Chapter

APPENDIX 2. Expenditure Efficiency—An Empirical Assessment

Author(s):
Kevin Fletcher, Sanjeev Gupta, Duncan Last, Gerd Schwartz, Shamsuddin Tareq, Richard Allen, and Isabell Adenauer
Published Date:
April 2008
Share
  • ShareShare
Show Summary Details

It will be essential for low-income countries to make efficient use of scaled-up aid to ensure sustained progress toward the Millennium Development Goals (MDGs). In particular, efficient spending in priority areas—for example, health care, education, public investment—will be critical. This appendix describes how a sample of Poverty Reduction and Growth Facility (PRGF)-eligible countries have fared in transforming inputs into outcomes in health and education; this may help to shed light on current absorptive capacities and the likely efficiency of using scaled-up aid (Table A2.1). In addition, this appendix attempts to identify factors that may help explain differences in expenditure efficiency across countries.

Table A2.1.Countries Included in the Efficiency Analysis1
1Angola
2Bangladesh
3Benin
4Bolivia
5Burkina Faso
6Burundi
7Cambodia
8Cameroon
9Central African Rep.
10Chad
11Congo, Dem. Rep. of
12Congo, Rep. of
13Côte d’Ivoire
14Djibouti
15Eritrea
16Ethiopia
17Gambia, The
18Ghana
19Guinea
20Guinea-Bissau
21Guyana
22Haiti
23Honduras
24India
25Kenya
26Lao People’s Dem. Rep.
27Lesotho
28Madagascar
29Malawi
30Mali
31Mauritania
32Mozambique
33Nepal
34Nicaragua
35Niger
36Nigeria
37Pakistan
38Papua New Guinea
39Rwanda
40Senegal
41Sierra Leone
42Sri Lanka
43Sudan
44Tanzania
45Togo
46Uganda
47Vietnam
48Yemen, Rep. of
49Zambia
50Zimbabwe

This list includes countries that were PRGF-eligible in September 2006, excluding some small island economies and transition countries. The list also excludes PRGF-eligible countries without available data on health and education spending. Countries with missing information on outcome measures were dropped from the analysis of that outcome measure.

This list includes countries that were PRGF-eligible in September 2006, excluding some small island economies and transition countries. The list also excludes PRGF-eligible countries without available data on health and education spending. Countries with missing information on outcome measures were dropped from the analysis of that outcome measure.

Expenditure Efficiency in Health and Education in Low-Income Countries—The Scoreboard

Expenditure efficiency is assessed here by measuring how effective countries are in producing health and education outcomes. An implicit assumption is that spending affects outcomes and that a relatively more efficient country achieves the same outcome with lower spending. The analysis is done in two stages. In the first stage, the spending efficiency for each country in the sample is measured using Data Envelopment Analysis (DEA).57 This methodology estimates the overall spending efficiency of the use of inputs (for example, health expenditure) to “produce” outputs (for example, health outcomes). The countries that provide the best combination (that is, the maximum outputs for a given level of inputs or, alternatively, the minimum inputs for the level of outputs) define the best-practice frontier. The countries that are not on the frontier are then ranked according to the distance from the frontier, which is a measure of relative efficiency expressed as the efficiency score. The second stage attempts to identify key factors that account for differences in the relative efficiency scores, using correlation coefficients and multivariate truncated regressions that relate relative efficiency scores to various control variables. The inputs used in the analysis are per capita health and education spending58 in purchasing power parity dollars, and the outcomes are indicators that are used to monitor progress toward the MDGs. Table A2.2 shows the different indicators used, their definitions, the corresponding MDGs, and the availability of data. Because of a relative paucity of data that directly measure outcomes, the analysis also uses intermediate indicators of outcomes, such as enrollment rates and the numbers of births attended by skilled staff.

