Professional Documents
Culture Documents
Empirical Tests
Author(s): Jim Granato, Ronald Inglehart and David Leblang
Source: American Journal of Political Science, Vol. 40, No. 3 (Aug., 1996), pp. 607-631
Published by: Midwest Political Science Association
Stable URL: http://www.jstor.org/stable/2111786 .
Accessed: 17/11/2013 18:13
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
Midwest Political Science Association is collaborating with JSTOR to digitize, preserve and extend access to
American Journal of Political Science.
http://www.jstor.org
Introduction
Do culturalfactorsinfluenceeconomic development?If so, can they
be measuredand theireffectcomparedwiththatof standardeconomicfac-
torssuch as savingsand investment?This articleexaminestheexplanatory
powerof thestandardendogenousgrowthmodel and comparesit withthat
of twotypesof culturalvariablescapturingmotivationalfactors-achieve-
mentmotivationand postmaterialist values. We believe thatit is not an
either/orproposition:culturaland economic factorsplay complementary
roles. This belief is borne out empirically;we use recentlydeveloped
econometrictechniquesto assess the relativemeritsof these alternative
explanations.
MaryBange,MikeBratton,
*We wouldliketothankPaulAbramson, DarrenDavis,Mark
Jones,KenMeier,andsomeanonymous
reviewers. withnamesinalpha-
Equal authorship,
beticalorder.
AmericanJournalofPoliticalScience,Vol. 40, No. 3, August1996,Pp. 607-31
? 1996bytheBoardof RegentsoftheUniversity ofWisconsin System
Culture,MotivationalFactors,and EconomicGrowth
We firstdiscusstheliterature thatviews achievementmotivationas an
essentialcomponentin theprocessof economicdevelopment,and thenwe
explorehow culturalmeasuresfromtheWorldValues Surveycan be used
to examinethe effectof motivationon growth.
The motivationalliteraturestressesthe role of culturalemphasison
economic achievement.It grows out of Weber's (1904-1905) Protestant
Ethic thesis.This school of thoughtgave rise to thehistoricalresearchof
Tawney(1926, 1955), case studiesby Harrison(1992), and empiricalwork
by McClelland et al. (1953) and McClelland (1961) on achievementmoti-
vation.Inglehart(1971, 1977, 1990) extendsthisworkby examiningthe
shiftfrommaterialist to postmaterialist
value priorities.Althoughprevious
workmainlyfocuses on the politicalconsequencesof these values, their
emergencerepresentsa shiftaway fromemphasison economicaccumula-
tionand growth.These "new" values could be viewedas theerosionofthe
Protestant Ethicamongpopulationsthatexperiencehighlevels ofeconomic
security.
We suggestthatWeberis correctin arguingthattheriseof Protestant-
ism is a crucialeventin modernizingEurope. He emphasizesthattheCal-
vinistversionof Protestantism encouragesnormsfavorableto economic
achievement.But we view therise of Protestantism as one case of a more
generalphenomenon.It is important, notonlybecause of thespecificcon-
tentof earlyProtestant beliefs,butbecause thisbeliefsystemundermines
a setofreligiousnormsthatinhibiteconomicachievementand arecommon
to mostpreindustrial societies.
Preindustrialeconomiesare zero-sumsystems:theyare characterized
The Data
The WorldValues Surveyasks representative nationalsamplesof the
publicsin a numberof societies,"Here is a listof qualitieswhichchildren
can be encouragedto learnat home. Which,if any,do you considerto be
especiallyimportant?"This listincludesqualitiesthatreflectemphasison
autonomyand economic achievement,such as "thrift,""saving money
and things,"and "determination."Otheritemson thelistreflectemphasis
on conformity to traditional
social norms,suchas "obedience," and "reli-
gious faith."
We constructan index of achievementmotivationthatsums up the
percentagein each countryemphasizingthefirsttwo goals minustheper-
centageemphasizingthelattertwogoals. This methodofindexconstruction
controlsforthetendencyofrespondents in some societiestoplace relatively
heavyemphasison all of thesegoals, whilerespondentsin othercountries
mentionrelativelyfew of them.
