
The probability of applying may be explained by other factors which are
not considered here because there is not enough information in the data.
One is the information the firm has: some might be unaware of the exist-
ence of the program. Another possibility is that some may find it not prof-
itable to apply if their project would not generate a private return, and the
expected subsidy does not cover all the costs of the project (recall that the
public agency requires firms to finance at least 30 % of ,the project).
Finally, some firms may be interested in strictly development projects,
while one of the public agency's conditions is that projects should have a
genuine innovative content. Such firms may then not apply.

The public agency

The public agency is assumed to have a systematic selection criterion,
therefore it does not choose randomly which firms and projects to subsi-
dize. In addition to selecting on the basis of the quality and feasibility of
the proposal, the agency may want to encourage R&D in small f h s
(because of fixed costs involved in R&D and of capital market imperfec- ,

* Managen' perceptions about the impoxtance of R&D in the short run may explain differ-
ential behavior across firms.These different perceptions may reflect to some extent differ-
ences in human capital.
t Some variables related to the type of research done (product or process, extent of basic
content, duration of RBtD projects, whether to apply for other sources of public R&D fund-
ing, such as EC Framework Program or Eureka) are available as well for this data set. They
may be jointly determined with the decision to apply for a subsidy. An additional equation for
each of them wuld be added to the system of equations above. However, because of the lim-
ited number of observations and of exogenous variables a more complex model is not esti-
mated here.
 124 ISABEL BUSOM

tions), or in some research fields (i.e., defense, health) where a higher gap
between private and social returns is assumed t~ exist*
In 1988 the CDTI financed projects in three broad areas: production
technology (information and communications technology, space, microe-
lectronics, new materials, robotics and automation), quality of life (toxi-
cology, pharmaceuticals, immunology, biotechnology), and natural
resources and agriculture (including food technology). The information
Downloaded by [University of Arizona] at 22:43 18 November 2012

and forms made available to applicants do not establish requirements for

application other than those related to the quality of the project (extent of
innovative content and technical, commercial, and economic feasibility)
and to the ability of the firm to carry it out (in terms of human capital) and
.tofinance at least 30 % of the total cost of the project.
Ractically all firms seem eligible, given the breadth of the technological
areas, and no explicit reference is made to requirements affecting the size
of the firm. Because the quality and feasibility of a project is likely to be
related to a Arm's experience in RBrD (measured by patents obtained in the
past) and general experience @roxied by the age of the firm)both varia-
bles will Be included in the selection rule. Finn size, presence of public or
foreign capital, and industry dummies will be included as well. The public
agency is assumed not to give any weight to the firm being an exporter,
nor to be concerned about competitive conditions in the product market, so
dummies for strategic variables will be excluded from the equation captur-
ing the agency behavior, in principle facilitating identification.
Whether an R&D project proposal has or does not have some basic
research content might be an indicator of potential research externalities,
and thus a possible justification for public support of R&D. Although a
basic research variable can be constructed, it will not be used as explana-
tory variable because in this cross-section data it is jointly determined with
the subsidy, and thus endogenous.
The error term will contain non-observed factors such as indicators of
the quality of the projects and of the human capital of the firm, which the
public agency is assumed to consider.

According to Navas (1984). the ability of the project to generate secondary technological
effects was also considered. This suggest that spillovers m y play a role in policy decisions.
 R&D SUBSIDIES 125

5. PARTICIPATION IN PUBLIC R&D PROGRAMS:

ESTIMATION AND RESULTS

Equations (5) and (6) of the model outlined in section 2 determine the par-
ticipation status of each h, with four possible outcomes. Since for a
given firm Ig is not observed unless If equals one, three outcomes are pos-
sible in theory. But the information contained in this sample is even more
Downloaded by [University of Arizona] at 22:43 18 November 2012

limited, because too few iirms declare applying and not obtaining a sub-
sidy. So in fact the reasonable equation to estimate is (7),the participation
status.*
I = If*& = g (Z.W, e )
where I = 1 iff A* > 0 and G* > 0 (the firm has a public subsidy), and I = 0
otherwise. There are two types of observations, participants and non-par-
ticipants, and it will be assumed that the probability of being a participant
is given by the bivariate normal density. The structural parameters associ-
. ated to vectors Z, W could still be estimated provided that identification
conditions hold. However, the particular configuration of the binary
explanatory variables may generate identification problemst as it turns
out to happen with this data set. Therefore, a simple univariate probit
model is estimated. This equation may then be interpreted as a reduced
form, where coefficients of explanatory variables capture the net effect of
each variable on the final status, participation. What is lost is the ability to
identify separately the different weight a particular variable may have on
the firm's application decision and on the agency's rule.
Estimation results are shown in Table IV,Appendix II. The first column
of coefficients shows the estimates obtained for a model with all explana-
tory variables, and the second column shows the.estimates of a restricted
model where non-significant variables have been dropped. The third col-
umn presents the marginal effect that each variable included in the
restricted model has on the probability of being a participant, that is,
6(Pr(I=1))/6xj = b *f(x'b) where x is the vector of explanatory variables
and f(-) the density function evaluated at the mean of vector x. Columns 4
and 5 show the results obtained when the variable SHORTRUN is
* This is one of the partial obscrvability cases discussed in Poirier (1980).
t Meng and Schmidt (1985) illustrate the implications that inclusion of b i explanatory
variables has for identification and estimation in discrete choice models.
 126 ISABEL BUSOM

