Ethnic or Political Fractionalisation? A District Level Analysis of the Provision of Public Goods in...

Ethnic or Political Fractionalisation? A DistrictLevel Analysis of the Provision of Public Goods

in Sri Lanka

ARUSHA COORAY

ABSTRACT This study investigates the influence of ethnic and political fractionalisation on the

provision of public goods at a district level in Sri Lanka. Evidence shows that political fractionalisation

has a greater negative impact on the provision of public goods compared with ethnic fractionalisation.

Similarly, political polarisation has a greater negative effect on the provision of public goods compared

with ethnic polarisation. The interaction between ethnic and political measures suggests that political

fractionalisation (political polarisation) exacerbates the negative effects of ethnic fractionalisation (ethnic

polarisation) on the provision of public goods. The ethnolinguistic fractionalisation (ELF) index assumes

greater statistical significance under fixed effects estimation suggesting greater within district effects in

the relationship between ELF and public good provision.

Introduction

T here has been an extensive literature on the negative effects of ethnic divisions oneconomic outcomes—Easterly and Levine (1997) and Collier (2000) on eco-

nomic growth; Alesina, Baqir and Easterly (1999), Alesina and La Ferrara (2000),Banerjee, Iyer and Somanathan (2005), Banerjee and Somanathan (2007), and Migueland Gugerty (2005) on public good provision; Canning and Fay (1993), Mauro (1995),and La Porta et al. (1999) on government activity and the quality of institutions. Thegeneral conclusion of these studies has been that ethnic divisions lead to suboptimaleconomic outcomes because of a lack of consensus among different ethnic groups.There are several channels through which ethnic fractionalisation could affect theprovision of public goods: political differences (Banerjee and Somanathan 2007),corruption (Dincer 2008), the quality of governance (Kimenyi 2006), governmentpolicies and conflict among ethnic groups which can adversely affect the delivery ofpublic goods (Easterly and Levine 1997), different preferences with regard to publicgoods which can lead to lower funds being channelled to public good provision(Alesina, Baqir and Easterly 1999; Alesina and La Ferrara 2000), income inequality

Arusha Cooray is a senior lecturer at the School of Economics, University of Wollongong, Wollongong,

NSW 2522, Australia and a research associate at the Centre for Applied Macroeconomic Analysis, Australian

National University, Canberra, ACT 0200, Australia. His e-mail address is: [email protected]. I wish to

thank two anonymous referees, Leelananda De Silva and S. Meegama for valuable suggestions.

bs_bs_banner

Growth and Change DOI: 10.1111/grow.12060Vol. 45 No. 4 (December 2014), pp. 640–666

Submitted May 2012; revised December 2012; accepted January 2013.© 2014 Wiley Periodicals, Inc

mailto:[email protected]

which can affect the distribution of public goods (Dincer and Lambert 2012), compe-tition between different ethnic groups for scarce resources (Bobo and Kluegel 1993).

A related literature questions as to what the best measure is of ethnic divisions. Thestudies of Alesina, Baqir and Easterly (1999), Alesina and La Ferrara (2000), Banerjee,Iyer and Somanathan (2005), and Banerjee and Somanathan (2007) use afractionalisation index. The most commonly used fractionalisation measure is based onthe probability that two randomly drawn individuals from a population come from twodifferent groups. The theoretical maximum is reached at a value of 1 when each personbelongs to a different group (see Alesina and La Ferrara 2000; Alesina, Baqir andEasterly 1999). As opposed to this, Esteban and Ray (1994) and Montalvo andReynal-Querol (2002) put forward a measure of polarisation according to which amaximum is reached when two equally sized groups face each other. The studies ofDincer (2008) and Dincer and Lambert (2012) use both measures of heterogeneity. Thepresent study uses both measures in an attempt to investigate the effects of heteroge-neity on public good provision.

The literature further suggests that political divisions often reflect social divides(Banerjee and Somanathan 2007). Collier (2000) observes that political institutions aremore important in deeply divided societies in comparison with relatively homogenoussocieties because of the fact that high levels of democracy in homogenous societieshave the ability to offset the adverse effects of high levels of fractionalisation. Aghion,Alesenia and Trebbi (2004) in an analysis of the degree to which societies choose todelegate power to their leaders, argue that political systems could insulate certainethnic groups from participating in the political process in more fragmented societies.Annett (2001) shows that higher levels of ethnic fractionalisation can lead to greaterpolitical instability to which governments may react, by increasing consumption expen-diture as a buffer against that political instability. The political science literaturemoreover, shows that ethnic divisions are closely related to the structure of politicalparty systems (Horowitz 1993; Mozaffar, Scarritt and Galaich 2003; Posner 2007).According to this literature, electoral institutions can either moderate or strengthen thenegative impact of ethnic divisions on public outlook towards democratic systems.

Accordingly, the literature suggests that political institutions and ethnic fragmenta-tion are inextricably interwoven. Political competition can therefore undermine theprovision of public goods in an ethnically divided society. The objective of the presentstudy is to investigate the influence of ethnic and political divisions on the provision ofpublic goods in Sri Lanka. Sri Lanka is particularly relevant for a study of this naturebecause: 1) it has a deeply divided society as reflected by 26 years of civil war betweenthe majority Sinhalese and the country’s largest minority group, the Tamils; 2) thepolitical party system appears to reflect this ethnic division (see Uyangoda 2010); and3) it has unusually high achievements in health and education despite being a low-income country. The life expectancy at birth currently stands at 74 years; the infantmortality rate is 13 per 1,000 live births; and the literacy rate is 89 percent (World Bank2011). Sri Lanka is considered as a model of a successful welfare state mainly because

ETHNIC OR POLITICAL FRACTIONALISATION? 641

of the Sri Lanka government’s commitment to the free provision of health services,education, and food subsidies since its independence. There are however, district-leveldisparities in public good provision, and the present study looks at ethnic and politicalconflict as possible explanations for this.

This is the first study to investigate the influence of ethnic and political divisions onthe provision of pubic goods in Sri Lanka. The data used in this study are drawnprimarily from the Department of Census and Statistics, Sri Lanka. Data for the censusyears 1953, 1963, 1971, 1981, 2001, and the year 2006/2007 covering the 25 districtsof Sri Lanka are used. Results are tested for robustness in a number of ways. Severaladditional control variables are used to capture a range of possible factors affectingpublic good provision, an interaction term for ethnolinguistic fractionalisation(ELF) × political fractionalisation is incorporated to see if political fractionalisationexacerbates the negative effects of ELF on public good provision, the estimation iscarried out by replacing the ELF and political fractionalisation indices with ethnicpolarisation and political polarisation indices, the existence of a non-linear relationshipbetween the provision of public goods and fractionalisation are also investigated (seeDincer 2008). Given the use of district-level data, fixed effects estimation is used toaccount for district-level fixed effects.

The rest of this study is structured as follows. The second section provides abackground to the Sri Lankan ethnic and political structure. The third section describesthe data. The fourth section evaluates the empirical results and the fifth section presentsthe conclusion.