Table A2.2.Spending and Outcome Indicators for the Efficiency Analysis
Spending Indicators1IndicatorsSpending Years (period averages)Indicator YearsNumber of CountriesType of IndicatorMDG2
Education
Public education expenditure3 (in PPP dollars)Literacy rate, youth (percent of people ages 15–24)1999-20024200439OutcomeMDG2
Primary school enrollment (percent net)1999-20024Avg.2000-04439OutputMDG 2
Persistence to grade 5 (percent of cohort)1999-20024Avg. 2000-03433OutcomeMDG 2
Ratio of girls to boys in primary and secondary (percent)1999-20024Avg. 2002-04454OutputMDG 3
Ratio of literate females to males (percent ages 15–24)1999-20024200436OutcomeMDG 3
Trained teachers in primary education (percent of total)1999-20024Avg. 2001-04432Outputn/a
Health
Public health expenditure (in PPP dollars)Immunization, measles (percent of children 12–23 mos.)1998-2001200478OutputMDG 4
Private health expenditure5 (in PPP dollars)Mortality rate, infant (per 1,000 live births)1998-2001200478OutcomeMDG 4
Total health expenditure5 (in PPP dollars)Mortality rate, under-5 (per 1,000)1998-2001200478OutcomeMDG 4
Maternal mortality ratio (per 100,000 live births)1998-20012000661OutcomeMDG 5
Births attended by skilled health staff (percent of total)1998-2001Latest 2000-0457OutputMDG 5
Health worker density (per 1,000 people)1998-2001Latest 2000-0339Outputn/a

Spending variables have been corrected for differences in purchasing power. PPP = purchasing power parity.

MDG 2 is “Achieve universal primary education,” MDG 3 is “Promote gender equality and empower women,” MDG 4 is “Reduce child mo rtality,” and MDG 5 is “Improve maternal health.”

Data on private spending on education are not available.

Years with available data.

Available for broadly the same number of countries and the same years.

World Bank model estimate.

Spending variables have been corrected for differences in purchasing power. PPP = purchasing power parity.

MDG 2 is “Achieve universal primary education,” MDG 3 is “Promote gender equality and empower women,” MDG 4 is “Reduce child mo rtality,” and MDG 5 is “Improve maternal health.”

Data on private spending on education are not available.

Years with available data.

Available for broadly the same number of countries and the same years.

World Bank model estimate.

First-stage results

First-stage results point to large variances in spending efficiency, implying that higher levels of spending do not always translate into better outcomes. More specifically—

  • Countries with the lowest per capita incomes tend to have the lowest efficiency scores for health (Table A2.3). This general conclusion holds broadly, irrespective of the outcome indicator used or whether total health spending or only public health spending is considered.59 Three outcome indicators—infant mortality, child mortality, and maternal mortality—were used in the analysis. Overall, only about 40 percent of the countries in the poorest half of the sample ranked in the top half with respect to their outcome efficiency scores.60

  • Relative efficiency analysis for education spending yields comparable results (Table A2.4). The two indicators used for this exercise are the primary enrollment rate and youth literacy rate. Only one-third of the poorest countries in the sample are ranked in the top half of efficient countries on the basis of the primary enrollment rate. The results are more favorable for youth literacy rates, with as many as 64 percent of the poorest countries in the top half based on efficiency scores.

Table A2.3.Percent of Countries in Top Half of the Efficiency Distributions for Health by Income Level1
Annual GDP Per Capita2Health Outcome Indicators
Infant mortalityChild mortalityMaternal mortality
≤ 1,45740.040.037.5
> 1,45760.060.062.5
Sources: World Bank, World Development Indicators database; and IMF staff estimates.

Countries in the first two quartiles of the efficiency distribution have better efficiency scores than the median country.

In PPP U.S. dollars. The median per capita income in the sample is $1,457 in PPP terms.

Sources: World Bank, World Development Indicators database; and IMF staff estimates.

Countries in the first two quartiles of the efficiency distribution have better efficiency scores than the median country.

In PPP U.S. dollars. The median per capita income in the sample is $1,457 in PPP terms.

Table A2.4.Percent of Countries in Top Half of the Efficiency Distribution for Education by Income Level1
Annual GDP Per Capita2Education Outcome Indicators
Primary school enrollmentYouth literacy
≤ 1,45733.364.3
> 1,45766.735.7
Sources: World Bank, World Development Indicators database; and IMF staff estimates.

Countries in the first two quartiles of the efficiency distribution have better efficiency scores than the median country.

In PPP U.S. dollars. The median per capita income in the sample is $1,457 in PPP terms.

Sources: World Bank, World Development Indicators database; and IMF staff estimates.

Countries in the first two quartiles of the efficiency distribution have better efficiency scores than the median country.

In PPP U.S. dollars. The median per capita income in the sample is $1,457 in PPP terms.

