language policy and economic development: delving into the...
TRANSCRIPT
Language Policy and Economic Development:Delving Into the Black Box of Fundamental
Institutions
David Laitin (Stanford University)Rajesh Ramachandran (Goethe University)
Noah Rosenberg (Stanford University)
SSDEV 2013July 16th, 2013
David Laitin, Rajesh Ramachandran and Noah Rosenberg
Motivation and question
I One of the defining feature dividing “developed” from“developing” states -choice of official language.
I “Developed” states - official language belongs to (majority)indigenous group/s in the country.
I “Developing” states official language belongs to no indigenousgroup in the country and distant.
I Literature highlights importance of institutions for growth:I However those identified are equilibrium outcomes (rule of law,
quality of govt., protection against expropriation) rather thanfundamental causes.
I In this paper try explore if language institutions are afundamental source for growth and development.
David Laitin, Rajesh Ramachandran and Noah Rosenberg
Current approach
I Construct a measure that captures distance from andexposure to the official language.
I Show that our empirical measure is robustly correlated withvariety of indicators of development.
I Theoretical channel - Argue that language choices imposevarying costs of participation depending on languagerepertoire of linguistic groups and hence:
I Determines who has access to power, wealth and prestige insociety.
I Provide numerous historical examples supporting the aboveclaim.
I In a cross-country regression framework, estimate impact ofdistance and exposure to language policy on GDP per capita.
David Laitin, Rajesh Ramachandran and Noah Rosenberg
Construction of measure
David Laitin, Rajesh Ramachandran and Noah Rosenberg
Measure of distance and exposure
I Language choices determines cost of participation.I Cost of participation for group i dependent on:
I Distance of group i from official language o.I Distance of other groups (i 6= j) from official language (the
exposure effect)
I Operationalize the measure using language family trees fromEthnologue.
David Laitin, Rajesh Ramachandran and Noah Rosenberg
Distance measure and example
I Distance between any two languages i and j defined as:dij =
1− ( # of common nodes between i and j12 (# of nodes for language i+# of nodes for language j)
)λ
I λ is a parameter that determines how fast the distancedeclines as the number of shared branches increases.
I Example Bawaen and IndonesianI Both belong to the Austronesian Language Family.I Share 3 common nodes.I Bawean 5 nodes; Indonesian 7 nodesI dij = 1− ( 36 )
λ
I Example Spanish and IndonesianI Different language families - dij = 0.
tree
David Laitin, Rajesh Ramachandran and Noah Rosenberg
Country level measure
I Consider all language groups comprising at least 1% ofpopulation (data from Fearon (2003))
I Calculate distance from official language (dio) for eachlinguistic group i in the country.
I Official language is the language in which the first organic lawsor constitution has been written.
I Alternatively language of secondary education and highercourts.
I The average distance from the official language for anycountry i is calculated as:Di = ∑j=1 nPijdjo
