Abstract

The legacy of the global financial crisis includes many important lessons for developing and low-income countries. The large fiscal buffers accumulated before the crisis opened up space to implement countercyclical policies that limited the negative impact on growth and employment (IMF 2010 and Celasun and others 2015). However, the easing was not followed by a timely withdrawal. As a result, buffers were depleted and in some cases debt reached unsustainable highs.

Introduction

The legacy of the global financial crisis includes many important lessons for developing and low-income countries. The large fiscal buffers accumulated before the crisis opened up space to implement countercyclical policies that limited the negative impact on growth and employment (IMF 2010 and Celasun and others 2015). However, the easing was not followed by a timely withdrawal. As a result, buffers were depleted and in some cases debt reached unsustainable highs.

Restoring the fiscal buffers required for protection from a new shock is challenging for these countries. This is especially so given high uncertainty about the global outlook, constraints on spending due to the servicing of debt, and long-term demographic pressures highlighted in the previous chapter. Although it is difficult to assess the exact size of buffers required to shield against contingencies, the experience of the global financial crisis highlights the value of building ample buffers.

The need to rebuild buffers runs parallel with the need to increase public spending to close gaps in the provision of social services. Improving health and education outcomes is critical to achieve this objective and boost human capital. Abundant literature on endogenous growth shows a close connection between accumulation of human capital and growth (Barro and Sala-i-Martin 2004). A healthy and educated population is more productive, which in turn raises income per capita.

How is the puzzle of restoring buffers and simultaneously increasing social spending solved? In short, by increasing the efficiency of public spending. Higher efficiency in the use of public resources allows buffers to be restored and essential public services to be increased. The focus in this chapter is on gauging the efficiency of social spending on health and education.

Calculated efficiency measures can be used to estimate potential expenditure savings. Due to the size of social spending, these savings are significant. On average, public spending in developing and low-income countries is about 23 percent of GDP, with the bulk allotted to health and education. Because of this, even small improvements in the efficiency of public spending provide sizable resources for use in other sectors with greater value for money. Tackling inefficiencies in public spending could help governments increase their delivery of public services while simultaneously keeping fiscal balances under control.

The chapter starts by describing the main trends in health and education spending in Central American countries and comparing how efficiently inputs translate into outputs using a benchmarking methodology detailed in Annex 7.1. Policy recommendations are proposed to improve the efficiency of social spending in several of the region’s countries.

Social Spending Trends

The empirical analysis starts with a review of the main trends in countries discussed in this chapter. As Figure 7.1 shows, social spending has been increasing slowly in Central America. Increases in health and education expenditure since 2000 are in both cases around 1 percentage point of GDP, from an average of around 4 to 5 percent of GDP. This trend is mirrored in Latin America as a whole, where public spending has also generally increased by 1 percentage point of GDP.

Figure 7.1.
Figure 7.1.

Social Spending Trends

Sources: The World Bank, WHO, and UNICEF.

Social spending is consistently higher in Costa Rica than in Central American peers, while this is the case for Honduras only in education spending. However, spending in Honduras is coming down, while in Costa Rica it seems to be on an increasing trend. On the other hand, social spending in Guatemala, in both health and education, is largely below the regional average.

Figure 7.2 shows some interesting comparisons, which are matched in the results of the benchmarking exercise. Honduras, Guatemala, and El Salvador (and Latin America in general) have large enrollment gaps in secondary education. This is consistent with the current demography of the region: a relatively young population, large dropout rates, and low average years of schooling. Population aging, while limited so far, is expected to accelerate over the coming decades creating the need to divert resources from lower to higher levels of education and achieving more efficiency of spending at the same time (Chapter 8). For example, pupil-teacher ratios in Honduras are very low relative to its peers. Guatemala also lags its peers in pupil-teacher ratios in secondary education. This points to a currently overfunded educational system and the need to rationalize resources with a focus on improving outcomes in secondary and tertiary education.

Figure 7.2.
Figure 7.2.

Selected Educational Input and Output Comparisons

Sources: World Bank; World Health Organization; United Nations Educational, Scientific, and Cultural Organization; and Barro-Lee database.Note: HALE = health-adjusted life expectancy.

As can be seen in Figure 7.2, public spending in education is highly concentrated toward primary grades, and more so in El Salvador, Guatemala, and Honduras than in other regional peers. If spending is raised to account for population aging, it should focus on higher, rather than primary, education.