included, but industry dummies are excluded.* Models 2 and 3 correctly

predict 80 % of non-participants and 70 % of participants. The predictive
performance of either of these two models is not negligible, given that a
naive model would have correctly predicted 52 % of non-participants and
none of participants.
According to these results, six factors had a significant net effect on the
probability of having national public funding in Spain in the late 1980s.
Downloaded by [University of Arizona] at 22:43 18 November 2012

Public participation in the ownership and being in the chemical or pharma-

ceutical sector are found to increase the probability of participation signif-
icantly. A firm's overall experience (proxied by age) and its experience in
R&D (proxied by the number of previous patents) increase this probability
as well. These variables could well affect both decision makers in the same
direction, as more experienced firms are more aware of the value of inno-
vation and may present better proposals, becoming thus more likely to be
selected by the agency. The fact that experience related variables are sig-
nificant suggests that subsidies might not be the best policy to induce firms
that did not have previous R&D activities to get them started.
'lbo factors reduce the probability of having a subsidy: large firm size
and the presence of foreign capital. In both caws the estimated net effect
may again be the result of a variable affecting both agents' decisions in the
same direction. Small firms may be more credit constrained, and the pub-
lic agency may be willing to favor small firms because of this. The nega-
tive sign of size also suggests that application costs probably were not
high. F m s with foreign capital ownership are found less likely to partici-
pate. One possible explanation is that core R&D activities may be located
in another country, and those carried out in Spain may be mostly focused
toward development, with littie innovative content, and thus these firms
don't apply for a subsidy. Or if they do, the public agency may reject their
application and prefer to promote mostly domestic firms.Finally, strategic
variables are not found to have any significant net effect on participation.

* Considering R&D an impoltant strategic variable in ttte s h m run might be carelated

with unobserved variables that also affect the participation status, such as managerial ability.
S h o is~thus instrumented with expolts, Electronics and Frival in order to obtain consist-
ent estimates. The only effect its inclusion has is that industry dummies become non signifi-
cant, suggesting some degree of multicollinearity.
 R&D SUBSIDIES 12i

A comment about participation in European-level R&D propuns

Model specification becomes more complex if firms have two sources of

public funding, and the decisions of the agencies are not independent. This
could happen if a firm having domestic funding is more likely to receive
funding from European institutions, or vice versa. lbenty seven percent of
the finns in this sample did participate in European-level programs
(Eureka or Framework Programs), and 16 % participated in both. A bivar-
Downloaded by [University of Arizona] at 22:43 18 November 2012

iate probit model to explain the joint probability of having domestic and
European funding wiis estimated. Because results suggest that the two
types of participation were independent at the time, consistent estimates
for participation in each program can be obtained by estimating a univari-
ate probit model for each.* Results obtained with the univariate probit
model to explain European-level participation are shown in the last col-
umn of Table IV. They suggest that contrary to the domestic program, the
probability of participating was positively related to firm size and that
experience related variables were not relevant at the time. The negative role
of foreign capital ownership is observed also in this case.'lnterestingly finns
that declared to compete in prices were less likely to participate. However,
the overall predictive capacity of the model is low, as more than half of par-
ticipants are missclassified by the model. Because the number of firms par-
ticipating in Ewpean R&D programs is small in our sample and because of
the novelty of these programs for Spanish finns (Spain had joined the EU
only in 1986), it may be difficult to find systematic patterns of behavior.
Results obtained here must therefore be interpreted with caution.

6. A COMPARISON OF R&D BEHAVIOR OF PARTICIPANTS

AND NON-PARTICIPANTS

Next step involves estimating R&D effort for participants and for non-par-
ticipants, taking into account that group membership is not random. The
purpose of estimation is to provide evidence regarding two questions:
1) does participation induce a higher R&D effort than would have been
made otherwise? and 2) does participation make the firm's choice of R&D
effort less conditioned by factors such as f h n size?
* Independence may not hold in posterior years, though. As information about Euro-
pean-level programs improves over time, and as the national agency encourages firms to par-
ticipate in them. firms with expcrieoce in participating in the national program may be more
likely to do so.
 128 ISABEL BUSOM