BackgroundSri Lanka is a low middle-income nation which ranks high in terms of human

development. It was first colonised by the Portuguese, then the Dutch, and finally theBritish. The Sinhalese make up 73.9 percent of the population, Tamils 12.7 percent,Indian Tamils 5.5 percent1, Moor (Muslim) 7.1 percent, Burgher 0.2 percent2, Malay0.3 percent, and other ethnic groups 0.3 percent (Department of Census and Statistics,2012). According to historical anecdotes, the Sinhalese–Tamil conflict dates back to theperiod of the kingdoms. When the Portuguese invaded Sri Lanka in 1505, the majorityof the Tamil people were concentrated in the Northern and Eastern parts of the countryand the Sinhalese resided in the rest of the country (DeVotta 2000; Singer 1992). TheBritish followed a divide and rule policy under which they favoured the Tamil minority.Therefore at independence, the Tamil community disproportionately held the majorityof jobs in public service. Independence led to a resurgence of nationalism with calls forthe need to restore Buddhism and the Sinhala language to their rightful place. Thisculminated in the enactment of the 1956 “Sinhala only” legislation by SWRDBandaranaike of the Sri Lanka Freedom Party (SLFP), according to which Sinhala wasestablished as the only official language for administrative purposes. This was aimed atcapturing the votes of the Sinhala educated community who felt marginalised by theEnglish-speaking elites of all ethnic groups (International Crisis Group 2007). The

642 GROWTH AND CHANGE, DECEMBER 2014

disastrous consequences that this had, leading to feelings of marginalism by the largestethnic minority group, the Tamils, resulted in a civil conflict which lasted 26 years, notat the time foreseen by Sri Lankan politicians.

Sri Lankan society is divided along race, religion, caste, and class lines. Since itsindependence Sri Lanka now has a two-party system: the conservative United NationalParty (UNP) formed in 1947 and the socialist SLFP in 1952, alternating with eachother for power. Since 1994 however, the two major parties have failed to gain a clearmajority and have formed coalition governments with smaller parties to come intopower. In 1994, the SLFP came into power by forming a seven-party, leftist coalitioncalled the People’s Alliance. In 2001, the UNP formed a coalition government with theSri Lanka Muslim Congress’s to take control of parliament. In 2004, the SLFP formedan alliance again with several leftist parties to come into power. The North and Easthave been represented by Tamil parties, the All Ceylon Tamil Congress, and the FederalParty. Subsequently, these two parties amalgamated to become the Tamil United Lib-eration Front (TULF). With the civil war in 1983, the TULF leadership went into exileand a number of Tamil militant groups emerged to fill the political vacuum created bythe TULF’s exile. Among these militant groups that emerged as political parties towork within the arena of provincial councils and parliament are the Eelam People’sRevolutionary Front, People’s Liberation Organisation of Tamil Eelam, and EelamPeople’s Democratic Party. These three parties have been active in parliamentarypolitics since 1987 (see Uyangoda 2010). The UNP and SLFP however, have not beenable to come to a consensus for power sharing with the Tamil community. Instead theyare engaged in a process of “ethnic outbidding” maintaining to be the better represen-tative of the majority (International Crisis Group 2007). Sri Lanka has beencharacterised by minority coalitions governments headed by the UNP and SLFP andthe presence of ideologically narrower parties since 1994. Coalition governments canundermine the provision of public goods by stalling the implementation of legislationrequired for public goods provision and placing pressure on the main governing party.

Sri Lanka has nine provinces and 25 districts in total. The Northern districts—Jaffna, Kilinochchi, Mullativu, Mannar, and Vavuniya; and the Eastern district,Batticaloa, have a Tamil majority, 99.9 percent, 99.9 percent, 99.9 percent, 51.4percent, 86.3 percent, 74 percent, respectively. The Eastern districts, Ampara andTrincomalee, have a Moor (Muslim) majority, 41.3 percent and 45.4 percent, respec-tively; and Nuwara-Eliya has an Indian Tamil majority of 50.6 percent (Department ofCensus and Statistics 2012). The rest of the districts have a Sinhalese majority (seeFigure 1 for a district-level division of the country).

Ethnic divisions are reflected in the party system, with the Tamil minority in theNorthern districts being represented primarily by the Tamil parties, and the Sinhaleseby the UNP and SLFP parties. According to Horowitz (1985), political parties individed societies which may initially reflect ethnic divisions and could contribute towidening the differences between groups because politicians may find it in their owninterest to make use of these divisions to pursue their own agendas.


Data and MethodologyThe study covers the census years, 1953, 1963, 1971, 1981, 2001, and the year

2006/2007 at the district level. Data at the district level are not easily accessible andrequired going through a number of reports primarily from the Department of Censusand Statistics. Data for all variables for all years were not available. Hence, the choiceof variables used in the study is severely restricted by the availability of data (seeTable 1 for data description and sources).

FIGURE 1. DISTRICT-LEVEL DIVISION OF SRI LANKA.Source: Urban Development Authority Sri Lanka (2014) Sri Lanka District Map:http://www.uda.lk/images/downloads/district-map.jpg (retrieved April 2014).


http://www.uda.lk/images/downloads/district-map.jpg

TABLE 1. DESCRIPTION OF VARIABLES USED IN THE STUDY AND DATA SOURCES.

Data Source

Public goods:

Number of government schools Department of Census and Statistics Statistical

Abstracts, various issues

Number of government school teachers Department of Census and Statistics Statistical


Maternal mortality rate per 1,000 live births

(used as proxy for provision of public health)

Department of Census and Statistics Statistical


Infant mortality rate per 1,000 live births (used

as proxy for provision of public health)



Public health medical officers Central Bank of Sri Lanka, Economic and Social

Statistics of Sri Lanka, various issues

Water availability of main water service line

near household

Department of Census and Statistics, Household

Income and Expenditure Surveys

Electricity availability of electricity line near

household



Road kilometerage Central Bank of Sri Lanka, Economic and Social

Statistics of Sri Lanka, various issues

Ethnic fractionalisation authors own calculation

using % of each ethnic group as classified by

the Department of Census and Statistics



Data for ethnicity Northern province 2006/2007

from Northern District reports

Political fractionalisation authors own calculation

based on the number of seats secured by each

party in each district

Department of Election Sri Lanka, results of

parliamentary general elections

Ethnic polarisation authors own calculation using

proportion of each ethnic group as classified

by the Department of Census and Statistics



Data for ethnicity Northern province 2006/2007

from Northern District reports

Political polarisation authors own calculation based

on the number of seats secured by each party

in each district

Department of Election Sri Lanka, results of

parliamentary general elections

Per capita mean monthly income Department of Census and Statistics, Household


Household Income and Expenditure Survey for

Northern and Eastern Province 2002/2003

Literacy rate Department of Census and Statistics Statistical


Income inequality—Gini coefficient of per capita

income



Sri Lanka Integrated Survey 1999/2000

Population (000) Department of Census and Statistics Statistical



The dependent variables in the study include the provision of public goods: 1)government schools, 2) government school teachers, 3) government health medicalofficers, 4) maternal mortality rate per 1,000 live births, 5) infant mortality rate per1,000 live births used as proxies for the provision of public health, 6) water availabilityof main water service lines near household, 7) electricity availability of electricity linesnear household, and 8) road kilometerage.

The main independent variables in the study are the ELF and the politicalfractionalisation index. The ELF index is calculated as one minus the Herfindahl indexof ethnolinguistic group shares that two people randomly drawn from the populationare from distinct groups as commonly used in the literature (Alesina, Baqir andEasterly 1999; Alesina 2004; Banerjee and Somanathan 2007 among others).

ELF = − ∑1 2( )ethnicii

n

Where ethnici denotes the population share of the ith group. Ethnic fractionalisationis calculated using the ethnic classification of the Department of Census and Statisticswhich divides the population into seven groups: Sinhalese, Tamil, Indian Tamil, Moor(Muslim), Burgher, Malay, and Other.

The other variable of interest is the political fractionalisation index. The politicalfractionalisation index is calculated using the Herfindahl index in the same manner asthe ELF index (see Banerjee and Somanathan 2007). The present study employs thenumber of seats held by each party in each district for the calculation of the politicalfractionalisation index rather than the vote share going to each party as in Banerjee andSomanathan (2007). The reason for this is that, if there are no political appointments toparliament from a particular party, a candidate will not be in a position to providepublic goods.

POLFRAC = − ( )∑1 2seatsii

m

Where seatsi denote the number of seats held by the ith party. Politicalfractionalisation is calculated using the number of seats held in parliament by eachparty in each district for the years under consideration.