Empirical Results

As explained in the technical analysis in Annex 7.1, both available methodologies for empirical analysis assume an underlying technological process to create certain social outcomes based on certain inputs (among them public spending). This framework requires the assumption of some degree of homogeneity across countries. The benchmarking literature identifies a shortcoming regarding this assumption; factor prices tend to be higher in wealthier than in medium or low-income countries.1 Higher prices usually imply higher spending; therefore, rich countries can be deemed inefficient only inasmuch as this effect applies. Since previous researchers have encountered a similar problem, in the current sample, all industrialized economies are excluded so that developing and low-income countries do not appear relatively more efficient only because of the higher factor prices paid in the industrialized world (see Gupta and Verhoeven 2001, or Herrera and Pang 2005).

Our input-output model specification is consistent with Herrera and Pang (2005), or Grigoli and Kapsoli (2013) in that it uses health-adjusted life expectancy (HALE) as output and public spending, private spending, and the educational level of adults as inputs.2 All spending variables are expressed in 2011 PPP US dollars. The educational attainment of adults is measured by the average years of schooling for population older than 15 years. For education, separate estimates are prepared for primary and secondary education. Net enrollment rates as output and public spending and the teacher-pupil ratio are used as inputs.3 In the case of education, a common critique is that enrollment rates do not adequately measure educational achievements. This is true, but unfortunately the standardized tests (PISA, TIMMS, and PIRLS) commonly used to measure achievement have limited country coverage, particularly for low-income countries.

Based on the input-output combinations for health and education spending, the efficiency scores and their corresponding confidence intervals are estimated using the bootstrapped data envelopment analysis (DEA) approach described in Annex 7.1. Specifically, efficiency scores are estimated for health and primary and secondary education spending using 2000 replications based on a sample of emerging and low-income countries. All variables are averages starting from 2000 until the last available observation. Figure 7.3 shows scatter plots for selected input-output combinations. To make viewing easier, labels are shown only for Latin American countries. As can be seen, all bijections seem positive.

Figure 7.3.
Figure 7.3.

Selected Inputs and Outputs for Social Spending

Sources: World Bank; World Health Organization; United Nations Educational, Scientific, and Cultural Organization; and Barro-Lee database.

Figure 7.4 shows the main results of the chapter, that is the estimated efficiency scores for Central American countries. Detailed efficiency scores and confidence bands for all countries in the sample are available in Annex 7.2.

Figure 7.4.
Figure 7.4.

Benchmarking Social Spending Results

Source: Authors’ estimations.

Honduras. The output-oriented (OO) score for health in Honduras is 0.955, showing limited room for getting better outcomes with inputs efficiently; however, the input-oriented (IO) score is 0.840 meaning that all inputs could be reduced by around 15 percent without a marked impact on output. In education, Honduras performed poorly in secondary education, ranking last among 88 countries in the IO score (0.208) and 66 out of 88 in the OO measure (0.522). The score is better for primary education, reaching 0.307 in the IO and 0.948 in the OO measures. These results imply potential efficiency savings in educational inputs between 70–80 percent. On the potential efficiency gains, they seem only significant in secondary education (around 50 percent).4

Guatemala. The OO score for health in Guatemala is 0.976, which also shows limited room for getting better outcomes with inputs efficiently, however, the IO score is 0.910 meaning that all inputs could be reduced by around 10 percent without a reduction in the output. In education, Guatemala performed poorly in secondary education, ranking 69 out of 88 countries with an input-oriented score of 0.403 and 65 out of 88 in the output-oriented measure (0.534). The result is better in primary education in the OO measure scoring 0.932, while it is worse for the input-oriented case (0.328). These results imply potential efficiency savings in educational inputs of between 60–70 percent. On the potential efficiency gains, they seem only significant in secondary education by around 50 percent.

El Salvador. The OO score for health is 0.918 in El Salvador, pointing to a margin of around 8 percent to increase outcomes by using inputs efficiently. The estimated IO score of 0.679 shows that significant room exists for efficiency savings, where all inputs could be reduced by around 33 percent without a change in the output. El Salvador performed better in secondary education, achieving higher IO and OO scores (0.681 and 0.733), but there are still efficiency savings and gains to be realized. The results are not as good in primary education, with scores of 0.419 for the IO measures but better for the OO measures, with scores of 0.921. These results imply potential efficiency savings in educational inputs of between 30–60 percent. Potential efficiency gains seem only significant in secondary education (at about 30 percent).