Severall equations will be estimated, because two measures for R&D

effort are used, R&D expenditures and R&D personnel. Each in turn is
estimated in levels and intensity (R&D expenditure per employee and
share of W&D personnel over employees, respectively.* One reason to use
both expenditure and p e r s o ~ e lis that expenditure may cap- differ-
ences in W&D equipment intensity or costs, while personnel may reflect a
more permanent, human capital component of R&D effort. R&D expendi-
Downloaded by [University of Arizona] at 22:43 18 November 2012

tures are total spending by the fum, which in the case of participants
includes phe subsidy received. The magnitude of the subsidy itself is not
known. Explanatory variables will once again capture the effect they have
on private effort plus the effect they have on the magnitude of the subsidy.
Because some firms received funding though participation in European
level R&D programs, an additional dummy variable (EUROPEAN)is
introduced to control for this fact, under the assumption that this variable
and the error term are uncorrelated. Each equation is estimated by four
procedures. By ordinary least squares, using the whole sample of firms,
and including a participation binary variable (CDTI), as in previous stud-
ies. This is the most restrictive specification, since it imposes equality of
coefficients for both groups of firms, as well as exogeneity of participa-
tion. The second procedure consists of splitting the sample, not imposing
equality of coefficients; but estimating by OU,thus imposing exogeneity
of participation. The third consists of using Heckman's two step procedure
to c o r m for endogenous sample selection; this will allow testing for exo-
geneity of participation.t Finally both equations are estimated by maxi-
mum likelihood, jointly with the participation equation.
Tables V through 8 in Appendix I show the. re~u1t.s.~ Columns 1,2 and 3
in each table show that running OLS on the whole sample of firms with the
participation dummy as explanatory variable, will generate biased esti-
mates because it imposes the restriction that the coefficients of the explan-
atory variables are the same for both types of fums, when this hypothesis

* Sales were not used because they were obtained as intervals in the original survey.
t The selection model is:
yl=bl'x+rl,ol&+el
yo=~x+rboo&+~
where ri, is is the correlation between the error term in each effort equation and the error term
in the participation or selection equation, and & is the inverse Mill's ratio.
$ Four firms were excluded from the sample, as they behaved as outliers and results were
sensitive to their inclusion.
 R&D SUBSIDIES 129

should be rejected.* This is so for all measures of R&D effort used. Once
different behavior is dowed for both types of firms, there are no sharp dif-
ferences between OLS estimates and Heckman's method estimates in the
expenditure equations. The coefficient for lambda in either of the sample
selection equations is not significant, suggesting that the hypothesis of no
selection bias cannot be rejected. Maximum likelihood estimates are very
close to the sample selection model, but there are some differences in esti-
Downloaded by [University of Arizona] at 22:43 18 November 2012

mated standard errors. The fit is in general better for R&D personnel equa-
tions than for RBd).expenditure (especially in case of the relative effort
equation), and better for participants than for non-participants.
Participants and non-participants differ in the way explanatory variables
affect total R&D effort. Because the hypothesis of selection bias based on
unobservables has been rejected, participants do not have an unobserved
comparative advantage at doing R&D (are not more likely to have higher
R&D effort because of unobserved factors) relative to non participants.
Differences in coefficients may be partly attributed to the public agency's
funding decision.
In absence of public funding, R&D expenditure is explained by firm size
(with a positive but smaller than one coefficient in the level equation, and
a negative sign in the relative effort equation) and by being either in the
chemicaYpharmaceutical industry or in the electricdelectmnic industries.
None of the remaining variables is found to have significant effects (expe-
rience, capital ownership, or strategic and spillover variables). These
results are consistent through all estimation methods, except for the case of
R&D personnel for non participants, where industry variables are some-
what sensitive to specification.
Participants show a different pattern. There are two main differences.
The first is that firms in the equipment and services to h s industries
spend more relative to participants from other industries. These firms are
equally likely as firms in traditional industries to be participants, but once
they are, they may be getting higher subsidies. The second difference is
that having obtained patents in the past makes firms spend more on R&D,
as do firms mostly oriented to the domestic markett A possible interpre-
* A Chow-test leads to rejection of the null hypothesis of equality of slopes in equations
for participants and non-mcipants.
t The share of e x p m has a negative sign. This result is conhary to what was expected,
but other studies using larger samples of Spanish finns (2400observations) have obtained
this result. See Labeaga and Martinez-Ros (1994).
 130 ISABEL BUSOM

tation is that the public agency may prefer subsidizing this type of firms,
which may make sense from the perspective the welfare of domestic con-
sumers.
There is some evidence that participants that declared to set production
plans first and then adjust prices spend less, in line with the prediction of
the theoretical model by Delbono and Deniwlo, but this doesn't hold
across dl estimations. The role of firm size on RBrD expenditure remains
Downloaded by [University of Arizona] at 22:43 18 November 2012

unchanged, the sign being negative in the relative effort equations, but
with a slightly smaller absolute value. Other strategic variables and exist-
ence of spillovers were not found significant in any of the equations.* As
a final remark, participating in a European-level R&D program does not
seem to have led to increased expenditure.
When effort is measured by R&D personnel, the same basic regularities
appear. No evidence of selection bias is found for participants, and relative
effort decreases with an increase in firm size or in export share, and
increases with previous patenting activity. Participating firms in the chem-
ical and electronic industries increase their R W personnel, although not
expenditure. Non-participants allocation of R&D personnel seems to be
mostly determined by firm size and industry.