Figure 1 shows the simple bivariaterelationshipbetweenthis index
and ratesofpercapitaeconomicgrowthbetween1960 and 1989.1 The zero-
pointon theachievementmotivationindexreflectsthepointwhereexactly
as manypeople emphasizeobedienceand religion,as emphasizethrift and
determination.As we move to theright,thelattervalues are givenincreas-
ing emphasis.A givensociety'semphasison thrift and determination over
obedienceand religiousfaithhas a strongbivariatelinkagewithits rateof
economic growthover thepast threedecades (r = .66; p = .001).
Though often stereotypedas having authoritariancultures,Japan,
China, and South Korea emergenearthepole thatemphasizesthrift more
'Data sourcesandvariabledescriptions
arecontained
in Appendix
Table 1.
EastAsia
Japanl
5
.0
0 4
C)
0 pi
d)
San Nrway Fini2
Baseline EndogenousGrowthModel
Neoclassical growthmodels today owe much to the work of Solow
(1956) and Swan (1956). The essentialfeatureofthesemodelsis theirfocus
on savings,populationgrowth,and shiftsin technology.Productionfunc-
tionsdependon shiftsin these "exogenous" variables.For example,one
could tracetheeconomicgrowthconsequencesresultingfroma shiftin the
rateof saving,thepopulationgrowthrate,or technology.The weaknessin
thesemodels,however,is thattheyshow a paradoxicalsteadystateresult.
In thesemodelsaggregatesavingsproducea level ofcapitalformation such
thatgross investment exceeds depreciation,and therebyincreasescapital
per worker.Consequently,at thelimit,themarginalproductof capitalde-
clines to the pointwhere the savings (revenue)generatedby the capital
falls to a level just large enough to replace old equipmentand provide
machinesfornew workers.The steadystateresultis an unchangingstan-
dard of living.2
This latterresultis clearlynot supportedby evidence fromthe real
world.In timeeconomistsbegan searchingforways to augmenttheneo-
classical model thatwould allow sustainablegrowthand increasesin the
standardof living. These models have been termedendogenousgrowth
models. At the heartof the endogenousgrowthliteratureis an emphasis
on the productivity of the population(Lucas 1988; Romer 1990). Unlike
the "old" neoclassical models,endogenousgrowthmodels show thatre-
produciblecapitalneed nothave decreasingreturns to scale. Growthcan be
sustainedin endogenousgrowthmodels.In particular, theyassumeconstant
returnsto scale to a broad rangeof reproducibleinputs,includinghuman
capital.
The two leadingschools of thought, however,differin theiremphasis.
MultivariateAnalysis
Our empiricalapproachis straightforward:
we beginby estimating
(via
OLS) a baselineendogenous modelthatincludesvariablesidenti-
growth
3Theindicator
forhumancapitalinvestment is thenumber ofstudents
enrolledat pri-
maryandsecondaryeducationinstitutions
relativetothetotalpopulation
ofthatage group.
Theindicator
forphysical
capitalinvestment
is theratioofrealdomestic
investmenttoGDP.
Thedefinitions
andsourcesforthisdata,whichwe uselater,arefoundinAppendix Table 1.
EmpiricalModels: Encompassing
ComparingCompeting Results
Both the economic and culturalmodels give similargoodness-of-fit
performance.Each model's regressorsare statisticallysignificant.Yet,
whichmodel is superior?Or do bothmodels possess explanatoryfactors
thatare missingin theother?In Table 1 theSchwarzcriterion(SC) favors
4Thenations included
inthemultivariate
analysis
are:Austria,Belgium, Brazil,Canada,
China,Denmark, Finland,
France,Germany, GreatBritain,
India,Ireland,
Italy,Japan,Korea,
Mexico,Netherlands, Nigeria,
Norway, SouthAfrica,Spain,Sweden,Switzerland, Turkey,
UnitedStates.