Costa Rica. Health and primary education show high OO efficiency scores, implying little room for efficiency gains. However, secondary education shows a margin for efficiency gains of around 25 percent. In IO scores, Costa Rica shows a better prospect for input savings particularly in primary education, where it ranked last in the Central American region.

Nicaragua. High OO efficiency is recorded in health and primary education but there is a lot of room for efficiency gains in secondary education (around 30 percent). Analyzing IO, Nicaragua shows large room for input savings in education.

In sum, the efficiency metrics show limited room to increase health outcomes without an increase in inputs; however, there is an opportunity for savings, particularly in El Salvador. In primary education, there is no space for sizable improvement in outcomes but all countries show ample room for savings. In secondary education, there is room for sizable efficiency improvements to achieve better outputs and to save on inputs.

For an idea of the policy implication of these estimations, potential long-term savings can be calculated for the scenario where all inefficient spending is removed.5 If Honduras and Costa Rica could remove all inefficiencies in education spending, they could attain savings worth around 3 percent of GDP. Savings in education in El Salvador and Guatemala would amount to 1¼ percent of GDP. In the health sector, savings from using inputs efficiently as a proportion of GDP would amount to 1½ percent in Costa Rica, ½ percent for Nicaragua and Honduras, and ¼ percent for Guatemala. It is important to highlight that these are medium- to long-term savings, which must be underpinned by deep structural reforms, as will be discussed in the next section.

Implications and Policy Recommendations

This chapter has shown that significant room exists to improve social spending efficiency with potentially large fiscal savings. From an input-oriented perspective, Guatemala and Honduras perform poorly on education spending efficiency. On health spending efficiency, Guatemala is the best performer in the region, while Honduras and El Salvador have room for improvement. From an output-oriented point of view, on health spending efficiency all countries appear to be in line with regional comparators and relatively efficient. There is some room to improve education spending efficiency, particularly in secondary education (to a lesser extent in El Salvador). Considering the rising social needs in these countries, but also the need to build fiscal buffers (El Salvador and Costa Rica) or maintain them (Honduras), improving spending efficiency can contribute to reducing the risk of fiscal stress.

Based on the identified efficiency gaps, the discussion turns to some possible measures to generate savings and align policies to best-performer countries.

El Salvador

The wage bill represents 68 percent of the education budget and 65 percent for health. This emphasizes the need to focus on compensation as part of a fiscal consolidation effort. In El Salvador, the main issue is the presence of a large public-private wage premium (García-Escribano and others 2015). Such a premium is mainly explained by structural rigidities stemming from different compensation frameworks (escalafones). Given its size and that it is the most inequitable of the many compensation frameworks, the priority should be to limit the fiscal pressure from health sector escalafón.6 Other—broader—alternatives could be explored such as wage bill limits or rightsizing employment but, from a fairness point of view, tackling the health sector wage bill is critical.

In education, the problem also seems to be the wage bill. The education budget has become more rigid due to wage increases based on the teachers’ escalafón and the incorporation of staff from the formerly community-based EDUCO program.7 The education wage rose by 82 percent from 2007 to 2014. As in Honduras, wage increases are unrelated to teachers’ performance evaluations. El Salvador ranked among the worst in math/science tests (49 out of 53 countries in the 2007 TIMSS, Trends in International Maths and Science Study, report). Also, teachers’ wages seem too compressed: the gross wage for teachers with graduate studies is only 10 percent more than for teachers with only undergraduate studies (García-Escribano and others 2015).

Pupil-teacher ratios in primary and secondary education are higher than regional peers (Figures 7.2 and 7.3). To account for the aging of the population, the number of primary teachers should decline in favor of those in secondary education. El Salvador has a sizable coverage gap in secondary and tertiary education, since the bulk of the spending in education is at the primary level (Figures 7.2 and 7.3).