7. DID PUBLIC SUBSIDIES CROWD OUT BWATE EFFORT?

Crowding out occurs when public funding drives pPivate funding out par-
tially or completely. Complete crowding out takes place when a recipient
firm would have privately spent the same total mount for a given project
It indicates that firms would have made the R&D effort anyway and that
private returns were high enough. Partial crowding out occurs when total
effort increases, but the private contribution is smaller than if the firm had
been on its own.
Testing the presence of crowding out effects involves constructing a
counterfactual for r&ipient firms, which will be done using the estimation
of effort of non-participants. If data on privately financed R&D effort are
available, then the extent of partial or complete crowding-out effects could
be estimated for each participant by computing an estimate of its expected
* In previous work with this same data set (Busom 1993) it was found that the measure of
spillovers affected negatively the probability of a firm doing basic research. and positively
the probability of doing process innovations.
 private effort as a participant, using the coefficients of the "Participants"
column (ylpredicted = E{y1[I1)) and comparing it to an estimate of its
expected potential effort, had the firm not been a pdcipant
(Yopotential - {Yo (I1)) ,obtained with the coefficients of the "Non-Par-

ticipants". If y 1mmd 2 Yopotential, then crowding out effects (total or par-

tial) can be ruled out.
In our case, however, only total R&D effort, which includes the subsidy
Downloaded by [University of Arizona] at 22:43 18 November 2012

received, is known for participants. The amount of subsidy is unknown. So

for each participant, .the expected hypothetical private effort had it not
received a subsidy can be computed (yo~tential), but it has to be compared
to total expected effort, which includes the subsidy. With this information
we can say that there are full crowding out effects if the effort a firm
makes after receiving the subsidy is not greater than the effort it would
have made had it not received the subsidy. The presence and extent of par-
tial crowding out cannot be estimated, however, because private spending
is not lcnown for participants.
TO calculate an estimate of y,predicted = E{ylll= 1) , the coeffi-
cients of OLS estimates in the participants' simplified regression are used.'
To calculate an estimate for participants' potential effort had they not
received the subsidy, yo~otentia'= E{Y,,~I= 1) ,the coefficients of the
non-participants regression are used. Because the goodness of fit of the
regression for non-participants suggests that relevant variables may be
missing, the following results must be taken with caution for policy assess-
ment purposes. Measuring effort with expenditure, it turns out that in
aggregate (calculating x ~ ~ P these ~ ~70 ~firms
~ would
~ ~ have ~ ) spent
,
15,368 MPta. instead of the estimated expected totalof 27,798 MPta., just
over one half. According to the public agency's annual report, that year it
subsidized one third of the total cost of the projects, so this means that
public funding induced an additional 20 % of private expenditure, in the
aggregate. Howwer, inspection of results for individual firms shows that
in 29 cases, firms would have spent at least as much on their own. It is pos-
sible that some of these lirms had received such a small subsidy that it did
not change the scope of their R&D plans. But they also suggest that there
might be cases where finns reduced private funds after receiving the sub-
sidy, even to the extent of full crowding-out.

* Non-significant variables, except for Quantity, were dropped Quantity was kept because
it bewme significant, witb a negative sign.
 132 ISABEL BUSOM

In the remaining 41 cases of subsidy recipients, firms did spend overall

more than they would have otherwise. But again, not knowing the size of
the subsidy rules out the possibility of inferring whether, for a particular
firm, this induced more private effort or not. All that can be said is that
complete crowding-out can be ruled out in those cases.
Results are similar if R&D personnel is used as a measure of effort. On
aggregate, and on their own, these 70 firms would have allocated 2,547
Downloaded by [University of Arizona] at 22:43 18 November 2012

people to R&D activities, instead of 3,730 (an increase of 46 %). For 30

firins, predicted potential effort is not smaller than the prediction with the
subsidy, suggesting as before that these h s would have carried out the
R&D projects anyway. For 20 of these firms (almost 30% of participants)
this prediction is obtained using both expenditure and R&D personnel.
Further analysis should attempt to find an explanation for these cases. This
could be useful for the public agency, helping it improve the selection rule
in order to fund only R&D activities that would not have been undertaken
otherwise.
Finally, the estimates for potential effort, yoPotenti", are positive for all
participants (for expenditure and for personnel), implying that all partici-
pants would have incurred some R&D expenditures even if they had not
received a subsidy. This is not surprising, since the sample only includes
firms doing R&D. But it also suggests that a more complete evaluation of
the subsidy program could be made if the sample included firms that did
not do in R&D in the past, so the ability of the program to induce new
comers to RBrD activities could be tested.