'This is a checkforspatialcorrelation
betweentheerrors ofthecases.
Table 2. EncompassingTests
Model 1 4 Model 2 Form Test Form Model 2 4 Model 1
3.34 t(l) JA-Test t(l) 2.44
5.42 F(2,17) JointModel F(4,17) 2.87
Notes: * = p > .05.
guishable(parameterencompassing).To testforparameterencompassing
we employa jointF-test.This testcombinesall instruments fromtwocom-
petingmodelsintoa largegeneralmodel(artificial nesting).Commonvari-
ables are removedfromthegeneralmodel to avoid multicollinearity. The
testimposeszero restrictions on theinstruments of each (sub)modelto de-
termineifeitherset of instruments altersthesum of squarederrorssignifi-
cantlyfromthegeneralmodel(see Appendix).In a testof Model 1 4 Model
2 acceptingthe null indicatesthe zero restrictions on Model 2 have no
statisticaleffecton thejoint model. Model 1 4 Model 2 in thiscase. In a
progressiveresearchstrategycontext,thisparametersubstitution testsif
one theoryexplains morethana rival's explanation.
Variance encompassing,on the otherhand,requiresthatthe "new"
parameterrestrictions be at least as efficient-intermsof theresidualvari-
ance-as theoriginalmodel.8In short,a progressiveresearchstrategy not
onlynecessitatesthatnew and novel factsbe putforthby thesuperiorthe-
ory,but thatthistheoryis also a moreaccurateexplanation.
Returning to themodelsin Table 1,culturalvalues clearlymatter.With
theencompassingprinciplein ourarsenal,we compareendogenousgrowth
and culturalexplanationsforeconomicgrowth.The encompassingresults
presentedin Table 2 are definitive: The JA-testforvarianceencompassing
is significant
and indicatesthatbothmodelsencompasseach other.Neither
model is an "efficient"substitute fortheother.In addition,bothmodels'
parameter encompasseach otheras indicatedbysignificant F-tests.In short,
bothmodelsexplainaspectsofgrowththattherivalcannot.The implication
is straightforward: growthratesare best understoodas a consequence of
botheconomicand culturalfactors.
Whathappenswhenwe combinetheeconomicmodelwiththecultural
model?The resultsof thisexperiment are containedin Model 3. Beginning
withtheendogenousgrowthvariables,addingthevariablesfromModel 2
8Theencompassing principle
is mostcommonly usedforlinearmodels.To assessvari-
ance encompassingwe use theJA-test(Davidsonand MacKinnon1981) as modified by
Godfrey(1984).Thistestinvolvesconstructing
Y-hat'sfortwocompeting models,andthen
sideof itsrival.
addingtherivalY-hat'sto theright-hand
SensitivityAnalysis
Table 1 contains an additional In Model4 we eliminate
specification.
thethreeinsignificantvariablesfromModel3-those forpostmaterialism,
investment, andsecondary schoolenrollment-to checkthestabilityofthe
remaining parameters. Model 4 is themostparsimonious and efficient
model,explaining 70% ofthevariance inpercapitagrowth rateswithonly
threevariablesandgenerating a Schwarzcriterionvalueof -.352. In addi-
tion,theresidualsare wellbehavedand themodelpassestestsforserial
(spatial)correlation,
normality, andheteroskedasticity.
Aretheresults inModel4 theconsequence ofeitherhighly influential
observations or theproductof specificvariablesselected?We ask these
questionsbecauseJackman (1987) demonstratesthatremovalof evena
singleinfluentialcase mayreduceparameter estimatesto insignificance.
Levineand Renelt(1992) takea different approachand showhowalter-
ationsinthesetofvariablesincludedina modelnotonlychangethestan-
darderrorof a variableof interest butalso cause theparameter estimate
tochangesigns.We usebothoftheseapproaches toevaluatetherobustness
of Model4.