Figure 2 illustrates the figures for ethnic and political fractionalisation by district for2006. Note that political fractionalisation exceeds ethnic fractionalisation in all dis-tricts for which data are available with the exception of Trincomalee, Nuwara-Eliya,and Batticaloa.

Figure 3 illustrates the movement of the ELF and political fractionalisation indicesover time. The figure shows that the two move very closely together, however,the political fractionalisation index is more volatile compared with the ELF index. Thepolitical fractionalisation index is higher than the ELF index in all districts with theexception of Trincomalee and Nuwara-Eliya. These two districts are the ethnically most


FIGURE 2. ETHNIC AND POLITICAL FRACTIONALISATION INDEX BY DISTRICT 2006.


fractionalised in Sri Lanka. Figures 4 and 5 plot the number of schools and number ofmedical officers, respectively against the ELF index. There is a negative relationshipbetween the number of schools and ELF index. There does not appear to be a discern-ible pattern between the ELF index and number of health medical officers.

Montalvo and Reynal-Querol (2002) argue that fractionalisation does not necessar-ily increase conflict. They find ethnic polarisation to have a significant effect on civilwars. This is particularly relevant in the case of Sri Lanka where the main ethnicminority, the Tamil minority, were engaged in a 26-year-old civil war with the ethnicmajority, the Sinhalese. The estimation is therefore also carried out by replacing thefractionalisation indices with the polarisation indices. Estaban and Ray (1994) andMontalvo and Reynal-Querol (2002) argue that a country comprising a number of

05

05

05

1970 1980 1990 2000 2010 1970 1980 1990 2000 2010 1970 1980 1990 2000 2010 1970 1980 1990 2000 2010 1970 1980 1990 2000 2010

Colombo Gampaha Kalutara Kandy Matale

Nuwara-Eliya Galle Matara Hambantota Jaffna

Batticaloa Trincomalee Kurunegala Puttlam Anuradhapura

Badulla Moneragala Ratnapura Kegalle

Ethnic Fractionalisation Political Fractionalisation

year

FIGURE 3. MOVEMENT IN ELF AND POLITICAL FRACTIONALISATION OVER TIME.


0.2

.4.6

.81

0 .2 .4 .6 .8Ethnic Fractionalisation Index

Number of Schools Fitted Values

FIGURE 4. ETHNIC FRACTIONALISATION INDEX VERSUS NUMBER OF SCHOOLS.Note: The number of schools has been normalised.

0.2

.4.6

.81

0 .2 .4 .6 .8Ethnic Fractionalisation Index

Health Medical Officers Fitted Values

FIGURE 5. ETHNIC FRACTIONALISATION INDEX VERSUS HEALTH MEDICAL OFFICERS.Note: Health medical officers has been normalised.


small groups may be more stable as opposed to one with two equally sized groups.Montalvo and Reynal-Querol (2002) put forward a polarisation index:

EPIs

sii

i

N

= − −⎛⎝⎜

⎞⎠⎟=

∑11 2

1 2

2

1

where si is the share of group i in the population. The Ethnic Polarization Index (EPI)reaches maximum when two equally sized groups confront each other and falls as therelationship between the groups diverges from this half and half divide. Montalvo andReynal-Querol (2002) show that this index is associated with ELF. A polarisation indexis similarly constructed for political polarisation (POLPI).

POLPIx

xii

i

N

= − −⎛⎝⎜

⎞⎠⎟=

∑11 2

1 2

2

1

Where xi denotes the number of seats held by the ith party. Figure 6 illustrates thefigures for ethnic and political polarisation by district for 2006. Note that as in Figure 2,political polarisation exceeds ethnic polarisation in all districts for which data areavailable with the exception of Trincomalee, Nuwara-Eliya, and Batticaloa. Figure 7illustrates the movement of the EPI and political polarisation indices over time. Asopposed to the fractionalisation indices, the two series tend to diverge a lot more.However, once again, the political polarisation index is more volatile compared to theEPI.

Table 2 presents the distribution of total and per capita public goods across districts.A preliminary examination of the data appears to support the hypothesis that greaterpolitical fractionalisation contributes to the lower provision of public goods. Thedistricts which rank highest in terms of political fractionalisation, Trincomalee (0.625),Colombo (0.615), Gampaha (0.581), Nuwara-Eliya (0.571), and Kalutara (0.540), alsoappear to have a lower number of schools, teachers, road kilometerage, and electricitylines per capita. Nuwara-Eliya, which also has a relatively high degree of ethnicfractionalisation, has a lower number of medical officers per capita and a high maternalmortality rate.

Other control variables used in the study are based upon the literature (Alesina,Baqir and Easterly 1999; Banerjee and Somanathan 2007). Per capita mean monthlyincome is used to measure the level of development of a district. District size iscaptured by the population in each district. As a more educated population is morelikely to choose better policies, the literacy rate for each district is used to measure thelevel of educational attainment of a district. The district level provision of public goodsmay be explained by the inequality in income distribution in that district. Therefore, theGini coefficient of the percentage of poor households are used to measure incomeinequality. A dummy variable is introduced for the war-affected Northern and Easterndistricts. This variable takes a value of 1 for the war-affected years and zero otherwise.


Dincer (2008) and Dincer and Lambert (2012) find evidence of non-linear effects ofELF on corruption and income inequality. In this case, an increase in fractionalisationbeyond a certain point could reduce the negative effect of ELF on the provision ofpublic goods. In order to capture any non-linear effects that might be present, the modelis also estimated with a squared term for ELF. Table 3 reports summary statistics for thedata used in the estimation.

FIGURE 6. ETHNIC AND POLITICAL POLARISATION INDEX BY DISTRICT 2006.


The empirical model takes the following form:

Y ELF POLFRAC xDt Dt Dt Dt D Dt= + + + + +α ϑ ϖ β ν ε

where YDt represents access to a particular public good in district D in period t. ELFDt

represents the ethnic fractionalisation index and POLFRACDt the politicalfractionalisation index for district D in period t. Robustness tests are carried out byreplacing the ELF index with the EPI and the POLFRAC index with a POLPI index.Correlation between the fractionalisation and polarisation indices are reported inTable 4. The correlation between the ELF and POLFRAC indices and the EPI and thePOLPI are low. There is however, a high degree of correlation between the ELF and EPIindices and the POLFRAC and POLPI indices. Therefore, the fractionalisation indicesand the polarisation indices are not used together but separately in the estimation thatfollows.

All control variables mentioned in the third section are captured by the vector xDt. VD

is a dummy variable which takes on a value of 1 for the war-affected Northern andEastern districts in the war-affected years. Banerjee and Somanathan (2007) argue that

0.5

10

.51

0.5

1

1960 1980 2000 2020 1960 1980 2000 2020 1960 1980 2000 2020 1960 1980 2000 2020 1960 1980 2000 2020

Colombo Gampaha Kalutara Kandy Matale

Nuwara-Eliya Galle Matara Hambantota Jaffna

Batticaloa Trincomalee Kurunegala Puttlam Anuradhapura

Badulla Moneragala Ratnapura Kegalle

Ethnic Polarisation Political Polarisation

Year

FIGURE 7. MOVEMENT IN PI AND POLITICAL POLARISATION OVER TIME.


TAB

LE

2.P

UB

LIC

GO

OD

S(T

OTA

LA

ND

PE

RC

AP

ITA

)B

YD

IST

RIC

TA

SO

F20

06.

Dis

tric

tTo

tal

gove

rnm

ent

scho

ols

Gov

ernm

ent

scho

ols

per

capi

ta

Teac

hers

Teac

hers

per

capi

taH

ealt

hm

edic

alof

fice

rs

Hea

lth

med

ical

offi

cers

per

capi

ta

Mat

erna

lde

aths

per

1,00

0

Col

ombo

410

0.17

14,9

856.

