Professional Documents
Culture Documents
eSI 2Li
Public Disclosure Authorized
POLICY RESEARCH
WORKING PAPER 1221
Public Disclosure Authorized
The WorldBank
PolicyResearchDepartnent
November1993
POLICY RESEARCH WORKING PAPER 1221
Summaryfindings
UsingUNESCOAatafor researchand developnient raitherthan vice versa.First, thtre is no evidencethat
(R&lD)expendiruresand personnel,Birdsalland Rhee R&D in the OFCD in the early yearsof 1970-85
document international differer.cesin R&l) a-:tivities contributedto growth durmngthe whiioleperiod.
and assessthe determinantsof these differencesand Second,the anralysisof deteriniiianitsof R&D activities
the link between R&I) and ecotiomic growth. suggeststhat levelof incomeaffectsR&D activities;
For a group of OECDcotiuitries,R&D activityand appairentlyR&D becomesimportantonly after a
economicgrowthiare correlatedfor one of their two coointryreachesa certain stage ot development.
proxies.Contrary to the findingsof Romerand For developingcountries,the authors' resultsare
Lichtenberg,however, they are not corielated acros, consistentwith the widespreadview,first proposed by
all (includingdeveloping)countries. (Gershenkron.that countries ti.it are b hind grow by
Moreover,even for OEC( countries,it appears catchingup technologic,ally, not by ad-ancing the
likelythit economicactivityaffectsR&D activity technologicalfrontier.
This paper - a product of rhe Officeof the Director,PolicvResearchDepartment- is part of a larger effort in the
departmentto assessthe sourcesof growthacrosscounitries.Copiesof tthepaper areavailablefreefromthe World Bank,
1818 H Street NW, Washington,DC 20433. Pleasecontact SushmaRajan, room NI 1-045, extension3,3747(25
pages).November 1993.
The Policy Research Working PaperSeriesdisseminatesthe findings of uworkin progressto e:courage the exchangeol ideasabout
developmentissues.An objectiveof theseriesis to get the findingsout quickly, eveni. thepresentations
are lessthan fully polished.The
paperscarry the namesof the authorsand shouldbeusedand citedaccordingly.Thefindings,interpretations.and conclusionsare the
authors' oumand should not beattributed to the World Bank,its ExecutiveBoardof Directors,or any' of its membercountries. J
Producedby the PolicyResearchDisseminationCenter
ConimentsWelcome
IN DEVELOPINGCOUNTRIES?
* World Bank and Universityof Rochester. This paper was completedwhile the first author,
now at Inter-AmericanDevelopmentBank, was Director, PolicyResearch Department,World
Bank and the second author was visiting the World Bank as a VisitingResearchFellow.
1. Introduction
from firm and industry level studies of healthy rates of private and social returns (Mansfield
basis of our adjusted data, we assess empiricallythe link betweenR&D activityand economic
growth in R&D activity and economicgrowth are positively correlated,but only across
Our cross country analysisdoes show a robust empirical link betweenR&D activitiesand the
The paper is organized as follows. SectionII describesour data. Section III sets out cross
country analysis of the growth-R&i) ink, and SectionIV sets out our analysisof R&D
(I. Data
Frascati, Italy, in 1963. The resulting "FrascatiManual" has provided the basis for OECD's
collecting and publishingR&D data on a regular basis. Since 1969, UNESCOhas published
3Pageand Pack(1993)distinguish
betweentechnological
catch-upand increasesin "technicalefficiency"and
showthat the most successfuleconomiesof East Asiahavegrownprimarilyon the basisof the former. Their
findingis consistentwith the view of Romer(1989)that growthis affectedby the transferof ideas, and the
empiricalfindingof KimandLau(1993)thattherehasbeenlittleor no increasein TFPgrowthin EastAsia,when
TFPgrowthis definednarrowlyas increasesin technicalefficiency.
'We are gratefulto Mr. MortonM. Brownof the statisticalofficein the UNESCO,Parisfor providinga
supplementarvtape.
2
We use measures of R&D inputs, rather than outputs. Within a country the difficultyof
measuringR&D inputs is well known. The concept itself is bard to define; according to the
OECD definition, the main criterion to distinguishR&D inputs from related scientific
(1981]). There are no adequate measuresof R&D output.' Much of R&D output is in the
deferse, and space, where extemalitiesare large and are difficultto measure and capture
statistically.
