Public and Private Funding of Basic Education in Zambia

Public and Private Fundingof Basic Educationin Zambia

Implications ofBudgetary Allocationsfor Service Delivery

Jishnu Das, The World Bank

Stefan Dercon, Oxford University

James Habyarimana, Harvard University

Pramila Krishnan, Cambridge University

Africa Region Human DevelopmentWorking Paper Series

The World Bank

Copyright © 2004Human Development SectorAfrica RegionThe World Bank

ii

The views expressed herein are those of the authorsand do not necessarily reflect the opinions or policies ofthe World Bank or any of its affiliated organizations.

Cover Design by Word ExpressInterior Design by Jennifer VitoCover photo: Drawing by Josephine Ngoma, Student, Namanongu School, Chongwe District.

Contents

iii

Foreword ...................................................................................................................................................... vAbstract ...................................................................................................................................................... viiAcknowledgments ..................................................................................................................................... viiiAbbreviations and Acronyms ...................................................................................................................... ixExecutive Summary ......................................................................................................................................x1. Introduction ........................................................................................................................................... 172. Zambian Education during the 1990s ................................................................................................... 193. Tracking Methodology ........................................................................................................................... 224. School Characteristics of the Sample .................................................................................................... 265. Expenditure Tracking at Provincial, District, and School Levels ........................................................... 296. Leakage of Nonsalary and Payroll Funds ............................................................................................... 377. Equity in Education Funding .................................................................................................................. 438. Private Expenditures for Education ...................................................................................................... 519. Discussion and Conclusions .................................................................................................................. 59Bibliography ............................................................................................................................................... 60Appendix .................................................................................................................................................... 62

TablesTable 1. Enrollment Rates, 1992–2001Table 2. Structure of Resource and Funding Flows in ZambiaTable 3. Enrollment, Urbanization, and PovertyTable 4. Mean Math and English Scores, by ProvinceTable 5a. School Enrollment and StaffingTable 5b. InfrastructureTable 5c. Key Performance IndicatorsTable 5d. Household IndicatorsTable 6. Cash Received by Province Education OfficesTable 7. Funding Sources at Province Education Office Level, June 2001–June 2002Table 8. Funding Disbursed to Districts, January 2002–June 2002Table 9. Cash Received by DistrictsTable 10. Funding Sources at the District LevelTable 11. Tracking of Resources, June 2001–June 2002Table 12. Tracking of Resources from the District Level to Schools, 2001–2002

iv

Table 13a. Disbursement of Fixed-Grant Allocations, by ProvinceTable 13b. Disbursement of Salaries, by ProvinceTable 13c. Disbursements of Rule-Based Allowances, by ProvinceTable 14. Percentage of Schools That Receive No Funds, by Source and LocationTable 15. Average Amounts Conditional on Receipts, by ProvinceTable 16. Discretionary Transfers to Teachers, by ProvinceTable 17. School and District Characteristics, by ProvinceTable 18. Enrollment, Urbanization, Wealth, and DistanceTable 19a. Equity in Rule-Based Funding at the District LevelTable 19b. Equity in Rule-Based Funding at the School LevelTable 20a. Equity in Discretionary Funding at the District LevelTable 20b. Equity in Discretionary Funding at the School LevelTable 20c. Equity in Discretionary Funding at the School Level, by LocationTable 21a. Equity in Per-Pupil Staff Remuneration at the School LevelTable 21b. Equity in Per-Pupil Staff Remuneration, Class Size and SalariesTable 22. Decomposing the Decline in FeesTable 23. Public and Private Sources of FundingTable 24. Equity in Public and Private Funding at the School LevelTable 25. Fee and Nonfee Expenditures, by ProvinceTable 26. Equity in Public and Private SpendingTable 27. Substitutability between Public and Private SpendingTable 28. Household Expenditure and Public FundingTable A1. Variables Used in Construction of the Asset Index

FiguresFigure 1. Funding Flows across Centralized and Decentralized ProvincesFigure 2. Characteristics of School Enrollment in ZambiaFigure 3. Funding Flows across Centralized and Decentralized ProvincesFigure 4. Funding Disbursement in Centralized and Decentralized ProvincesFigure 5. Variation in District ReceiptsFigure 6. Inequalities in Public Funding in Zambian Basic EducationFigure 7. Declining Fees since 2000Figure 8. Fee Decline by Province and LocationFigure 9. The Relationship between PTA Fees and Wealth, 2000–2002Figure 10. Household Spending on Education by Wealth TercilesFigure 11. Lorenz Curves of Inequalities in Public and Private FundingFigure A1. Asset Index Characteristics for All-SampleFigure A2. Asset Index Characteristics for Rural SampleFigure A3. Distribution of Wealth in All-Sample and Separated by LocationFigure A4. Comparison of All-Sample and Rural Index

Foreword

v

Education is at the center of many poverty re-duction strategies. Yet too often, educationservices fail to improve outcomes for a vari-

ety of reasons. Budgetary allocations often favor thebetter-off, limiting poor people’s access to servicesor preventing improvements in quality. In many poorcountries improvements in educational outcomes callfor institutional—not merely managerial—reforms.Such reforms include bottom-up measures to giveusers a stronger voice and more power over pro-viders. They also include top-down measures toensure better monitoring of providers and introduceeffective incentives for improving staff performance.Both types of reform depend on a body of system-atic information on performance, incentives, andother aspects of frontline service delivery. This in-formation is indispensable for catalyzing and guidingthe institutional reforms needed to improve educa-tion and educational outcomes—yet little of theseessential data are currently available.

To help fill this gap, the Development ResearchGroup of the World Bank is carrying out, in collabo-ration with local institutions and Bank operations, amulticountry study of education service delivery inLaos, Pakistan, Papua New Guinea, Uganda, as wellas Zambia, the subject of this paper. The purpose ofthe research is to understand the relationship be-tween funding and educational outcomes by track-ing funding and resources and relating them at dif-ferent levels of the delivery system to outcomes indifferent institutional and organizational contexts.

The education research program pilots the Quan-titative Service Delivery Survey (QSDS), in which the

school is the primary unit of observation. Beyond itsuse in analyzing provision of services, the QSDS fitswell into the larger goal of impact evaluation. Whencombined with household surveys, it allows explo-ration of interactions between frontline providersand users of services. By adding surveys of local poli-ticians and officials, it can also shed light on the po-litical economy of service delivery and on interac-tions between providers and policymakers.

The Education Service Delivery Study (ESD) inZambia carefully examines the structure of fundingand service delivery to better understand the rela-tionship between educational inputs and outcomes.It specifically takes into account linkages that mayarise among different players involved in the deliv-ery of education (for instance, the province educa-tion office, the district education office, schools, par-ent teachers associations, and pupils) and seeks toexamine the ways through which these players re-spond to changes in the institutional setting and fund-ing structure. The ESD addresses these importantissues through the careful measurement of schoolinputs both at the school and the household level, aswell as educational outcomes, through tests admin-istered to the same pupils at two different times (incollaboration with the Examination Council of Zam-bia).

The Government of the Republic of Zambia (GRZ)has long recognized the importance of education.Nevertheless, a sharp decline from the mid-1980sonward in copper prices has led to a commensuratedecline in GNI per capita, from $590 in 1975 to $300in 2000, and this has presented several obstacles for

vi

the education sector. The government, along with aconsortium of donors, has been working closely toimprove educational outcomes. The Education Ser-vice Delivery Study addresses important issues re-garding the link between this effort and educationaloutcomes.

This report provides a detailed evaluation of pub-lic expenditure. It includes both a public expendi-ture tracking and a funding-equity exercise. Thetracking exercise provides information on the ex-tent to which educational expenditures earmarkedunder the central budget actually reach schools, andthe funding-equity exercise examines whether suchfunding can be regarded as progressive—to whatextent, if any, do poorer schools receive a greatershare of public funding than their richer counterparts,and if so, how does this funding impact inequality ineducational expenditure? The evaluation provides acomplete picture of funding to primary and basicschools based on allocated and discretionary bud-getary components, and household contributions toeducation expenses.

This report makes three contributions to theunderstanding of education funding in the Zambiancontext. First, it shows that funding characteristicsare closely linked to the type of disbursement. Fundsallocated according to a clear, predefined rule reachschools with little evidence of diversion. A signifi-cant proportion of funds allocated at the discretionof district and provincial educational offices, how-ever, are retained at higher levels of the hierarchy.

Second, rule-based allocations have led to greaterper-pupil funding for poorer and more rural schools.

These allocations, however, are the only progressivedisbursements in the survey. The disbursement ofdiscretionary allocations is neutral with respect towealth, and per-pupil staff allocations are higher inurban and richer schools. Once all sources of publicfunding are factored into the analysis, public schoolfunding in Zambia is regressive, with almost 30 per-cent higher allocations to richer schools.

Third, private expenditures at the level of thehousehold have a critical impact on equity in educa-tional funding. In fact, the inequality in public expen-diture is dominated by the large share of privateexpenditures in overall educational spending. More-over, these expenditures adjust to school funding,decreasing when funding rises. The magnitude ofthese adjustments appear to be the same across richand poor households, suggesting that higher publicfunding may be useful in decreasing inequality across,but not within, communities.

The findings are highly relevant for educationalreforms being undertaken in many developing coun-tries to improve educational outcomes. My hope isthat by offering a new perspective from the frontlines of education, this paper will make a useful con-tribution to the reform agenda for improving educa-tion services for poor people in Zambia and else-where.

Birger J. FredriksenSenior Education Advisor

Human Development DepartmentAfrica Region

Abstract

T his report presents findings from a surveyof 182 primary (grades 1-7) and basic (grades1-9) schools carried out in Zambia in 2002.

It describes and analyzes resource flows to theseschools from three sources: rule-based funding fromthe center, discretionary funding from district andprovincial education offices, and household spend-ing on education.

Rule-based funds reached schools exactly as ear-marked. The rule-based component of funding washighly progressive, as the same amount was dis-bursed to all schools irrespective of enrollment. Sincesmall schools tend to have poorer student bodies,the rule-based allocation per pupil translated to morefunding for poorer students.

Discretionary funds, controlled by the provinceand district education offices, reached only 25 per-cent of schools. The discretionary funds were wealthneutral. Even shares were distributed to schools withpoor and nonpoor students.

Household educational expenditures show thatnonfee expenditures by the family are seven times

the corresponding expenditure on fees, makingthem the main source of inequalities in private ex-penditure. Once private expenditure is factoredinto the analysis, the nonprogressive nature of thepublic education funding system worsens: the shareof educational expenditures for the poorest 50 per-cent of the population declines from 40 to 34 per-cent. Furthermore, when examining substitutionbetween private expenditures and public funding,there is strong evidence that households decreaseexpenditures when public funding increases.

Although public funding could address inequali-ties in educational spending with progressive allo-cation across villages and schools, the desired im-pact of such redistributions may be less than imag-ined due to the crowding-out of private expendi-tures. The report suggests that increases in fund-ing may not be the optimal way to improve educa-tional attainment. It might be more beneficial toconcentrate instead on providing inputs that house-holds cannot supply on their own, such as high-quality teachers.

vii

Acknowledgments

This study would not have been possiblewithout the support and guidance of anumber of institutions and individuals. We

would especially like to thank Hinh Dinh and BruceN. Jones for their keen interest and suggestions atevery stage of the project, Ritva Reinikka for hercontinued support and encouragement, and KeithWood and Richard Arden at DFID for funding thesurvey. The study relied on critical support from ourZambian counterparts, particularly ClementSiamatowe at the World Bank; RuralNet Associates

and Professor Mwanawina at the University of Zam-bia for their excellent support and survey work in thefield; Joe Kanyika, Teza Nakazwe, C. T. Sakala, andArnold Chengo at the Examination Council of Zam-bia and Ministry of Education; the Permanent Secre-tary of the Ministry of Education, Barbara Y. Chilangwa;and J. Chiwele from the Accountant General’s office.Finally, Mushiba Nyamazana has been an integral partof this study from its inception; he helped design thesurveys used in the study and provided invaluable in-sight into the educational system.

viii

Abbreviations and Acronyms

BESSIP Basic Education Sub-Sector Investment ProgramDEO District Education Office (Officer)ESD Education Service DeliveryESDS Education Service Delivery SurveyGDP Gross Domestic ProductGNI Gross National InvestmentGRZ Government of the Republic of ZambiaHIPC Debt Initiative for Heavily Indebted Poor Countries (World Bank)IRT Item Response TheoryLCMS Living Conditions Monitoring SurveyNGO Nongovernmental organizationPAGE Program for the Advancement of Girls’ EducationPEO Province Education Office (Officer)PTA Parent Teachers AssociationQSDS Quantitative Service Delivery Survey

ix

Executive Summary

The link between education and develop-ment provides a strong case for allocatingpublic expenditures to the education sector.

However, allocating more central budgetary re-sources to the education sector will not necessarilydeliver better outcomes. Budgetary resources maynot reach the intended beneficiaries, and even if theydo, these resources may not lead to the desiredoutcomes. This report examines the structure offunding and service delivery in Zambia using a re-cently completed Educational Service Delivery Sur-vey (ESDS).

The Government of the Republic of Zambia(GRZ) has long recognized the importance of edu-cation. Nevertheless, a sharp decline from the mid-1980s onward in copper prices has led to a com-mensurate decline in GNI per capita from $590 in1975 to $300 in 2000, and this has presented sev-eral obstacles for the education sector. The govern-ment and a consortium of donors have been work-ing together to improve educational outcomes inZambia. The survey and study ask two questions:What is the current record of the education sectorin delivering services to the intended recipients? Havethe changes undertaken by the Ministry of Educa-tion had the desired consequences in terms of edu-cational outcomes?

The report focuses on a detailed evaluation ofpublic expenditure—through both a tracking and afunding-equity exercise—to gauge the extent towhich educational expenditures earmarked under

the central budget actually reach schools, as well asthe extent to which such funding can be regarded asprogressive. Of the four issues that concern us dur-ing this exercise, the first three relate to the publicfunding of education and the last examines privateeducational funding and its relationship to public ex-penditure. These issues are as follows:

• What percentage of resources is spent at eachlevel of the administrative hierarchy?

• To what extent is variation in funding acrossdistricts and schools explained by funding for-mulas that relate budgetary allocation to dis-trict and school characteristics such as enroll-ment or the number of schools?

• To the extent that there is variation in fundingacross districts and schools, what is the rela-tionship between this variation and character-istics of districts and schools? (For instance,doricher districts and schools systematically re-ceive more funds?)

• How important is private funding in the provi-sion of education and how does this compo-nent of funding relate to equity in the overallfunding of education?

In Zambia public funding for education flows toschools through a three-tiered administrative hierar-chy involving the province offices, the district offices,and schools. Moreover, funds do not necessarily flowin a top-down manner; at each tier of the hierarchy,funds may come directly from the government or

x

from international and national donors. In our analy-sis, we classify these funds into four categories:

• Rule-based allocations to schools. A fixed-grant of $600 or $650 is allocated to eachschool, depending on the type of school. Thegrant is not a per-pupil allocation and it is inde-pendent of the enrollment in the school.

• Discretionary allocations to schools. Suchfunds are allocated at the discretion of districtsand provinces and are over and above the rule-based allocation.

• Rule-based allocations to teachers. Teachersare paid directly via deposit slips through a cen-tralized payroll in Zambia. This money does notpass through the province or district offices.

• Discretionary allocations to teachers. In ad-dition to salaries and monthly allowances,teachers are entitled to one-time benefits,such as leave benefits, transfer benefits (paidwhen a teacher switches schools), and funeralbenefits. The payment of such benefits is left

to the discretion of the provincial and district-level administration.

It is also necessary to distinguish between prov-inces with and without district education boards. Al-though districts receive money directly from the GRZ,both for rule-based and discretionary allocations, allmoney for discretionary allocations is first transferredto the concerned province, and from there moves on-ward to the district. For this reason, provinces withdistrict education boards are referred to as decentral-ized provinces and those without district educationboards are referred to as centralized provinces. In thestudy sample there are two of each: Lusaka andCopperbelt provinces are decentralized and Northernand Eastern province are centralized. Figure 1 showshow funds are allocated across both types of provinces.

EXPENDITURE TRACKING

The tracking exercise follows all nonsalary fundingflows through the administrative hierarchy (recall that

xi

Figure 1. Funding Flows across Centralized and Decentralized Provinces

Note: Decentralized provinces are Lusaka and Copperbelt. Centralized provinces are Northern and Eastern.

Ministry of Education

(Centralized) (Decentralized)

District(Centralized)

District(Decentralized)

SchoolsHouseholdsDiscretionary flows

Rule-based flows

Case IV donors (primarily the Program

for Advancement of Girl’s Education






Rule-based flows



Province Province

Leakage =

This report, however, goes beyond an estimateof leakage based on allocated funds to provide a com-plete picture of the funding of educational institu-tions. It examines all sources of funding for the school.For example, in the case of discretionary compo-nents the rule-based methodology fails (since thereis no rule about the amounts that schools are sup-posed to receive, it is not possible to determine whatconstitutes leakage in the system). After tracking theamounts that schools receive, it’s possible to exam-ine the equity implications of such allocations. Thesection on leakage establishes the following:

• Rule-based allocations to schools in theZambian educational system seem towork efficiently. There is little evidence tosuggest that funds earmarked for disburse-ments do not reach intended beneficiaries.Specifically, more than 90 percent of allschools (95 percent in all provinces exceptLusaka) had received the rule-based allocationat the time of the survey, and delays in disburse-ment rather than leakage of funds was a morelikely explanation in the case of schools that hadnot received the grant (this grant was disbursedtwo months prior to the survey).

• Rule-based allocations to teachers—sala-ries and clearly defined allowances—aredisbursed efficiently. There is some evidenceof delays in the updating of the payroll sys-tem, as well as significant arrears in the caseof allowances that are not clearly specified. Inthe case of salaries, 95 percent of teachers hadno outstanding amounts. For allowanceswhere there is a clear specification based onlocation (hardship allowance in rural locations)or status (teacher trainee allowance), less than15 percent of payments were overdue by sixmonths or more. However, in the case of over-time allowances (which must be filed everyterm), and allowances that resulted from achange in the status of the teacher (such asadded tasks with commensurate allowances),50 percent of payments were overdue. For allallowances there is considerable evidence that

payroll goes from the center directly to teachers). Theexpenditure flows can be characterized as follows:

• On average, K 28,000 per pupil enters the edu-cational system for the four provinces surveyed.This amount hides a significant degree of varia-tion among provinces with, for instance, East-ern Province (K 44,300) receiving more thandouble the per-pupil funding received byCopperbelt (K 19,000).

• Of this amount, discretionary funds at the levelof the province and district account for 70 per-cent of all funding and rule-based funds accountsfor the remaining 30 percent. Thus rule-basedfunds allocated through the fixed-school grantof $600 ($650) account for less than one-thirdof all the funding received.1

• Between one-sixth and one-third of total fund-ing in the system eventually reaches the schools.Again, there are significant differences betweenprovinces, with 34 percent of provincial-levelfunds reaching schools in Copperbelt, and only14 percent reaching schools in Lusaka.

• Although decentralization has shifted spendingfrom the provincial to the district level, it hasnot resulted in greater disbursements toschools. The differences between centralizedand decentralized provinces up to the level ofthe district are as follows: centralized provincesspend more than decentralized provinces at theprovincial level. Thus, in Eastern (38 percent)and Northern (18 percent) province a far higherpercentage of funding is spent at the provinciallevel compared to Copperbelt (9 percent) andLusaka (5 percent). However, extra funding thatreaches the districts in decentralized provincesresults in higher spending only at the districtlevel and is not associated with greater fundingto schools.