103,

145

1.28

0.03

Gam

paha

537

0.25

14,0

106.

551,

160

0.54

0.07

Kal

utar

a40

60.

379,

670

8.70

480

0.43

0.19

Kan

dy64

30.

4715

,318

11.1

01,

197

0.87

0.1

Mat

ale

307

0.64

5,36

511

.25

237

0.50

0.43

Nuw

ara-

Eli

ya51

70.

706,

180

8.33

172

0.23

0.61

Gal

le42

20.

4010

,850

10.3

179

40.

750.

15M

atar

a36

30.

459,

539

11.7

335

00.

430.

34H

amba

ntot

a30

80.

566,

859

12.4

318

40.

330.

13Ja

ffna

410

0.68

5,74

69.

5924

60.

410.

32V

avun

iya

188

1.13

1,80

610

.88

770.

460.

25M

ulla

tivu

103

0.70

987

6.71

n.a.

n.a.

0.42

Kil

inoc

hchi

960.

66n.

a.n.

a.n.

a.n.

a.0.

93B

atti

calo

a32

30.

624,

250

8.13

252

0.48

0.15

Am

para

390

0.63

6,54

110

.64

180

0.29

0.31

Tri

ncom

alee

258

0.73

3,93

511

.08

186

0.52

0.37

Kur

uneg

ala

881

0.58

18,5

5512

.18

587

0.39

0.08

Put

tlam

340

0.45

6,25

38.

3224

70.

330.

08A

nura

dhap

ura

550

0.69

9,17

511

.45

311

0.39

0.12

Polo

nnar

uwa

232

0.59

3,46

78.

7817

40.

440.

14B

adul

la56

90.

679,

788

11.5

238

80.

460.

34M

oner

agal

a26

20.

624,

632

10.9

018

70.

440.

14R

atna

pura

579

0.53

10,3

999.

5833

50.

310.

33K

egal

le52

50.

659,

396

11.7

237

10.

460.

52


TAB

LE

2.(C

ON

TIN

UE

D)

Dis

tric

tIn

fant

mor

tali

type

r1,

000

live

birt

hs

Roa

dki

lom

etre

Roa

dki

lom

etre

per

capi

taW

ater

line

sW

ater

line

spe

rca

pita

Ele

ctri

city

line

sE

lect

rici

tyli

nes

per

capi

ta

Col

ombo

15.0

823

0.34

89.4

0.04

99.8

0.04

Gam

paha

6.1

1,61

40.

7545

.70.

0299

.40.

05K

alut

ara

4.1

1,06

80.

9636

.80.

0394

.10.

08K

andy

15.4

2,01

11.

4661

.00.

0494

.70.

07M

atal

e10

.385

31.

7941

.90.

0986

.80.

18N

uwar

a-E

liya

15.6

1,10

31.

4964

.10.

0991

.10.

12G

alle

10.8

1,15

01.

0933

.50.

0392

.70.

09M

atar

a8.

493

41.

1546

.20.

0695

.10.

12H

amba

ntot

a6.

71,

073

1.94

75.2

0.14

89.7

0.16

Jaff

na5.

71,

230

2.05

n.a.

n.a.

n.a.

n.a.

Vav

uniy

a8.

060

43.

64n.

a.n.

a.n.

a.n.

a.M

ulla

tivu

2.5

567

3.86

n.a.

n.a.

n.a.

n.a.

Kil

inoc

hchi

1.2

281

1.92

n.a.

n.a.

n.a.

n.a.

Bat

tica

loa

21.1

652

1.25

17.0

0.03

93.1

0.18

Am

para

5.9

738

1.20

26.1

0.04

85.4

0.14

Tri

ncom

alee

3.4

856

2.41

Kur

uneg

ala

14.2

2,34

51.

5413

.70.

0184

.20.

06P

uttl

am6.

41,

127

1.50

45.5

0.06

82.5

0.11

Anu

radh

apur

a17

.42,

215

2.77

32.7

0.04

84.9

0.11

Polo

nnar

uwa

27.6

926

2.34

38.5

0.10

880.

22B

adul

la9.

51,

877

2.21

77.2

0.09

890.

10M

oner

agal

a2.

11,

023

2.41

43.8

0.10

78.8

0.19

Rat

napu

ra13

1,83

61.

6952

.80.

0578

.60.

07K

egal

le7.

51,

468

1.83

44.9

0.06

93.6

0.12


TABLE 3. SUMMARY STATISTICS FOR FULL SAMPLE.

Variable Observations Mean Standarddeviation

Minimum Maximum

Ethnolinguisticfractionalisationindex

130 0.322 0.199 0.001 0.694

Tamil % 130 19.13 30.28 0.3 99.95Moor % 130 9.90 11.80 0.04 49.80Indian Tamil % 125 7,93 12.97 0.00 59.20Political

fractionalisationindex

101 0.416 0.231 0.00 0.777

Ethnic polarisationindex

126 0.513 0.021 0.002 0.927

Political polarisationindex

101 0.708 0.343 0 1.00

Government schools(number)

104 393.89 215.81 74 1,074

Government schoolteachers (number)

104 6,208.14 4,598.76 413 19,940

Government healthmedical officers(number)

80 428.40 582.23 20 3,145

Maternal deaths (per1,000 live births)

118 0.88 0.85 0.03 3.70

Infant mortality (per1,000 live births)

120 26.45 19.26 1.2 100

Electricity (availabilityof electricity linesnear household)

72 82.67 8.85 63 100

Water (availability ofwater lines nearhousehold)

72 40.47 19.44 7 89.4

Road kilometerage 91 1,104.69 527.09 267 2,500Population (‘000) 137 619.35 510.99 35.1 2,672.3Mean per capita

income (rupees)68 3,154.72 2,113.57 800 10,165

Literacy rate 72 87.08 5.10 75 96.3Gini coefficient 74 0.46 0.09 0.31 0.75


political heterogeneity could reflect social heterogeneity. This suggests that politicalfractionalisation could further exacerbate the negative effects of ethnicfractionalisation. In Sri Lanka, ethnic divisions are reflected in the party system. TheNorthern districts in which there is a Tamil majority, are represented mainly by Tamilparties and the Sinhalese mainly by the UNP and the SLFP. There are, however, severaldifferent Sinhala and Tamil parties which the index of political fractionalisation cap-tures which cannot be captured by the measure of ethnic fractionalisation. In order toexplore the effects of political divides on ethnic divisions, the political fractionalisationindex is also interacted with the ethnic fractionalisation index. εDt is a random errorterm that captures all other variables.

ResultsAs mentioned above, the data cover the years, 1953, 1963, 1971, 1981, 2001, and

2006. The preliminary model is estimated using pooled OLS. Fixed effects panel dataestimation is used additionally to capture unobserved district-specific effects.

Pooled ordinary least squares (OLS). The results for the regression using pooledOLS are reported in Table 5 where the dependent variables are: the number of govern-ment schools (column 1), government school teachers (columns 2), government healthmedical officers (column 3), the number of maternal deaths (column 4), the infantmortality rate (column 5), electricity availability of main electricity service lines nearhousehold (column 6), water availability of main water service lines near household(column 7), and road kilometerage (column 8). The data comprises an unbalancedpanel. The main independent variables are the ELF index, POLFRAC index, and theinteraction term between ELF and POLFRAC. A number of control variables are alsoincluded in the model. Population is used, to control for the size of the district, and percapita income and the literacy rate are included to capture the level of development ofthe district. As pointed out by Alesina, Baqir and Easterly (1999), fractionalisationcould be a fractionalisation of income rather than ethnicity. Therefore, the Gini

TABLE 4. CORRELATION BETWEEN THE FRACTIONALISATION AND POLARISATION

INDICES.