Second-stage results

The second-stage analysis is limited, because of data constraints, to efficiency of health spending. The control variables used in this stage are listed in Table A2.5. Second-stage results point to several factors that may help to explain differences in spending efficiency in the sample:

  • Governance and the quality of fiscal institutions have a strong positive correlation with efficiency in health. Several indicators of governance and institutions were used in the analysis, such as the International Country Performance Rating (ICPR), the average Country Policy and Institutional Assessment (CPIA) score, and some of the latter’s components for the quality of fiscal institutions. Table A2.6 presents those correlation coefficients between the relative efficiency scores and the control variables that are robust.61 Thus, on average, countries with better governance and fiscal institutions achieve higher health outcomes at lower levels of spending.62

  • The level of aid and aid volatility are not correlated with health efficiency scores.63 This is because aid volatility does not translate into similar changes in health spending (see Appendix 1) or health outcomes. In other words, short-term changes in aid do not affect the relationship between spending and outcomes in the health sector, and therefore do not affect spending efficiency.

  • The efficiency of health sector spending is correlated positively with outcomes in the education sector and with infrastructure, and negatively with the prevalence of HIV/AIDS. Higher adult literacy rates and improved access to sanitation are associated with higher efficiency of health spending (Table A1.6). These results reflect the importance of adequate sanitation infrastructure to health outcomes and the well-known fact that better education and health outcomes reinforce each other: better education leads to better decisions on health-related matters, and improved child health promotes investment in education (Miguel and Kremer, 2004). The prevalence of HIV/AIDS in a country lowers the relative expenditure efficiency in the health sector.

Table A2.5.Control Variables
GroupFactors
Income and human developmentGDP (in PPP dollars per capita)
Prevalence of HIV
Adult literacy rate
Infant mortality rate
ConflictCountry at war anytime between 1995 and 2005
Military expenditure (percent of GDP)
InfrastructureImproved sanitation facilities access (percent of population)
Improved water source access (percent of population)
Urban population (percent of total)
Level and volatility of aid1Total ODA received
Technical cooperation aid
Development food aid
Emergency aid
Other aid
Total loans (net)
Grants
Governance and fiscal institutionsICPR: Governance rating
CPIA 12: Property Rights and Rule-Based Governance
CPIA 13: Quality of Budget and Financial Management
CPIA 15: Quality of Public Administration
CPIA 16: Transparency, Accountability, and Corruption
Control
CPIA 12–16: Average
Note: CPIA = Country Policy and Institutional Assessment; ODA = official development assistance; ICPR = International Development Agency Country Performance Rating; PPP = purchasing power parity.

The level of aid received is measured as aid received as a percent of GDP for each type of aid. Volatility of aid received is measured in three ways: standard deviation of aid, coefficient of variation of aid, and the relative variance of aid to revenue.

Note: CPIA = Country Policy and Institutional Assessment; ODA = official development assistance; ICPR = International Development Agency Country Performance Rating; PPP = purchasing power parity.

The level of aid received is measured as aid received as a percent of GDP for each type of aid. Volatility of aid received is measured in three ways: standard deviation of aid, coefficient of variation of aid, and the relative variance of aid to revenue.

Table A2.6.Correlation Matrix of Relative Efficiency Scores and Control Variables1
Relative Efficiency Scores
Immunization, measlesInfant mortalityChild mortalityMaternal mortalityBirths attended by skilled health staff
Control VariablesCoefficientNCoefficientNCoefficientNCoefficientNCoefficientN
Prevalence of HIV0.30**440.29*440.39**43
Adult literacy rate–0.35**36–0.43**36–0.28*36–0.31*35
Improved sanitation facilities–0.25*50–0.26*50–0.43**46
ICPR: Governance rating–0.40**50–0.30**50–0.26*50
CPIA 12: Property Rights and Rule-Based Governance–0.33**50–0.30**50–0.29*50
CPIA 15: Quality of Public Administration–0.40**50–0.33**50–0.31**50
CPIA 16: Transparency, Accountability, and Corruption Control–0.29**50–0.24*50–0.27*46
CPIA fiscal indicators average
–0.40**50–0.31**50–0.25*50
Sources: World Bank; and IMF staff estimates.

A negative sign means that more of the control variable is negatively correlated with the efficiency score and hence positively correlated with level of efficiency. * and ** denote significance at the 10 and 5 percent levels, respectively.

Sources: World Bank; and IMF staff estimates.

A negative sign means that more of the control variable is negatively correlated with the efficiency score and hence positively correlated with level of efficiency. * and ** denote significance at the 10 and 5 percent levels, respectively.