Few studies have used econometric modeling to test the systematic effects
that different technology policy tools have on firms; yet ex-post evaluation
of any policy at the microeconomic level is a necessary step for checking
its effectiveness and for detecting unanticipated effects. Using a sample of
Spanish b s , where half of them received public subsidies for R&D, this
paper provides evidence on some factors explaining participation in a
domestic R&D subsidy program, and on the effects that participation has
on R&D effort. Although the information available in this sample does not
allow for separate estimation of the application decision by a lirm and the
 R&D SUBSIDIES 133

granting decision by the public agency, a reduced form of the determinants

of participation status can be estimated.* Results show that smaller firms
have a higher chance of being participants, which, given that R&D doers
tend to be large, means that smaller firms are more likely to apply for and
be granted a subsidy. This conclusion can also be inferred from annual
reports from CDTI or the Ministry of Industry. But results obtained here
also show that other variables have an important weight in the participa-
Downloaded by [University of Arizona] at 22:43 18 November 2012

tion status, in particular foreign capital participation. Whether this is

mostly the result of.the application process or of the granting process
deserves further attention, as over time, and as a result of economic inte-
gration, more foreign capital participation in Spanish iirms is to be
expected. The estimated weights that experience (age and past patents) and
industry variables have for participation could be used by the public
agency to assess the performance of its selection rule.
The analysis developed here also shows that firm size remained related
to R&D intensity, whether or not a firm received public funding, but that
R&D intensity diminishes somewhat more slowly with size for partici-
pants. This effect could not have been observed with the simple regression
model pooling all firms and adding a participation indicator.
Finally, in the aggregate, public funding induced more effort by firms in
this sample, but for a sizeable proportion (about 30%)of participants com-
plete crowding out effects cannot be ruled out.
There are several ways to improve the analysis carried out here if better
data could be obtained. An obvious way is by having observations on the
same firm for several periods, including before, during and after receiving
a subsidy for participants. This should allow a better control for unob-
served firm specific effects. It would also be helpful to have better varia-
bles, in special those related to the human capital of the firm,to a h ' s
strategy and markets, and to organizational characteristics. The analysis
presented here is just a step for a better understanding of the effects of
R&D subsidies to private firms.

References
Antonelli, C. (1989). "Afailure inducement model of research and development expenditure:
Italian evidence from early 19809". Journul of Economic Belurvior and Oganirafwn,
12(2). 159-80.

* If the firm anticipated c o m t l y the agency's selection criterion, then the detenninantsof
application would be exactly the same as those of the granting decision, and there would be
no identification problem. 'he coefficients would capture entirely the agency's selection rule.
 134 ISABELBUSOM

Busom. L (1993)."Los proyectos de I+D de las empresas: Un an5isis empiriw de algunas de

sus ~ c a s nRevista . W&la de Econoin&a, Monogrdfrco Investigaci6n
- y
Desarrollo. 3945.
Capron, H. (1992). Economic QuantitativeMethodsfor the Eva!uazion of the Impact of R&D
Pmg~rvnr,Comission of the European Communities, MonitorISpear, EUR 14864EN.
CDTI (1988). Memoria 1988. Ministerio de Industria y Energia, Madrid, Spain.
Cum%, P., (1984), Timpact de la recherche et developpernent sur la productivit6 industri-
elle", Econornie et Starictique, 164,S18.
DeIbono, F. and V. Denicolo (1990)."R&D Investment in a symmetric and homogeneous oli-
gopoly: B e m d versus Coumotn. International Joumal of Industrial Organization,
Downloaded by [University of Arizona] at 22:43 18 November 2012