ELF POLFRAC EPI POLPI

ELF 1.00 - - -POLFRAC 0.107 1.00 - -EPI 0.977 0.131 1.00 -POLPI 0.06 0.921 0.101 1.00

ELF, ethnolinguistic fractionalisation; EPI, ethnolinguistic polarisation; POLFRAC,political fractionalisation; POLPI, political polarisation.


TAB

LE

5.P

OO

LE

DO

LS

.

Dep

ende

ntva

riab

les

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Inde

pend

ent

vari

able

s

Gov

ernm

ent

scho

ols

Gov

ernm

ent

teac

hers

Gov

ernm

ent

heal

th

med

ical

offi

cers

Mat

erna

lde

aths

Infa

ntm

orta

lity

Ele

ctri

city

avai

labi

lity

Wat

erav

aila

bili

tyR

oad

kilo

met

erag

e

EL

F−

0.00

1(0

.021

)−

0.00

4(0

.001

)***

−0.

004

(0.1

02)

0.00

4(0

.005

)0.

002

(0.0

06)

−0.

005

(0.0

02)*

*−

0.00

4(0

.002

)*−

0.00

7(0

.005

)

PO

LF

RA

C−

0.01

3(0

.007

)*−

0.00

6(0

.002

)***

−0.

008

(0.0

03)*

*0.

007

(0.0

09)

0.00

4(0

.002

**−

0.00

8(0

.001

)**

−0.

007

(0.0

04)*

−0.

008

(0.0

09)

EL

F×

PO

LF

RA

C−

0.01

1(0

.005

)*−

0.04

0(0

.023

)*−

0.00

7(0

.002

)***

0.02

0(0

.013

)0.

026

(0.0

13)*

*−

0.03

6(0

.018

)*−

0.02

1(0

.013

)*−

0.04

8(0

.023

)**

Popu

lati

on0.

027

(0.0

11)*

**0.

104

(0.0

12)*

**0.

049

(0.0

21)*

*0.

079

(0.0

50)

0.04

0(0

.034

)0.

029

(0.0

38)

−0.

031

(0.4

24)

0.03

8(0

.010

)***

Lit

erac

yra

te0.

015

(0.0

11)

0.01

9(0

.011

)*0.

033

(0.0

17)*

*−

0.04

5(0

.021

)−

0.03

2(0

.110

)0.

091

(0.0

93)

0.03

1(0

.039

)0.

082

(0.1

31)

Gin

ico

effi

cien

t−

0.00

7(0

.009

)0.

004

(0.0

02)*

**0.

006

(0.0

02)*

**0.

005

(0.0

10)

0.00

5(0

.006

)−

0.00

6(0

.002

)***

−0.

006

(0.0

18)

0.00

7(0

.005

)

Per

capi

tain

com

e−

0.01

7(0

.003

)***

0.00

2(0

.016

)0.

017

(0.0

05)*

**−

0.02

3(0

.008

)***

−0.

002

(0.0

01)*

*0.

012

(0.0

04)*

**−

0.00

7(0

.002

)**

−0.

028

(0.0

15)*

*

Dum

my

for

Nor

th

and

Eas

t

−0.

012

(0.0

08)*

−0.

010

(0.0

05)*

*−

0.00

9(0

.005

)*0.

005

(0.0

06)

0.00

7(0

.004

)*−

0.01

0(0

.006

)*−

0.00

9(0

.005

)**

−0.

005

(0.0

04)

R2

0.81

0.96

0.91

0.28

0.22

0.77

0.46

0.51

Obs

erva

tion

s60

6060

6060

6060

60

Not

e:R

obus

tst

anda

rder

rors

repo

rted

wit

hin

pare

nthe

sis.

*,**

,**

*,si

gnifi

cant

atth

e10

%,

5%,

and

1%le

vels

,re

spec

tivel

y.

EL

F,et

hnol

ingu

isti

cfr

acti

onal

isat

ion;

PO

LF

RA

C,

poli

tica

lfr

acti

onal

isat

ion.


coefficient is used to capture income distribution. A dummy variable is included for thewar-affected Northern and Eastern districts for the war-affected years.

The results reported in Table 5 show that ELF has a statistically significant negativeeffect on government school teachers, the availability of electricity, and water lines. Aone unit increase in the ethnic fractionalisation index in column (2) for example, willlead to a 0.4 percent fall in government school teachers. Note that the politicalfractionalisation index has a much greater negative statistically significant impact onthe provision of public goods compared with the ELF index. A one unit increase in thepolitical fractionalisation index leads to a 0.6 percent fall in government school teach-ers in column (2). Political fractionalisation has a significant negative impact on thenumber of government schools, school teachers, health medical officers, the availabil-ity of electricity and water lines, and a significant increase in infant mortality. Theinteraction of ELF with the POLFRAC index has a negative impact on the same publicgoods suggesting that political fractionalisation acts to increase the negative effects ofethnic fractionalisation.

The statistically significant coefficients on district size as measured by populationindicate that larger districts have more government schools, teachers, health medicalofficers, and road kilometerage. The coefficients on the literacy rate in columns (2) and(3) are statistically significant implying that more literate districts are more likely tohave more school teachers and medical officers. The estimates on the Gini coefficientsindicate that school teachers and medical officers are higher in districts in which thereis greater inequality, and the supply of electricity is lower in districts characterised bygreater inequality. Per capita income is statistically significant in all columns with theexception of column (2). The dummy variables on the Northern and Eastern districtsare negative and significant in all columns with the exception of columns (4) and (8)suggesting that the war had a negative impact on the provision of public goods in thesetwo regions.

Alesina, Baqir and Easterly (1999) observe that the share of ethnic minorities couldbe strongly correlated to the ELF index so that the ELF index could be proxying for theshare of minority groups. If the ethnic minority groups were proxying for the ELFindex, or alternatively the political fractionalisation index, then these indices wouldcease to be statistically significant with the inclusion of ethnic minority groups in theestimation. Hence, in order to investigate if the ELF or political fractionalisation indexare proxying for ethnic minority groups, the percentage of Tamil, Indian Tamil, andMoors are included in the estimation in Table 6. In Table 5, the ELF index wasstatistically significant in columns (2), (6), and (7). The ELF index gains greaterstatistical significance in Table 6, showing that an increase in ELF leads to a fall ingovernment schools, teachers, health medical officers, electricity, and water availabil-ity. The political fractionalisation index continues to be statistically significant withrespect to the same variables as before. It can be concluded that the percentage ofethnic minority groups are not proxying for the ELF index, as the ELF gains greatersignificance rather than losing statistical significance with the incorporation of the


TAB

LE

6.P

OO

LE

DO

LS

INC

LU

DIN

GP

ER

CE

NTA

GE

OF

RA

CE.

Dep

ende

ntva

riab

les

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Inde

pend

ent

vari

able

sG

over

nmen

t

scho

ols

Gov

ernm

ent

teac

hers

Gov

ernm

ent

heal

th

med

ical

offi

cers

Mat

erna

lde

aths

Infa

nt

mor

tali

ty

Ele

ctri

city

avai

labi

lity

Wat

er

avai

labi

lity

Roa

d

kilo

met

erag

e

Tam

il−

0.19

0(0

.078

)***

−0.

057

(0.1

13)

0.33

7(0

.302

)−

0.23

0(0

.104

)**

−0.

219

(0.3

99)

−0.

051

(0.0

14)*

**0.

296

(0.2

42)

−0.

214

(0.1

18)*

Indi

anTa

mil

−0.

025

(0.0

35)

−0.

029

(0.0

16)*

−0.

203

(0.1

00)*

−0.

010

(0.0

58)

−0.

130

(0.1

56)

−0.