Multivariate truncated regressions confirm these findings.64 Efficiency scores for infant mortality were regressed on the prevalence of HIV, adult literacy rate, access to sanitation services, and the average CPIA scores for fiscal institutions. The coefficients for all variables are significant and of the expected sign (Table A2.7). It is worth noting that the coefficient for the CPIA indicator is significant in each of the three alternative specifications.

Table A2.7.Truncated Regressions of Expenditure Efficiency Scores1
Specification 1Specification 2Specification 3
Prevalence of HIV0.008*0.002**0.001*
(0.108)(0.022)(0.051)
CPIA fiscal indicators average–0.018**–0.018**–0.012*
(0.026)(0.027)(0.096)
Adult literacy rate–0.001**
(0.014)
Access to improved sanitation facilities–0.001**
(0.020)
Constant1.1061.1391.109
(0.000)(0.000)(0.000)
Sigma20.0250.0240.023
N403240
Sources: World Bank; and IMF staff estimates.

A negative sign means that more of the control variable is negatively correlated with the efficiency score and hence positively correlated with level of efficiency. * and ** denote significance at the 10 and 5 percent levels, respectively.

Sigma is the standard error of the regression.

Sources: World Bank; and IMF staff estimates.

A negative sign means that more of the control variable is negatively correlated with the efficiency score and hence positively correlated with level of efficiency. * and ** denote significance at the 10 and 5 percent levels, respectively.

Sigma is the standard error of the regression.

This analysis of the efficiency of education and health spending should be interpreted with some caution. Health and education outcomes are influenced by a host of factors beyond spending that can only be partially captured by the use of controls in the second stage of the analysis. In addition, the methodology focuses on quantifiable inputs and outcomes, and only partially takes into account harder-to-measure factors such as quality. Finally, efficiency is measured in relative terms, implying that if a country is on the frontier, it is relatively more efficient than other countries in the sample. In relatively small samples, such as for this analysis, this may result in some bias in the result. Nevertheless, the thrust of the findings presented here is consistent with those reported in the literature on expenditure efficiency.65

Fiscal Policy Implications for Scaled-Up Aid

The above analysis points to two important implications for expenditure policy in the context of scaled-up aid: (1) improving efficiency of spending in low-income countries is critical to achieving the MDGs; and (2) in most low-income countries, effective utilization of scaled-up aid will require a further strengthening of fiscal institutions. These reforms will contribute to enhancing the efficiency of spending in low-income countries.

The DEA methodology derives from the literature on the estimation of production functions (for a detailed exposition of DEA and other methods of assessing efficiency, see Zhu, 2003). DEA has the advantage of being sparse in its assumptions about the characteristics of the production technology. This is particularly important for assessing spending efficiency, because little is known about the nature of the relationship between spending and outcomes.

Health spending includes both public and private spending. However, education spending data relate to public spending only, because private spending data are not available.

Taking into account both private and public health spending is important for assessing health outcomes. Some countries with low public health spending have relatively better health outcomes. It would be reasonable to assume that this could reflect higher private spending; however, no direct relationship was found between the share of private spending in health and relative efficiency scores.

The first-stage efficiency scores computed with the three health outcome indicators are strongly correlated, indicating that the results are robust.

A control variable is considered correlated with the health efficiency scores when the correlation coefficient of that variable is statistically significant at the 10 percent level or higher and with the expected sign. To be considered robustly correlated, the relationship has to hold for at least three out of five efficiency score indicators.

The sample size does not allow computing correlation coefficients for education with sufficient confidence.

However, Herrera and Pang (2005) find that countries with high ratios of aid to fiscal revenues tend to score lower on efficiency.

The number of control variables that could be included was limited by the number of observations available.

For example, Baldacci and others (forthcoming) find that increased public spending has a lower effect on outcomes when the quality of spending and the governance and institutional arrangements are weak. A paper by the IMF’s Policy Development and Review Department (IMF and World Bank, 2004) states that improved country policies, institutions, and public expenditure management in low-income countries are important for aid to be more effective. Estache, Gonzalez, and Trujillo (2007) find that low-income countries have lower expenditure efficiency in achieving health and education outcomes than do lower-middle-income, upper-middle-income, and high-income countries. Finally, Gupta and Verhoeven (2001) find that efficiency of education spending is lower in African countries than in Asian and Western Hemisphere countries.

    Other Resources Citing This Publication