Guatemala

Guatemala presents a more complex issue as its social spending level is systematically below the regional average and other comparators. Public education spending in Guatemala has been virtually flat since 2006 at around 3 percent of GDP, being the lowest in the region even though in 1991 the national law on education established the earmarking of 35 percent of total revenues for financing education. The law also establishes that the government should target an increase in spending on education to 7 percent of GDP. Spending on education is also extremely rigid with most of it allocated for paying teachers’ salaries leaving only a small fraction for infrastructure investment. The education wage bill went up from 53 percent of total education expenditure in 2007 to 70 percent in 2013.

Similarly, public spending on health has increased only slightly as a share of GDP since 2000, remaining by far the lowest in the region at an average around 2.2 percent of GDP. Spending rose only slightly, from 1.9 percent of GDP in 2000 to 2.3 percent of GDP in 2013 (Figure 7.1). More than half of spending is allocated to personnel salaries/benefits and another 40 percent to medical supplies.

Therefore, achieving better outcomes in health and education will require an increase in inputs, notably in public spending. While efficiency improvements could be achieved, they are small and any efforts could go hand-in-hand with increasing spending, which in turn requires ensuring a stable source of additional revenues. The situation in Guatemala is dramatic as it has the lowest tax burden in the region (CEPAL 2017). The necessity of higher revenues is more pressing considering that the poverty rate is close to 60 percent.

Honduras

In health and education, the priority for Honduras is to tackle the disconnection between compensation rules and productivity. As mentioned, the wage bill represents 80 percent of the education budget and 60 percent of the health budget. Therefore, achieving sizable savings in both sectors would necessarily require reforms in their compensation policies. IMF (2016) has identified the fragmentation of the public compensation framework as the main problem of the wage bill. These different compensation frameworks result mainly from pressures from powerful interest groups, particularly teachers.

Generous wage increases, and indexation benefits granted to teachers in 2009 have been the main drivers of the increase in education spending. They established the indexation of teachers’ wage to the minimum wage, and instituted performance evaluations that were never implemented. Given the size of the education sector in the budget, revision of this policy is suggested in accordance with the original agreement to align compensation to performance and ultimately to the performance of students in standardized tests. Additionally, as suggested by Arcia and Gargiulo (2010), the required affiliation to a trade union for being a teacher grants excessive bargaining power and should be revised.

Over the coming years, Honduras also needs to adjust its public policies to tackle the impact of population aging. This will entail moving resources from primary to secondary and tertiary education and changing the composition of the teaching profession to reduce the number of primary teachers and increase the roll in higher levels of education. In Honduras, pupil-teacher ratios indicate over-staffing. Honduras has a sizable coverage gap in secondary and tertiary education while the bulk of the spending in education is at the primary level (Figures 7.2 and 7.3). At least part of these coverage gaps is explained by the relatively short instructional time received by students.8 The estimations in this chapter show that the process of moving resources from primary to higher levels of education can be done through efficiency savings, therefore preventing a dramatic short-term adjustment on primary education spending.

The issue of compensation fragmentation is particularly severe in the health sector. The health sector has six of the eight compensation frameworks in Honduras. Although some favor only a small group of workers, they are still largely inequitable.9 These frameworks should be revised in line with the need to expand coverage as stated in the law of social protection. Also, since the provision of health care services is goods-intensive, administrative measures could be implemented to exploit economies of scale stemming from the size of the public sector as a purchaser (World Bank 2015).

Nicaragua

As in other Central American countries, the wage bill represents a large share of the budget leaving limited room for investment and strengthening capacities. Compensation of employees corresponds to 71 and 55½ percent of the education and health budgets, respectively.

In education, spending is highly concentrated on the primary level (about 45 percent of the budget) where the efficiency score shows more room for savings. That the registration of primary students has been declining due to population aging also points toward rationalizing this spending. Additionally, the number of teachers without certification is high, particularly in secondary education where uncertified teachers could reach 50 percent of the teachers’ population.10

In health spending, efficiency scores show little room from efficiency gains, which highlights the need to provide higher inputs to attain better outcomes. Nicaragua in particular needs more medical facilities in rural areas. Also, a limited coverage partnership with the private sector could be an option in urban areas.