David Laitin, Rajesh Ramachandran and Noah Rosenberg
Reduced from evidence
DOM
KOR
JPN
CYP
HRV
HTI
MUS
PRT
ITA
EGY
LAO
URY
BIH
IRLNORSGP
BGD
ISRDNKAUS
BRA
CHN
JAMTUN
POL
SVNARG
COLCUB
ALB
BEL
VEN
JOR
CRICZE
GRC
SWE
YEM
DEU
UKR
CAN
BLR
ARM
NLD
NIC
CHEAUTGBRFIN
KHM
LBYCHLLBNPAN
RUS
AZEVNM
USA
ESP
HND
LTU
SYR
FRA
SLV
HUN
BGR
TKM
ROU
MNG
DZA
SAU
SVK
NZL
MRTUZB
TUR
NPL
MKDMAR
MDA
THA
OMN
LVA
IRQIDN
TTO
GEO
BHR
TJK
MEX
KGZ
EST
IRNGUY
MYS
KWT
BTN
ECUGTMPER
KAZ BOL
PAKSDN
LKA
INDCMR
ARE
ZAFFJI NAM
SLE
GAB
DJIZWE
LBR
BWA
TZA
GNB
PHL
NGAZMB
NERMDGRWAMLI
SENAGO
BDIMWIMOZBEN
ERICAFBFA
LSOCIV
GINGHATGO
COGPRY
ETH
UGATCD
SWZ
COD
KENGMB
46
810
12Lo
g GDP
per c
apita
2000
US c
onsta
nt $
0 .2 .4 .6 .8 1average distance from official language
Figure: Log GDP per capita in 2000 US constant dollars and averagedistance from official language
David Laitin, Rajesh Ramachandran and Noah Rosenberg
Reduced form evidence on channel
CHN
BGD
PRTPRKITADOMHRV
HTI
NORBRA
MUS
AUS
LAO
URY
CYP
JPN
SGP
DNKISRKOR
JAM
POL
SVNARGCUB
COLALBBELVENCRICZE
GRC
SWE
DEU
UKR
CAN
BLR
ARMNLD
NICCHEAUT
GBRFINCHL
PAN
RUS
AZE
VNM
USA
ESP
HNDLTU
FRASLV
BGR
TKM
ROUMNG
DZA
SVK
NZL
MRT
TUR
NPL
MKD
MMR
MDATHA
LVA
IDN
TTO
GEOBHR
TJK
MEX
KGZ
ESTIRN
GUY
MYS
AFG
BTN
ECU
GTM
PER
KAZ
BOL
LKA
IND
CMRZAF
FJI
NAMSLE
GAB
DJI
ZWE
LBR
TZABFASWZ
ETHLSO
KENZMB
RWA
CAF
PRY
NGA
MWIGNB
TGO
GIN
GHABEN
AGO
TCD
MOZ
MDGSENBDIGMBMLINER
COG
COD
CIV
UGA
0.2
.4.6
.81
Prop
ortion
spea
king o
fficial
lang
uage
0 .2 .4 .6 .8 1average distance from official language
Figure: Proprtion speaking official language and average distance fromofficial language
Sources: Ethnologue, Leclerc (2011), Albaugh (2012), La Francaphonie(2007), Crystal(2003)
I In fact in more than 25 African countries less than 20% speakthe official language - Average less than 10%.
David Laitin, Rajesh Ramachandran and Noah Rosenberg
Link between indicators of interest and average distancefrom official language
David Laitin, Rajesh Ramachandran and Noah Rosenberg
Reduced form evidence on effect on educational indicators
CYP
AUS
PRTCHN
KOR
EGY
BGD
HRVITA
JPN
HTI
DNK
DOM
LAO
BRA
SGPURY
ISRNOR
MUS
IRL
JAM
TUN
POL
SVN
ARG
COL
CUBALBBEL
VEN
JORCRI
CZE
GRC
SWE
YEM
DEU
UKR
CAN
ARMNLD
NIC
CHEAUTGBR
FIN
KHM
LBY
CHLPAN
RUS
VNM
USA
ESP
HND
LTU
SYR
FRA
SLV
HUN
BGRROU
MNG
TWN
DZA
SAU
SVK
NZL
MRT
TUR
NPLMMRMAR
THA
LVA
IRQIDN
TTOBHRTJK
MEX
KGZ
EST
IRNGUY
MYS
KWT
AFG
ECU
GTM
PER
KAZ
BOL
PAK
SDN
LKA
IND
CMR
AREZAF
FJI
NAM
SLE
GABZWE
LBR
BWA
TZA
MLI
TGO
BEN
PHL
UGA
CAFRWAGMB
MOZ
CIV
PRY
COG
BDI
SWZ
NER
KEN
COD
MWI
LSOZMB
SEN
GHA
0.00
5.00
10.00
15.00
Avg.
years
of Sc
hooli
ng
0 .2 .4 .6 .8 1average distance from official language
Panel A
MUSCHN
BRA
ISRITA
DOM
URYNOR
EGY
CYPSGP
HRVKOR
BGD
AUSDNK
PRT
JPN
LAOHTI
IRLJAM
TUN
POLSVNARGCOLCUBALBBEL
VEN
JOR
CRICZEGRCSWE
YEM
DEUUKRCANARMNLD
NIC
CHE
AUTGBRFIN
KHM
LBY
CHLPAN
RUS
VNM
USAESP
HND
LTU
SYR
FRAHUN
SLV
BGRROUMNGTWN
DZA
SAU
SVKNZL
MRT
TUR
NPL
MMR
MAR
THA
LVA
IRQ
IDN
TTO
BHR
TJK
MEX
KGZEST
IRN
GUY
MYSKWT
AFG
ECU
GTM
PER
KAZ
BOL
PAK
SDN
LKA
IND
CMR
AREZAF
FJI
NAM
SLE
GAB
ZWE
LBR
BWA
TZA
MOZ
SENMWI
PHL
KEN
NER
BDI
MLI
CIV
BEN
LSO
GHA
COG
SWZ
RWA
GMB
UGA
TGO
PRY
COD
ZMB
CAF
0.00
20.00
40.00
60.00
80.00
% of
pop.