(i) Personnel: As a r%roxyfor the accumulatedstock of R&D capital, we use the number
(ii) Expenditures:As a proxy for the changein R&D stock, we use gross domestic
observations)if their specificationis too different from the irternationalnorm. For example,
we exclude the expenditureand manpowerdata for Greece and the manpowerdata for
Norway since they do not include R&D activitiesin private productivesectors. We exclude
Fiji, French Polynesiaand Tonga since their data relate to one research instituteonly.
At the same time, to avoid excludingtoo many countries, we permit some variation in
the R&D input definitions. The most severe exampleof this is the manpowervariablefor
developing countries, whose sample periods differ widelyacross countries. Table I shows the
sample averagesand the growth rates of our R&D inputproxies for each :ountry. The first
two columnslist the average and the annualgrowth rates of R&D expendituresas a
percentage of GDP between 1970 and 1985. The third column shows the average of the
1985. The last two coiBinnsare foi the numberof scientistsand engineersnot normalizedby
population.'
Followingthe OECD convention,Table 1 puts OECD countries into three groups (major,
medium, and small R&D performner)on the basis of the amountof R&D expenditures. The
groups based on R&D expendituresare not necessarilythe same as the groups based on R&D
funds and b,- type of R&D activities. The first two columns in Table 2 report the average
expenditurcs.'" The last two columnsreportthe awerag. ratios of basic R&D expenditures
countries are negligible.The average R&D expenditureas a percentageof GDP between 1970
aa,d 1985 is only 0.36 percent for developingcountries,just 25 percent of the average for
OECD countries of 1.42 percent. Between1970and 1975, the average number of scientists
and engineers employed in R&D activitiesis 172 per million populationfor developing
Moreover, given that the developingcountriesin our sampleare almost certainly those with
more R&D activity than those not in the sample, the average R&D activity for all developing
Second, manpowerproxies indicatethat both the average growth rates and the variance of
and 1985, the average growth rates of the number of scientistsand engineersper rnillion
populaticnare 3.84 percent for 18 OECD countriesand 6.04 percent for 11 developing
countries. S..andarddeviations are 1.97 and 4.86, respectively.During that period, the growth
rates vf R&D expendituresas a percentageof GDP are not very different between OECD
countries and developingcountries. They are 2.27 percent for 18 OECD countries and 2.42
and Korea are similar in size to those for low incomeOECD countriessuch as Spain,
Portugal, and Iceland. The growth rates of the former are greater.
Thirdly, though high R&D investmentis often regarded as a major factor explaining
even Japan's ratio of applied R&D is not higher than the OECD average. Its ratio of
6
Though not reported in Table 1, the time series of our data shows that R&D expenditures
have grown as rapidly a. or more rapidly than G!)P in the majorityof countries; the R&D
data also reveal that R&D expendituresstarted to accelerate from the late 1970s,after
where qi, ki, RI, and IIare the average annual growth rates of real GDP, capitalstock, R&D
capital stock, and employment,respectively,of country i. Y* denotes initial per capita CDP
catch-up effects. The reduced form (1) can be easily derived by introducingR&D capital
stock in Dowrick and Nguyen's [1989] productionfunction. Equation (1) can be rewrittento
"Thisapproximation
is exactif capital-output
ratiosareconstantacrosscountriesandovertime.Withthesame
logic, if R&D capital-outputratios are constant, we can proxy R in equation(2) by the averageR&D expenditures
as a percentageof GDP.
7
(2) q! = ao + a,(I/Q); + a2 R, + a31i - a4 InY1j*+ e;
variables are summarizedin the appendix.For R1, we use three measures: the growth rates of
(RNDi_Y). Since our R&D proxies are availableonly after 1970, the data are the sample
averages or the growth rates between 1970and 1985.LNDIFF70 is the log of the gap in
The first three columns of Table 3 report the regressionresults for the full sample. The
next four columns report the results for OECDcountriesand developingcountries. Consistent
[1991]), we find that high investmentrates are stronglyassociatedwith high GDP growth,
As for the growth-R&Dnexus, our results are mixed. The average growth rates of the
GSC17085),are significantin explainingGDP growth rates for the full sample and O2CD
countries, but not for the developingcountries. I' e take this evidence at face value, it
of economicdevelopment.