Costa Rica

In Costa Rica, increased focus on raising the efficiency, rather than the level, of expenditure on education is critical to achieve higher educational outcomes. While efficiency of expenditure on health care appears high, continued gains will be important given the long-term pressures from population aging. Costa Rica’s general government budget is dominated by education and health care spending. Expenditure on these two sectors represents over 60 percent of the budget, more than double the share in emerging markets and the OECD on average. Within Latin America, Costa Rica has the largest expenditures on both education and health care as a share of GDP. Compensation of employees corresponds to 76 percent of the education budget and close to 70 percent of the health budget. Attaining efficiency gains in these sectors could yield large benefits in expenditure rationalization over the long term.

Education outcomes do not reflect the significant public expenditure outlay on schooling. Even though the country spends less per capita than only Denmark and Sweden, education outcomes are not remarkably better than in other emerging markets. Pupil-teacher ratios are much lower than the average of Latin America or emerging markets, pointing to overstaffing and relatively elevated teaching salaries by international standards. Regarding policy outcomes, while school enrollment ratios in primary education are in line with OECD countries, they are not significantly higher than in other emerging markets. OECD standardized test (PISA) scores for secondary school students broadly match those of emerging economies that spend much less on education, suggesting there is significant scope for improving efficiency. PISA scores on reading, math, and science are consistently well below those of advanced economies.

Looking at policy recommendations, authorities in Costa Rica should move away from an emphasis on increasing spending toward establishing better educational outcomes as their main policy target. Given the large share of wages in the education budget, and salaries that are much higher than for workers in the private sector, payroll savings would be important to support needed fiscal consolidation efforts. Rationalization of bonus schemes that have contributed to total salary increases well above inflation could be particularly important. Specific recommendations made by the OECD in the context of accession included improving evaluation mechanisms and enhancing accountability for teachers, as well as improving professional development and harmonizing their qualifications. Reinforcement of the vocational technical track would also contribute to reducing dropouts in secondary schools and tackling high youth unemployment.

The scope for early gains in efficiency appears much more limited in the health care sector. Public spending on health care as a share of GDP is high relative to Costa Rica’s level of development, significantly higher than the emerging markets average, and higher than in richer countries like Chile, Uruguay, and Mexico. Efficiency in public spending on health care also appears to be strong, with near universal coverage and health outcomes close to OECD levels despite significantly lower health expenditure per capita.

These favorable outcomes are not explained by significant additional private sources of health spending, as Costa Rica has one of the lowest shares of out-of-pocket and other private health expenditures in the region. Notwithstanding this favorable assessment, since Costa Rica is one of the countries in the region where population aging is more advanced, spending on health care is projected to double over the next 50 years if universal access and current service levels are to be maintained (see Chapter 8 for details). This highlights the need for measures to gradually increase efficiency. As in the education sector, there is scope to contain the wage bill. OECD recommendations include updating information systems to better monitor performance indicators and the forward-looking allocation of resources to consider changing demographic patterns and disease trends and introducing diagnosis-related funding schemes that provide stronger incentives to control spending than fee-for–service schemes that can result in service oversupply (OECD 2016).

References

  • Arcia, Gustavo and C. Gargiulo, 2010, “Análisis de La Fuerza Laboral en Educación en Honduras”, Banco Interamericano de Desarrollo, Notas Técnicas N°7.

    • Search Google Scholar
    • Export Citation
  • Baltodano, Eduardo and O. Pachecho, 2016, “La Eficiencia del Gasto Público en Educación y Salud en Nicaragua, 2003–2013”, Banco Interamericano de Desarrollo, Notas Técnica N° IDB-TN-977.

    • Search Google Scholar
    • Export Citation
  • Banco Interamericano de Desarrollo, 2014, “Mejorando la Eficiencia de los Recursos Humanos del Estado: Informe sobre Empleo Público y Política Salarial en Centroamérica, Panamá y la República Dominicana”.

    • Search Google Scholar
    • Export Citation
  • Banco Interamericano de Desarrollo, 2016, “La Eficiencia del Gasto Público en Educación y Salud en El Salvador, 2003–2013”.

  • Barro, Robert, and Xavier Sala-i-Martin. 2004. Economic Growth, Second edition. Cambridge, MA: The MIT Press.

  • Celasun, Oya and others, 2015, “Fiscal Policy in Latin America: Lessons and Legacies of the Global Financial Crisis”, IMF Staff Discussion Note SDN/15/06, International Monetary Fund, Washington DC.