LEAKAGE

Following the methodology used in Uganda (Abloand Reinikka 2000), the definition of leakage in theZambian educational system is the ratio of whatschools actually receive to what they were supposedto receive. The equivalent of per-pupil funding inUganda is the fixed-school grant in Zambia. Thus leak-age for rule-based allocations is defined as follows:

xii

amount received by school$600 ($650 for basic schools)

lags in updating the payroll regularly result indelays of one to three months in payment.

• For discretionary allocations to schools,the positive results obtained earlier nolonger hold: less than 25 percent ofschools receive any funding from discre-tionary sources. The crucial importance ofrule-based funding at the level of the school isalso highlighted by the change in the relativeshare by source as we move down the admin-istrative hierarchy. At the level of the province,the share of rule-based allocations in total fund-ing ranges from 8 to 40 percent, with a medianof 12 percent. Moving down to the districts,this share increases to between 19 and 63 per-cent, with a median of 44 percent. Finally, atthe level of the school this share ranges from 2to 100 percent, with a median of 99 percent.More than 75 percent of all schools receivedcash resources only from rule-based sourcesin the current year.

• For discretionary allocations to teachersthere are substantial amounts overduefor one-time benefits and payments. Insome provinces the overdue paymentsamount to three times the monthly salary ofthe teacher.

• Discretionary funding disbursed to schoolstend to be one of two types. Small sums aredisbursed under the UNICEF-administeredProgram for the Advancement of Girls’ Educa-tion (PAGE) or larger sums are received fromdistrict offices. The high variation in the amountsreceived is significant. Conditional on a schoolreceiving such funding, it could account for asmuch as 60 times what it would get under thefixed-allocation grant.

This suggests two potential explanations for the pat-terns of discretionary funding and school expenditure:

• Schools are provided discretionary funds forlarge expenditures on a rotating basis. For in-stance, if a school were rehabilitated every10 years only 10 percent of schools in anygiven year would receive money for capitalexpenditures. In addition, the amounts ofsuch expenditures received would be large

compared to amounts received for recurrentspending. According to this hypothesis, thepattern of funding and expenditure observedis a reflection of the lumpy nature of big invest-ments.

• Since rule-based funding is clearly defined andtransparent, it is extremely difficult for any politi-cal economy considerations (such as capture offunds by elites) to impact school allotments. Dis-cretionary funding is not associated with any suchrule. So the pattern of funding observed is a re-flection of the difference between rules and dis-cretion. The few schools that receive largeamounts are special schools that have greater bar-gaining power within the administrative structure.

The section on equity turns to exactly this con-cern and relies on the following observation: if thefirst explanation is correct it is unlikely that the flowof discretionary funds to schools will be correlatedwith wealth. If schools are provided with capitalexpenditure funds on a rotating basis, it seems natu-ral to assume that the sample of schools receivingsuch funding will be a mix of schools in rich andpoor areas.2 Consequently, systematic differencesin funding by wealth levels would lead to an expla-nation based on the difference between rules anddiscretion, rather than one based on a distinctionbetween one-time and recurrent expenditures.

EQUITY

The examination of funding equity in the educationsystem follows rule-based and discretionary allocationsseparately, and focuses on the relationship betweenthe type of funding and following related variables:

• Urbanization. Do urban schools and urban-ized districts receive more than their rural coun-terparts?

• Wealth. Do richer schools and districts receivemore than their poorer counterparts?

• Distance to administrative offices. Do schoolsthat are closer to administrative offices receivemore than those farther away?

The main findings are summarized below.• Equity of rule-based funds. Rule-based funds

are progressive, with greater per-pupil amounts

xiii

allocated to less urban and poorer schools anddistricts. This result is entirely due to the inverserelationship between school size and wealthand urbanization. Thus, schools in rural areastend to be smaller and poorer whereas schoolsin urban areas tend to be larger and richer. Sincerule-based funds are allocated per school ratherthan per student, poorer and more rural schoolsreceive greater funding per pupil. The same logicapplies at the district level, where those withsmaller average enrollments will receive higherfunding.3 Thus, poorer schools receive four timesthe rule-based allocation of richer schools andrural schools receive three times the allocationof urban schools.

• Equity of discretionary funds. There is noevidence that discretionary funds are distrib-uted in a progressive manner, with poorerschools receiving more than their richercounterparts. If anything, discretionary fund-ing within rural areas has a higher probabil-ity of being disbursed to wealthier schools,while allocations in urban areas are wealthneutral.

• Equity of staff compensation. The data showthat the per-pupil compensation of staff ishigher in richer and more urban schools, andthis arises primarily from higher class sizes andthe extensive use of teacher-trainees in ruralareas.

• Overall equity in funding. Since discretion-ary disbursements can be very large, oncethese are accounted for the progressive na-ture of the rule-based allocations disappears.At best, funding that flows from the district tothe school is wealth neutral (with no signifi-cant differences in received amounts acrossrich and poor schools), and at worse it is re-gressive for rural and progressive for urbanschools. Once staff compensation is factoredinto the analysis, the results show that the onlyprogressive component of the education sys-tem is the rule-based allocation. Once per-pupil teacher funding is added in, the entireeducation funding system becomes regressivewith poorer schools (K 14,531 per pupil) re-ceiving less than richer schools (K 19,826 perpupil).

PRIVATE SPENDING

This then opens up the question of household in-puts into education. If there are huge funding differ-entials between schools (for instance, the rule-basedcomponent implied that per-pupil funding could varyfrom K 1,889 to over K 8,000 depending on schoolsize), do households adjust contributions to accountfor the level of school receipts from the public fund-ing system? If yes, how does this household spendingimpact on funding equity in the education system? Thereport presents some preliminary results relating toboth household contributions to school funding, aswell as household private expenditure on educationthat complements the construction of the public ex-penditure system presented here.

The issue of household contributions to school fund-ing is at an important juncture in Zambia. Anecdotal evi-dence from a number of studies during the 1990s sug-gested that schools had started charging higher ParentTeachers Association (PTA) fees, and a concurrent de-cline in net enrollment during the same time period ledto an correlation between lower enrollment and highPTA fees. Consequently, in April 2002 (three monthsprior to the fielding of the ESDS) PTA fees were abol-ished for primary and basic schools and the governmentreiterated its commitment to free basic education.

With this background in mind, the key findingsregarding household contributions to school fund-ing follow:

• The announcement seemed to have had thedesired effect. PTA fees decreased to less than30 percent of 2000 values, although this declineis concentrated primarily in the urban areas ofLusaka and Copperbelt. The gradient betweenPTA fees and school wealth has also declinedsharply during the same period.

• For all provinces, but especially for Eastern andNorthern provinces, public funds are the mostimportant source of financial flows to theschool, comprising 96 percent of the total fund-ing of the schools in Eastern province (the mostreliant on public funds) and 82 percent in Lusakaprovince (the least reliant).

• Since the contribution of households to schoolfunding is small, the addition of private fundsshould not alter the results regarding equity inschool funding obtained in the previous section,and this is indeed the case. Thus, the difference in

xiv

private contributions between schools withpoorer and richer households is only K 1,300 perpupil compared to K 3,100 for public funding.

• Examining only private contributions to schoolfunds can be misleading if most inequality inprivate expenditure is at the level of the house-hold, i.e., through private, household-level nonfeespending (purchase of textbooks). The final sec-tion examines this aspect of educational funding.

• Nonfee expenditures are seven times the cor-responding expenditure on school fees. Thus,nonfee expenditures are the main source ofinequalities in private expenditure. When suchprivate spending is included in the total fundingpicture, the nonprogressive nature of the pub-lic funding system worsens: the share of edu-cational expenditures for the poorer 50 per-cent of the population declines from 40 to 34percent once private expenditures are factoredinto the analysis.

• There is strong evidence that households decreasespending when public funding increases. This sug-gests that, although public funding could addressinequalities in educational spending by progres-sive allocation across villages and schools, the de-sired impact of such redistributions may be lessthan imagined due to the crowding-out of privateexpenditures.

DISCUSSION AND CONCLUSIONS

This report makes three contributions to theunderstanding of education funding in the Zambiancontext. First, it shows that funding characteristicsare closely linked to the type of funding and the wayit is disbursed. In the case of rule-based funding, thedisbursement system seems to works efficiently andthere is no evidence that such funds are divertedfrom their stated purpose. The majority of discre-tionary funds, however, are spent at the district andprovincial levels, and less than 20 percent is allocatedto schools.

Second, analyzing funding equity using thewealth of pupils in the school shows that the spe-cific rule used in the case of rule-based allocationshas led to greater per-pupil funding for poorer andmore rural schools. However, these allocations arethe only progressive disbursements. Per-pupil staffallocations are higher in urban and richer schools.

For discretionary allocations there is evidence ofhigher disbursements to richer schools within ru-ral areas, and wealth-neutral allocations within ur-ban areas. Once all sources of public funding arefactored into the analysis, public school funding inZambia is regressive, with almost 30 percent higherallocations to richer schools.

Third, the report shows how private expendi-ture at the level of the household impacts equity ineducation funding. It argues that nonfee expendi-tures incurred by households, rather than contri-butions to school funds through PTA and other fees,are the major source of inequalities in the currentenvironment. Moreover, it find evidence that house-holds decrease private contributions when publicfunds to the schools increase.

If the government wishes to allocate higher fund-ing to poorer schools, these findings suggest that agreater percentage of all funding allocation shouldbe rule-based. It was initially thought that the pro-cess of decentralization would partially fulfill thisneed, since more money would flow directly to thedistricts, making accountability and therefore dis-bursements higher. Unfortunately decentralizationseems to have only shifted spending from the prov-ince to the district, and, in terms of funding equity,it is precisely at the district level that richer schoolsare receiving higher discretionary funds than theirpoorer counterparts.

Even if rule-based funding were to increase,three subsidiary implications need to be carefullyevaluated. First, the current rule-based allocationimplies that schools fare better in terms of per-pu-pil funding if they decrease enrollments. A morecommon funding rule (used, for instance, in Uganda)is based on transfers that increase the number ofenrolled children (such as $1 per enrolled child).There is unfortunately no guarantee such a schemewould work as well as the current, unambiguousrule. One suggestion would be to continue with thecurrent rule (which also has the desired equity im-plications), but to monitor enrollment carefullythrough regular data collection under the schoolcensus.

Second, an increase in public funds to schoolscrowds out private spending. Although the resultspresented here are preliminary, there is evidencethat this crowding-out can be fairly large. Thus,

xv

public funds may be far more effective at address-ing inequalities across rather than within villages. Thiswould suggest targeting at the level of schools, withgreater funding to poorer regions.

Third, the crowding-out of private expendituresby public funding implies that rule-based funds maynot have the desired impact on learning achieve-ment. Since for every dollar of school funding, pri-vate funding decreases, the true increase in totalfunds available for education is much less than theadditional dollar. This would suggest that increasesin funding may not be the optimal way to improveeducational attainment. It might be more benefi-cial to concentrate instead on providing inputs thathouseholds cannot supply on their own, such ashigh-quality teachers.4

NOTES

1. In Kwacha terms, schools receive either K 2.6 mil-lion or K 3.0 million. We use the dollar equivalents at theexchange rate of $1=K4,400.

2. Unless of course, rich schools systematically de-preciate their infrastructure faster than poor schools, butthis does seem unlikely a priori.

3. Consider two districts each with 100 schools butaverage enrollments of 50 (district A) and 100 (districtB). Since the rule-based allocation is a fixed per-schoolgrant, both districts receive exactly the same allocation.However, since district A has a smaller number of stu-dents than district B, the average per-pupil grant will bemuch higher in district A compared to district B.

4. This argument will be developed fully in anotherreport on the relationship between schooling inputs andlearning achievement.

Introduction1

The importance of education for developmentprovides a strong case for allocating publicexpenditures to the education sector. But

simply allocating more central budgetary resourcesto the education sector may not deliver better out-comes. Budgetary resources may not reach the in-tended beneficiaries, and even if they do, these re-sources may not lead to the desired outcomes. Thisreport examines the structure of funding and ser-vice delivery in Zambia using a recently completedEducational Service Delivery Survey (ESDS).

The Government of the Republic of Zambia(GRZ) has long recognized the importance of edu-cation. It set up universities immediately followingindependence and has focused more recently onattaining equity and quality in primary education.Nevertheless, a sharp decline in copper prices sincethe mid-1980s (the main export) and a commensu-rate decline in GNP per capita (from $590 in 1975to $300 in 2000) has presented several obstacles forthe education sector.

To overcome these obstacles, the government anda consortium of donors have been working togetherto improve educational outcomes in the country. Someof these changes are directly related to overall financ-ing of education and others aim to improve the deliv-ery of education by changing the administrative andinstitutional structure of education delivery.

The Education Service Delivery Study (ESD) isdesigned to examine the structure of funding andimplications for service delivery, and the relation-ship between public expenditure and educational

outcomes. Specifically, it tries to answer two ques-tions: (1) What is the current record of the educa-tion sector in delivering services to the intended re-cipients? (2) Have the changes undertaken by theMinistry of Education had the desired consequencesin terms of educational outcomes?

The study specifically takes into account linkagesthat may arise among the different players involvedin the delivery of education (for instance, the districteducation office, Parent Teachers Associations, theprovincial education office, schools, and pupils) andexamines the way these players respond to changesin the institutional setting and funding structure.5 TheESD addresses these issues through the careful mea-surement of school inputs (the subject of this re-port), and educational outcomes through tests ad-ministered to the same pupils at two different timesin collaboration with the Examination Council ofZambia (another report).

This report focuses on public and private expen-diture at the school level. It provides a detailed evalu-ation of public expenditure through a tracking and afunding-equity exercise, which gauges the extent towhich educational expenditures earmarked under thecentral budget actually reaches schools, as well as theextent to which such funding can be regarded as pro-gressive. A related body of work in Uganda (Ablo andReinikka 2000; Reinikka and Svensson 2002) showsthat the traditional view of the government as a be-nevolent agent is highly simplistic—public funding maynot matter simply because it does not reach grassrootlevels of the administrative structure.

17

18 • AFRICA REGION HUMAN DEVELOPMENT WORKING PAPER SERIES

Of the four questions that repeatedly concern usduring this exercise, the first three relate to the publicfunding of education and the last examines privateeducational funding and its relationship to public ex-penditure.

• What percentage of resources is spent at eachlevel of the administrative hierarchy?

• To what extent is variation in funding acrossdistricts and schools explained by funding for-mulas that relate budgetary allocation to dis-trict and school characteristics, such as enroll-ment or the number of schools?

• To the extent that there is variation in fundingacross districts and schools, what is the rela-tionship between this variation and character-istics of districts and schools? (For instance, doricher districts and schools systematically re-ceive more funds?)

• How important is private funding in the provi-sion of education and how does this compo-nent of funding relate to equity in the overallfunding of education?

The findings from this report provide the basisfor the second component of the ESD study (an-other report), which relates the availability of fund-ing to improvements in learning outcomes in Zam-bian schools. The main concern in addressing thisset of issues arises from the observation thatlower school funding does not in itself imply

worse outcomes. Problems of inappropriate usemay be worse at the level of the school or higherlevels of the administration may be better able touse available resources. The second component ofthe study on enrollment and learning outcomes willmake some welfare statements regarding the find-ings of the current report.

This report is structured as follows. Section 2 pro-vides a brief history of education in Zambia duringthe 1990s. Section 3 describes the methodology, andsection 4 describes the school sample. Funding flowsare examined in section 5, and section 6 discussesleakage. Section 7 addresses the equity aspects ofeducation funding and examines the association be-tween funding and school and district characteris-tics. Section 8 presents preliminary results from therelated household survey on private expenditures.Section 9 concludes.

NOTES5. Todd and Wolpin (2003) illustrate this kind of link-

age. An intriguing aspect of the link between public ex-penditure and education is that it is extremely hard todemonstrate any effect of increased funding to schoolson learning by pupils (see for instance, Hanushek 1986).Todd and Wolpin argue that a potential reason for this(weak) relationship is that households change their be-havior with the level of public funding; when public ex-penditure is high, households spend less and vice versa,keeping the overall level of schooling inputs for the childconstant.

Zambian Education during the 1990s2

Adecline in copper prices has been accompa-nied by a commensurate decrease in incomeand government resources. Since 1998 do-

mestic financing fell from 2.3 percent of GDP to only2.1 percent in 1999, and although the share of pri-mary education has increased from 54 to 57 per-cent over the same time period, net funding for pri-mary education has declined (World Bank 2001;Siamatowe 2002). Equally important, actual overallexpenditure has declined during the past six years,

from US$ 97 million in 1996 to US$ 74 million in2000 (Siamatowe 2002). As a result, average real percapita government education expenditure in 1996–98 was only 73 percent of its 1990–92 level, declin-ing further to an average of 60 percent of this levelby 1999–2000.6 This decline in funding seems to havehad a number of undesirable impacts.

First, primary school enrollment is currently low,and there is strong evidence that it has fallen in thelast decade. Table 1 shows that net enrollment

Table 1. Enrollment Rates, 1992–2001(percent)Enrollment rates 1992 1996 2001Net (Gross) enrollment rate (Primary)Female 66 57 60

(92) (77) (78)Male 65 54 60

(107) (86) (90)Net (Gross) enrollment rate (Secondary)Female 6 12 16

(9) (16) (24)Male 8 11 16

(16) (23) (29)School participation rate (Primary)Female 77 68 67Male 77 66 67School participation rate (Secondary)Female 46 45 48Male 65 61 66Data sources: Analysis based on Demographic and Health Surveys (1992, 1996, and 2000). The net enrollment rate andgross enrollment rates follow standard definitions. The school participation rate (primary) is defined as the percentage ofchildren between 7 and 13 years of age enrolled in school at the time of the survey. The secondary participation rate uses theage group 14 to 18 instead.

19


stood at 60 percent in 2001, with limited genderdisparity at the primary school level.7 These levelsare similar to Kenya, higher than Mozambique, butbelow levels typically attained in other Southern Af-rican countries.8 Net primary school enrollment ratesfor both girls and boys over the last decade havedeclined by 5 percent. Although declines of a similarmagnitude are found in other African countries likeBotswana and Lesotho, the decline in Zambia islarger than in comparable countries, such asMozambique or Tanzania (UNESCO Edstats 2002).

Figure 2 shows further characteristics of schoolenrollment in Zambia based on age-specific enroll-ment rates (Filmer and Pritchett 1999). The indi-vidual curves show the proportion of children en-rolled in school at the time of survey, where the pri-mary data is based on the Demographic and HealthSurveys from 1992 and 1996. Net enrollments aresubstantially higher in urban areas, and the differ-ences in enrollment between the poor and the rich

are also substantial.9 Moreover, the curves clearlyshow delayed enrollment among all income groupsin Zambia. Even among the rich (1996 survey), en-rollment is less than 50 percent for children agedseven or less. Finally, in every income group acrossrural and urban locations there has been a decline inenrollment between 1992 and 1996, and this declineappears to be particularly severe for middle-incomegroups.

Other characteristics based on the Living Condi-tions Monitoring Survey (LCMS) data (World Bank1998) show further problems with grade-specificenrollment rates in grade 1 and grade 7, and pro-gression rates between grade 8 and 9. The LCMSdata show rapid declines in enrollment over theyears, suggesting substantial dropout. While 90 per-cent net enrollment at grade 1 is achieved, this fig-ure is 34 percentage points lower by grade 7. Fur-ther, the progression rate from grade 7 to 8 ap-pears to be only 33 percent. Data comparability

Figure 2. Characteristics of School Enrollment in Zambia

Source: Based on Filmer and Pritchett (1999).