-+1 6
0
0) +1.2 Canada
-
._E Finland
o +0.8
C - Norway
o United 0Sweden
a) +0.4 - States
Great
Denmark
,o0
S
co Britaln.Belgium Austria
Switzer - *
0. 0 Italy land /
co 0
o - Spain Turkey Netherlands
a- ~~0 0 0
d) -0.4 -
. Mexico
_ Bra
O _ Germany
0 -0.8 *-
Ireland India
I- 1.2
. 0 Nigeria
France
a) SouthAfrica
E -1.6 l IlI l I I I I I I I I I I I I I
-60 -40 -20 0 + 20 + 40 +60 +80 +100
Public's score on 4-itemAchievement MotivationIndex
'?Chatterjee
andHadi(1988) suggest thecutoff forCook'sDistanceis defined as Di >
41(n - k - 1). To be more cautious (due to the small sample size) we use Di > 41n.
"A relateddiagnostic,
DFFITS,confirms theinfluenceofthesetwoobservations. The
DFFITS valueforKoreais 1.54andfortheUnitedStatesis .87.Thecutoff pointforDFFITS,
as suggested byBollenandJackman (1985) is 2* (k/n)"2.
'2Alongwiththeseoverallmeasures therearediagnostics thatexaminetheinfluence
ofindividualcasesonspecificparameter estimates.Themostpopularis theDFBETA which
looksat theeffecton eachcoefficient ofdeleting theobservations one at a time.We found
thatno case exertsundueinfluence on thevariablewe areprimarily interestedin,thefour
itemachievement motivationindex.
< .34 andequal to .34/1DFFITSI forIDFFITSI > .34. The cutoff of .34
is usedbecauseWelschsuggeststhatit providesforapproximately 95%
asymptoticefficiency.Theresultsofusingthisbounded-influence estimator
are in Equation4d. Again,we findthattheparameter estimatesare not
muchdifferent fromthoseobtainedusingOLS. Note,however, thatthe
weightsassignedby thebounded-influence estimatorare a resultof the
valueforDFFITS.
The finalestimation technique we use is a nonparametric approach:
thebootstrap.Bootstrap resampling treatsthesampleas a population and
resamplestheresidualswithreplacement a specifiednumberof times15
(Mooneyand Duval 1993;Stine1990).Equation4e is basedon residual
resampling using1,000replications.16 The parameter estimates and stan-
darderrorsbasedontheresidual resampling areveryclosetothoseobtained
withOLS.
Thesecondrobustness checkis concerned withtheeffect ofotherpos-
sibleexplanatoryvariablesonourparameter estimates.
Inordertoascertain
whether theseparameter estimatesare"robust"toalterations inthecondi-
tioningsetofinformation, we followLevineandRenelt(1992) andinclude
a setof variablesin Model4 anddetermine whetherthese"conditioning
alterthecoefficients
variables"significantly orstandard errorsonourvari-
ablesofinterest.
Theconditioning variablesweuse,as suggested byLevine
andRenelt(1992),arethegrowth rateofdomestic thestandard
credit, devi-
ationof domesticcreditgrowth, theaverageinflation rate,thestandard
deviationoftheinflation rate,thegrowth in government consumption ex-
theaveragenumber
penditure, ofrevolutions andcoups,anda dummy vari-
able indicating
exportorientation."7
The findings,notreported here,indicatethatwhilethecoefficient on
thefouritemindexofachievement motivation doesdecreaseto 1.73,itis
(t = 2.8).18 In short,we have a greatdeal of
still statisticallysignificant
in theparameter
confidence estimates
andstandard in Model4.
errors
Conclusion
Theideathateconomicgrowthis partly
shapedbyculturalfactorshas
encountered resistance.
considerable is be-
One reasonforthisresistance
"Notethatthereis a significant
difference
between resamplingwithrandom regressors
andresampling withfixedregressors.
We resample becausewe haveassumedthat
residuals
regressorsin ourmodelarefixedin repeatedsamples.See Stine(1990) fora discussion.