    • Search Google Scholar
    • Export Citation
  • Bogetoft, Peter and L. Otto, 2011, “Benchmarking with DEA, SFA, and R”, Springer.

  • Bruns, Barbara and J. Luque, 2014, “Great Teachers: How to Raise Student Learning in Latin America and the Caribbean”, The World Bank.

    • Search Google Scholar
    • Export Citation
  • CEPAL. 2017. Panorama Fiscal de América Latina y el Caribe 2017. Santiago: Economic Commission for Latin America and the Caribbean.

  • De la Fuente, Angel, 2011, “Human Capital and Productivity”, Nordic Economic Policy Review, N°2, pp. 103132.

  • García-Escribano, Mercedes, E. Flores, J. Kapsoli, and M. Soto, 2015, “Gasto en Salarios Gubernamentales: Análisis y Desafíos”. Retrieved from http://www.mh.gob.sv/portal/page/portal/PMH/Documentos_O_M/Fondo_Monetario_Internacional/Documentos/2016/GASTOS_EN_SALARIOS_GUBERNAMENTALES_ANALISIS_Y_DESAFIOS_2016.PDF

    • Search Google Scholar
    • Export Citation
  • Grigoli, Francesco and J. Kapsoli, 2013, “Waste Not, Want Not: The Efficiency of Health Expenditure in Emerging and Developing Economies”, IMF Working Paper 13/187, International Monetary Fund, Washington DC.

    • Search Google Scholar
    • Export Citation
  • Gupta, S. and M. Verhoeven, 2001, “The Efficiency of Government Expenditure: Experiences from Africa”, Journal of Policy Modelling, Vol. 23, pp. 43367.

    • Search Google Scholar
    • Export Citation
  • Farrell, M. J., 1957, “The Measurement of Productive Efficiency,” Journal of the Royal Statistical Society, Vol. 1203, pp. 25390.

    • Search Google Scholar
    • Export Citation
  • Herrera, S. and G. Pang, 2005, “Efficiency of Public Spending in Developing Countries: An Efficiency Frontier Approach”, World Bank Policy Research Working Paper N° 3645

    • Search Google Scholar
    • Export Citation
  • International Monetary Fund, 2010, “Emerging from the Global Crisis: Macroeconomic Challenges Facing Low-Income Countries”, Policy Paper. IMF, Washington.

    • Search Google Scholar
    • Export Citation
  • International Monetary Fund, 2014, World Economic Outlook, “Legacies, Clouds, Uncertainties: Is It Time for an Infrastructure Push? The Macroeconomic Effects of Public Investment”.

    • Search Google Scholar
    • Export Citation
  • International Monetary Fund, 2016, “Managing Government Compensation and Employment: Institutions, Policies, and Reform Challenges”.

    • Search Google Scholar
    • Export Citation
  • Kumbhakar, Subal, H. Wang, and A. Horncastle, 2015, “A Practitioner’s Guide to Stochastic Frontier Analysis Using Stata”, Cambridge University Press.

    • Search Google Scholar
    • Export Citation
  • Obstfeld, Maurice and K. Rogoff, 1997, “Foundations of International Macroeconomics”, The MIT Press.

  • Simar, Léopold and P. W. Wilson, 1998, “Sensitivity Analysis of the Efficiency Scores: How to Bootstrap in Nonparametric Frontier Models,” Management Science, Vol. 44, N°1, pp. 4961.

    • Search Google Scholar
    • Export Citation
  • International Monetary Fund, 2000, “A General Methodology for Bootstrapping in Non-Parametric Frontier Models,” Journal of Applied Statistics, Vol. 27, N°6, pp. 779802.

    • Search Google Scholar
    • Export Citation
  • World Bank, 2015, “Honduras. Estudio de Gasto Público Social y sus Instituciones”.

Annex 7.1 Benchmarking Methodology

Benchmarking is the systematic comparison of the performance of one unit against peers. It involves comparing units implementing the same transformation processes consuming inputs to produce goods and services (outputs). These units could be firms, industries, and so on, but for this chapter, they are countries. This comparison is done based on performance evaluations. Because of this, any benchmarking exercise is intimately related to the concept of efficiency. In the benchmarking literature, efficiency is measured by identifying the best- performing units and using them to build a frontier. This frontier is called the “efficiency frontier.” Once the frontier is obtained, the performance of all units is assessed by measuring their distances relative to the efficiency frontier.