199219961992199619921996

Low income

00.20.40.60.8

11

6 7 8 9 10 11 12 13 146 11 12 136

Rural

Prop

ortio

n cu

rrent

ly en

rolle

d

00.20.40.60.8

11

6 7 8 9 10 11 12 13 146 11 12 136Age

00.20.40.60.8

11

6 7 8 9 10 11 12 13 146 11 12 136

Urban

Middle incomeHigh income

00.20.40.60.8

11

6 7 8 9 10 11 12 13 146 11 12 136Age

00.20.40.60.8

11

6 7 8 9 10 11 12 13 146 11 12 136Age

Perc

ent c

urre

ntly

enro

lled

AgeAge

199219961992199619921996

Low income

00.20.40.60.8

11

00.20.40.60.8

11

6 7 8 9 10 11 12 13 146 11 12 136

Rural

Prop

ortio

n cu

rrent

ly en

rolle

d

00.20.40.60.8

11

00.20.40.60.8

11

6 7 8 9 10 11 12 13 146 11 12 136Age

00.20.40.60.8

11

00.20.40.60.8

11

6 7 8 9 10 11 12 13 146 11 12 136

Urban

Middle incomeHigh income

00.20.40.60.8

11

00.20.40.60.8

11

6 7 8 9 10 11 12 13 146 11 12 136Age

00.20.40.60.8

11

00.20.40.60.8

11

6 7 8 9 10 11 12 13 146 11 12 136Age

Perc

ent c

urre

ntly

enro

lled

AgeAge

PUBLIC AND PRIVATE FUNDING OF BASIC EDUCATION IN ZAMBIA • 21

appears difficult, but this latter figure appears tobe lower than most other African countries, es-pecially in Southern Africa.10 This suggests that inZambia, in addition to declining enrollments overthe last decade, there seems to be a particularproblem with pupil retention within the primaryeducation system and progression after grade 7.

Since 1998, however, the government in col-laboration with a consortium of donors hasworked actively to reverse steadily declining out-comes in the education sector. Specifically, thegovernment’s Basic Education Sub-Sector Invest-ment Program (BESSIP) was the subject of a JointAppraisal in September 1998, and donors madenew financing commitments in support of BESSIPbeginning in 1999. This collaboration has resultedin changes both in the funding of education andthe administrative structure of education delivery(decentralizing decisions and funding to the districtlevel through district education boards is one suchexample), teacher deployment, and the ability of

schools to levy fees and contributions. This re-port examines whether further changes can bemade to the institutional structure of educationdelivery to improve outcomes in the educationsector.

NOTES

6. World Bank data based on Government of ZambiaFinancial Statements.

7. Depending on the sources used, net enrollmentvaries from 68 to 72 percent, so that 72 percent repre-sents an upper bound.

8. See, for example, online educational databases suchas World Development Indicators (World Bank) andEdstats (UNESCO)).

9. These two statistics are not unrelated, since ruralpoverty is still substantially higher than urban poverty (83versus 56 percent) despite some recent increases in ur-ban poverty (World Bank 2001).

10. For instance, this is lower than Mozambique for1990 (39 percent), but higher than Tanzania in 1998 (14percent).

Tracking Methodology3

The tracking exercise was designed to collectinformation on the effectiveness of fundingprograms in reaching their intended benefi-

ciaries, and the relationship between increased fund-ing at the central level and educational outcomes inschools. The survey was carried out at various lev-els of the administrative structure, including the prov-ince office (immediately below the Ministry of Edu-cation), the district office, schools, and households.The tracking exercise follows the flow of funds andeducational resources from the Ministry of Educa-tion to the province education office and then on-ward to the district education office and the schools.While in theory such an exercise may seem relativelystraightforward, the complexity of funding flows andthe multiplicity of sources at each level of the ad-ministration for educational funding and materialsmakes this exercise challenging.

In the Zambian education sector, funding toschools has three separate components. The first isa per-school grant of $600 (for primary) and $650(for basic) schools.11 The second is the money dis-bursed by districts (and in some cases provinces),presumably using some notion of need at the levelof the school. The third is remuneration of staffthrough salaries and benefits. What does this fund-ing structure imply for the tracking of public expen-ditures?

In the Uganda study (Ablo and Reinikka 2000),the government used a clearly defined rule to allo-cate funds to schools. Schools were allocated $1 foreach enrolled child. In this context, leakage in publicfunding was defined as the ratio of what a school

actually received to what the school was supposedto receive. The per-school grant in Zambia corre-sponds exactly to the per-pupil grant in Uganda.Analogous to the Uganda study, leakage in rule-basedallocations in Zambia can be defined as follows:

Leakage =

Unlike the Uganda study, however, this reportgoes beyond an estimate of leakage in allocatedfunds. It uses an innovative approach to provide acomplete picture of the funding for educational in-stitutions by examining all sources of funding for theschool. In the case of the fixed grant and the salarycomponent, the Uganda methodology is used to de-termine leakage in the system. But in the case ofdiscretionary components, this methodology fails sincethere is no rule about the amounts that schools aresupposed to receive. Therefore, it is not possible tobenchmark the amounts that schools actually receiveor determine what constitutes leakage in the sys-tem.

This problem is addressed by classifying fundingflows into the rule-based component (the $600 fixedgrant), the discretionary component (infrastructureand other grants), and the payroll component, andfocusing on three related issues:

• The pure tracking exercise first asks how muchmoney per pupil is available at each level of thedelivery chain (see section 5). This addresses theissue of the percentage of spending undertaken

22

amount received by school$600 ($650 for basic schools)


at different levels of the hierarchy and theamount that finally reaches the schools.

• The estimate of leakage focuses on the rule-based allocations and payroll components (seesection 6).12

• Equity in funding examines the degree to whichvariation in funding across districts and schoolsis explained by funding rules (see section 7). Tothe extent that there is excess variation after ac-counting for differences due to funding rules, therelationship between this excess variation anddistrict and school characteristics is analyzed.

The entire equity exercise is made possiblethrough the careful data collection on the householdassets of pupils in the school. Such data is normallynot available as part of expenditure tracking surveys.This report shows that the regular collection andanalysis of this information has important repercus-sions for policies regarding the delivery of educa-tion.

Educational funding in Zambia derives both fromdomestic (GRZ) funding and funding from donors.In the case of donors, funds are divided into fourcategories: donor funds bracketed as Case I, Case IIor Case III, which are controlled by the Ministry ofEducation, and Case IV donors that administer fundsdirectly.13 While most funding is theoretically ob-tained from the GRZ, a clear distinction at all levelsof the administration is maintained throughout thereport between BESSIP funding (the program linkedto the consortium of donors that form Case I–CaseIII) and other GRZ funding.

To better understand the flow of funds and re-sources, it is useful to build classifications based onthe type of resource (fund) and the underlying ad-ministrative structure, and the level of discretion ateach level of the administration to allocate and dis-tribute such flows. Resources and funds are dividedinto three categories: educational materials (such astextbooks), monetary (only cash), and payroll (re-muneration of staff). The four provinces of the sur-vey are divided in two categories: provinces with dis-trict education boards (henceforth decentralized prov-inces—Lusaka and Copperbelt in the sample) andprovinces without district education boards (hence-forth centralized provinces—Northern and Easternin the sample).14

Using this typology, school funding is composedof five types of funds and resources:

• Cash flows: fixed-grant component. Thefixed-grant component (also referred to as theBESSIP grant) is the per-school allocation of ei-ther $600 or $650, and no level of the adminis-tration has discretion over the amount dis-bursed. No information is required for the dis-bursement of funds as in the per-pupil allotmentin Uganda, where data is required on enroll-ments in the school, and which may be subjectto distortions at the level of the school or dis-trict. The fixed-grant is a rule-based componentof school funding.

• Cash flows: infrastructure grants. Schools alsoreceive money for rehabilitation of classroomsor for new construction (infrastructure). Thismoney is disbursed through the microprojectsunit of the Zambian Social Investment Fund, butthe district retains considerable discretion overdisbursement. Such grants are part of the dis-cretionary component of school funding.

• Cash flows: other grants. Schools may re-ceive additional money from the GRZ or CaseIV donors (mostly funding through the Programfor Advancement of Girls’ Education), and theseresources are distributed entirely at the discre-tion of the district. Consequently, as with (2)above, these grants are classified as a discre-tionary component of school funding.

• Flows of educational materials. Schools mayreceive educational materials such as textbooksand chalk, again allocated at the discretion ofthe district, with some input from the provincesand the Ministry of Education. Although it istechnically possible to track such resourceflows, the procurement of educational materi-als had not yet been completed at the time ofthe survey (midway through the academic year).As a result, schools received such materials spo-radically and in very small quantities. This com-ponent is excluded from the exercise.

• Payment of staff. Staff remuneration can bethought of as arising from rule-based and dis-cretionary components. Specifically, the salariesand allowances of teachers are paid through thecentralized payroll system. Neither provincesnor districts have discretion over the salary or


allowances that a teacher receives, so this isidentified as a rule-based component of staffpayment. On the other hand, payments ofone-time benefits (detailed below) are left tothe discretion of districts and provinces, so thisis identified as a discretionary component.Since most staff remuneration is a direct flowfrom the center to the teachers payroll fundsare omitted from the tracking exercise,15 and

this component in discussed in the sections onleakage and funding equity. The structure of fund-ing flows is summarized in table 2.

SUMMARY

Funding flows to schools in Zambia are dividedinto a rule-based component and a discretionarycomponent to track and analyze the level of expen-ditures at each level of the administrative hierarchy.

a. Northern and Eastern province in the ESDS.

Table 2. Structure of Resource and Funding Flows in Zambia

Ministry of Education/ Ministry

of Finance

Does province

have discretion?

(centralized)a

Does province have

discretion? (decentralized)

Does district have

discretion? (centralized)

Does district have

discretion? (centralized)

What can the school/teachers

expect to receive?

Flows of educational materials

(discretionary)

Handles procurement and distributes to provinces/districts

YES Decides how

much to pass on to district

NO Materials

passed directly to districts

YES Decides how

much to pass on to schools

YES Decides how


Depends on district

Payment of staff: salaries and allowances

(rule-based)

Completely centralized and are made directly to teachers

NO NO NO

NO

Teachers receive salaries and allowances according to the salary scale

Payment of staff:

One time benefits

(discretionary payments)

Disburses funds to provinces and districts for such payments

YES YES YES YES Payments for one-time benefits such as leave and transfers

Cash flows from

GRZ: fixed grant (rule-based funding)

An allocation of either $600 or $650 is made to every primary-basic school.

NO NO NO NO $600 or $650

Cash flows from GRZ: (discretionary funding)

Allocations are made for recurrent and capital expenditures, as well as disbursement to districts/schools

YES Decides how

much to pass on to districts

NO Districts

receive most money directly

YES Decides how


YES Decides how


Depends on district: Schools may expect to receive money for recurrent/ infrastructure expenditures

Cash flows from

case IV donors (discretionary

component)

No discretion. Most money from case IV donors is for PAGE (Program for Advancement of Girls’ Education)

YES Province

decides how much to pass on to districts

NO Districts

receive most money directly

YES District

decides how much to pass on to schools

YES District

decides how much to pass on to schools

Depends on district


For the rule-based and payroll components, leakage isidentified as the ratio of what the school received towhat the school was supposed to receive. The analysisof equity in funding flows examines total funding perpupil, including the two components above, as well asremuneration of staff, and studies the correlates of suchfunding with particular emphasis on the wealth of stu-dents’ households. In order to analyze equity in fund-ing, it is critical to establish a tracking exercise (see sec-tion 5). One far-reaching conclusion from the Ugandastudy was that budgetary allocations cannot be used asa proxy for funding at the school level.

NOTES

11. In current terminology, a “lower-and-middle ba-sic school” offers grades 1–7 and a “full-basic school”offers grades 1–7 and continues to grades 8-9. This re-port refers to them as primary schools, (grades 1–7), andbasic school, (grades 1–9).

12. For discretionary components, we remain ag-nostic about whether low/high receipts at the schoollevel constitute good or bad service delivery. This is-sue is addressed in a follow-on report on outcomesand efficiency of public spending.

13. DFID, Ireland, the Netherlands, and Norwaycontribute to the Case I pool; IDA credit is Case II;the African Development Bank is Case III; and USAID,JICA, and Denmark follow Case IV.

14. The prevalence of both systems of adminis-tration arises because district education boards havebeen phased in since 2000. Decentralization was in-troduced to grant greater autonomy to districts andto reduce their dependence on the province for fund-ing.

15. The money for salaries and allowances are dis-bursed directly to teachers so that no money passesthrough the administrative chain. Although it is theo-retically possible to track one-time benefits, thesefunds are not earmarked for such disbursements andhence cannot be traced through the system.

School Characteristics of the Sample4

This section provides a brief introduction tothe sample. The schools surveyed were cho-sen from a list frame of primary and basic

schools in four Zambian provinces: Lusaka,Copperbelt, Northern, and Eastern. A randomsample stratified on the basis of urban and rural lo-cations included 184 schools in 33 districts. As partof the re-testing exercise, 3,200 pupils formed theinitial sample for the administration of tests in En-glish, mathematics, and vernacular (Icibemba orNyanja).16 The choice of these four provinces wasdictated primarily by the variation in educational at-tainments, regional incomes, and administrativestructures. Specifically, Lusaka and Copperbelt arethe two richest provinces in Zambia, with commen-surately high enrollment rates, and Northern and

Eastern provinces are the poorest, with enrollmentrates only marginally better than the worst perform-ing Central province (table 3). Learning outcomesconfirm the wide disparities in performance amongthe provinces studied: apart from Northwesternprovince, Northern and Eastern provinces reportedthe worst English and mathematics scores, whileLusaka reported the best (table 4).

The schools in the sample are grouped into fourdescriptive categories: school enrollment and staff-ing, school infrastructure, school performance, andcharacteristics of households in the school. Since sig-nificant differences are likely to emerge betweenurban and rural schools, the summary statistics aredisaggregated by school location.

Table 5a shows significant differences betweenrural and urban schools in enrollment and staffing.Urban schools tend to have large enrollment (an av-erage of 1,440 pupils per school compared to 554

Table 3. Enrollment, Urbanization, and Poverty(percent)

Net Povertyenrollment head Population

Province age 7–13 count Rural shareCentral 75 77 66 10Copperbelta 76 65 23 18Easterna 49 79 91 13Luapala 61 81 86 7Lusakaa 79 53 19 15Northerna 60 82 84 12North Western 66 76 86 6Southern 73 75 80 12Western 64 89 90 7All Zambia 68 73 63 100a. Provinces covered in the ESD sample.Source: World Bank (1998).

Table 4. Mean Math and English Scores, by ProvinceProvince English MathCentral 9.02 14.94Copperbelta 9.06 13.60Easterna 9.02 13.56Luapula 9.68 13.90Lusakaa 10.34 15.16Northerna 8.68 13.21North Western 7.76 13.37Southern 9.43 14.37Western 8.95 13.62a. Provinces covered in the ESD sample.Source: ESDS team calculations based on ExaminationCouncil Data.

26


for rural schools), but the difference in size is morethan compensated for by the difference in staffing.The pupil-teacher ratio in urban schools at 42 is muchcloser to the Zambian goal of 40 pupils per teacherthan the rates in rural schools at 61.17 Also notewor-thy is the large difference in female teachers betweenurban and rural schools. Female make up 70 per-cent of the teaching workforce in urban schools,and 31 percent in rural schools. Conversations withteachers suggest that safety and the lack of teacher’shousing is a major concern in rural areas. If there isa significant difference in performance by the gen-der of the teacher, this is an area for future policyconsideration.

Table 5b shows the distribution of key infrastruc-ture variables in urban and rural schools. All threevariables reported—number of pupils per classroom,number of students per toilet, and number of teach-ers per staff house—suggest considerable over-crowding, although the lack of data from other coun-tries does not allow us to establish a suitable bench-mark. With an average of 6.4 teachers per staffhouse, it appears there is a considerable shortage of

staff housing in the system. Interestingly, for all threeindicators of infrastructure, rural schools performbetter than their urban counterparts, and this differ-ence is significant for toilets per pupil and teachersper available staff house.

The next two tables provide performance indi-cators for, and household characteristics of, childrenin the school. Table 5c shows four key indicators ofperformance: children repeating classes, dropouts,and performance on examinations for male and fe-male students. Two key points emerge. First, ruralschools perform significantly worse than urbanschools in terms of repetition and dropouts. Closeto 10 percent of students in rural schools repeat agrade, whereas 5 percent of students in urban schoolsrepeat a grade. Similarly, dropouts in the previous

year as a percentage of current enrollment was 5percent in rural schools and 2 percent in urbanschools. The poorer performance of rural schools,however, did not extend to performance on the na-tionwide grade 7 examination in 2000. In fact, ruralschools performed slightly better, achieving a passrate of 51 percent (males) and 45 percent (females),compared to 44 (males) and 43 (females) percentfor their urban counterparts.18

Table 5d presents households indicators for ruraland urban schools. Two characteristics emerge fromthis table. First, children attending rural schools comefrom poorer households compared to their urbancounterparts (the average difference exceeds onestandard deviation in the wealth distribution). Sec-ond, although averages are very high, school atten-dance for single and double orphans is the same in

Table 5a. School Enrollment and StaffingVariable Urban Rural DifferenceSchool size 1439.5 553.4 886.56a

(number of pupils) (74.23) (38.19)Number of 37.5 12.8 21.98 a

teachers (1.67) (1.15)Female teachers as 69.0 30.9 –38.10 a

percent of total (5.6) (4.3)Pupil-teacher 42.25 60.96 –18.70 a

ratio (2.93) (3.76)a. Difference is significant at the 5 percent confidence level.Standard errors in brackets.Source: ESD sample.

Table 5b. InfrastructureVariable Urban Rural DifferenceNumber of pupils per 103.4 96.7 6.73 classroom in good condition (7.14) (4.76)Number of pupils per 198.9 113.2 85.72* toilet in good condition (39.9) (15.2)Number of teachers per 11.39 4.34 7.04* staff-house in good condition (2.01) (0.827)a. Difference is significant at the 5 percent confidence level.Standard errors in brackets.Source: ESD sample.

Table 5c. Key Performance IndicatorsVariable Urban Rural DifferenceRepeating the same 4.7 9.45 –4.74* grade (percent) (0.49) (0.62)Dropouts as ratio of current 1.67 4.56 –2.88* enrollment (percent) (0.32) (0.47)Pass-rate in 2000 Grade 44 50.5 –6.50 VII examination (males) (6.1) (4.8)Pass-rate in 2000 Grade 42.5 44.5 –2.00 VII examination (females) (6.0) (4.8)a. Difference is significant at the 5 percent confidence level.Standard error in brackets.Source: ESD sample.


rural and urban. On average, one in every 13 pupils is asingle orphan (has lost a single parent) and one out ofevery 25 pupils is a double orphan (has lost both par-ents). From this table one would expect a priori thatvulnerability is a significant problem in rural schools com-pared to urban schools in terms of overall poverty, butnot in terms of the orphan status of the child.

SUMMARY

This section provides insights regarding theequity aspects of the rule-based component of

funding. Because schools in rural regions tend tohave lower enrollment than their urban counter-parts, and the rule-based funding is allocated ona per-school basis, as long as schools receive theallotted amounts, one would expect rural schoolsto receive higher per-pupil amounts. This has di-rect implications for equity. Since rural schoolsare also poorer than urban schools, in theory,rule-based funding should be highly progressive.