8(2), 297-313.
Folster, S. (1992),"Should Cooperative R&D be subsidized: an empirical analysis". mimeo,
presented at the EARE 1992 Meetings. Stuttgart.
Griliches, Z. (1986). "Productivity, R&D and Basic Research at the F i Level in the
1970's".American Economic Review. 76(1). 41-154.
Oliliches, 2. and F. Lichtenberg (1984), "RBU) and Productivity Growth at the Industry
Level: Is there still a Relationship?", in Z Griliches (de). W&D, Patents und P h t i v -
iry, Chicago, Chicago University Press, 465-96.
Hall, B. (1993). "R&D Tax Policy During the Eighties: Success or Failure?", in J. Poterba
(ed),Tax Policy and the Economy 7.NBER and MlT Press, 1-35.
Hall, B.H. and J. Mairesse (1995)."Exploring the relationship between R&D and productiv-
ity in French manufacturing firms", Joumal of Ecommetrics, 65,263-293.
Hall, B.K.J. Mairesse, L. Branstetter and B. Crepon (1998)."Does Cash Flow Cause Invest-
ment and R&D: An Exploration using panel data for French, Japanese and United States
ScientificF i n ,forthcoming in D. Audretsch and A. Thurik (eds), Innovation, Indus-
try Evolution and Employment. Cambridge University Press.
Heckman. J. and R Robb (1985),"Alternativemethods for evaluating the impact of interven-
tions. An overview", Journal of Econometrics, 30.
Holemans, B. and L. Sleuwaegen (1988). "Innovation expenditures and the role of govern-
ment in Belgium". Research Policy. 17.375-379.
Kauko. K. (19%), "Effectivenessof R&D subsidies -a skeptical note on the empirical litera-
me".Research Policy 25,321-323.
Klette, J. And J. Moen (1997), "R&D investment responses to BBrD subsidies: A theoxtical
analysis and a microeconometric studyn,mimeo, presented at TSER workshop. CEMFI.
Madrid.
Labeaga, J.M. and E. Martinez-Ros (1994). "Estimaci6n de un modelo de ecuaciones simul-
theas con variables dependientes limitadas: una aplicaci6n con dams de la industria
espa6ola", Investigaciones Econdmicas, XVIII (3).465439.
Levin, R. and P. Reiss (1984). "Tests of a Schumpeterian Model of R&D and market struc-
ture", in Z. Griliches (ed)R&D, Parents and Piducdviry, University of Chicago Ress.
175-204.
Lichtenberg, E (19841,"Lhe Relationship between Federal Contract R&D and Company
W, American Economic Review; 74(2),May 1984.73-78.
Lichtenkrg, F. (1987). "The Effect of Govenunent Funding on Rivate Industrial Research
and Development: a Re-assessment". The Journul of Industrial Economics, 36 (1). 97-
104.
Lichtenberg, F. (1988). "The Private R&D Investment Response to Federal Design and Tech-
nical Competitions". American Economic Review. 78.3.550-59.
Lichtenberg, F. (1992)."R&D investment and international productivity differences". NBER,
Working Paper No. 4161.
Mowery, D. (1995),"The h t i c e of Technology Policy", in P. Stoneman (ed), Handbook of
the Economics of lnnovarion and Technological Change, Oxford: Basil Blackwell, 513-
58.
 RBD SUBSIDIES 135

Meng, Chun-Lo and F! Schmidt (1985). "On the cost of partial observability in the Bivariate
Robit Model", lnremutionul Economic Review, 26(1). 71-85.
Nadiri, M.I. and T.P. Mamuneas (1994). "The Effects of Public infrastructure and RBtD capi-
tal on the cost structun and performance of US manufacturing industries", Review of
Economics and Statistics. Vol. LXXVI(1). 22-37.
Mamuneas, Th. P. and M.I. Nadiri (1995). "Public R&D Policies and Cost Behavior of the.
US Manufacturing Industries". NBER Working Paper 5059.
Park, W.G. (1995). "InternationalRBiD Spillovers and OECD Economic Orowth". Economic
Inquiry, XWCIKI, 571-591.
Navas, JE. (1984), "a financiaci6n pdblica de las actividades empresariales en investigacidn
y d e m l l o " . Moneda y Credilo. 37-53.
Downloaded by [University of Arizona] at 22:43 18 November 2012

Poirier, D. (1980), "Partial Observabilily in Bivariate Robit Modelsn.Joumal of Economet-

rics, 12,209-17. .
Scott, J.T. (1984). " F iversus Industry Variability in W , in Grilicbes, Z,R&D, Patcnts
ond Productivity. University of Chicago Press.
Veugelers. R and P. Vanden-Houte (1990), "Domestic R&D in the presence of multinational
enterprises", Intemationul Journal of Industrial Organization, 8. 1-15.
 BABEL BUSOM

APPENDIX IB
The data originated in a survey conducted in 1988 for a report on the par-
ticipation of Spanish firms in European research programs. Researchers at
the Universitat Autonoma de Barcelona and the Hnstitut dYAnalisiEconom-
ica (CSIC) designed the questionnaire. The sample was designed to con-
tain firms active in R&D and/or technology acquisition, including among
Downloaded by [University of Arizona] at 22:43 18 November 2012