8
However, the average R&D expenditureas a percentageof GDP (RND _Y) is
expenditures into sub-fieldsdoes not alter the rcsult. table 4 eeportsregressionresults when
How can our mixed result be best interpreted?One might argue that the expenditure
variable is a better proxy since manpowerirr ices couldbe proxying the changesin human
capital instead 3f R&IDcapital. To check the validityof this argument,we include proxies for
human capital accumulationused in Barro [1991], the primary and the secondaryschool
enrollment rates, in the equation (2). The second column in Table 4 showsthat both the
R&D activities is thus probablynot simplyplaying the role of (less educated)human capital.
Of course our proxies may represent highlyeducatedhuman capital, but it would be hard to
9
As mentioned in Section 11,since our selectioncriterion ziows a certaindegree of
invariant, comparing the growth rates of our proxies would be less problematic. This
4
argument suggests that GSCP7085or GSC17085are better proxies than RNDIyY."
At the same time, the results in Table 3 cannot be interpretedas demonstratinga causal
capital stock could simply followeconomicgrowth. In order to control for this possible
Following Barro [1991] and other cross country studies, we regressedthe growth rates of
real GDP per capita between 1970and 1985on initiallevel of real GDP (GDP70), initial
primary and secondaryschool enrollmentrates (PRIM, SEC), and the ratio of government
spending to GDP (GOV), together with a proxy for initial R&D capital stock,'5 the average
We use the average value between 1970and 1975 becausesome of our sample countriesdo
10
not have the data for the year 1970. Usingaverage valueshas the additionaladvantageof
The resuit in column 1, Table 5 showsthat SCI_P is not significantin explainingthe per
capita GDP growth rat-. In order to check whetherthis result is due to a different
ratio and GSCP7085in the regression. We confirmthe previous result that the average
stock, SCI P, is insignificant.The third columnreports the same conclusionfor the OECD
countries only.'6
The last three columns in Table 5 test parameterstabilityover time. Given a possible lag
in the effect of R&D on economicgrowth, the fourthcolumn regresses per capita GDP
growth rates between 1975and 1985on the sameset of independentvariables, whose initial
values are for the year 1970. Also, from previouscross country studies of growth,
we know that the fit of the regression is muchbetter if the sample periods are extended back
GDP growth rate for the period 1960 to 1985. and the initialvalues for independentvariables
(except SCI_P) are the flgures in the year 1960. Note that the estimatedcoefficientof SCI_P
In both cases, however, SCI P is not correlatedwith economicgrowth. The results for OECD
The finding that our proxies for initial R&D activities are not correlated with economic
growth contrasts sharply with Romer [1989] and Lichtenberg[1992] who use the same
UNESCO data set. Romer [1989] finds a positive correlationbetweenthe initial number of
Lichtenberg [1992]finds that the ratio of privately funded R&D expendituresto GNP is
that have just one annualobservationon the R&D expenditure/GNPratio, and uses the one
observationas an average for the whole sampleperiod. Availabledata for other countriesare
often too concentratedaround certainperiods to represent an average value for the period
(Table 3) is capturing the positive effect of economicgrowth on R&D activities-- tne reverse
[1977])At face value, our results suggestthat high private returns to R&D expendituresin
micro-studiesdo not reflect high social returns. They may result from enhancedmonopoly
power of a firm at the cost of its competitors.Alternatively, it may be that evidenceof high
returns fromr,micro studies of firms and industriesin highly developedcountries, mostly the
UnitedStates, does reflect social gains, but such gains to R&D activities become important
in R&D expenditures as a percentageof GDP. The issue itself is interesting. In addition, this
differences in national specificationsare a major reason for our previous finding, it is less
13
In Table 6, the dependentvariablein the first two regressionsis average R&D
expendituresas a percentageof GDP between 1970 and 1985 (RNDI_Y). The results show
that the initial level of per capita GDP (GDP70)and the initial numberof scientists and
country differentialsin R&D expenditures.We find that GDP70 or SCI_P alone can explain
61 percent or 75 percent of the cross country variation. It is interestingto note that the
average share of physical investmentin GDP is not a significantfactor once the initial level of
in explainingthe variation of RN'D1_Y,and their inclusiondoes not alter the results. The
The finding that the R&D expenditureshare is strongly relatedto the level of income and
the level of initial R&D stock is not surprising. A country with higher per capita GDP (larger
productivityof physical capital. In contrast, the positive effect of initialR&D stock suggests
that the marginalproductivityof R&D capital is not diminishing,i.e., there are constant or
increasingreturns to R&D investment. This is consistentwith the view that huge fixed costs
The third and the fourth columnsreport the results when privatelyfunded and government
funded R&D expendituresto GDP ratios are used as dependentvariables, respectively. They
show qualitativelythe same results with those for total R&D expenditures.The finding is not
14
surprising, consideringthe high correlationcoefficient(0.70) betweenprivatelyfunded and
The last two colurrns in Table 6 show typicalregression results when the dependent
variable is the growth rates of the numberof scientistsand engineers per millionpopulation
(GSCP7085). As expectedfrom our previous regressionsin Table 3 and 5, the per capita
we used in this paper are not robustlyrelated to GSCP7085using Levine and Renelt's [1992]
of income, GDP70, which is a major determinantof R&D expenditures.Ex post, this finding
proxies for R&D activities(see footnote 14). We believe that R&D activitiesare income-
elastic and led by high incomecountries, consistentwith the results in columns 1-4.