NOTES

16. Detailed notes on the sampling procedure can befound at http://econ.worldbank.org/programs/public_services/topic/tools.

17. This does not imply that classes are bigger in ruralschools. Data from the survey suggests that teachers inrural schools teach more hours than their urban counter-parts, and as a result the average class size is actually lowerin rural schools.

18. This is probably due to greater ability-based selec-tion into higher grades in rural areas. For instance, in ruralareas 40 percent of students do not make it to grade 7. Ifthese children are worse than the average enrolled child,test scores in rural areas will reflect this selection process.

Table 5d. Household IndicatorsVariable Urban Rural DifferenceAverage value of wealth index of households with 0.74 –0.44 1.18* children in the school (0.067) (.05)Percentage of children who 7.6 7.7 –0.10 are single orphans (3.2) (2.5)Percentage of children who 4.72 4.79 –0.70 are double orphans (2.5) (1.9)a. Difference is significant at the 5 percent confidence level.Standard error in brackets.Source: ESD sample.

Expenditure Tracking at Provincial, District,and School Levels

5

This section examines whether schools actu-ally receive the fixed-grant component, andwhether provinces and districts further sup-

port poorer districts and school through discretion-ary cash grants. The tracking exercise for nonsalary

funding assesses the total amount of cash availableat the province and district levels, and then tracksthe movement of funds from provinces to districtsand from districts to schools.19 Figure 3 highlightsthe distinction between centralized and decentralized

29

Figure 3. Funding Flows across Centralized and Decentralized Provinces

Note: Decentralized provinces are Lusaka and Copperbelt. Centralized provinces are Northern and Eastern.






Rule-based flows








Rule-based flows



Province Province


provinces in terms of expenditure flows. The arrowsdivide funding flows into rule-based and discretionarycomponents. One characteristic of the funding sys-tem becomes immediately clear—funding in the sys-tem does not flow in a top-down manner. Instead,fresh money enters the hierarchy at each level.

In both centralized and decentralized provinces,money given under the BESSIP program (Case I–CaseIII donors) enters directly at the district level. In de-centralized provinces other cash flows move directlyto districts. Finally, fresh money enters at the levelof the school through household contributions. Anaccurate assessment of the resources available ateach stage involves tracking monetary flows fromthe funding body immediately above as well as theflows that directly enter the system at each level.

Each of these funds will be reported in terms ofKwacha per pupil. Total funds are divided by total en-rollment in the province (and similarly for districts andschools).20 The data on enrollment is based on the2001 schools census, using grade 1–9 pupils of allschools for which the GRZ has some funding respon-sibility (this includes schools aided by grants and com-munity schools, but excludes five private schools).21

THE PROVINCIAL LEVEL

Cash receipts. Table 6 shows the cash receivedper pupil at the level of the province averaged over

one year (June 2001 to June 2002). The pattern ofreceipts at the provincial level are clearly in line withexpected receipts from the decentralization process.The centralized provinces receive sums that are or-ders of magnitude higher than their decentralizedcounterparts, with Eastern province (K 29,000 perpupil) receiving more than seven times the receiptsfor Copperbelt (K 4,133 per pupil).

The amounts received in the centralized prov-inces give a rough idea of the money passing to theprovinces from the ministry. From the Ministry ofEducation “Financial Statistics 1986–2000” and theBESSIP Financial Statement 2000, total per-pupil re-current expenditure for basic schools (not includ-ing salaries) is K 29,550 and total per-pupil capitalexpenditure is K 36,519. Since money given to the

Table 6. Cash Received by ProvinceEducation Offices

Cash receipts TotalProvince (Kwacha primarytype Province per pupil) enrollmentCentralized Northern 19,779 235,012

Eastern 29,308 192,321Decentralized Copperbelta 4,133 286,710

Lusakaa 8,034 206,318a. Decentralized provinces with district education boards.Source: ESD survey data and total primary enrollment datefrom the Zambian School Census (2001).

Table 7. Funding Sources at Province Education Office Level, June 2001–June 2002(Kwacha per pupil)

GRZ funding:Case I– Case III pool GRZ funding Case IV

(external sources) (domestic sources) funding

Rule-based Discretionary Discretionarycomponent components components Discretionary

Province type Province (BESSIP) (HIPC Funds) (other GRZ) components TotalCentralized Northern 7,882 4,011 5,250 2,636 19,779

(40) (20) (27) (13) (100)Eastern 2,206 8,947 14,801 3,354 29,308

(8) (31) (51) (11) (100)Decentralized Copperbelta 475 1,764 1,300 593 4,133

(12) (43) (31) (14) (100)Lusakaa 704 5,828 1,350 152 8,034

(9) (73) (17) (2) (100)Note: Percentages (in brackets) may not add up to 100 due to rounding.a. Decentralized provinces with district education boards.Source: ESDS.


province includes money for recurrent and capitalexpenditures, an upper bound is that 95 percent oftotal per-pupil expenditure is available at lower lev-els in the case of Eastern province, and 65 percent isavailable at lower levels in the case of Northern prov-ince.

Table 6 also shows vast differences within cen-tralized and decentralized provinces, with almost a50 percent difference in cash received per pupil be-tween Eastern and Northern provinces, and a 100percent difference in cash received between Lusakaand Copperbelt. Table 7 examines these differencesfurther by looking at the receipts per province fromthe four sources of funding: BESSIP, HIPC (Debt Ini-tiative for Heavily Indebted Poor Countries of theWorld Bank), and other GRZ and Case IV donors.

Two characteristics stand out. First, the fundingfrom nondomestic sources (BESSIP and HIPC), con-sistently exceeds 60 percent of total funding for allprovinces (except Eastern with 49 percent). Second,there is no clear pattern in terms of a single prov-ince receiving more or less than others from anysingle source. Some receive more funding fromBESSIP sources (Northern), but less from Case IVand domestic GRZ sources, while others (Lusaka)receive very little from BESSIP, domestic GRZ, andCase IV sources, but large amounts from HIPC.

Cash disbursements. From these financial re-ceipts provinces disburse cash to districts. In the caseof decentralized provinces (which receive cash di-rectly from the ministry), one would expect disburse-ments to be very low. Table 8 confirms this with dataon disbursements to the districts over a six-monthperiod from January 2002 to June 2002. In per-pupilterms, this is virtually nil for all cash in decentralizedprovinces and for capital expenditures in centralized

provinces. However, in the centralized provincesthese disbursements represent a small proportionof what the province actually receives. In Northernprovince, disbursements to districts correspond onlyto one-third of the annual funding that the provincereceived, and in Eastern province to only one-tenth.22

THE DISTRICT LEVEL

Cash received. Table 9 shows the cash re-ceived at the district level.23 The data for one year,from June 2001 to June 2002, are enrollment-weighted averages across all districts in the prov-ince. On average, districts in the Eastern andNorthern provinces receive much more (districtsin Eastern for example receive almost 60 percentmore than Copperbelt districts) than those inLusaka and Copperbelt, but this hides substantialdifferences within provinces, an issue explored insection 6.

Similar to cash flows at the provincial level (table7), there are wide differences between district re-ceipts across provinces from the four funding sources.As table 10 shows, the higher level of funding per pu-pil in districts in Northern and Eastern province re-sulted primarily from the rule-based component ofthe funding system, with the average district in East-ern province receiving more than twice (219 percent)the receipts of the average district in Copperbelt.

This variation follows entirely from the specific-ity of the rule-based component, which takes theform of lump-sum payments per school. In provinceswhere the average school size is smaller, the fixed-grant funding per student will be higher. Accordingto the 2001 school census data, Northern and East-ern schools have, on average, about 315 pupils, whileCopperbelt schools have about 877 pupils, andLusaka schools have 1,037 pupils. The difference in

Table 9. Cash Received by Districts(weighted average by province)

Cash receipts at district levelProvince type Province (Kwacha per pupil)Centralized Northern 24,196

Eastern 27,437Decentralized Copperbelta 17,263

Lusakaa 20,690a. Decentralized provinces with district education boards.Source: ESDS.

Table 8. Funding Disbursed to Districts,January 2002–June 2002(Kwacha per pupil)Province Recurrent Capitaltype Province expenditures expenditures TotalCentralized Northern 3,381 264 3,645

Eastern 1,390 0 1,390Decentralized Copperbelta 0 0 0

Lusakaa 397 0 397a. Decentralized provinces with district education boards.Source: ESDS.


average enrollment almost entirely accounts for thelarge variation in per-pupil funding at the district level.

Turning to discretionary funds, it is striking thatCopperbelt and Lusaka receive more in per-pupilterms via discretionary sources than the centralizedprovinces, which depend on funding initially routedthrough the provincial level. Thus in Northern andEastern province, discretionary sources account foronly 20 percent of total resources, whereas this in-creases to 45 percent for the decentralized provinces.

Tracking results (province to district). Thetracking exercise of public funds from the provinceto the district level takes into account the complexi-ties of the funding arrangements discussed in theprevious section.24 It is assumed that total rule-basedcash received by districts is fresh money in the sys-tem, since there is no movement of rule-based fund-ing from the province to the district. Discretionaryfunding at the district level in centralized provinces(including both GRZ and Case IV sources) is routedvia the province; in decentralized provinces fundingis not routed via the provinces and constitutes newresources in the system. From this the net total fundspassed on to the province and the district via theMinistry of Education can be calculated. Table 11gives the results.

On average, K 28,000 per pupil enters the sys-tem for nonsalary spending from the center. Upto one-third of total cash in the system is earmarked

for schools via the rule-based funding component,and the remainder is spent at the discretion of theprovinces and districts. As table 11 shows, decen-tralization seems to have decreased the spendingat the provincial level. Thus, spending at the pro-vincial level is between K 5,000 per pupil (North-ern) and K 16,800 per pupil (Eastern) in the cen-tralized provinces (18 and 38 percent of total fundsin the system, respectively) and this decreases to K1,500 per pupil for the decentralized provinces(around 7 percent of total funds in the system). As aresult the proportion of money available at the levelof the district increases in the decentralized prov-inces from 70 percent to more than 90 percent (95percent in the case of Lusaka) of total funds in thesystem.

This data seem to suggest that the system of de-centralization improves the flow of funds by decreas-ing provincial spending, but it leaves unanswered thequestion of whether decentralized districts arespending more at the district level, or passing ongreater amounts to the schools.

THE SCHOOL LEVEL

District disbursements and school receipts.Schools are supposed to receive either $600 or $650through the rule-based component. In addition, theymay also receive cash from infrastructure and otherspending grants, through GRZ spending or from Case

Table 10. Funding Sources at the District Level(Kwacha per student)

GRZ funding: GRZ funding from Case IVCase I – Case III pool domestic sources funding

Rule-based Discretionary Discretionary DiscretionaryAverage district component components components components

Province type in province (BESSIP) (HIPC Funds) (other GRZ) (Case IV funding) TotalCentralized Northern 9,576 1,278 4,656 1,547 24,196

(44) (6) (17) (7) (100)Eastern 15,019 921 7,364 1,534 27,437

(63) (4) (21) (5) (100)Decentralized Copperbelta 6,850 1,049 8,074 652 17,263

(43) (6) (43) (4) (100)Lusakaa 4,057 263 9,751 731 20,690

(19) (1) (45) (3) (100)Note: Percentages (in brackets) may not add up to 100 due to rounding.a. Decentralized provinces.Source: ESD data.


IV donors. Regardless of the original funding body, thismoney is eventually disbursed through the district. Inisolated cases schools may also receive funds from othersources (churches, local donors, nongovernmental or-ganizations (NGOs), or directly from the Ministry ofEducation). Since the data contains only one such ex-ample, public funding to the school is considered as aone-way flow from the district.

The results of the expenditure tracking at theschool level are presented in two parts. First, incontinuance of the previous tracking exercise theper-pupil amounts available to each school in dif-ferent districts provides an estimate of the per-centage of funds that reaches schools. Since therule-based component of funding is a per-schoolamount, we cannot ascertain leakage through thisexercise. The per-school amounts are thus post-poned to the next subsection, where school fund-ing is disaggregated in terms of rule-based versus

Table 11. Tracking of Resources, June 2001–June 2002(Kwacha per pupil per year)

Rule-based + Rule- Rule-based +discretionary based Discretionary Discretionary discretionary

(district) (district) (district) (province) (total)

Totaldiscretionary

Total Funds Funds Additional fundsfunds received Funds received funds availableat the directly received directly from at the Spending: Totaldistrict from from the from province provincial provincial funds in

Province type Province level ministry province ministry to district level level the system(1) (2) (3) (4) (5) (6) (7) (8)

Centralized Northern 24,196 9,576 7,812 0 6,808 19,779 5,159 29,355(82) (33) (27) (0) (23) (67) (18) (100)

Eastern 27,437 15,019 2,780 0 9,638 29,308 16,890 44,327(62) (34) (6) (0) (22) (66) (38) (100)

Decentralized Copperbelta 17,263 6,850 0 8,074 2,339 4,133 1,793 19,056(91) (36) (0) (42) (12) (22) (9) (100)

Lusakaa 20,690 4,057 916 9,751 5,966 8,034 1,152 21,842(95) (19) (4) (45) (27) (37) (5) (100)

Note: Percentages of total funds in system are bracketed. All total funds in the system refer to nonsalary resources.a. Decentralized provinces.Sources: ESD survey data. Column (1) data from table 9. Column (2) data from table 10 total rule-based funding to districts.Column (3) extrapolated to yearly spending from data on funding to districts, January 2002-June 2002. Column (4) from table10, discretionary funds allocated directly to the district in the case of Lusaka and Copperbelt. Column (5): (5)=(1)-(2)-(3)-(4).This correction assumes that funds are reported correctly at the district level and the province underreports transfers. (Thesums involved correspond closely to the sum of HIPC and PAGE resources received at the provincial level, which wereearmarked to be passed on to lower level, but which may not have been recorded as cash passed on to district.) Column (6)data from table 8. Column (7): (7)=(6)-(5)-(3). Column (8): (8)=(1)+(7).

discretionary funding, which directly tackles theissue of leakage.

To check for robustness, three options are usedto calculate the amounts passed on to schools. Op-tion 1 in table 12 is direct measurement of the cashpassed on from the district to the schools in thesample, with the associated problem that the data isonly for a sample of schools (those covered by thesurvey for a six-month recall period). Option 2 is ameasure of spending patterns at the district level andis based on a one-month recall of spending in May2002. Option 3 uses data collected at the school level,with a six-month recall period. The high correlationobserved between all three measures confirms thereliability of the reported financial data. Although thereis some variation in the amounts depending on theoption used, option 3 always lies between options 1and option 2. One strategy therefore is to use theamount calculated from option 1 as the upper bound


of school receipts and amount calculated from op-tion 2 as the lower bound.

The results of this per-pupil tracking exercise areshown in table 12 (and summarized in figure 4). Col-umns 1, 2, and 3 show the funds available in the en-tire system at the province and district level (identicalto table 11). Column 4 shows the total amount passedon to the schools using option 1 (district report).Column 5 shows the spending on schools using sharesof spending in May (option 2), and column 6 showsthe funds reported by the school (option 3).

Option 1 (the upper bound) suggests that typi-cally between 25 and 48 percent of total funding(between K 5,000 and K 13,600 per pupil) in thesystem is passed on to schools; option 2 (thelower bound) suggests 11 to 33 percent (betweenK 2,500 and K 8,000) is passed on. Typically lessthan half the funding in the system reaches theschool level, and in Eastern and especially Lusakaprovince the share seems even lower.

Gains from decentralization in terms of reducedspending at the provincial level now disappear:

there is no evidence that increased funding to dis-tricts in decentralized provinces is passed on toschools. As a percentage of total funds in the sys-tem, schools in centralized provinces receivearound 30 percent of total funds and this actu-ally decreases to (approximately) 25 percent inthe case of decentralized provinces. Thus, interms of funding it appears that decentralizationhas merely shifted spending from the provinceto the district, with no improvement in disburse-ments to schools.

SUMMARY

The tracking exercise established the followingcharacteristics of the education funding system inZambia.

• Discretionary funds at the level of the prov-ince and district account for 70 percent ofall funding. On average, about K 28,000 perpupil enters the educational system at the levelof the province. Approximately 30 percent ofthis amount is earmarked for rule-based funds

Table 12. Tracking of Resources from the District Level to Schools, 2001–2002(enrollment weighted by Kwacha per pupil)

Rule-based + Discretionary Rule-based + Rule-based + discretionarydiscretionary (province) (district) (schools)

Total discretionary Funds Option 1: Option 2: Option 3:Total funds at the reportedly spending on fundsfunds available at district passed schools reportedin the the provincial education on to (based on by

Province type Province system level office level schools shares) schools(1) (2) (3) (4) (5) (6)

Centralized Northern 29,355 19,779 24,196 13,618 7,985 9,179(100) (67) (82) (46) (27) (31)

Eastern 44,327 29,308 27,437 15,528 7,408 11,925(100) (66) (62) (35) (17) (27)

Decentralized Copperbelta 19,056 4,133 17,263 9,216 6,215 6,384(100) (22) (91) (48) (33) (34)

Lusakaa 21,842 8,034 20,690 5,382 2,483 3,121(100) (37) (95) (25) (11) (14)

Note: Percentages of total funds in brackets. Data excludes two schools in Kafue district, with substantial reported grants notreflected in district data, and most likely outliers linked to measurement problems.a. Decentralized provinces.Source: ESD data. Columns (1), (2) and (3): taken from table 11. Column (4) reported by district, extrapolated from a recallperiod January 2002-July 2002 and from data specifically related to the ESDS sample of schools explicitly asking how muchcash was allocated to each school in our sample (for recurrent and capital expenditures). Column (5) reported by district,extrapolated from actual spending data for May 2002. Column (6) based on data collected at the schools level, recall periodJanuary 2002–June 2002.


through the fixed-school grant of $600 ($650),and 70 percent is in the form of funding allo-cated at the discretion of provinces and districts.

• Between one-sixth and one-third of totalfunding in the system eventually reachesschools. Of this total amount, between 14 and34 percent eventually reaches the schools as acombination of rule-based and discretionaryspending, and the rest is spent at the provincialand district levels.

• Decentralization has shifted spending fromthe provincial to the district level, but hasnot resulted in greater disbursements toschools. There are important differences be-tween centralized and decentralized provincesup to the level of the district. Centralized prov-inces spend more at the provincial level com-pared to those that have been decentralized.However, the extra funding that reaches thedistricts in decentralized provinces results inhigher spending at the district level and is notassociated with greater funding to schools.

These findings are summarized in figure 4, whereK per pupil funding is plotted at each level of theadministrative hierarchy. The amount that finallyreaches the school is much less than the totalamount available in the system and this decline isthe sharpest in Eastern province. Moreover, a largefraction of the difference in funds arises from the

nature of the rule-based component of funding.Hence, disbursement curves in the decentralized dis-tricts are consistently below those in the centralizeddistrict as a direct consequence of the rule-based com-ponent, in conjunction with the difference in averageschool enrollment between the provinces discussedpreviously.

Already the results of the tracking exercise pro-vide suggestive evidence regarding the leakage offunds. In particular, the correspondence between thefunding received by schools and the amount of thefixed-school grant indicate that almost all rule-basedfunds are passed on, while little of the available dis-cretionary funding actually reach the schools. In thenext section we explicitly address this issue for boththe nonsalary funding discussed in this section, aswell as payroll funds that do not pass through theadministrative hierarchy.