with n
o sch
oolin
g0 .2 .4 .6 .8 1
average distance from official language
Panel B
Figure: Educational indicators and average distance from official language
David Laitin, Rajesh Ramachandran and Noah Rosenberg
Distance from official language and effects on education
I Most poignant example Sub-Saharan Africa:I No country, except Ethiopia and Tanzania, provides the entire
span of primary schooling in a local language.I At the same time the lowest levels of educational attainment
in the world, despite huge increase in resources allocated(Devarajan and Fengler 2013)
I Studies from developed world (finding little effects of mothertongue provision) might be inapplicable to the developingworld:
I Exposure effect very different.I Supply side constraints - eg. Teacher language skills
I Namibia - 98% teachers had inadequate English skills.I Tanzania - 89% teachers had inadequate English skills.
David Laitin, Rajesh Ramachandran and Noah Rosenberg
Language and health
I Disease prevention activities and accessing medical care ofteninvolve the use of various forms of interpersonalcommunication and mass media.
I Report of institute of medicine of United States (2003)analyzing racial and ethnic disparities in health finds:
I Language barriers and limited English proficiency were a majorsystem-level determinant of poor health outcomes.
I Led to the first Ferderal policy (2000) of language in healthcare.
I Martinez (2007) finds language an important cause fornegative health outcomes for Lationos.
I Similar findings by Health Canada (2001)I Striking lack of basic preventive care in developing countries
(Madajewicz et al. 2007, Jalan Somanathan 2008, Dupas2011, Cohen et al. 2011).
I Role of lack of information has been highlighted - specific roleof language barriers remains unstudied.
David Laitin, Rajesh Ramachandran and Noah Rosenberg
Average distance from official language and health
SOM
EGY
LAO
KORBRA
PRK
BGD
BIHCHNURYDOM
HTI
JAMTUNARGCUBCOLALBVENJORCRIGRCDEUBLRARMNIC
KHM
LBYCHLLBNPANAZE
VNM
USAESP
HNDSYRSLVBGRMNGDZASAU
MRT
UZBTUR
NPL
MKD
MMR
MARMDATHAOMNIRQ
IDN
GEO
TJK
MEXKGZIRNGUYMYS
KWT
AFG
BTNECU
GTMPERKAZBOL
PAKSDN
LKA
IND
CMRZAF
NAM
SLEDJI
ZWE
LBR
BWATZA
NER
BFA
SENMOZ
NGA
AGOGMBPHLCIV
GHA
PRY
TCD
RWA
GINCAF
GNB
SWZ
ZMB
ETH
COG
MDG
TGOKENLSO
BDI
BENMWI
MLI
UGA
010
2030
4050
Malnu
trition
prev
alenc
e
0 .2 .4 .6 .8 1average distance from official language
Panel A
BGD
BIH
SGPDNKJPN
HTI
KOR
MUS
ISRNORHRVBRA
CYP
EGYPRK
IRL
CHNURYDOM
LAO
PRTAUS
SOM
ITA
JAMTUNPOLSVNARGCOL
CUBALBBEL
VENJOR
CRICZEGRCSWEDEU
UKR
CAN
BLRARM
NLD
NIC
CHEAUTGBRFIN
KHM
LBYCHL
LBNPAN
RUSAZEVNMUSAESP
HNDLTUSYRFRA
HUNSLVBGR
TKMMNGDZASAUSVK
NZL
MRT
UZBTUR
NPL
MKD
MMR
MARMDATHAOMNLVAIRQIDNTTO
GEOBHR
TJK
MEX
KGZESTIRNGUYMYSKWT
AFG
BTN
ECUGTMPER
KAZBOLPAKSDN
LKA
IND
CMR
ARE
ZAF
FJI
NAM
SLE
GABDJI
ZWE
LBRBWATZAGMB
NGACIVAGOGNBBDITCDBFAGIN
SEN