them firms that had participated in some public R&D program. The
national public agency, CDTI, provided a list of applicants for subsidies,
from which a random sample was drawn. The rest of the sample was drawn
from listings of firmsdoing R&D. The sample is not meant to be representa-
tive of the whole population of Spanish h s , because a l l firmsin this Sam-
ple conduct R&D, while at that time at most 20 % of all Spanish firms did
so. This research focuses on the sub-population of R&D doers. In the sample
compiled by the Institute Nacional de Estadistica in 1988, which consisted
of 1232 firms doing R&D, about 19 % had more than 500 employees
(large), 34 % between 100 and 499 (medium), and 47 % less than one hun-
dred employees (small).In the sample used here, h o s t 50 % of the firms
are large, 20 % are medium and 30 % are small firms. Large firms are thus
over represented in our sample.
In addition to the descriptive statistics shown in Tables 11 and IJI for the
whole sample, it is of interest to compare the sub-samples of participants
and non-participants in the R&D program, as the second sub-sample will be
used to construct estimates of the potential RBLB effort of participants.
Participants Non-Participants

Mean Median Mean Median

R&D expenditure (M. Pta)

R&D personnel

Total employment

Age
Patents
Exports share (%)
 RBtD SUBSIDIES

APPENDIX II
TABLE IV Detembants of the Probability of Participation
Participation in the National RdrD subsidy program
Participmion

Model I MmW2 Tz:z M0&13

Effects
European-kvel
hgmnu
Downloaded by [University of Arizona] at 22:43 18 November 2012

Constant -0.61 -0.39 -0.67 1.71

(-1.2) . (0.9) (-1.4) (-3.14)
Employment -0.29 -0.30 0.12 -0.30) -0.12 0.24
(-3.2) (-3.3) (-3.3) (2.7)

Age 0.52 0.54 0.21 0.57 0.23 -0.11

(2.7) (2.9) (3.1) (-0.6)
Exportsh -0.01 -0.01

(-0.1) (-0.2)
Patents 0.44 0.46 0.18 0.45 0.18 0.05
(3.7) (4.0) (3.9) (0.4)
Public 0.64 0.55 0.22 0.67 0.27 0.44
(1.7) (1.5) (1.8) (1.3)
Foreign -0.91 -0.76 -0.30 -0.79 -0.32 -0.87
(-2.9) (-2.7) (-2.8) (-2.6)
Rice 0.31 -0.48

(1.1) (-2.6)
Regul -0.10 0.51
(-0.2) (1.2)
FRival 0.26

(0.3)
Dchemical 0.50 0.50 0.20 -0.20
(1.4) (1.6) (-0.5)
Detronics 0.51 0.51 0.20 0.79
(1.5) (1.5) (2.2)
Dequipment -0.09 0.39

(-0.4) (1.0)
 138 ISABEL BUSOM

Parlicipatwn in the National RBtD subsidy pmgmm

PartiEipatr.On in
Eumpean-level
Mod.1 l M0&12 7;;: Mo&l3 Mawid
Effects
pmg-
Downloaded by [University of Arizona] at 22:43 18 November 2012

LogLikelihood -75.5 -76.9 -76.9

Restricted L -101.8 -101.8 -101.8

x2 52.5 49.6 49.7

Pseudo It2 0.26 0.24 0.24
N 147 147 147
(t-values shown in parenthesis)
Downloaded by [University of Arizona] at 22:43 18 November 2012
Downloaded by [University of Arizona] at 22:43 18 November 2012

Ordinary Least Squams Sample Selection Likelihood

Mauaumwn

Participants NOPI-
Pafficipunts AN Participants Non- Participants Parficipants Non- Panicipants

Frival 0.22 -0.12 0.12 0.22 -0.17 0.13 -0.27

(0.30) (0.46) (0.28) (0.23) (0.41) (0.30) (0.57)

Detronics 0.45 1.85.. 1.17'. 0.45 1.84'. 0.47 1.82"
(0.33) (0.47) (0.29) (0.30) (0.42) (0.30) (0.49)
Dequipmt 0.89'. 0.61 0.74.' 0.89'. 0.58 0.86 0.55
(0.42) (0.54) (0.35) (0.37) (0.49) (0.71) (0.51)
den erg^ 0.10 0.83 0.62'~ 0.11 0.83 0.13 0.89
(0.40) (0.64) (0.36) (0.35) (0.53) (0.42) (0.84)
Dservice 1.74' 0.47 0.90.. 1.74.~ 0.46 1.79'' 0.49
(0.47) (0.55) (0.37) (0.42) (0.50) (0.47) (0.59)
Lambdaa -0.01 -0.19
(0.43) (0.79)

* In this and all remaining tables, standard errors are in parenthesis. One asterisk denotes a 10 % significance level; two asterisks a 5 %.
Downloaded by [University of Arizona] at 22:43 18 November 2012

RBtD SUBSIDIES
Downloaded by [University of Arizona] at 22:43 18 November 2012

TABLE VI Relative R&D Effort: R&D Expenditurelemployee

.-
Onfinary Least Squares Sample Selection Marimum Likelihood

Participants Non- Participanrs All Participants Non- Participants Participants Non- Participants

Constant 0.69 0.04 0.18 0.70 -0.11 0.95. -0.39

(0.30)
Employmt -0.38'.
(0.07)
Exportsh -0.01..