NOTES

19. Recall that the payroll is centralized and paymentsare made directly to teachers from the center. Thus thereis no notion of money being distributed from provincesonward. Of course, the important question of whetherteachers receive their salaries remains, and is taken up insection 5.

20. An alternative measure could rely on per-childfunds. This measure would have the attractive propertythat differential dropout rates would not affect the over-all funding statistic. Ideally we would like to report both,

Figure 4. Funding Disbursement in Centralized and Decentralized Provinces

Centralized Provinces

Funds at administrative level

Decentralized Provinces

0


10000

20000

30000

40000

50000

Total Province District Schools(LB)

Schools(UB)

Fund

ing (K

per

pupil

)

10000

20000

30000

40000

50000


Schools(UB)

Fund

ing (K

per

pupil

)

Lusaka ProvinceCopperbelt

Northern Province

Eastern Province

Centralized Provinces


Decentralized Provinces

0


10000

20000

30000

40000

50000


Schools(UB)

Fund

ing (K

per

pupil

)

10000

20000

30000

40000

50000


Schools(UB)

Fund

ing (K

per

pupil

)

Lusaka ProvinceCopperbelt

Northern Province

Eastern Province


but there are two problems. The first is that the popu-lation census is not currently available. The second andmore serious problem is that to report this statistic atthe school level one would have to define a catchmentarea, which is often arbitrary, especially for urbanschools. Without a clearly defined catchment, it is un-clear that this concept would be useful at the level ofthe individual school.

21. The enrollment figures include all districts forwhich data are available. In the calculation of the financialfigures per pupil, two districts were dropped because ofincomplete financial data, Nakonde in Northern Provinceand Chongwe in Lusaka Province. Recall also that theseare all nonsalary transfers.

22. One option could be that this low amount is linkedto low cash receipts from January to June. According tothe timing of funds, most funding reaches provinces from

September to November and could thus have alreadybeen disbursed by January.

23. Aspects of the tracking exercise to keep in mindinclude two crucial points. First, the fixed grant is directlypassed on from the Ministry of Education to the districtsand thus enters the system only at the district level. Sec-ond, in decentralized provinces discretionary resourcesfor recurrent expenditure flow directly to the districts,without passing through the provinces.

24. The nature of the data available still implies theneed for some assumptions to interpret the results. Forexample, recall periods for funds received are a full year,but for resources passed on the only workable recall wasthe preceding half-year. For spending at the province ordistrict level, a one-month recall period was used. Thepilot study revealed that these recall periods provide themost accurate information.

Leakage of Nonsalary and Payroll Funds6

For both nonsalary and payroll, funding sourcesare divided into rule-based and discretionaryflows. Since the division for payroll is like the

division for nonsalaried funding explored in somedetail earlier, the same classification is used. Specifi-cally, staff remuneration has three distinct compo-nents: salary, allowances, and one-time benefits.25

Both salary and allowances are paid individually toeach teacher through a centralized payroll system,whereas one-time benefits are paid by the districtaccording to the availability of funds. As withnonsalaried funding, salary and allowances are clas-sified as the rule-based component of the payroll sys-tem, and one-time benefits are classified as the dis-cretionary component.

RULE-BASED FUNDING

The disbursement of the rule-based componentsof school funding and remuneration of staff yields apositive finding. There is little evidence of leakage,either for disbursements to schools or for paymentsto staff. Table 13a shows the provincial breakdownof the fixed-grant receipts by schools. In every prov-ince except Lusaka, more than 90 percent of schools

received the funds allocated, with the remaining 10percent evenly divided among schools that eitherreceived no money or received less than the allo-cated sums. In both cases it is likely that there areother explanations relating to delays in disbursementand measurement error, rather than leakage in thedelivery system. In the case of schools that receivedno cash (particularly the 28 percent in Lusaka), theactual disbursement of this cash started only in May,five months into the academic year. Since the surveywas carried out in July, one possibility is that theschools that received no money were experiencingdelays in disbursement (i.e., head teachers had notyet gone to the district headquarters to receive thecheck), rather than the money being diverted at ahigher level of the administrative hierarchy.26 Forschools that reported receiving less than the allot-ted amount (only two schools each in Eastern andNorthern provinces), the amounts reported are be-tween $363 and $500, which may reflect errors inreporting or recording rather than actual variationin the amounts disbursed. Thus, in the case of thisfixed-grant allotment it would appear that the ad-ministration is fairly efficient and unbiased in the

Table 13a. Disbursement of Fixed-Grant Allocations, by Province(percent)

Centralized provinces Decentralized provincesRule-based funds Northern Eastern Copperbelt LusakaWho received either $600 or $650 90.39 94.87 94.12 71.3Who received nothing 5.77 0 5.88 28.57Who received less than $600 3.84 5.13 0 0Source: ESD data.

37


allocation of funding resources to schools. Given thesmall numbers of schools that had not received thisfunding at the time of the survey, there does notseem to be evidence of systematic leakage or diver-sion of funds from the schools.

A similar story emerges with the salary part ofrule-based components of payroll (table 13b). Over95 percent of those on salary were paid on time. Inall provinces, the percentage of staff that are morethan six months overdue is less than 3 percent.27

The record of disbursements for allowances isworse, and seems to depend on the kind of allow-ance the teacher was supposed to receive. Consid-erable lags in updating the centralized payroll for newallowances is reflected in delays in disbursement. Dis-bursements for four different types of allowancesare examined: double-class allowances, hardship al-lowances, student-trainee allowances, and other al-lowances.

Double-class allowance. This allowance is anadditional amount, over and above the regular sal-ary, given to teachers who teach more than theircontracts stipulate. This allowance category has sig-nificant problems in disbursement. While, on aver-age, less than 6 percent of teachers in all four prov-inces are supposed to receive this allowance, morethan 75 percent of those designated to receive thisallowance are at least six months overdue. On aver-age, less than 20 percent of teachers are up to dateon such receipts. It’s likely that the double-class al-lowance was originally conceived of as an overtimeallowance, where extenuating circumstances, suchas the sudden departure of a teacher, forced staffmembers to take on extra duties. As a result, it isnecessary for a teacher to apply every semesterfor renewal of this allowance. Large delays in the

renewal process have led to problems in the timeli-ness of disbursements, and in some cases teachershave stopped applying for this particular allowance.28

Hardship allowance. This allowance is a clearlydefined additional amount given to teachers as anincentive to teach in rural areas. This is clearly re-flected in the large differences between those whoare supposed to receive the allowance in predomi-nantly rural (45 percent in Northern and 59 percentin Eastern) compared to urban provinces (14 per-cent in Copperbelt and 16 percent in Lusaka). Forall provinces (except Copperbelt, with 42 percent)more than 80 percent of the teachers are up to dateon the receipts, and less than 15 percent are morethan six months overdue.

Student-trainee allowance. This allowance isgiven to teachers, during the second year of theteacher training program, who have been deputedto a rural school. Again, problems with delays ap-pear. On average, only 22 percent are up to date inNorthern and Eastern provinces and 46 percent inCopperbelt and Lusaka provinces. But data on thestructure of delays does not support the hypothesisthat this money is being diverted for other purposes.Very few student trainees are more than six monthsoverdue, and as with the salary for new staff, thisseems to reflect delays in updating the payroll sys-tem in Lusaka.

Other allowances. Other allowances includeprimarily a responsibility allowance—an amountgiven to senior teachers who take on additional re-sponsibilities, such as sports coordinator. On aver-age, 70 percent are up to date on receipts, whereas30 percent are at least six months overdue.

These systematic delays in the disbursement of al-lowances might suggest leakage in the payroll system if

Table 13b. Disbursement of Salaries, by Province(percent)

Centralized provinces Decentralized provincesSalaries Northern Eastern Copperbelt LusakaThose who are supposed to receive 79.46 81.18 89.11 96.88Up to date (of those who receive) 97.78 94.29 91.80 95.481–3 months overdue 0 5.72 4.92 2.58 > 6 months overdue 2.22 0 3.28 1.94Note: Two types of teachers—student trainees and volunteers—are not supposed to receive a salary, but are paid an allow-ance either by the government or by the community.Source: ESD data.


the money not disbursed earned interest during thedelay period. This possibility could be investigatedfurther, especially by the Ministry of Finance and theAccountant-General’s office, it is not possible to as-certain at this point whether these delays arise fromleakage or reflect the time needed to update thecentralized payroll system. As discussed above, thedouble-class allowance, both in conversation withteachers and through the evidence presented here,seems to be completely nonfunctional. The require-ment that teachers file for this allowance at the be-ginning of every school term combined with the longlag time in disbursement implies that applications forthis allowance often go untendered.

In general the data on the rule-based componentsof both school and payroll disbursements suggest thatleakage problems are small (table 13c). Most schoolshave received the allotted fixed grant and most teach-ers have received salaries and well-defined allowancepayments (hardship and responsibility) on time. There

is some indication of delays in disbursement due to thelag in updating the payroll (particularly with regard tonew staff), but it is hard to benchmark these figureswithout comparable data from other countries. On thewhole, the Ministry of Education and the Ministry ofFinance have evolved an efficient system of deliveryfor rule-based allocations, with few distortions or di-version of resources to unintended recipients.

DISCRETIONARY ALLOCATIONS

The positive results from the previous sectionare completely altered when examining discre-tionary allowances. Both in the case of schoolfunding and payroll disbursements (recall that thepayment of all one-time benefits is classified asdiscretionary) the record of school and teacherreceipts of such allocations is very poor. On av-erage, schools receive no such funds, and thereare large outstanding amounts of these benefitsdue to teachers.

Table 13c. Disbursements of Rule-Based Allowances, by Province(percent)

Centralized DecentralizedAllowance Type Salaries Northern Eastern Copperbelt LusakaDouble classa Those who are supposed to receive 4.0 13.0 9.0 11.0

Up to date (of those who receive) 0 27.3 5.26 11.111–6 months overdue 25.0 0 10.53 16.67> 6 months overdue 75.0 72.73 84.21 72.22Average Monthly Amount Due (Median) 60,000 45,000 52,500 45,000

Hardshipb Those who are supposed to receive 45.0 59.0 14.0 16.0Up to date (of those who receive) 82.0 82.0 42.3 80.01–3 months overdue 0 4 7.75 0> 6 months overdue 18.0 14.0 50.0 20.0Average Monthly Amount Due (Median) 46,500 40,000 59,000 46,500

Student traineec Those who are supposed to receive 16.0 10.0 4.0 2.0Up to date (of those who receive) 22.2 22.2 55.5 50.01–3 months overdue 66.6 77.7 44.4 50.0> 6 months overdue 11.1 0.0 0.0 0.0Average Monthly Amount Due (Median) 100,000 100,000 100,000 108,500

Otherd Those who are supposed to receive 20.0 17.0 14.0 13.0Up to date (of those who receive) 69.5 60.0 39.3 66.71–3 months overdue 8.7 6.7 25.0 0> 6 months overdue 21.7 33.3 35.7 33.3Average Monthly Amount Due (Median) 50,000 40,000 50,000 41,500

Note: a. Double-class allowance is for teachers teaching more than the stipulated number of hours in their contract.b. Hardship allowance is for teachers teaching in rural areas.c. Student-trainee allowance is for trainee-teachers deputed to teach in rural schools.d. Other allowance is primarily the “Responsibility Allowance” for teachers who have taken on additional responsibilities.Source: ESD data.


Discretionary school funding. Table 14 showsthe percentage of schools that receive no funds, bysource and province. The first row shows the per-centage for the rule-based component. The remain-ing rows examine each of four potential fundingsources: the Case I—Case III donor pool and do-mestic funding; Case IV donors administering thePAGE program; Case IV donors administering otherprograms, and non-GRZ and donor sources, whichmay include churches and NGOs. The shocking re-sult from this disaggregation is that while 90 percentof schools receive the rule-based funding, less than6 percent of schools receive money from all foursources of discretionary funding. Although such lowreceipts could be justified in the case of the Case IVpool (one could argue that Case IV donors supportonly a small number of schools), the fact that justover 10 percent of schools receive money from dis-cretionary GRZ sources is troubling.

An instructive exercise decomposes receipts inschools by different sources in terms of average sharesand average amounts. While the former provides in-formation on whether a school receives money froma certain source or not, the latter provides informa-tion on how much schools receive from each source.

In this example, three schools each receive $10from Source 1, but while schools 1 and 2 each receive$0 from source 2, school 3 receives $90. In this case,source 1 accounts for 100 percent of the share offunding in schools 1 and 2, but only 10 percent in

school 3. Thus, the average share of source 1 is (thesum of the share of source 1) divided by 3 whichequals 70 percent. Of the total contribution fromeach source, source 1 contributes $30 to all schools,while source 2 contributes $90. Thus, the averageamount from share 1 is ($30 + $90)/$120 whichequals 25 percent while share 2 accounts for theremaining 75 percent. In this example, the high-average shares but low-average amounts fromsource 1 indicate that this source of funding is re-ceived by most schools, but the amounts receivedare relatively small compared to source 2.

Similar to the example above, rule-based fundingaccounts for a large proportion of average funding-shares (88 percent), but a much smaller proportionof average funding-amounts (43 percent). Hence,public funding of schools can be thought of as theoutcome of two distinct processes. Rule-based fund-ing tends to reach every school, thus accounting forthe large, average funding-share of this component.Discretionary funding reaches few schools (low fund-ing-shares), but conditional on a school receiving suchfunds, average amounts are likely to be large (highfunding-amounts). 29 This is further explored in table15, which shows the average disbursement by prov-ince of rule-based and discretionary funding condi-tional on receipt. Except for Northern province, dis-cretionary allocations are orders of magnitude higherranging from six times (Copperbelt) to 30 times(Lusaka) their rule-based counterparts. Thus, rule-based funding consists of smaller funding amountsdisbursed to all schools, while discretionary fundingconsists of larger (mean) funding amounts disbursedto few schools.

Discretionary transfers to teachers. A similarstory emerges about the disbursement of discre-tionary funds to teachers.30 Teachers may receivethree basic categories of one-time benefit transfers:

Table 14. Percentage of Schools That Receive No Funds, by Source and LocationCentralized Decentralizedprovinces provinces

Funding category Source Northern Eastern Copperbelt Lusaka All provincesRule-based Fixed–grant allocation 5.70 0 5.88 28.57 10.93Discretionary Case I–III donor pool and domestic sources 94.20 94.8 80.00 89.10 89.20

Case IV pool: PAGE 71.15 97.44 100.00 97.06 90.66Case IV pool: other sources 98.00 92.31 96.00 85.29 93.96Other sources 98.00 97.44 98.00 85.29 95.60

Source: ESD survey data.

Example

Share of source 1School Source 1 Source 2 in school funding

School 1 10 0 100School 2 10 0 100School 3 10 90 10


leave benefits, transfer benefits (allowances given toteachers when they are transferred across schools),and other benefits (notably funeral benefits). For allthree categories, the percentage of teachers whohave not received their allotted amount are high.Moreover, the amounts due represent 200 percentor more of their net salaries. Table 16 shows thepercentage of teachers overdue for such benefitpayments, as well as the average amount owed. Al-though the proportion of teachers with benefits out-standing are similar to those with outstanding allow-ances, the significant difference is in the amountsowed. While the allowances seldom account formore than 25 percent of net salary (table 13c), inthe case of these benefits, outstanding amounts arebetween 200 percent (Northern) and 500 percent(Lusaka) of net salary.

SUMMARY

The discussion of leakage in public expenditurehas relied on the classification of payroll and schoolfunding into rule-based and discretionary compo-nents. For rule-based components of school andpayroll funding, the Zambian administrative systemis working efficiently, and there is little evidence thatfunds earmarked for disbursements are not reach-ing intended beneficiaries (schools or teachers).

The reverse is true for discretionary compo-nents. In the case of school funding, less than 20percent of schools receive any funding from all dis-cretionary sources combined, while for teachers

there are substantial amounts overdue for one-timebenefits and payments. The crucial importance ofrule-based funding at the level of the school is alsohighlighted by the change in the relative shares bysources when moving down the administrative hier-archy. At the level of the province, the share of rule-based funding ranges from 8 to 40 percent, with amedian of 12 percent. Moving down to the districts,this share increases to between 19 and 44 percent,with a median of 44 percent. Finally, at the level ofthe school, this share ranges from 2 to 100 percent,with a median of 99 percent. More than 75 percentof all schools received cash resources only from rule-based sources in the current year.

One explanation for these results advanced inthe study of Uganda is based on political economyconsiderations (Reinikka and Svensson 2002). Sincerule-based funding is clearly defined with a simpleallocation rule, the capture of funds is difficult. Be-cause discretionary funds are not associated withany rule, the pattern of funding thus reflects thedifference between rules and discretion. The fewschools that receive large amounts have greaterbargaining power with higher levels of the admin-istration.

NOTES

25. One-time benefits include leave, transfer, and fu-neral benefits.

26. This still constitutes a problem since it suggeststhat delays in Lusaka are much higher than for other

Table 15. Average Amounts Conditional on Receipts, by Province(Kwacha per pupil)

Centralized provinces Decentralized provincesType of funding Northern Eastern Copperbelt LusakaRule-based 6,951 7,326 3,305 2,132Discretionary 7,593 25,679 21,008 62,236Source: ESD survey data.

Table 16. Discretionary Transfers to Teachers, by ProvinceCentralized provinces Decentralized provinces

Salaries Northern Eastern Copperbelt LusakaPercent who have no benefits outstanding 73.68 55.17 65.85 60.25Percent who have at least one benefit outstanding 26.32 44.83 34.15 39.75Average amount of outstanding benefit (K) 579,033 1,292,744 1,303,629 1,511,751Note: Benefits include leave, transfer and other benefits (mostly funeral benefits).Source: ESD data.


provinces, despite these schools being closest to the Min-istry of Education.

27. Iterviews with teachers suggest that this is prob-ably a reflection of delays in adding new teachers to thecentralized payroll rather than outright leakage in the dis-bursement of salaries.

28. In a number of interviews, teachers were surprisedthat the double-class allowance was still operative. Theyhad not applied for this allowance despite being eligible.

29. One particular case amounted to more than 60times the rule-based allotment. As the introduction men-tions, however, not all discretionary amounts are large.Some schools receive exceptionally large sums, while oth-ers receive smaller amounts.

30. In this case the use of the term discretionary maybe a bit misleading. Teachers are supposed to receive thesebenefits, but the disbursement of these amounts is left tothe discretion of the district.

Equity in Education Funding7

This sections concentrates on the equity con-siderations of school funding. It examines theimpact of rule-based funding on per-pupil re-

ceipts in rich and poor schools, as well as the flow ofdiscretionary funds. Figure 5 demonstrates why ananalysis of equity is critical to the understanding ofschool funding.31 The average per-pupil receipts perdistrict in the sample shows immense variation. Themost-funded district, at K 80,000 per pupil, receivesalmost eight times the funding received by the least-funded district, and this difference arises primarilydue to variation within provinces. A simple variancedecomposition confirms that while 9 percent of varia-tion in district funding is due to variation across prov-inces, 90 percent is due to variation within provinces.A similar result obtains at the school level. Themost-funded school receives more than 3,000times the amount received by the least-fundedschool, and again, most of the variation in schoolfunding (83 percent) arises from differenceswithin districts, with only 19 percent explainedby variation across districts.

The following variables are used to exam-ine funding differences across districts andschools, and the equity aspects of rule-basedfunding, discretionary funding, and staff remu-neration. The analysis focuses on correlationsbetween district and school funding, and threeimportant variables.