MOZNER
LSO
RWA
ZMB
ERI
SWZ
PHL
UGABEN
GHA
KEN
PRY
MDG
ETH
MLI
TGOCOGMWICAF
4050
6070
80Lif
e Exp
ectan
cy at
birth
0 .2 .4 .6 .8 1average distance from official language
Panel B
LAO
JPN
BGDCHNHTIHRVITABIH
EGYDOM
DNK
MUS
ISRURYCYPKORAUSIRLPRTBRANOR
SGPJAMTUNPOLSVNARGCOL
CUB
ALB
BEL
VEN
JORCRICZE
GRCSWE
YEM
DEU
UKR
CAN
BLRARM
NLDNICCHEAUTGBRFINKHM
LBY
CHLLBNPANRUSAZEVNM
USA
ESPHNDLTUSYR
FRA
HUNSLVBGR
TKM
MNGDZASAU
SVKNZL
MRTUZBTURNPLMKD
MMR
MAR
MDA
THAOMN
LVAIRQ
IDNTTO
GEO
BHRTJKMEXKGZESTIRNGUY
MYSKWT
AFG
BTN
ECUGTMPERKAZ
BOLPAK
SDNLKAINDCMR
ARE
ZAF
FJI
NAM
SLE
GAB
DJI
LBR
BWATZAMLI
MDG
UGAPRYBFARWA
NERSWZMOZMWITGO
PHL
NGA
LSO
ZMBGNBCAFGHATCDKENERICIVGINBENCOGAGOGMBETH
BDI
SEN
05
1015
20He
alth E
xp. a
s % of
GDP
0 .2 .4 .6 .8 1average distance from official language
Panel C
BGDBIHBRA
IRL
SGPCYP
DNK
EGY
KOR
NOR
MUS
PRT
URYHTICHN
ITA
DOM
JPN
LAO
ISRHRV
AUS
JAMTUNPOL
SVN
ARGCOLCUBALB
BEL
VENJORCRICZEGRCSWE
YEM
DEU
UKR
CAN
BLRARM
NLD
NIC
CHEAUTGBRFIN
KHMLBYCHLLBNPANRUSAZEVNM
USA
ESP
HNDLTUSYR
FRA
SLV
HUNBGRTKMMNGDZASAUSVKNZL
MRTUZBTURNPLMKDMMRMARMDATHAOMNLVAIRQIDN
TTOGEO
BHRTJKMEXKGZ
ESTIRNGUYMYSKWT
AFGBTNECUGTMPERKAZBOLPAKSDNLKAINDCMR
AREZAF
FJI NAM SLEGABDJILBRBWATZAGNBNGAMWITGOGMBMOZBENZMBKENSWZETHPHLERICOGPRYGINMDGRWASENMLIGHANERUGACAFTCDAGOBDIBFALSOCIV0
2000
4000
6000
8000
Per c
apita
healt
h Exp
. in 20
05 PP
P doll
ars0 .2 .4 .6 .8 1
average distance from official language
Panel D
Figure: Health indicators and expenditure and average distance fromofficial language
David Laitin, Rajesh Ramachandran and Noah Rosenberg
Historical examples of role of language policy
I Tool for elite manipulation leading to conflict - Sri Lanka(1956 decree)
I Tool to preserve privileges - Austro-Hungarian Empire (1899decree)
I Tool to impose new barriers to mobility - United Pakistan(1947 law)
I How gains might not materialize due to changes in onlyeducation - Algeria (1967 law)
David Laitin, Rajesh Ramachandran and Noah Rosenberg
Language policy and final effects on majority
LAO
DOM
BGD
CHN
EGY
BRA
BIHURYHRV
HTI
JAMTUNPOLSVN
ARG
COL
ALB
VEN
JORCRIUKRBLR
ARM
NIC
KHM
CHL
PAN
RUS
AZE
VNMHND
LTUSYRHUN
SLV
BGRSVK
MRT
TUR
NPL
MKD
MARMDA
THALVA
IDN
GEOTJK
MEX
KGZ
ESTMYS
BTN
ECU
GTM
PER
KAZ
BOL
PAK
LKACMR
ZAFFJI
NAM
SLE
DJI
TZA
MWIMOZ
MLI
ETH
RWA
SEN
AGO
BEN
UGA
SWZ
GMB
GNB
TCDNGA
PHLCIV
PRY
MDG
GIN
LSO
ZMB
BFA
CAF
020
4060
80Pr
oport
ion of
peop
le livi
ng un
der $
1.25
0 .2 .4 .6 .8 1average distance from official language
Figure: Proportion of population living below $1.25 a day and averagedistance from official language
David Laitin, Rajesh Ramachandran and Noah Rosenberg
Empirical methodology
David Laitin, Rajesh Ramachandran and Noah Rosenberg
The method
I We run the reduced form regression given by:
I Log GDP per Capitai =α ∗ Average Dist. From Official Languagei + Bi ∗ Xi + εi
I Xi refers to a vector of controls.