(0.00s)
Patent 0.22'.
(0.08)
Public 0.03
(0.28)
Foreign 0.38
(0.30)
Qquantity -0.37
(0.25)
 Ordinary Least Squares Sample Selection Ma*Lnum Likelihood
Downloaded by [University of Arizona] at 22:43 18 November 2012

Participants Non- Participants AN Participants Non- Participants Participants Non- Participants

Frival

IdeaRiv

Dchemical

Detronics

kquiprnt

Denergy

Dservice

Lambda
Downloaded by [University of Arizona] at 22:43 18 November 2012

ISABEL BUSOM
 TABLE VII Absolute R&D effort: R&D Personnel

Ordinary Least Squams Sample Selection L.ikelihood

MmPnmwn
Downloaded by [University of Arizona] at 22:43 18 November 2012

Participants Non- pariicipants All Participants Non- Participants Pmicipants Non- Participants

Constant -0.51 -0.71 -0.98" -0.71. -1.18. -0.67 -1.56.

(0.22)
Employmt 0.54"

(0.05)
Exports -0.01.'
(0.005)
Patent 0.19"

(0.07)
Quantity -0.1
Downloaded by [University of Arizona] at 22:43 18 November 2012

ISABEL BUSOM
 TABLE VIII Relative Effort:R&D personneYemployee
Oniinary Least Squans Sample Selection Maximum Liklihood
Downloaded by [University of Arizona] at 22:43 18 November 2012

Participants Non- Participants All Pam'cipants Non- Panicipanfs Participants Non- Pam'cipants

Constant 4.10 3.88" 3.62.. 3.91- 3.43- 3.93- 3.04.

(0.35) (0.50) (0.33) (0.36) (0.62) (0.43) (0.m
European -0.37 0.001 -0.10 -0.31 -0.07 -0.33 -0.08
(0.22) (0.34) (0.20) (0.21) (0.31) (0.30) (0.43)
Employmt -0.46" -0.55" -0.49.. -0.49.. -0.51" -0.48'. -0.48"
(0.05) (0.07) (0.09 (0.06) (0.c"O (0.07) (0.07)
Exportsh -0.01** 0.007 0.002 -0.01' 0.01 -0.01** 0.01
(0.005) (0.0'm (0.004) (0.004) (0.07) (0.005) (0.rn)
Patent 0.19.' 0.19 0.17.~ 0.26.' 0.04 0.24'. -0.06
(0.07) (0.16) (0.07) (0.09) (0.20) (0.12) (0.21)
Quantity -0.11 -0.21 -0.15 -0.10 -0.16 -0.10 -0.03
(0.20) (0.30) (0.18) (0.18) (0.28) (0.21) (0.31)
Frival 0.22 -0.01 0.11 0.28 -0.06 0.26 -0.09
(0.19) (0.29) (0.18) (0.18) (0.26) (0.23) (0.30)
IdeaRiv -0.18 -0.13 -0.13 -0.14 -0.13 -0.15 -0.10
(0.18) (0.28) (0.17) (0.17) (0.25) (0.22) (0.35)
Downloaded by [University of Arizona] at 22:43 18 November 2012

Onlinary Least Squams Sample Selection Maximum Likelihood E

Pam'cipants Non- P~Iikipants All Participants Non- Participants Participants Non- Pam'cipants

Dchemical 0.67" 0.57 0.70. 0.69~' 0.61' 0.69" 0.68'

(0.25) (0.37) (0.24) (0.22) (0.33) (0.26) (0.42)
Detmnics 0.88.. 1.58.' 1.19~' 0.84.. 1.59'. 0.86'. 1.46**
(0.27) (0.38) (0.24) . (0.25) (0.34) (0.32) (0.37)
Dequipmt 1.06'. 1.14.. 1.04.. 1.07.. 1.06.. 1.03" 0.87.
(0.34) (0.44) (0.29) (0.31) (0.4) (0.34) (0.52)
Denergy 0.25 0.77 0.55' 0.25 0.71 0.25 0.64 !
P
(0.32) (0.50) (0.30) (0.28) (0.45) (0.37) (0.52) m
Dservice 1.41°0 -0.68 0.14 1.39- -0.68' 1.39- -0.56 3z
(0.38) (0.45) (0.32) (0.34) (0.41) (0.33) (0.49)
Lambda 0.32 -0.58
(0.29) (0.57)
CDTI 0.47'.
(0.18)
R2 adjust. 0.68 0.58 0.60 0.68 0.58
Loglikelib. -238.5
N 70 73 143 70 73 143