• Urbanization. The analysis examines twoaspects of the relationship between loca-tion and funding. First, at the school and

district level, it looks at whether significant ru-ral and urban biases in per-pupil fundingemerge. Second, at the school level, it looks atwhether there are different patterns of fund-ing within rural and urban schools. If there aresystematic differences in the cost of monitor-ing between urban and rural schools, this maylead to differences in funding equity dependingon the location of the school. The measure ofurbanization at the school level is based on thelocation variable in the school census (whichcategorizes a school as rural or urban). A dis-trict is defined as urban if 50 percent or moreof the children are enrolled in a school in anurban location.

43

Fund

ing (K

per

pup

il)

Variation in district receipts0

20000

40000

60000

80000

Per -pupil receipts

Fund

ing (K

per

pup

il)

Variation in district receipts0

20000

40000

60000

80000

Per -pupil receipts

Figure 5. Variation in District Receipts

Source: ESDS.


• Distance. Preliminary discussions with dis-trict officers and head teachers suggested thatanother important source of variation in fund-ing is the distance of the district office from theprovince office, and distance of the school fromthe district office. The analysis looks at the roleof physical distance in explaining funding varia-tion. The measure of distance is based on ques-tions regarding the time taken (and the physi-cal distance) to the school and the district of-fice administering to district officers and headteachers.

• Wealth. Perhaps the most important corre-late is the wealth of the recipients. Althoughthe urbanization and distance variables can helpexplain the system of funding, correlations withschool funding do not allow us to make anyclaims regarding the equity of the funding sys-tem. One major innovation of the current studyis the use of an index based on the assets ofstudents’ households (obtained through a sur-vey), rather than on proximate determinants,such as district-level wealth or school infrastruc-ture, which suffer from severe endogeneity bi-ases and measurement error.32, 33

Table 17 reports the summary statistics for thesethree variables. As expected, Copperbelt and Lusakaprovinces are predominantly urban, and Eastern andNorthern provinces are mostly rural. Furthermore,urban schools are richer than rural schools, with morethan a one standard deviation difference in the valueof the index.34 Moreover, schools in Northern andEastern province tend to be further from district andprovince offices than their counterparts. Finally,Copperbelt and Lusaka are the richest provinces, even

after controlling for school location. The asset indexfor pupils in schools located in the former two prov-inces are one standard deviation higher than inNorthern and Eastern province. Although the dif-ferences are smaller after conditioning on the loca-tion of the school, they remain significant.

EQUITY IN RULE-BASED FUNDING

Another aspect of equity involves the variation inrule-based funding at the district and school level,and how this relates to urbanization, distance, andwealth. Given that there is no evidence of leakage inthe distribution of rule-based funds, the equity as-pects of rule-based funding will depend crucially(both at the district and the school level) on the re-lationship between school size and these three vari-ables. At the district and school levels, schools withsmaller average enrollments will receive higher fund-ing.

Table 18 examines the relationship between en-rollment and the three variables of interest. The vari-ables are summarized by three terciles of enrollment.At the district level, schools with low enrollment aresignificantly poorer and less urban. When movingfrom those districts with the lowest average enroll-ments to those with the highest average enrollments,the standard deviation difference in the asset indexis more than one, and there is a 64 percent increasein the number of urban schools. Schools with lowerenrollments are almost one and one-half standarddeviations poorer than those with high enrollments.Low-enrollment schools are also predominantly ru-ral (95 rural percent compared to 18 percent ur-ban) and tend to be further from administrative of-fices (less than 5 percent are within 5 km comparedto 62.3 percent for the high-enrollment schools).

Table 17. School and District Characteristics,by ProvinceCentralized provinces Decentralized provinces

Variables Northern Eastern Copperbelt LusakaPercentage of urban schools 7.69 5.13 72.70 59.46Percentage of rural schools 92.31 94.87 27.30 40.54Districts with at least 50 percent urban schools (percent) 0.00 0.00 60.00 50.00Schools with the district office less than 5 km away (percent) 23.00 17.90 33.90 29.70Schools with the province office less than 5 km away (percent) 7.70 5.20 16.00 24.30Average Wealth Index of Urban Schools 0.64 0.34 0.80 0.71Average Wealth Index of Rural Schools –0.50 –0.67 –0.32 0.15Source: ESD data.


Note the large differences in enrollment across thesethree categories. Moving from the lowest to the high-est average enrollment terciles increases enrollmentby almost a factor of six. The table thus allows us tonaturally think of districts and schools as falling inone of two clusters: (1) low enrollment, poor, andrural; and (2) high enrollment, rich, and urban.

This immediately suggests that the rule-based al-location, as implemented in the Zambian context,will be progressive. Rich (and urban) districts andschools will receive less per-pupil funding than poor(and rural) schools, since they will tend to have higheraverage enrollments. Tables 19a and 19b confirm thisintuition. Districts with low average wealth receivemore than double the amount of high average wealthdistricts, and urban districts receive half the amountreceived by rural districts. At the school level, thesedifferences are even larger: schools with childrenfrom low-income households receive three timesmore than those with children from high-incomehouseholds, and rural schools receive two and one-half times more than urban schools. A simple test ofdifferences in means shows that these differencesare significant. Further, the result that poorer schoolsreceive higher per-pupil funding continues to hold ina multiple regression framework, where the assetindex is interacted with the location of the school.Thus even within rural (urban) locations, rich rural(urban) schools receive significantly less than poorrural (urban) schools (the coefficient on the interac-tion varies from K 1,800 to K 2,100 depending onthe specification used and is always significant).

Thus, rule-based allocations have a progressive im-pact on school funding, at least in terms of equity, andis propoor even after controlling for rural-urban differ-ences. Because of the inverse relationship between

wealth and enrollment, as well as urbanization andenrollment, a one standard deviation decrease in theasset index is associated with almost a 100 percentincrease in per-pupil funding. In combination withthe findings on the allocation of this funding compo-nent, it appears that rule-based funding has had apositive impact on school funding in Zambia. Theamount allocated has reached almost every school,and poorer schools have received higher per-pupilamounts than their richer counterparts.

EQUITY IN DISCRETIONARY FUNDING

The previous section demonstrates the progres-sive impact of rule-based funding. But as the track-ing exercise previously established, such funding ac-counts for only 30 percent of the total funds in thesystem. Overall equity in the educational system willalso depend on the allocation of discretionary funds.If these funds are allocated primarily to poor dis-tricts (schools), the progressive nature of rule-basedfunds will be further accentuated. But if discretion-ary funds are primarily distributed to richer districtsand schools, the overall progressive nature of the

Table 18. Enrollment, Urbanization, Wealth, and DistanceEnrollment in schools

Level Urbanization, wealth, and distance Low Medium HighDistrict Average district wealth –0.52 0.14 0.67

Schools in urban locations (percent) 2 4 66School Average enrollment 275 719 1,695

Urban schools (percent) 5 25 82Schools less then 5 km from the district office (percent) 4.84 22.11 62.3Schools less than 5 km from the province office (percent) 5.2 13.1 21.5Average wealth index –0.66 –0.07 0.75

Source: ESD data.

Table 19a. Equity in Rule-Based Funding at theDistrict Level(Kwacha per pupil)

District levelUrbanization, wealth, and distance fundingLow average wealth 13,429Medium average wealth 11,392High average wealth 5,754Districts with 50 percent or more schools urban 5,206Districts with 50 percent or more schools rural 11,394Note: District averages weighted by primary enrollment.Source: ESD data.


funding system may be severely undermined. To ad-dress this concern the methodology used is similarto the previous section. Attention is restricted tothe relationship between discretionary funding andurbanization, distance to administrative offices, andthe household wealth of children attending theschool.

The results from this exercise are presented intables 20a and 20b. First, at the level of the district,discretionary funding remains progressive, butmuch less so than rule-based allocations. While lowaverage wealth districts receive almost equalamounts from discretionary and rule-based alloca-tions, districts with high average wealth now re-ceive nearly twice as much discretionary fundingcompared to rule-based funding. Further, districtsthat are more urbanized receive less than those thatare more rural, but the difference is smaller than

that for rule-based funding. Thus, discretionary flowsfrom the province to the district continue to favorpoor districts, but funding differences between richand poor districts are much reduced. Consequently,at the district level it is not possible to reject thehypothesis that discretionary transfers from prov-inces and the center are the same across all districts.

At the level of the schools, equity in discretion-ary funding is more complicated. Table 20b showsthat differences in the probability of receipts by dis-tance from an administrative office and the locationof the school tend to be small, and if present, tendto favor more disadvantaged schools.35 These results,however, are not statistically significant. In terms ofthe asset index, schools with children from middle-income (34 percent) households have a much higherprobability of receipt compared to those from ei-ther low- (15 percent) or high-income households(23 percent), although again, these differences arenot statistically significant.

Table 20c further decomposes the allocation ofdiscretionary funding within urban and rural schools.

Table 20b. Equity in Discretionary Funding at the School LevelDiscretionary receipts Probability of receiving any

Urbanization, wealth, and distance (Kwacha per pupil) discretionary fundingPupils from low-income households 1,229 0.15Pupils from middle-income households 11,759 0.34Pupils from high-income households 5,986 0.23Urban schools 4,076 0.24Rural schools 7,594 0.24Less than 5 km from the district office 1,330 0.31More than 5 km from the district office 8,375 0.21Less than 5 km from the province office 962 0.18More than 5 km from the province office 7,055 0.25Source: ESD data.

Table 19b. Equity in Rule-Based Funding at theSchool Level(Kwacha per pupil)

Rule-basedUrbanization, wealth, and distance fundingPupils from low-income households 8,276Pupils from middle-income households 4,063Pupils from high-income households 1,868Urban schools 2,150Rural schools 6,223Less than 5 km from the district office 2,380More than 5 km from the district office 5,716Less than 5 km from the province office 1,911More than 5 km from the province office 5,137Source: ESD data.

Table 20a. Equity in Discretionary Funding at theDistrict Level(Kwacha per pupil)

District levelUrbanization, wealth, and distance fundingLow average wealth 13,046Medium average wealth 9,101High average wealth 8,341Districts with 50 percent or more schools urban 7,984Districts with 50 percent or more schools rural 10,671Note: District averages weighted by primary enrollment.Source: ESD data.


The disaggregation between rich and poor schoolswithin urban (rural) locations suggests that forschools in urban areas there is no difference in prob-abilities of receiving discretionary funding by eitherdistance to administrative offices or the wealth ofthe pupils.36 Among rural schools, however, there isa statistically significant difference between schoolswith richer students compared to those with poorerstudents, and a positive (but not significant) differ-ence between schools that are closer to the districtoffice compared to those further away. This differ-ence in receipt probabilities converts into a signifi-cant difference in the amounts received as discre-tionary payments in rich (K 13,893 per pupil) com-pared to poor (K 1,295 per pupil) rural schools.

In fact, the difference in discretionary fundingbetween rich and poor schools in rural areas is largeenough to overturn the progressive nature of rule-based allocations, with the rich (K 18,129 per pupil)receiving almost double the amount received by thepoor (K 9,507 per pupil) in rural areas. Contrast thiswith urban schools. Although the progressive natureof rule-based funding is reduced once discretionaryallocations are taken into account, greater fundingof poorer schools is maintained with richer schools(K 5,068 per pupil) receiving marginally less thanpoorer schools (K 7,430). When discretionary andrule-based components are combined it alters thenature of funding equity in the Zambian educationalsystem. Note, however, that all nonsalaried fundingin basic and primary schools accounts for a smallpercentage of total funding, with the rest paid out instaff compensation. The discussion of equity now

turns to the issue of staff remuneration using thesame three variables.

EQUITY IN STAFF COMPENSATION

Although teachers are allocated to schools, theymay request transfers or leave the workforce.37 As aresult, rural schools (and schools that are far from theroad) have a smaller number of teachers than theirurban counterparts.38 As a result, differences in stafffunding stem either from systematic salary differen-tials or systematic class-size differentials (or a combi-nation of both) across different schools. These issuesare examined using data on teacher’s salaries and al-lowances collected as part of the survey.39

The results are presented in tables 21a and 21b.The first column of table 21a shows only the salarycomponent of staff remuneration. Along all dimen-sions, per-pupil funding of more privileged schoolsis higher. Richer schools receive more than doublethe amount (K 8,318 per pupil) poor schools re-ceive (K 3,963 per pupil). Similarly, schools in urbanareas and those closer to administrative offices re-ceive one and one-half times their counterparts. Thisdifference is partially mitigated when adding in al-lowances that teachers are supposed to receive (col-umn 3), but re-emerges when teacher allowancesactually received are considered (column 4).

Part of this inequity arises from salary differen-tials across teacher types. The greater use of teachertrainees in rural and poorer schools (14.4 percent inrural schools compared to 6.4 percent in urbanschools) implies that the average salary bill per stu-dent is lower in rural schools and in schools with

Table 20c. Equity in Discretionary Funding at the School Level, by LocationRural locations Urban locations

Discretionary Probability of Discretionary Probability ofreceipts receiving any receipts receiving any

Urbanization, wealth, and distance (Kwacha per pupil) funding (Kwacha per pupil) fundingPupils from poorer households 1,295 0.16 5,003 0.22Pupils from richer households 13,893 0.32 3,178 0.25Less than 5 km from the district office 1,442 0.33 1,271 0.29More than 5 km from the district office 8,772 0.23 7,364 0.17Less than 5 km from the province office 0 0 1,113 .21More than 5 km from the province office 7,803 0.24 5,355 0.25Note: The classification of schools by wealth is done within location in this table, so that schools that are rich in the rurallocations may be middle-income in the entire sample, and schools that are poor in the urban locations may similarly be middle-income in the entire sample.Source: ESD data.


children from poorer households, an effect that isfurther exacerbated by higher class sizes. Table 21bshows the interaction of these two effects. The av-erage class size is significantly different for rural andpoorer schools. Moving along a continuum fromschools with children from low- to high-incomehouseholds, average class size decreases from 64 to39, and moving from urban to rural schools, classsize increases by almost 50 percent from 43 to 61.Significant differences in salaries also exist. The av-erage teacher salary in a low-income school is al-most 25 percent lower than a high-income school,and in rural schools salaries are16 percent less thanin urban schools.40

SUMMARY

This section explores the equity aspects of rule-based and discretionary funding in the Zambian

education system. The results show that the rule-based component of educational funding is progres-sive with significant positive differences in per-pupilreceipts both at the district and school levels andbetween rich and poor schools. The findings for dis-cretionary funding suggest a disturbing trend. Expen-diture flows from provinces to districts are onlymildly progressive, and hence total receipts at thedistrict level retain a progressive nature, althoughthis is smaller than what it would be under rule-basedfunding only.

Once the money reaches the district, discre-tionary funding in rural schools is disbursed towealthier schools with higher probabilities. Be-cause the discretionary disbursements are largerthan the rule-based allotment, the progressivenature of the funding system disappears. Fundingthat flows from the district to the school is wealth

Table 21b. Equity in Per-Pupil Staff Remuneration, Class Size and SalariesAverage teacher salary

Urbanization, wealth, and distance (Kwacha per month) Average class sizeSchools with pupils from low-income households 212,135 64Schools with pupils from middle-income households 235,606 60Schools with pupils from high-income households 277,750 39Schools that are urban 270,107 43Schools that are rural 228,195 61Schools less than 5 km from the district office 273,929 41Schools more than 5 km from the district office 228,823 60Schools less than 5 km from the province office 286,344 41Schools more than 5 km from the province office 235,976 56Source: ESDS.

Table 21a. Equity in Per-Pupil Staff Remuneration at the School Level(Kwacha per month)

Per-pupil total(salary + allowances)

Per-pupil What teachers are What teachersUrbanization, wealth, and distance salary supposed to receive actually receivePupils from low-income households 3,963 5,128 4,445Pupils from middle-income households 5,190 6,334 5,755Pupils from high-income households 8,318 8,997 8,662Urban 7,308 7,755 7,423Rural 5,065 6,344 5,728Less than 5 km from the district office 7,422 8,062 7,654More than 5 km from the district office 5,198 6,332 5,750Less than 5 km from the province office 7,702 8,169 7,760More than 5 km from the province office 5,602 6,660 6,113Source: ESD data.


neutral (and potentially regressive in rural regions).Factoring in staff compensation, the results show thatthe only progressive component of the Zambian edu-cation system is the rule-based allocation. Onceteacher funding is added in, the entire educationfunding system becomes regressive, with poorerschools receiving less than richer schools.

These findings are summarized in figure 6. Thehorizontal axis is the cumulative wealth distributionand the vertical axis is the share of public fundingthat each group receives. Each curve in the graphsshows the share of public funds that accrues to (forinstance) the poorest percent of the population. Theline of perfect equality is the diagonal that showsfunding shares if all funds were distributed equallyamong the population.

The graphs in figure 6 clearly demonstrate thefollowing:

• Rule-based funding is always progressive withhigher shares for the poorer schools. Thecurve always lies above the diagonal so that (forinstance) in the entire sample, the poorest 20percent of the population receives 38 percent

of all rule-based funding. Such funding is pro-gressive not only for the entire sample ofschools, but also within urban and rural schools.

• Staff compensation is always regressive, withthe curve lying below the diagonal for all threesamples (the entire sample, and within urbanand rural).

• Discretionary funding is regressive in theentire sample. A closer look at such fundingwithin urban and rural schools shows that dis-cretionary funding is progressive within urbanschools and regressive within rural schools.

If there are huge funding differentials betweenschools (for instance, the rule-based componentimplied that per-pupil funding could vary from K1,889 to over K 8,000 depending on the school), andsome schools receive a lot less than others, doeshousehold spending adjust? Evidence that householdspending is crucial for schools can be obtained froma survey of head teachers regarding potential fundsthat the school could access during a normal schoolyear. For every issue that was asked in the survey

Figure 6. Inequalities in Public Funding in Basic Zambian EducationAll schools

0

Rural schools

Urban Schools

Shares: DiscretionaryShares: Rule-Based

Shares: Staff PaymentsPerfect Equality

0.2

0.4

0.6

0.8

1

Wealth

Shar

e of f

undi

ng

Poorest 20%

Poorest 40%

Poorest 60%

Poorest 80%

All households

0.2

0.4

0.6

0.8

1

Wealth

Shar

e of f

undi

ng

Poorest 20%

Poorest 40%

Poorest 60%

Poorest 80%

All households

0.2

0.4

0.6

0.8

1

Wealth

Shar

e of f

undi

ng

Poorest 20%

Poorest 40%

Poorest 60%

Poorest 80%

All households

All schools

0

Rural schools

Urban Schools

Shares: DiscretionaryShares: Rule-Based

Shares: Staff PaymentsPerfect Equality

0.2

0.4

0.6

0.8

1

Wealth

Shar

e of f

undi

ng

Poorest 20%

Poorest 40%

Poorest 60%

Poorest 80%

All households

0.2

0.4

0.6

0.8

1

Wealth

Shar

e of f

undi

ng

Poorest 20%

Poorest 40%

Poorest 60%

Poorest 80%

All households

0.2

0.4

0.6

0.8

1

Wealth

Shar

e of f

undi

ng

Poorest 20%

Poorest 40%

Poorest 60%

Poorest 80%

All households

0.2

0.4

0.6

0.8

1

Wealth

Shar

e of f

undi

ng

Poorest 20%

Poorest 40%

Poorest 60%

Poorest 80%

All households

0.2

0.4

0.6

0.8

1

Wealth

Shar

e of f

undi

ng

Poorest 20%

Poorest 40%

Poorest 60%

Poorest 80%

All households

0.2

0.4

0.6

0.8

1

Wealth

Shar

e of f

undi

ng

Poorest 20%

Poorest 40%

Poorest 60%

Poorest 80%

All households


(ranging from money to repair a borehole to moneyfor buying chalk and textbooks), at least 60 percentof all head teachers cited community sources andparents as the primary source of funding. Tradition-ally, these funds were raised through Parent Teach-ers Association (PTA) fees. However, with the re-cent ban on such fees, parents may have shiftedspending on education to within the household. Priorto the ban, funding might have consisted of contri-butions to the school fund. After the ban, house-holds are now spending more on private educationalgoods, such as textbooks.