I Robust standard errors
I Clustering at country level does not change results
David Laitin, Rajesh Ramachandran and Noah Rosenberg
Table: Effect of average distance on GDP per capita
Dependent variable - Log GDP per capita in 2000 US constant dollars(1) (2) (3) (4) (5) (6) (7) (8) (9)
Average Distance from Official Language -2.494*** -2.665*** -2.366*** -2.266*** -2.121*** -1.949*** -1.854*** -1.470*** -1.492***(0.227) (0.390) (0.428) (0.361) (0.409) (0.456) (0.425) (0.480) (0.479)
Ethno-Linguistic Fractionalization -0.0293 1.134** 1.168*** 1.209*** 1.117** 0.914* 0.589 0.956*(0.523) (0.544) (0.409) (0.429) (0.510) (0.492) (0.547) (0.543)
Avg. Protection against 0.540*** 0.516*** 0.532*** 0.499*** 0.470*** 0.372*** 0.359***Expropriation risk (0.0519) (0.0640) (0.0612) (0.0648) (0.0588) (0.0758) (0.0787)
Tempreature Controls Yes Yes Yes Yes Yes Yes
Humidity Controls Yes Yes Yes Yes Yes Yes
Soil Quality Yes Yes Yes Yes Yes
Landlocked Dummy Yes Yes Yes Yes Yes
Natural Resources Yes Yes Yes Yes
Legal Origin Dummies Yes Yes Yes
Continent Dummies Yes Yes
Size of Largest Ethnic Group Yes
Observations 149 117 102 99 99 99 99 99 99R-squared 0.387 0.477 0.754 0.797 0.822 0.841 0.884 0.900 0.903
Robust standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1
David Laitin, Rajesh Ramachandran and Noah Rosenberg
Residual analysis
GAB
COGJPNISR
ETH
BWA
MWI YEM
SGP
BGD
TURCOL
JORMAR
TTO
PHLUGA DEU
MEX
GUY
GMB ITA
HTI ARG
CHN
SLV
ESP
MOZ
FRA
TZA
BRA
IRL
URYNAM
NLDBOL DZA CANTHA
SWEZWENGA
VEN AUTEGYKENGHA
TGO
AGO
GRC
NOR
PRY KOR GBR
IDN FINBGR CHEVNMIND LKA
BFA SDN
MLI
HUN
PERSLE
PAN
LBR
HND
ECUCMR
MYS
POL PRT
BELDNK
SYR
GIN
PAK
ZMB TUNSENMDG CRIJAM
GTMMNGGNB USANIC LBYNER CHL AUSNZLCIV DOMZAF
-20
24
Jack
-knife
d res
iduals
4 6 8 10 12Fitted values
Figure: Plot of Jack-knifed residuals and fitted values
David Laitin, Rajesh Ramachandran and Noah Rosenberg
Table: Effect of ELF on GDP per capita controling for averagedistance from official language
Dependent variable - Log GDP per capita in 2000 US constant dollars(1) (2) (3)
Ethno-Linguistic Fractionalization -2.934*** -1.545*** 1.134**(0.333) (0.383) (0.544)
Avg. Protection against 0.615*** 0.540***Expropriation risk (0.0579) (0.0519)
Average Distance from Official Language -2.366***(0.428)
Observations 146 110 102R-squared 0.291 0.672 0.754
Robust standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1
David Laitin, Rajesh Ramachandran and Noah Rosenberg
ELF and link to language policy
I Show how ethno-linguistically fractionalized polities likely tochoose languages belonging to no indigenous group.
I Rationale - nation building, avoiding conflict.
I Ethnically neutral policies however might be not neutral whenlooked from class perspective.
I Path-dependence implies former colonial language choices.
I Is negative effect of ELF found in the literature an artifact of thepositive correlation between distance and ELF?
David Laitin, Rajesh Ramachandran and Noah Rosenberg
Methodological concerns
I Omitted variable bias
I Apply AET (2005) methodology
I Results suggest selection on unobservables would need to be 1.7 to2 times stronger than selection on observables.
I Endogeneity
I Instrument - use genetic distance of indigenous population frompopulation to which official language belongs.
I Does it still meet exclusion restriction?
David Laitin, Rajesh Ramachandran and Noah Rosenberg
Other work in progress
I Compliment with analysis at the sub-national level
I Compare South vs. North India?
I Hindi belongs to Indo-European family; South Indian languages tothe Dravidian family.
I Other suggestions?
I Why sub-optimal langauge policy?
I An elite based capture model.
I Compare average incomes of top 1% or 0.05% - slope with averagedistance should be zero or turn positive?
David Laitin, Rajesh Ramachandran and Noah Rosenberg
THANK YOU
David Laitin, Rajesh Ramachandran and Noah Rosenberg
Figure: Family Tree
back
David Laitin, Rajesh Ramachandran and Noah Rosenberg