Therefore, at least for the OECD countries, expendituredata appearsto provide a better
proxy for growth in R&D activitiesthan manpowerdata. At the same time, the finding in
columns 5 and 6 confirmsour earlier suppositionof reverse causality, i.e., growth of income
VI. Conclusions
Using the UNESCOdata for R&D expendituresand personnel, this paper documents
economic growth, and the deterninants of R&D differentials.Within our group of OECD
15
countries, R&D activity and economicgrowth are correlatedfor one of our two proxies.
Contrary to the findings of Romer [1989] and Lichtenberg[1992], however, they are not
is likely that economicgrowth affects R&D activityrather than vice versa. First, there is no
evidence for OECD countries that R&D activityat the beginningof the period 1970-85
contributedto subsequentgrowth during the period. Second, our analysisof the determinants
of R&D activities shows a robust link betweenthese activitiesand the level of income,
suggestingthat R&D activities become importantonly after a country reaches a certain stage
view, first proposedby Gerschenkron, that countriesthat are behind grow by catchingup
16
Appendix:Definitionsof Variables in Table 1-6
AVG EX70: Averageratio of exports to GDP for the years 1960-85;from World
Bank BESD.
GrP70: GDP per capita in 1980 $US, for the year 1970; from World Bank BESD.
17
References
Mankiw, Gregory, David Romer, ani David Weil, "A Contribution to the Empirics
of EconomicGrowth," QuarterlyJournal of Economics107, (May 1992), 407-38.
1o
Page, John and Howard Pack, "Accumulation,Exports, and Growthin the High
PerformingAsian Economies," presentedat the Carnegie-RochesterConferenceon Public
Policy, 1993.
19
TABLE : R&D INPUTS
OECD
Countries
Major R&D
Performer
U.S. 2.44 0.79 2523.6 1.30 531516 2.27
Medium R&D
Performer
Italy 0.84 1.85 605.3 4.45 33054 4.83
Small R&D
performer
Austria 0.99 4.93 617.4 4.67 4640 4.78
LDCs
Argencina 0.38 3.23 289.7 1.84 7300 3.38
20
tl] Expenditure in (2] Researchers per (33 Number of
% of GDP mill. population Researchers
21
TABLE3: R&D and Economic Growth(1)
22
i.able 4: R&D and Economic Growth
GSCP7085 0.0197
(2.93)
SEC 0.0197
(1.97)
PRIM -0.004
(-0.367)
R2 0.0683 0.81Q
D. W 1.63 1.305
L of observation 18 18
23
Table 5: Growth 4 Initial R&D Capital
AVGI7085 0. 1257
(2.496) _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
GSCP7085 0.0027
(2.810)
R_ _ 0.363 0.601 0.485 0.327 0.530
#of obs. 36 29 20 36 36
24
Table 6: Determinants of R&D Expenditures
2
R 0.65 0.84 0.48 0.63 0.3 _ 0.42
# of obs. 29 25 21 21 26 26
25
Policy Research Working Paper Series
Contact
Thtle Author Date for paper
WPS1218 Implementation
of TradeReformin John Nash November1993 D. Ballantyne
Sub-SaharanAfrica: HowMuchHeat 37947
and HowMuchLight?
Contact
Title Author Date for paper