041

(0.0

10)*

**−

0.18

1(0

.100

)*−

0.17

0(0

.091

)*

Moo

r−

0.05

0(0

.072

)0.

099

(0.0

81)

−0.

564

(0.2

72)*

*−

0.12

3(0

.025

)***

−0.

074

(0.3

66)

−0.

080

(0.0

39)*

*−

0.03

9(0

.422

)−

0.24

9(0

.384

)

EL

F−

0.00

4(0

.002

)**

−0.

002

(0.0

01)*

−0.

002

(0.0

01)*

0.02

3(0

.042

)0.

022

(0.0

12)

−0.

012

(0.0

06)*

*−

0.00

4(0

.002

)*−

0.00

3(0

.004

)

PO

LF

RA

C−

0.00

4(0

.004

)*−

0.00

8(0

.002

)***

−0.

006

(0.0

03)*

*0.

002

(0.0

02)

0.00

2(0

.001

)*−

0.00

4(0

.002

)**

0.00

4(0

.002

)**

−0.

001

(0.1

01)

EL

F×

PO

LF

RA

C−

0.00

6(0

.003

)**

−0.

008

(0.0

04)*

*−

0.00

6(0

.003

)*−

0.00

8(0

.003

)***

−0.

004

(0.0

02)*

*0.

006

(0.0

03)*

0.00

8(0

.004

)*−

0.00

3(0

.003

)

Popu

lati

on0.

042

(0.0

14)*

**0.

107

(0.0

15)*

**0.

116

(0.0

33)*

**0.

083

(0.0

23)*

**0.

146

(0.1

44)

0.05

8(0

.035

)*−

0.10

8(0

.197

)0.

034

(0.0

18)*

Lit

erac

yra

te0.

325

(0.1

90)*

0.26

5(0

.111

)***

0.21

7(0

.144

)*−

0.15

0(0

.006

)**

−0.

265

(0.5

13)

0.19

4(0

.166

)0.

295

(0.5

42)

0.02

1(0

.414

)

Gin

ico

effi

cien

t0.

008

(0.0

04)*

0.00

4(0

.001

)***

−0.

003

(0.0

06)

0.00

4(0

.002

)*0.

004

(0.0

04)

−0.

007

(0.0

02)*

**−

0.00

3(0

.004

)0.

009

(0.0

09)

Per

capi

tain

com

e−

0.12

8(0

.050

)***

0.12

0(0

.060

)**

0.15

4(0

.039

)***

−0.

199

(0.1

40)

−0.

092

(0.0

43)*

*0.

069

(0.0

24)*

**−

0.09

8(0

.055

)*−

0.16

4(0

.054

)***

Dum

my

for

Nor

th

and

Eas

t

−0.

013

(0.0

07)*

*−

0.01

0(0

.006

)*−

0.01

0(0

.008

)0.

005

(0.0

06)

0.00

8(0

.004

)**

−0.

011

(0.0

06)*

−0.

008

(0.0

05)*

−0.

005

(0.0

05)

R2

0.80

0.91

0.90

0.51

0.33

0.85

0.70

0.80

Obs

erva

tion

s60

6060

6060

6060

60

Not

e:R

obus

tst

anda

rder

rors

repo

rted

wit

hin

pare

nthe

sis.

*,**

,**

*,si

gnifi

cant

atth

e10

%,

5%,

and

1%le

vels

,re

spec

tivel

y.

EL

F,et

hnol

ingu

isti

cfr

acti

onal

isat

ion;

PO

LF

RA

C,

poli

tica

lfr

acti

onal

isat

ion.


percentage of minority groups. Once again, political fractionalisation has a largernegative statistically significant impact on the provision of public services comparedwith the ELF index. The interaction of the ELF × POLFRAC has a negative significantimpact on the provision of public schools, school teachers, medical officers, water,electricity supply, and on the maternal and infant mortality rates. The coefficient onTamil is negative and statistically significant, in columns (1), (4), (6), and (8), suggest-ing that the number of government schools, electricity availability, and road connec-tions to other cities are lower if belonging to the Tamil minority group. For Indian Tamilcommunity, government teachers, health medical officers, electricity, water availability,and road connections are lower, and for the Muslim community, health medical officersand electricity availability are lower and maternal deaths are also lower.

The coefficient on population has a statistically significant positive effect onschools, school teachers, medical officers, maternal deaths, and road kilometerage.Larger districts as measured by population have more schools, school teachers,medical officers, and are better connected to other cities. They also have a largernumber of maternal deaths. The coefficients on the literacy rate in columns (1)–(4)are statistically significant suggesting that more literate districts are more likely tohave more schools, school teachers, medical officers, and fewer maternal deaths. Theestimates on the Gini coefficient are statistically significant in columns (1), (2), (4),and (6). The provision of schools and school teachers are higher, and the supply ofwater lower, the higher the inequality in income distribution. Higher levels of incomeinequality lead to more maternal deaths. Per capita income is not statistically signifi-cant. The dummy variables for the Northern and Eastern districts are negative incolumns (1), (2), and (5)–(7) suggesting that the provision of schools, teachers, elec-tricity, and water were lower in these areas in the war-affected years and infant deathswere higher.

Additional control variables which included population density, the percentage ofurban and rural population, and land ownership were also explored with. The coeffi-cients on population density indicate that higher levels of population density lead to ahigher provision of public goods. The coefficients on land ownership and urban-ruralpopulation were not statistically significant.

Table 7 reports results for non-linearity effects in ELF. The ELF index is statisticallysignificant in columns (2), (6), and (7). The political fractionalisation index continuesto be statistically significant in all columns except for columns (4) and (8). The squaredterm on the ELF index is statistically significant at the 10 percent level only in the caseof health medical officers. There is in general, no evidence of non-linear effects of ELFon the provision of public goods.

Table 8 replicates the estimation carried out in Table 5, replacing the ELF index withthe EPI and the political fractionalisation index with a political polarisation index. Thecoefficients on the EPI are statistically significant at the 5 percent level in columns (1)and (3) indicating a negative effect on government school and health medical officers.The political polarisation index assumes greater statistical significance compared with


TAB

LE

7.T

ES

TIN

GF

OR

NO

N-L

INE

AR

EF

FE

CT

S.

Dep

ende

ntva

riab

les

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Inde

pend

ent

vari

able

s

Gov

ernm

ent

scho

ols

Gov

ernm

ent

teac

hers

Gov

ernm

ent

heal

th

med

ical

offi

cers

Mat

erna

lde

aths

Infa

nt

mor

tali

ty

Ele

ctri

city

avai

labi

lity

Wat

er

avai

labi

lity

Roa

d

kilo

met

erag

e

EL

F−

0.00

2(0

.003

)−

0.00

6(0

.003

)**

−0.

001

(0.0

02)

0.00

2(0

.004

)0.

005

(0.0

04)

−0.

004

(0.0

02)*

−0.

006

(0.0

02)*

**−

0.00

4(0

.004

)

PO

LF

RA

C−

0.00

5(0

.002

)**

−0.

006

(0.0

03)*

−0.

007

(0.0

02)*

**0.

001

(0.0

01)

0.00

6(0

.002

)***

−0.

002

(0.0

01)*

−0.

006

(0.0

02)*

**0.

002

(0.0

03)

EL

F2

0.00

2(0

.003

)0.

002

(0.0

02)

0.00

2(0

.001

)*−

0.00

3(0

.002

)−

0.00

1(0

.001

)0.

003

(0.0

04)

0.00

1(0

.001

)0.

002

(0.0

038)

R2

0.06

0.09

0.10

0.05

0.07

0.08

0.05

0.06

Obs

erva

tion

s65

6567

7373

6161

66

Not

e:R

obus

tst

anda

rder

rors

repo

rted

wit

hin

pare

nthe

sis.

*,**

,**

*,si

gnifi

cant

atth

e10

%,

5%,

and

1%le

vels

,re

spec

tivel

y.