'6Duetothesmallsamplesize,we "fattened" bydividing
theresiduals by((1 -k)ln)
We also used 10,000replicationsandfoundalmostidentical results.
"We couldnotincludeChinainthisexercisesincedataformostoftheseconditioning
variablesarenotavailable.
'8Thespecificsofthissensitivity as wellas theobtained
analysis, coefficient
estimates
andstandard areavailablefromtheauthors
errors, uponrequest.
20Institutional
factors
suchas regime
typeandproperty rights
havealsobeensuggested
as importantdeterminants
ofeconomicgrowth (Helliwell1994;Leblang1996).
APPENDIX
The F-test
To show thatModel 1 4 Model 2 do the following:
Al) Estimatethe joint specificationof Model 1 and Model 2 below. Save the
"unrestricted" residual sum of squares (RSSu):
Y = aX + FZ + j*
(RSSr - RSSu)lk2F(k2, n - k)
RSSul(n - k)
APPENDIX
Table 1. Data Used in EconomicGrowthRegressions
Country Growtha GDPb PrimarycSecondarydInvestmenteFourItem' Postmaterialg
Austria 3.141 3.908 1.05 0.5 0.24373 0.46 2.11
Belgium 3.0639 4.379 1.09 0.69 0.19595 0.22 2.02
Brazil 3.2383 1.313 0.95 0.11 0.20599 -0.32 1.67
Canada 3.0608 6.069 1.04 0.52 0.201 0 2.14
China 5.5 0.567 0.75 0.41 0.20163 0.9 1.36
Denmark 2.4935 5.49 1.03 0.65 0.21627 0.2 1.99
Finland 3.5184 4.073 0.97 0.74 0.25217 0.38 2.23
France 2.9729 4.473 1.44 0.46 0.2224 0.09 2.04
Germany 2.7082 5.217 1.33 0.53 0.20923 0.52 2.14
GreatBritain 2.1637 4.97 0.95 0.67 0.15317 -0.01 2
India 1.9398 0.533 0.61 0.2 0.19982 -0.46 1.58
Ireland 2.9652 2.545 1.1 0.35 0.22252 -0.44 1.96
Italy 3.5253 3.233 1.11 0.34 0.22909 -0.1 2.07
Japan 5.5539 2.239 1.03 0.74 0.31723 0.82 1.81
Korea 6.6378 0.69 0.94 0.27 0.2493 0.47 1.66
Mexico 2.26 2.157 0.8 0.11 0.20675 -0.15 1.86
Netherlands 2.3531 4.69 1.05 0.58 0.19853 0.13 2.26
Nigeria .7517 0.552 0.36 0.03 0.147 -1.24 1.67
Norway 3.551 5.001 1.18 0.53 0.29782 0.1 1.81
SouthAfrica 1.428 2.627 0.89 0.15 0.2555 -0.46 1.73
Spain 3.6954 2.425 1.1 0.23 0.22484 -0.24 1.94
Sweden 2.542 5.149 0.98 0.55 0.21237 0.5 2.09
Switzerland 1.9991 6.834 1.18 0.26 0.25747 -0.03 2.1
Turkey 2.8506 1.255 0.75 0.14 0.19792 -0.19 1.95
UnitedStates 2.0976 7.38 1.18 0.86 0.13906 -0.28 2.06
aGrowth:Growth rateofrealpercapitaGDP from1960to 1989.Source:LevineandRenelt(1992).
bGDP:The 1960valueofrealpercapitaGDP (1980 base year).Source:LevineandRenelt(1992).
cPrimary:Thenumber ofstudentsenrolled
inprimary schoolgradelevelrelative
tothetotalpopulation
ofthatage groupin 1960.Source:LevineandRenelt(1992).
dSecondary:Thenumber ofstudents
enrolledinsecondaryschoolgradelevelrelativetothetotalpopula-
tionofthatage groupin 1960.Source:LevineandRenelt(1992).