NOTES

31. A concern that arises from the restricted empha-sis on cash flows is the following: if provinces and dis-tricts systematically provide more in-kind transfers (ratherthan cash) to a particular type of district or school, theresults would not accurately reflect the equity implica-tions of school funding. For instance, if districts providedonly cash to rich schools nearby, but educational materi-als to poor schools further away, by focusing only on cashwe would find large differentials. The data on the distri-bution on educational materials shows that this is not thecase. The receipts of such materials has been very small,and if anything, this information further strengthens theresults.

32. For instance, one measure of wealth often used isthe infrastructure of the school. As the preceding sectionclearly shows, since school infrastructure is a combina-tion of household and public contributions, if the govern-ment has a stated policy of systematically subsidizingpoorer schools, measures of wealth based on this vari-able will severely understate the difference between richand poor schools.

33. The constructed asset index is a standardized(mean 0 and standard deviation 1) distribution constructedusing item response theory methods. The appendix pro-vides details of the index and its properties. While therehave been some concerns about the use of asset indexes,these methods allow us to gauge the accuracy of the in-dex through computation of the standard errors at eachpoint of the wealth distribution. The results from this

exercise are presented in the appendix. The main findingis that the index is fairly reliable with standard errors lessthan 0.2 for most values of the index, reaching a maxi-mum of 0.6 for the poorest households. Across prov-inces, the asset index suggests that Eastern and North-ern provinces are the poorest, with a difference of morethan one standard deviation from Copperbelt and Lusaka.This is also in line with the LCMS (World 1998) that sug-gested poverty rates of 53 percent in Lusaka, 65 percentin Copperbelt, 79 percent in Eastern, and 81 percent inNorthern.

34. Recall that the index is a standardized distributionwith a mean of zero and standard deviation of one.

35. For instance, 21 percent of schools less than 5 kmfrom the district office receive such funds compared to31 percent of schools further away.

36. Wealth terciles are computed within urban andrural samples, so that a school classified as middle-incomein the full sample may be classified as rich within ruralareas (or poor within urban areas).

37. The lack of housing—one reason to request atransfer—is the norm rather than the exception, withonly 32 percent of teachers reporting that they are livingin staff housing.

38. This has been a consistent cause for concern andthe Ministry of Education has taken several steps, includ-ing the establishment of hardship allowance and field-based training for those in training college, to increasethe supply of teachers to rural regions.

39. As part of the survey, teachers for grades 5 and 6were interviewed. We compute staff funding by taking theaverage salary of all teachers interviewed in a given school.This is then multiplied by the total number of teachers toprovide the total staff bill, and finally divided by total schoolenrollment to provide the per-pupil staff bill for the school.This estimate will be biased if there are systematic differ-ences between grades 5 and 6 teachers and those teach-ing other grades. This issue is currently under investiga-tion using data from the school census.

40. This is not due to differential salaries per se, butto the significant presence of trainee teachers with lowernet payments in rural areas—for example, they accountfor 16 percent of all teachers interviewed in Northernprovince.

Private Expenditures for Education8

This section explores how household spend-ing impacts funding equity in the Zambianeducation system, and presents preliminary

results from a household survey that complementsthe public expenditure tracking method.

HOUSEHOLD CONTRIBUTION TO SCHOOLS

The issue of school funding from household con-tributions is at an important juncture in Zambia.Anecdotal evidence from a number of studies dur-ing the 1990s suggested that schools had startedcharging higher PTA fees and were using direct andmore subtle means to ensure that students whocould not pay these fees stayed home. The concur-rent decline in net enrollment during the same timeperiod led to an association between enrollment andPTA fees. Parents were not sending their childrento school because they could not afford PTA fees. InApril 2002 (three months prior to the fielding of theESDS), PTA fees were abolished for primary andbasic schools (although secondary schools could anddo continue to charge fees) and the government re-iterated its commitment to free basic education. Thiswas the final step in a reversal from the former eraof user fees, and schools became limited to fund-raising though community fetes and events, whereprivate funding could not be tied to any specific child.

Since the announcement was made in the middleof the school term (which starts in January), somefamilies had already paid the PTA fees and this isreflected in figure 7 as the collection of positive feesby most schools. As the graph shows, the result of

this announcement has been dramatic. From a highpoint of K 10,554 in 2000, fees per pupil havedropped to an average of K 3,269 per school at thetime of the survey.41

However, the experience of different provincesand locations has been very different from the aver-age experience shown in figure 7. A disaggregationin the decline in PTA fees by province and location(figure 8) shows that major declines have occurredin Lusaka and Copperbelt provinces, primarily inurban areas. In the rural areas of Northern and East-ern provinces, PTA fees were already minimal to startwith so the abolishment of such fees had little impact.The major declines in urban Copperbelt and Lusakaaccount for 67 percent of the total decline in fees.

Figure 7. Declining Fees since 2000

Source: ESDS.Fees per student

0

2000

4000

6000

8000

10000

2000 2001 2002Fees per student

0

2000

4000

6000

8000

10000

2000 2001 2002

51


Adding in the rural schools in these provinces takesthe total up to 87 percent, with less than 14 percentof the total decline being accounted for by Easternand Northern provinces (table 22).42

Finally, anticipating the discussion on equity, fig-ure 9 shows the relationship between the averagewealth of children and PTA fees in the school.43 The

vertical axis corresponds to predicted PTA fees andthe horizontal axis plots the value of the averagewealth index for the school. By 2002 this relation-ship had completely changed from 2000 (and to aslightly lesser degree in 2001), where a one stan-dard deviation increase in the average wealth of chil-dren attending the school led to almost a K 10,000increase in PTA fees. In 2002 this relationship is muchweaker with a similar increase in average wealth re-sulting in less than a K 1,000 increase in PTA fees.

The changes in household contributions is sum-marized in table 23, which provides a detailedbreakdown, by province, of every funding sourceavailable to schools. For all provinces, but especiallyfor Eastern and Northern provinces, public funds arethe most important financial flow to the school. Pub-lic funds comprise 96 percent of the total funding inEastern province (the most reliant on public funds)and 82 percent in Lusaka (the least reliant on publicfunds). Of these public funds, between 24 and 40percent arrive in the form of staff remuneration,with the rest split between rule-based and discre-tionary allocations. Once private funding is takeninto account, schools in Northern and Copperbeltprovinces seem to have the lowest access to fundsat the school level (K 15,400 per pupil), while thosein Eastern province are slightly better off (K 18,085per pupil). Finally, schools in Lusaka are the best-funded, receiving an average of K 35,000 per pupil,and this is a result primarily of high discretionary(K 18,300 per pupil) and private funding (K 6,810per pupil) at the school level.

The decreased importance of household contri-butions to school funding suggests that the addition

Table 22. Decomposing the Decline in FeesProvince Urban RuralCopperbelt 46.63 7.77Lusaka 19.27 13.77Eastern 0 8.4Northern 0.77 3.38Note: This table shows that percent decline in PTA fees thatcan be attributed to each province/location combination.Source: ESD data.

Table 23. Public and Private Sources of Funding(percent of total in brackets)

Public funding Household funding(Kwacha per pupil) (Kwacha per pupil)

Staff Discretionary Rule-based Cash raised from TotalProvince remuneration funds funds households fundingCopperbelt 6,759 3,296 3,111 2,546 15,711

(43.0) (20.9) (19.8) (16.2) 100Lusaka 8,465 1,8331 1,523 6,811 35,130

(24.1) (52.2) (4.3) (19.4) 100Eastern 5,477 4,609 7,326 673 18,085

(30.3) (25.5) (40.5) (3.7) 100Northern 5,031 2,629 6,551 944 15,154

(33.2) (17.3) (43.3) (6.2) 100

Figure 8. Fee Decline by Province and LocationFe

es p

er pu

pil

Province and location

0

5000

10000

15000

20000

25000

Fees per student in 2000Fees per student in 2001Fees per student in 2002

COP-RUR

COP-URB

EAS-RUR

EAS-URB

LUS-RUR

LUS-URB

NOR-RUR

NOR-URB

Fees

per

pupil

Province and location

0

5000

10000

15000

20000

25000

Fees per student in 2000Fees per student in 2001Fees per student in 2002

COP-RUR

COP-URB

EAS-RUR

EAS-URB

LUS-RUR

LUS-URB

NOR-RUR

NOR-URB


of such sources of funds should not alter results re-garding the equity of school funding obtained in theprevious section. Table 24 shows that this is indeedthe case. Using the same variables as before, we findthat richer schools closer to administrative offices andlocated in urban areas raise more money through pri-vate sources. However, the difference between anytwo categories in private funding is always much lessthan that across public funding. For instance, the dif-ference between schools with poorer and richerhouseholds in private funding is K 1,300 per pupil, and

this increases to K 3,100 in thecase of public funds. Similarly,schools that are closer to thedistrict office raise K 300 moreper pupil, but this differenceis dwarfed by the K 8,200 dif-ference in public funds withinthe same categories. Conse-quently, the addition of pri-vate contributions to schoolsin educational funding doesnot exacerbate differences infunding between rich andpoor schools by any signifi-cant amount (or alter resultsfrom section 7).

This does not imply thathousehold expenditures haveno impact on equity. And thekey point to recognize is theimportant difference be-tween household contribu-

tions to school funds, and household expenditureon education. Even if the former is small comparedto total funding in the system, the latter can be largeand have significant equity implications. In fact, thelatter would be expected to increase preciselywhen the former is small. An example clarifies thispoint. In a system with PTA fees, these fees maybe used to buy textbooks that are then distributedto students. In a system with zero PTA fees, how-ever, the household may still buy textbooks. So itcould well be that textbooks are now bought and

Figure 9. The Relationship between PTA Fees and Wealth, 2000–2002

Source: ESD study.

Pred

icted

fees

Linear regression

Predicted fees 2002Predicted fees 2001Predicted fees 2000

-1.12436 1.70723

0

5000

10000

15000

20000

25000

Pred

icted

fees

Linear regression

Predicted fees 2002Predicted fees 2001Predicted fees 2000

-1.12436 1.70723

0

5000

10000

15000

20000

25000

Table 24. Equity in Public and Private Funding at the School Level(Kwacha per pupil)

Public funding Private funding(Rule-based, discretionary (PTA, registration, sports

Urbanization, wealth, and distance and staff allocations) and other fees)Pupils from poorer households 13,863 1,371Pupils from richer households 16,919 3,102Urban 19,473 3,202Rural 13,203 1,720Less than 5 km from the district office 21,266 2,474More than 5 km from the district office 13,019 2,134Less than 5 km from the province office 10,424 3,338More than 5 km from the province office 16,038 2,085Note: This table shows the per-pupil funding raised at the school level from public and private sources.Source: ESD survey.


consumed at the household rather than the schoollevel. While private spending in both systems couldbe higher for the rich compared to the poor thiswould be picked up at the school level in the firstcase, but similar data in the second case wouldlead to the erroneous conclusion that the contri-bution of household spending to funding equity issmall. To examine the role of private educationalspending on equity it is necessary to disaggregatespending at the household level.

This immediately creates two problems. Thefirst is that data on household expenditures can-not be obtained from a survey of schools. Suchinformation must be gathered at the level of thehousehold. And to create a total picture of publicand private expenditure, it is critical to havematched data between households and schools, sostandard household surveys will not work. Wewould need to know not only the level of house-hold expenditures on education, but also the levelof public funding in the school where the child iscurrently enrolled.44

The current survey deals with this problem inan innovative way. Using Geographical InformationSystems the location of every school was plottedand households were surveyed only in those ar-eas where schools were sufficiently far apart. Thusthe data is based on the household survey con-ducted alongside the school survey in a limitedsample of schools.45

This technique solves both problems arisingfrom the need for matched data between schoolsand households, and the more technical problemdetailed in note 45. One problem with this tech-nique is the lack of widely separately schools inurban areas. The sample of households is exclu-sively chosen from rural and remote regions of theschool sample; hence the results should be under-stood as relating only to the sample of schools

matched to households and not for the entire sur-vey.

HOUSEHOLD-LEVEL SPENDING ON EDUCATION:METHODOLOGY

Our concern regarding the inability of householdcontributions at the school level to accurately cap-ture differences in household-level spending on edu-cation is laid out in table 25 and figure 10. Table 25decomposes household expenditure on fee andnonfee items at the level of the household.46 As thisdecomposition shows, nonfee expenditures are sev-eral times higher than current expenditure on fees,and this difference is consistent across provinces.Households in Copperbelt province (which reportthe smallest proportional difference) spend threetimes as much on nonfee items compared to feeitems, and this increases to a factor of nine in North-ern province. Finally, as figure 10 shows, these ex-penditures are a large potential source of inequali-ties in schooling inputs, with the rich spending farhigher amounts than the poor.

The first step to incorporating private expendi-ture into the analysis of equity in school funding isto continue with the previous methodology of ex-amining the association between funding and wealthfor the households in the household survey afterincorporating spending on education at the house-hold level.47 The second step is to examine a re-lated issue. To the extent that there is substitution

Table 25. Fee and Nonfee Expenditures, by Province(Kwacha per pupil)

Mean fees Mean nonfeeProvince per child expenditure per childCopperbelt 5,825 18,458Lusaka 5,083 20,459Eastern 2,453 15,018Northern 1,063 9,854Source: ESD survey.

Figure 10. Household Spending on Education byWealth Terciles

Source: ESD study.

Per-c

hild

spen

ding

0

5000

10000

15000

20000

25000Per-child fees Per-child non-fee spending

Poor Middle Rich

Per-c

hild

spen

ding

0

5000

10000

15000

20000

25000Per-child fees Per-child non-fee spending

Poor Middle Rich


between nonfee spending at the household leveland public funding at the school level, it may bepossible to increase funding equity by increasing dis-bursements of such funds to schools. The table be-low shows two different scenarios of public and pri-vate expenditures.

Under both scenarios, moving from a low to ahigh level of public funding decreases expenditureat the household level. However, under the first sce-nario, this decline is the same for both rich and poorhouseholds. Households decrease expenditures byexactly the amount of the increase in public funds.In this case increased public funding does not increasetotal expenditure for education and does not changethe equity of public funding. Under the second sce-nario, however, while again the level of householdexpenditures decreases with an increase in publicfunding, the decline does not equal the increase inpublic funds. As a result total funding in the systemincreases, and since the poor increase total expen-diture on education (public + private) more than therich, this improves equity in the funding system.Clearly the impact of public funding on equity willthus depend on the exact relationship between pub-lic funds and household expenditures, and theoreti-cally either scenario is possible.48 We therefore lookat two related questions: (1) will an increase in ex-ternal funding decrease household schooling expen-ditures? and (2) will the extent of this decrease dif-fer by the wealth of the household?

The next two sections examine the implication ofprivate spending for equity, and the issue of substitut-ability between private spending and public funds.

THE IMPACT OF PRIVATE SPENDING ON EQUITY

Table 26 shows the results of incorporatinghousehold-level spending on funding equity. The first

two columns show public funding per child in thehousehold survey, while the next two columns showprivate funding, disaggregated by nonfee and feeexpenditures per child. One effect of the sample ofremote schools is that there is a much smaller dif-ference between the poorest (K 3,951 per pupil) andthe richest (K 4,530 per pupil) quintile in salariedfunding per pupil. Furthermore, there is also no dif-ference in the rule-based component of fundingacross quintiles, although differences in discretion-ary funding result in greater public funding for thetop 40 percent. The fee component of private fund-ing increases fourfold as we move from the poorestto the richest quintile (K 1,164 to K 4,109), but theseamounts tend to be relatively small compared toother sources of funds in the system. The single larg-est source of inequality comes from nonfee house-hold expenditures, which doubles from K 8,659 perchild to K 19,563 per child. Since this component isalmost six times as large as the fee component, itaccounts for a larger share of inequality in the fund-ing system.

Figure 11 clearly shows the relationship betweenfunding and wealth. It plots the modified Lorenz curvesfor public and private funding, where the horizontalaxis represents the share of the population and thevertical axis represents the share of funding. Whilethere is inequality in both public and private funding,inequality in the private funding dominates (the Lorenzcurve is always further below the diagonal).

The addition of private expenditures to publicfunds exacerbates funding inequalities in the educa-tional system. For the sample of schools with house-hold survey data, the bottom 50 percent receive justover 40 percent of all funding, if we don’t add privatefunds. Once we factor in spending at the householdlevel, this share decreases to 34 percent, suggesting

Example

Equity: percent ofLevel of Per-pupil Per-pupil Per-pupil total spendingpublic public private private accounted for

Scenario funding funding spending (rich) spending (poor) by the rich

Scenario 1: Increase in public Low 1,000 9,000 4,000 66 funding does not change equity High 3,000 7,000 2,000 66Scenario 2: Increase in public Low 1,000 9,000 4,000 66 funding increases equity High 3,000 8,000 3,000 64


that even among the relatively poor households inthis particular sample, funding inequalities are main-tained and further strengthened through house-hold-level inputs. The next section examines therelationship between public and private funding,and whether these inequalities could be offsetthrough increases in public funding.

THE SUBSTITUTABILITY OF PRIVATE FOR PUBLIC

FUNDING

This section examines the relationship betweenpublic funding and private expenditures, with tworelated questions in mind. First, to what extent arepublic and private funds substitutable (so that house-holds decrease spending on education when publicfunding increases)? Second, to what extent does anincrease in public funding differentially impact privatespending across rich and poor households? Thehousehold data is used to examine the impact ofschool funding on household nonfee expenditures.The sample is divided into those that receive highper-pupil funding and those that receive low per-pupil funding.49

Table 27 shows the startling difference in themeans of household spending when we move froma school with low per-pupil funding to one with highper-pupil funding: nonfee expenditures more thanhalves from K 20,000 to K 9,400 when funding in-creases from K 3,600 to K 11,500. Moreover, totalfunding in the system (the sum of private and publicfunding) actually decreases across schools with lowand high public funding.50 Similar results obtain in amultiple regression framework:

Log of household nonfee expenditure =Constant + a. Wealth + b. School Fund-ing + c. School Funding x Wealth + d.Village Wealth

Here the wealth variable is used to classify house-holds into different income groups based on the as-set index of household possessions. The term schoolfunding x wealth captures the interaction of publicfunds with wealth. If this term is positive and signifi-cant it would imply there are differential impacts ofpublic funding across different income groups. Fi-

nally, the village wealth variableis the average wealth of the vil-lage that captures, to a limitedextent, other characteristics ofthe village that may be corre-lated with household expendi-ture on schooling.51

Table 28 shows the resultsfrom this estimation. Threespecifications are presented.The first column uses wealthterciles and the second columnuses a continuous wealth index.Since some households reportzero fungible expenditures, andthe dependant variable is the log

Figure 11. Lorenz Curves of Inequalities in Public and Private Funding

0

0.2

0.4

0.6

0.8

1

Wealth

Shar

e of t

otal

fund

ing

Share of public funding

Share of private funding

Line of perfect equality

Poorest 20%

Poorest 40%

Poorest 60%

Poorest 80%

All households

0

0.2

0.4

0.6

0.8

1

Wealth

Shar

e of t

otal

fund

ing

Share of public funding

Share of private funding

Line of perfect equality

Poorest 20%

Poorest 40%

Poorest 60%

Poorest 80%

All households

Table 26. Equity in Public and Private Spending(Kwacha per child)

Public funding Private fundingNonfee Fee

Public (nonsalary) Salary expenditure expenditure TotalWealth quintile funding funding on child on child fundingPoorer households 14,831 3,951 1,164 8,659 28,605Richer households 23,089 4,530 4,109 19,563 51,291Source: ESDS.


of this variable, the linear regression treats thesehouseholds as if they were spending K 1.00. Thismay result in biased estimates, and to check for ro-bustness the third column presents the results fromthe Tobit specification with left censoring at zero.