EL

F,et

hnol

ingu

isti

cfr

acti

onal

isat

ion;

PO

LF

RA

C,

poli

tica

lfr

acti

onal

isat

ion.


TAB

LE

8.U

SIN

GT

HE

PO

LA

RIZ

AT

ION

IND

ICE

S.

Dep

ende

ntva

riab

les

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Inde

pend

ent

vari

able

s

Gov

ernm

ent

scho

ols

Gov

ernm

ent

teac

hers

Gov

ernm

ent

heal

th

med

ical

offi

cers

Mat

erna

l

deat

hs

Infa

nt

mor

tali

ty

Ele

ctri

city

avai

labi

lity

Wat

er

avai

labi

lity

Roa

d

kilo

met

erag

e

EP

I−

0.00

6(0

.003

)**

−0.

001

(0.0

01)

−0.

004

(0.0

02)*

*0.

004

(0.0

04)

0.00

2(0

.002

)−

0.00

2(0

.002

)−

0.00

1(0

.001

)−

0.00

3(0

.005

)

PO

LP

I−

0.01

0(0

.005

)**

−0.

004

(0.0

02)*

*−

0.00

6(0

.003

)**

0.00

5(0

.006

)0.

005

(0.0

02)*

*−

0.00

4(0

.001

)***

−0.

006

(0.0

03)*

−0.

006

(0.0

05)

EP

I×

PO

LP

I−

0.01

0(0

.005

)*−

0.01

0(0

.005

)*−

0.00

6(0

.002

)***

0.00

8(0

.010

)0.

006

(0.0

03)*

*−

0.00

7(0

.003

)**

−0.

006

(0.0

03)*

−0.

005

(0.0

02)*

*

Popu

lati

on0.

005

(0.0

02)*

**0.

031

(0.0

10)*

**0.

021

(0.0

10)*

*0.

013

(0.0

06)*

*0.

040

(0.0

34)

−0.

020

(0.0

11)*

*−

0.02

3(0

.024

)0.

016

(0.0

10)*

Lit

erac

yra

te0.

014

(0.0

10)

0.02

0(0

.010

)*0.

021

(0.0

07)*

**−

0.01

5(0

.007

)*−

0.02

4(0

.012

)**

0.04

5(0

.043

)0.

020

(0.0

19)

0.03

2(0

.031

)

Gin

ico

effi

cien

t−

0.02

1(0

.021

)0.

027

(0.0

12)*

*0.

031

(0.0

15)*

*0.

054

(0.0

10)*

**0.

035

(0.0

16)*

*−

0.00

6(0

.005

)−

0.00

4(0

.007

)0.

021

(0.0

25)

Per

capi

tain

com

e−

0.01

2(0

.004

)***

0.01

0(0

.005

)**

0.01

4(0

.007

)**

−0.

020

(0.0

08)*

**−

0.00

3(0

.001

)**

0.00

9(0

.004

)***

−0.

008

(0.0

02)*

*−

0.01

3(0

.007

)**

Dum

my

for

Nor

th

and

Eas

t

−0.

007

(0.0

03)*

**−

0.00

5(0

.002

)**

−0.

008

(0.0

04)*

0.00

3(0

.003

)0.

003

(0.0

04)

−0.

011

(0.0

05)*

*−

0.01

0(0

.005

)**

−0.

004

(0.0

04)

R2

0.78

0.77

0.75

0.65

0.68

0.66

0.56

0.59

Obs

erva

tion

s60

6060

6060

6060

60

Not

e:R

obus

tst

anda

rder

rors

repo

rted

wit

hin

pare

nthe

sis.

*,**

,**

*,si

gnifi

cant

atth

e10

%,

5%,

and

1%le

vels

,re

spec

tivel

y.

EP

I,et

hnic

pola

risa

tion

;P

OL

PI,

poli

tica

lpo

lari

sati

on.


the ethnic EPI as with the political fractionalisation index.The interaction terms betweenthe EPI and political polarisation index is statistically significant in all columns with theexception of column (4) suggesting that political polarisation magnifies the effects ofethnic polarisation having an adverse effect on public good provision.

Fixed effects estimation. Next fixed effects estimation is carried out to account fordistrict-level time invariant unobservable influences on the provision of public goods.The fixed effects estimation is replicated on the estimation in Table 5. The results arereported in Table 9.

The fixed effects estimation yields better results with the coefficient on the ELFindex assuming statistical significance in columns (1), (2), (5), (6), (7), and (8),suggesting greater within district effects in the relation between ELF and public goodprovision. Greater ELF leads to a lower provision of schools, teachers, electricity, andwater, and higher infant mortality. Increased political fractionalisation leads to a lowersupply of schools, teachers, health medical officers, electricity, water, and higher infantmortality rates. Larger districts as measured by population are better off with respect toschools, teachers, medical officers, and electricity. Per capita income is statisticallysignificant in all columns except (3) and (7) suggesting that higher income leads toincreased schools, school teachers, electricity, and road kilometerage and lower mater-nal and infant mortality.

ConclusionsThis study investigates the effect of ELF and political fractionalisation on the

provision of public goods in Sri Lanka. The impact of ethnic polarisation and politicalpolarisation on public good provision are also examined. The political fractionalisationand political polarisation indices have a greater adverse effect on the provision ofpublic goods compared with the ELF and EPI. While the political fractionalisationindex has a consistently negative effect on public good provision under both estimationmethods, the ELF index assumes greater statistical significance under fixed effectsestimation. The results suggest that the relationship between ELF and public goodprovision is stronger within given districts suggesting the importance of district-specific effects in the relation between ELF and public good provision.3 While theresults are consistent with the previous literature (for example, Alesina, Baqir andEasterly 1999; Alesina and La Ferrara 2000; Banerjee, Iyer and Somanathan 2005;Banerjee and Somanathan 2007), the present study offers a new dimension to theexisting literature with regard to interpretation. The results also suggest that ethnic andpolitical fractionalisation cannot be considered in isolation of one another. Since 1994,the two main political parties in Sri Lanka have had to accommodate the needs of anumber of smaller less moderate parties which may have potentially slowed down theprovision of public goods in particular districts. Therefore, policymakers should con-sider the problems arising from fractionalised (or polarised) party systems in theirattempts to achieve sustainable growth. With the cessation of the war in 2009, Sri


TAB

LE

9.PA

NE

LF

IXE

DE

FF

EC

TS

ES

TIM

AT

ION

WIT

HA

DD

ITIO

NA

LC

ON

TR

OL

VA

RIA

BL

ES.

Dep

ende

ntva

riab

les

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Inde

pend

ent

vari

able

sG

over

nmen

t

scho

ols

Gov

ernm

ent

teac

hers

Gov

ernm

ent

heal

th

med

ical

offi

cers

Mat

erna

l

deat

hs

Infa

ntm

orta

lity

Ele

ctri

city

avai

labi

lity

Wat

erav

aila

bili

tyR

oad

kilo

met

erag

e

EL

F−

0.00

9(0

.002

)***

−0.

003

(0.0

01)*

**−

0.00

5(0

.004

)0.

001

(0.0

01)

0.00

4(0

.002

)**

−0.

007

(0.0

02)*

**−

0.00

5(0

.003

)*−

0.00

4(0

.002

)**

PO

LF

RA

C−

0.01

0(0

.004

)***

−0.

004

(0.0

02)*

*−

0.00

4(0

.002

)*0.

004

(0.0

03)

0.00

5(0

.002

)**

−0.

004

(0.0

02)*

−0.

004

(0.0

02)*

−0.

001

(0.0

03)

EL

F×

PO

LF

RA

C−

0.00

6(0

.003

)*−

0.00

5(0

.002

)*−

0.00

6(0

.002

)**

0.00

5(0

.004

)0.