'lnvestment:
Averagefrom1960to 1989oftheratioofrealdomestic investment (private
pluspublic)
to realGDP. Source:LevineandRenelt(1992).
'Fourltem:FourItemAchievement MotivationIndexcomprisedof(Thrift+ Determination) - (Obedi-
ence + ReligiousFaith).Source:WorldValuesSurvey(1990).
Meanscoreofpostmaterialism.
8Postmatefialism: Source:WorldValuesSurvey(1990).
APPENDIX
Table 2. Diagnosticsand Case Weights
Country Rstandarda Rstudentb Cooks Distc DFFITSd RobustWe BoundWI
Austria -0.9495 -0.947182 0.019 -0.2736804 0.9344327 1
Belgium -0.2202 -0.2152011 0.001 -0.0504492 0.9968488 1
Brazil 0.0193 0.0188692 0.000 0.0078119 0.9990563 1
Canada 1.6618 1.740317 0.110 0.6955425 0.7999728 0.48882705
China 0.1369 0.1336649 0.003 0.1037808 0.98891 1
Denmark -0.1206 -0.1177412 0.000 -0.0409734 0.9985626 1
Finland 0.1797 0.1755344 0.001 0.0541456 0.9949721 1
France -1.1715 -1.182587 0.150 -0.7829324 0.9146342 0.43426483
Germany -1.8565 -1.981786 0.146 -0.8162843 0.7501034 0.41652156
Great Britain -0.1063 -0.1037699 0.000 -0.0342944 0.998759 1
India -1.0073 -1.007698 0.055 -0.4691486 0.9372047 0.72471705
Ireland 0.2852 0.278949 0.005 0.1360241 0.9923928 1
Italy 0.5576 0.5483339 0.007 0.1696072 0.9692861 1
Japan 0.6540 0.6448598 0.027 0.3214996 0.946776 1
Korea 2.6414 3.154506 0.417 1.54147 0.4755642 0.22056864
Mexico -0.8926 -0.8881056 0.015 -0.2475407 0.9416614 1
Netherlands -0.6922 -0.6834447 0.007 -0.1679137 0.9590508 1
Nigeria 0.3351 0.3279361 0.029 0.3349275 0.994396 1
Norway 0.9402 0.9375238 0.017 0.2619067 0.9240916 1
South Africa -1.2503 -1.268319 0.039 -.4015923 0.8742813 0.84662985
Spain 0.7417 0.7335156 0.024 .3084299 0.9493236 1
Sweden -0.9887 -0.9882439 0.049 -0.4408704 0.9296764 0.77120165
Switzerland 0.1883 0.1839838 0.002 0.0804552 0.9991782 1
Turkey -0.3363 -0.3291797 0.003 -0.1098775 0.9953059 1
UnitedStates 1.5000 1.549266 0.176 0.8659872 0.8714244 0.39261552
aStandardizedResiduals
IStudentizedResiduals
CCook's Distance
dDFFITS
eRobust RegressionWeights
'Bounded-InfluenceWeights
REFERENCES
Abramson,Paul, and Ronald Inglehart.1995. Value Change in Global Perspective.Ann
of MichiganPress.
Arbor:University
and Using Regression.BeverlyHills: Sage.
Achen, Christopher.1982. Interpreting
Alwin,DuaneF. 1986."ReligionandParental
Child-Rearing Evidenceof a
Orientations:
Catholic-Protestant
Convergence." AmericanJournalof Sociology 92:412-40.
Baker,KendallL., RussellDalton,andKai Hildebrandt.
1981.GermanyTransformed.
Cam-
bridge:Harvard UniversityPress.
Barro,Robert.1991."EconomicGrowth ina CrossSectionofCountries."
Quarterly
Jour-
nal of Economics 106:407-44.
andRobert
Bollen,Kenneth, Jackman.
1985."Regression AnExpository
Diagnostics: Treat-
mentof Outliersand InfluentialCases." Sociological Methodsand Research 13:510-
42.