The results confirm that households significantlyreduce spending on nonfee expenditures if their chil-dren go to schools that receive high public funding.Further, this difference is similar across incomegroups with a small and insignificant coefficient onthe interaction term between wealth and public

funding. To characterize the coefficients, recall thatthe asset index is a standardized (0,1) distribution,so that the roughly equal size of the coefficient onexternal funding and the asset index implies thathouseholds who send their children to a school thatmoves from low to high external funding reducespending, similar to what it would be if their wealthwas reduced by one standard deviation.

The results show household spending is an im-portant source of inequality across households.There is also evidence of substitution between

Table 27. Substitutability between Public and Private SpendingMean public Mean household

funds received expenditure Mean Average netPublic funding (rule-based) (Kwacha per child) (Kwacha per child) wealth enrollment rateLow 3,633 20,085 .005 0.892High 11,547 9,440 –.139 0.890Source: ESDS.

Table 28. Household Expenditure and Public Funding(1) (2) (3)

Linear regression: Linear regression: Totalwealth categories asset index regression

High external funding –0.860 –0.746 –0.885(1.50) (3.67)b (2.34)a

Medium wealth household 0.821 0.853(1.41) (2.43)a

Rich household 1.554 1.582(2.96)b (4.54)b

Interaction: high funding and medium wealth 0.341 0.359(0.48) (0.77)

Interaction: high funding and high wealth 0.071 0.086(0.11) (0.19)

Mean village assets 0.484 0.400 0.489(2.02) (1.61) (2.12)a

Asset index 0.731(3.32)b

Interaction: high funding and asset index 0.151(0.56)

Constant 8.129 8.929 8.101(16.29)b (47.55)b (25.54)b

Observations 436.00 436.00 436.00R-squared 0.20 0.21Robust t=statistics in parentheses clustered at school level, provincial level dummies included.a. Significant at 5 percentb. Significant at 1 percentNote: This table shows the relationship between external funding and private household expenditures using three differentspecifications: a linear regression with wealth as a categorical variable, a linear regression with wealth as a continuous assetindex and a tobit regression, left censored at 0.Source: Based on data from the ESD survey.


public funding and household expenditures, withhigher public funding reducing household expendi-ture on education, but retaining the difference be-tween rich and poor households. The net result isthat well-allocated funds could address inequalitiesacross villages. However, the impact of allocatedfunds on inequalities within villages may be smallerthan imagined due to the substitutability betweenpublic funds and household spending. Nevertheless,to the extent that such public funding will free upother resources at the household level, it is still pos-sible to argue that the overall utility of the poor willincrease relative to the rich with increases in suchfunds.

NOTES

41. These fees are for registration, sports, PTA, andthe like. Note that this may be an underestimate if headteachers felt that the survey information would be usedagainst them to enforce the government’s stated policyof zero PTA fees.

42. Note that the provinces with problems of lowenrollment are Northern and Eastern, and these prob-lems tend to be concentrated predominantly in rural ar-eas, exactly the locations that have seen almost no changefrom their already very low PTA fees.

43. This relationship is based on predictions of a lin-ear regression of PTA fees on wealth and province anddistrict-level controls.

44. An equally severe problem arises when householdshave a choice of schools. In this case, a school-choice equa-tion needs to be estimated to ensure that the relationshipbetween public and private funding is not biased by house-hold-level variables related to the choice of the school.But this would require that all schools the child could po-tentially enroll in are surveyed for the matched data.

45. For every school in the sample catchment areaswere constructed as Thiessen polygons around eachschool. Based on this catchment area, 36 schools at least5.5 km from a neighboring school were selected. Thisguaranteed that in each of the villages surveyed, house-hold-level variables could be perfectly matched withschool-level variables, since choice in schooling would beextremely limited. In other words, households were sur-veyed only in areas where schools are relatively remoteso there are no viable alternatives to the specific school

surveyed as part of the sample. For each school, onenearby village or community was selected and 15 house-holds with children were randomly chosen for the house-hold survey. The weights applied in these tables and re-gressions are individual child weights, essentially calcu-lated without applying further weights, from a child-levelfile in which household-level variables were introduced.There may be an issue of sampling: for direct comparisonwith school characteristics, each village should get equalweight irrespective of village size, but no such weightswere applied. Since the same numbers of householdswere sampled in each village, this may be such a prob-lem, but note that larger households with more childrenwill get a higher weight.

46. Nonfee items include expenditures on textbooks,uniforms, and the like.

47. Note that the other two variables of interest—the location and the distance from administrative offices—cannot be used due to lack of variation in the data. Thisfollows directly from the sampling methodology.

48. With homothetic preferences, the expenditure oneducation, E = aY after normalizing Pe = 1. An educa-tional subsidy with an interior solution will simply shiftthe budget constraint outward by the amount of the sub-sidy, S, so that the new Enew = a(Y+S). However, thehousehold expenditure is Enewhh= a(Y+S) – S = aY+S(a–1). The gradient of expenditures with income remainsunchanged, and the parameter measures the degree ofsubstitutability with the subsidy.

49. We use only rule-based funding for this exercise,since arguably this is uncorrelated to any other schoolcharacteristics besides school size, and is an amount thatcan be budgeted for by households prior to making theirspending decisions.

50. There could be other systematic differences be-tween households sending children to a high- or low-funding school. The sampling methodology reducesthese problems since there is no school choice in thesample, and further, all households are based in similarlocations. For instance, the third and fourth columns oftable 27 show that, at least for average wealth and netenrollment rates, villages with schools that receive highfunding are not significantly different from those thatreceive low funding (the difference of 0.15 standard de-viations in wealth is insignificant).

51. Since we control for village wealth our analysisrelates to within and not across school inequalities.

Discussion and Conclusions9

This report makes three contributions to un-derstanding education funding in the Zam-bian context. First, it shows that funding

characteristics are closely linked to the type of fund-ing that is disbursed. In the case of rule-based fund-ing, we find that the administrative system works ef-ficiently, and there is no evidence that such funds arediverted from their stated purpose. For discretionaryfunding, however, the majority of such funds are spentat the district and province levels, and the rest is allo-cated to less than 20 percent of all the schools.

Second, using the wealth of pupils in the schoolwe find that rule-based allocations have led to greaterper-pupil funding for poorer and more rural schools.However, these allocations are the only progressivedisbursements in the survey; staff allocations per pu-pil are higher in urban and richer schools. For discre-tionary allocations we find evidence of higher disburse-ments to richer schools within rural areas and wealth-neutral allocations within urban areas. Once all sourcesof public funding are factored into the analysis, publicschool funding in Zambia is regressive, with almost30 percent higher allocations to richer schools.

Third, the report shows how private expendi-tures at the level of the household impacts equity ineducational funding. It argues that nonfee expendi-tures incurred by households, rather than contribu-tions to school funds through PTA and other fees,are the major source of inequalities in the currentenvironment. The report also presents evidence thathouseholds decrease private contributions whenpublic funds to the schools increase.

These findings suggest that rule-based allocationsensure that schools will receive a larger share of funds(either in cash or kind) in the system. It was initiallythought that the process of decentralization would par-tially fulfill this need. Since more money would flow di-rectly to districts, accountability and therefore disburse-ments would be higher. Unfortunately the evidence doesnot support this assumption. Decentralization seems tohave only shifted spending from the province to the dis-trict, and, in terms of funding equity, it is precisely at thedistrict level that richer schools are now receiving higherdiscretionary funds than their poorer counterparts.

Even if rule-based funding were to increase, two sub-sidiary implications would need to be carefully evalu-ated. First, the current rule-based allocation has the un-fortunate implication that schools would fare better interms of per-pupil funding if they could decrease enroll-ments. The more common school-funding rule is basedon transfers that increase the number of enrolled chil-dren. Unfortunately, there is no guarantee this wouldwork as well as the current, unambiguous rule. One sug-gestion would be to continue with the current rule, whichhas the desired equity implications), but monitor enroll-ment carefully through the regular data collected underthe school census.

Second, an increase in public funds to schools crowdsout private spending. Although the results presented hereare preliminary, there is evidence that this crowding-outcan be fairly large. Thus, public funds may be far moreeffective at addressing inequalities across rather thanwithin villages. This would suggest targeting at the levelof schools with greater funding to poorer regions.

59

Bibliography

The word “processed” describes informally repro-duced works that may not be commonly availablethrough library systems.

Ablo, Emmanuel, and Ritva Reinikka. 2000. “Do Bud-gets Really Matter: Evidence from PublicSpending on Education and Health inUganda.” Policy Research Working Paper1926. World Bank, Washington, D.C.

Filmer, Deon, and Lant Pritchett. 1999. “The Effectof Household Wealth on EducationalAttainment: Evidence from 35 Countries.”Population and Development Review 25(1):85–120.

Government of the Republic of Zambia. 1997.“Decentralization of the Education Systemin Zambia: Institutional Roles, Functions, andRelationships.” Ministry of Education,Lusaka.

Government of the Republic of Zambia. 2001. “BasicEducation Sub-Sector Investment Program,2000: Program Performance Indicators.”Ministry of Education, Lusaka.

Hanushek, Eric. 1986. “The Economics of Schooling:Production and Efficiency in Public Schools.”Journal of Economic Literature 24 (3):1141–77.

Hanushek, Eric, and Javier Luque. 2001. “Efficiencyand Equity in Schools Around the World.”Processed.

International Consultative Forum on Education forAll. 2000. “Education for All: Status and

Trends 2000. Assessing LearningAchievement.” UNESCO, Paris.

Jensen, Peter, and H. S. Nielsen. 1997. “Child Labouror School Attendance? Evidence from Zam-bia.” Journal of Population Economics 10: 407–24.

Jones, Bruce N. 2000. “Zambia: Public ExpenditureReview Education Sector.” World Bank,Africa Region Human Development,Washington, D.C. Processed.

Kelly, Michael J., ed. 1999. The Origins andDevelopment of Education in Zambia: FromPre-Colonial Times to 1996. Lusaka: ImagePublishers Limited.

Kelly, Michael J., and Joe Kanyika. 2000. “LearningAchievement at the Middle Basic Level.” TheExamination Council of Zambia on behalf ofBESSIP. Ministry of Education, Lusaka.

Oxfam-Zambia and Jesuit Center for TheologicalReflection. 2001. “Will the Poor Go toSchool?” Jesuit Center for TheologicalReflection, Lusaka.

Oxford Policy Management. 2002. “Unit Cost Studyof Education in Zambia.” Oxford PolicyManagement, Oxford. Processed.

Reinikka, Ritva, and Jakob Svensson. 2002.“Explaining Leakage of Public Funds.”Discussion Paper No. 3227. Center forEconomic Policy Research, London.

Siamatowe, Clement. 2002. “Teacher Deploymentand Compensation.” World Bank, CountryOffice, Lusaka. Processed.

60


Todd, Petra, and Kenneth Wolpin. 2003. “On theSpecification and Estimation of the ProductionFunction for Cognitive Achievement.”Economic Journal 113(485): F3–33.

UNESCO. Various years. “Edstats.” [Retrieved onSeptember 2, 2003, from http://devdata.worldbank.org/edstats/about_data.asp.]

World Bank. 1998. “Living Conditions MonitoringSurvey II.” World Bank, Washington, D.C.

[Retrieved on September 2, 2003, fromhttp://www4.worldbank.org/afr/poverty/databank/survnav/show_survey.cfm?ID=153.]

World Bank. 2001. “Zambia Public Expenditure Re-view: Public Expenditure, Growth and Pov-erty—A Synthesis.” World Bank, Washing-ton, D.C.

World Bank. Various years. World Development Indi-cators. Washington, D.C.

Appendix

This appendix describes the asset indexes createdfor the ESDS exercise using the pupil questionnairefor 2001.

ITEM RESPONSE METHOD

Item Response Theory (IRT) methods are usedto generate the asset indexes. The findings showthat the IRT asset index performs more or less simi-larly to the principal components methods (com-parisons based on Monte Carlo simulations). Thecorrelation between the IRT index and the princi-pal components method was more than 0.98 in allthe countries looked at. The main advantage of us-ing IRT methods is that, in addition to providingthe index, it also provides standard errors of theindex at each wealth level. This is useful for evalu-ating the accuracy of the index at different levels ofwealth.

The IRT method assumes a structural relation-ship (the logistic function) that defines the probabil-ity of ownership (in this case) with wealth. The curveis defined by three parameters:

• Difficulty. The level of the curve—an asset ismore difficult to own if the probability of own-ership is 0.5 at a higher level of wealth.

• Discrimination. The slope of the curve—an as-set has higher discrimination if the slope issteeper, measured at the level of difficulty ofthe asset. Thus, an asset with greater discrimi-nation makes it easier to distinguish betweenthe two groups on either side of the difficultyparameter.

• Guessing parameter. The lower asymptote ofthe logistic curve—the probability of owner-ship for the lowest wealth level in this case.

The main assumption in IRT is that difficulty andwealth share the same dimensionality, an “x” pointincrease in difficulty is identical to an “x” point in-crease in the value of the index.

INDEX COMPONENTS

Unfortunately, the pupil questionnaire has veryfew assets for constructing the index (Table A1). The

62

Table A1. Variables Used in Construction of theAsset IndexAsset number(for graphs) Description of asset Used foritem1 Is house made of brick? All sampleitem2 Is their electricity? All sampleitem3 Does hh own:TV All sampleitem4 Does hh own: Radio-Cassette All sampleitem5 Does hh own: Radio All sampleitem6 Does hh own: Video All sampleitem7 Does hh own: Sewing Machine All sampleitem8 Does hh own: Stove/Cooker All sampleitem9 Does hh own: Fridge/Freezer All sampleitem10 Does hh use: Plough Rural onlyitem11 Does hh use: Crop sprayer Rural onlyitem12 Does hh use: Hammer mill Rural onlyitem13 Does hh use: Hand-grinding mill Rural onlyitem14 Does hh use: Tractor Rural onlyitem15 Does hh own: Cattle Rural onlyitem16 Does hh own: Goats/Sheep Rural only


wealth index created for the entire sample uses nineassets, and the rural wealth index uses 16 compo-nents. A potential seventeenth component (doeshousehold own donkeys?) had to be discarded, sinceonly 29 out of 3,656 reported yes. The reason forcreating a separate index for rural and all-sample is,of course, based on the fact that nonownership ofthe last seven assets reveals no information aboutwealth in an urban area. Potential questions regard-ing the regularity of meals had absolutely no discrimi-nation across wealth levels, and was also discarded.

RESULTS

The all-sample wealth index seems to be reason-ably accurate in the range of the sample. Copperbeltcomes out as the richest province, followed by Lusaka,Northern, and Eastern provinces. Further, the indexwas much higher for children going to private schools,and correlated fairly well with 2001 test scores. Therural index, however, does much worse in terms ofmeasurement error, since the statistical precision ofthe index is additive with respect to the number ofassets, and we use a super set. This difference arisesentirely from differences in the sample, and the factthat we are now trying to predict asset informationfor a much tighter distribution (compared to that ofthe all-sample). The goal is to divide the poor in theall-sample into further subdivisions of poor, middle,and rich, and the information is not good enoughto make the distinctions finer. The graphs for thedifferent samples are presented below, where the

vertical lines show the maximum and minimum rangeof the wealth index in each sample.

The standard errors are much worse in the caseof the rural sample, arising primarily from the factthat we are trying to predict wealth for a muchsmaller range of wealth values than the all-sample.This does not imply that the estimation of the wealthindex is worse; if we construct the index using onlythe limited asset set, the prediction worsens. Pre-dictions for the all-sample and rural samples in therange of the data set seem reasonably precise (seeFigures A1 and A2).

Figure A3 (panel a and b) show the distributionof wealth in the all-sample. The left part of the dis-tribution is driven almost entirely by rural house-holds, and the right part is driven by urban house-holds (see bottom panel). This is the main problemwith this asset index. Since the weights of the assetsare computed from the all-sample, the estimates areessentially a reflection of the variation across the ru-ral and urban samples rather than within them. As aresult, both within urban and rural samples the pre-cision of the index falls dramatically. Figure A4 showshow the index computed from the all-sample com-pares with the index computed from the rural samplefor rural households only. The correlation betweenthe two indexes is 0.82, and the graph shows thatthe lower correlation arises due to wealth index val-ues in the lower part of the distribution. The axesdo not match up since the index is standardized overdifferent populations.

Figure A2. Asset Index Characteristics for Rural Sample

.009035.0090353.99

Sample range: minimum Sample range: maximum

Stan

dard

erro

r

Inde

x inf

orm

ation

Index Standard Error Index Information

-3.99 4.106006

10.5205 88.9898

- 4.106006

10.5205 88.9898

Wealth index characteristics

.009035.0090353.99

Sample range: minimum Sample range: maximum

Stan

dard

erro

r

Inde

x inf

orm

ation

Index Standard Error Index Information

-3.99 4.106006

10.5205 88.9898

- 4.106006

10.5205 88.9898

Wealth index characteristics

Figure A1. Asset Index Characteristics for All-SampleIndex InformationIndex Standard Error

Sample range: minimum Sample range: maximum-3.99 4

.256171 .029421-3.99 4

.256171 .029421

Index Standard Error Index Information5.83006 15.23845.83006 15.2384

Stan

dard

erro

r

Wealth index characteristicsIn

dex i

nfor

mati

on

Index InformationIndex Standard Error

Sample range: minimum Sample range: maximum-3.99 4

.256171 .029421-3.99 4

.256171 .029421

Index Standard Error Index Information5.83006 15.23845.83006 15.2384

Stan

dard

erro

r

Wealth index characteristicsIn

dex i

nfor

mati

on


Figure A3. Distribution of Wealth in All-Sample and Separated by LocationDe

nsity

Kernel density estimate

Wealth index: ML estimate-1.36055 2.08283

.073001

.684049

Dens

ity


Wealth index: ML estimate-1.36055 2.08283

.073001

.684049

b.

Wealth index: ML estimate

Wealth index for urban household

Wealth index for rural household

-1.5 -1 -.5 0 .5 1 1.5 20

1.32242

Wealth index: ML estimate

Wealth index for urban household

Wealth index for rural household

-1.5 -1 -.5 0 .5 1 1.5 20

1.32242

a.


Figure A4. Comparison of All Sample and Rural Index

Inde

x: all

sam

ple

Index: rural sample-.65 -.15 .35 .85 1.35 1.85 2.35

-1.19223

1.91451

Inde

x: all

sam

ple

Index: rural sample-.65 -.15 .35 .85 1.35 1.85 2.35

-1.19223

1.91451

Business

Public and Private Funding of Basic Education in Zambia