006

(0.0

03)*

−0.

008

(0.0

02)*

**−

0.00

4(0

.002

)*0.

003

(0.0

04)

Popu

lati

on0.

010

(0.0

05)*

*0.

014

(0.0

05)*

**0.

011

(0.0

03)*

*0.

014

(0.0

12)

0.02

6(0

.010

)***

0.01

3(0

.007

)***

0.01

3(0

.016

)0.

026

(0.0

24)

Lit

erac

yra

te0.

019

(0.1

30)

0.02

7(0

.009

)***

0.01

0(0

.005

)**

−0.

188

(0.1

06)

−0.

013

(0.1

09)

0.01

7(0

.014

)0.

036

(0.0

16)*

*0.

029

(0.0

09)*

**

Gin

ico

effi

cien

t−

0.01

6(0

.100

)0.

018

(0.0

14)

0.01

5(0

.055

)0.

032

(0.0

15)*

*0.

016

(0.0

48)

0.03

6(0

.070

)0.

016

(0.1

22)

−0.

003

(0.0

50)

Per

capi

tain

com

e0.

014

(0.0

05)*

**0.

012

(0.0

04)*

**0.

010

(0.0

13)

−0.

018

(0.0

07)*

−0.

016

(0.0

04)*

**0.

010

(0.0

05)*

*0.

017

(0.0

48)

0.01

0(0

.005

)**

Dum

my

for

Nor

th

and

Eas

t

−0.

011

(0.0

06)*

−0.

009

(0.0

05)*

−0.

008

(0.0

07)

−0.

005

(0.0

05)

−0.

006

(0.0

04)*

−0.

010

(0.0

06)*

−0.

008

(0.0

05)*

−0.

003

(0.0

04)

R2

0.41

0.46

0.52

0.33

0.26

0.51

0.26

0.34

Obs

erva

tion

s60

6060

6060

6060

60

Not

e:R

obus

tst

anda

rder

rors

repo

rted

wit

hin

pare

nthe

sis.

*,**

,**

*,si

gnifi

cant

atth

e10

%,

5%,

and

1%le

vels

,re

spec

tivel

y.

EL

F,et

hnol

ingu

isti

cfr

acti

onal

isat

ion;

PO

LF

RA

C,

poli

tica

lfr

acti

onal

izat

ion.


Lanka now faces the task of achieving a sustainable growth rate. While Sri Lanka isregarded as a model of a welfare state, it can be argued that public service provision canbe enhanced to a greater degree, particularly districts in which political and socialdivides are high; if the two main political parties in consultation with other parties areable to reach common ground concerning successful democratization. This woulddepend to a large degree upon the construction of institutions that promote the rule oflaw and restrain the concentration of power. “The design of these institutions needs tobe based on a realistic appreciation of the nature and power of the interests involved”(Barnes 2001).

NOTES1. The Indian Tamil people were brought in by the British colonisers to work on the tea plantations.

2. Descendants of the Dutch and Portuguese.

3. This is supported by Forbes (2000) who finds that when country effects are incorporated into a pooled

OLS model, the relationship between inequality and growth becomes positive and significant suggesting

the importance of country-specific effects.

REFERENCESAghion, P., A. Alesenia, and F. Trebbi. 2004. Endogenous political institutions. Quarterly Journal of

Economics 119(2): 565–611.

Alesina, A. 2004. Ethnic diversity and economic performance. NBER Working Paper no. 10313.

Alesina, A., and E. La Ferrara. 2000. Participation in heterogeneous communities. Quarterly Journal of

Economics 115(3): 847–904.

Alesina, A., R. Baqir, and W. Easterly. 1999. Public goods and ethnic divisions. Quarterly Journal of

Economics 114(4): 1243–1284.

Annett, A. 2001. Social fractionalization, political instability, and the size of the government. IMF Staff

Papers 48(3): 561–592.

Banerjee, A., and R. Somanathan. 2007. The political economics of public goods: Some evidence from India.

Journal of Development Economics 82(2): 287–314.

Banerjee, A., L. Iyer, and R. Somanathan. 2005. History, social divisions and public goods in rural India.

Journal of the European Economic Association 3(2–3): 639–647.

Barnes, S. 2001. The contribution of democracy to rebuilding postconflict societies. American Journal of

International Law 95(1): 86–101.

Bobo, L., and J.R. Kluegel. 1993. Opposition to race targeting: Self interest, stratification ideology, or racial

attitudes. American Sociological Review 58(4): 443–464.

Canning, D., and M. Fay. 1993. The effects of transportation networks on economic growth. Columbia

University Working Paper.

Collier, P. 2000. Ethnicity, politics and economic performance. Economics & Politics 12(3): 225–245.

Department of Census and Statistics. 2012. Population atlas of Sri Lanka 2012. Department of Census and

Statistics, Ministry of Finance and Planning, Sri Lanka.

DeVotta, N. 2000. Control democracy, institutional decay, and the quest for Eelam: Explaining ethnic conflict

in Sri Lanka. Pacific Affairs 73(1): 55–76.

Dincer, O. 2008. Ethnic and religious diversity and corruption. Economics Letters 99(1): 98–102.

Dincer, O., and P. Lambert. 2012. Taking care of your own ethnic and religious heterogeneity and income

inequality. Journal of Economic Studies 39(3): 290–313.


Easterly, W., and R. Levine. 1997. Africa’s growth tragedy: Policies and ethnic divisions. Quarterly Journal

of Economics 112(4): 1203–1250.

Esteban, J., and D. Ray. 1994. On the measurement of polarization. Econometrica 62(4): 819–851.

Forbes, K. 2000. A reassessment of the relationship between inequality and growth. American Economic

Review 90(4): 869–887.

Horowitz, D. 1985. Ethnic groups in conflict. Berkeley: University of California Press.

———. 1993. Democracy in divided societies. Journal of Democracy 4(4): 18–38.

International Crisis Group. 2007. Sri Lanka: Sinhala nationalism and the elusive Southern consensus. Asia

Report no. 141.

Kimenyi, M. 2006. Ethnicity, governance and the provision of public goods. Journal of African Economies

15(1): 62–99.

La Porta, R.F., A. Lopez de Silanes, R. Shleifer, and R. Vishny. 1999. The quality of government. Journal

of Law, Economics and Organization 15(1): 222–279.

Mauro, P. 1995. Corruption and growth. Quarterly Journal of Economics 110(3): 681–712.

Miguel, E., and M. Gugerty. 2005. Ethnic diversity, social sanctions, and public goods in Kenya. Journal of

Public Economics 89(11–12): 2325–2368.

Montalvo, J.G., and M. Reynal-Querol. 2002. Why ethnic fractionalization? Polarization, Ethnic Conflict and

Growth, UPF Working Paper 660.

Mozaffar, S., J. Scarritt, and G. Galaich. 2003. Electoral institutions, ethnopolitical cleavages, and party

systems in Africa’s emerging democracies. American Political Science Review 97(3): 379–390.

Posner, D. 2007. Regime change and ethnic cleavages in Africa. Comparative Political Studies 40(1):

1302–1327.

Singer, M. 1992. Sri Lanka’s Tamil-Sinhalese ethnic conflict. Asian Survey 32(8): 712–722.

Uyangoda, J. 2010. Working and outcomes of democracy in Sri Lanka: Paper prepared for the Project on

State of Democracy in South Asia as part of the Qualitative Assessment of Democracy Lokniti (Pro-

gramme of Comparative Democracy). Centre for the Study of Developing Societies Delhi.

World Bank. 2011. World Development Indicators.


Ethnic or Political Fractionalisation? A District Level Analysis of the Provision of Public Goods in...

Documents

Transcript of Ethnic or Political Fractionalisation? A District Level Analysis of the Provision of Public Goods in...