The modern discussion of gauging efficiency started with Farrell’s seminal paper in 1957. The paper defines two types of efficiency, technical and allocative. Figure 7.1.1 illustrates both concepts using the familiar isoquant diagram assuming a production function with two inputs x1 and x2. To simplify the analysis, inputs are normalized relative to the output so that the level of production is always 1. The YY isoquant represents the optimal (minimum) combination of normalized inputs required to produce one unit of output. The point P represents a suboptimal production bundle because it produces one unit of output, but with more inputs relative to Q (which is part of the isoquant). As point Q represents the optimal consumption of inputs required to efficiently produce one unit of output, the ratio QP/OP would be a measure of technical inefficiency, which means that distance QP could be saved if inputs were used efficiently.

Annex Figure 7.1.1.
Annex Figure 7.1.1.

Technical and Allocative Efficiency

The latter is a view of efficiency entirely based on the technical capacity to obtain the higher level of output with the minimum consumption of inputs. However, one can see efficiency also from a cost minimizing perspective. Let p1 and p2 be the prices of inputs x1 and x2 then the slope of line AA’ would be –p2/ p1 and Q’ would be the optimal bundle assuming such price levels. For the production bundle P, the ratio OR/OQ would be a measure of the allocative or cost efficiency. Allocative efficiency measures the amount of resources that could be saved if, given input prices, the consumption of inputs would be used to reduce the unit’s total cost to a minimum. Because of the lack of comparable multicoun-try data on prices, this chapter has focused entirely on the estimation of technical efficiency.

Technical efficiency could be estimated based on input- or output-oriented models. In input-oriented models, the efficiency scores are the proportional amount by which input consumption could be reduced while leaving outputs unchanged. On the other hand, output-oriented (OO) efficiency scores are defined as the proportional amount by which outputs could be increased while leaving inputs’ consumption unchanged.

There are two families of methodologies—parametric and non-parametric— to estimate technical efficiency. Each methodology has advantages and disadvantages. Parametric methods require several assumptions on the errors’ stochastic distribution and the functional form underpinning the model. At the same time, parametric methods assume a stochastic relationship between inputs and outputs allowing separation in the efficiency estimation of the part that is real inefficiency from the part explained by measurement errors or other noise in the data.11 The flagship of the parametric methods is the stochastic frontier model (SFA).12

Non-parametric methods, on the other hand, are based on linear programming and therefore do not require any errors’ distributional assumption or assumptions relative to the functional form of the relation between outputs and inputs. However, non-parametric models do not include randomness, so therefore all the data by construction provide information on the inefficiency or the technological frontier. This assumption makes non-parametric models very sensitive to the presence of outliers or noise in the data. Data envelopment analysis (DEA) is—by far—the most widely method used in the benchmarking literature. DEA is a mathematical programming method that can solve the two main tasks involved in a benchmarking exercise: (a) calculate the frontier based on the best performer units, and (b) evaluate performances relative to such frontier. A DEA model requires some basic assumptions about the frontier: (1) free disposability, (2) convexity, (3) returns to scale, and (4) additivity. Note that returns to scale cover DEA specifications ranging from constant to variable returns to scale. The constant returns model has only one best performer unit, while the variable returns model allows for several best performers operating on a convex efficiency frontier.

As mentioned, the DEA model has drawbacks. First, it is a purely deterministic method that ignores the presence of noise in the data, such as measurement errors. Such noise is very common in the case where the units under analysis are countries, and in particular emerging or low-income countries. Second, DEA estimations are biased. They estimate the efficiency frontier based on “best performer” units, which do not necessarily represent the true frontier. The SFA model also has drawbacks beyond the many assumptions required to set up the model. It assumes inefficiency as one of the parameters to estimate. This assumption needs a prior on the statistical distribution of such inefficiency term. The most popular distributional assumptions are half-normal, truncated normal, and exponential. All of them involve a zero or constant average for the inefficiency parameter, therefore, resulting in an underestimation of it (that is, the efficiency scores are overestimated).

Simar and Wilson (1998, 2000) developed a methodology using bootstrapping to add a layer of randomness to the DEA model to overcome these drawbacks. They pointed out that a DEA frontier is an estimation of the true frontier based on a single sample drawn from an unknown population. Because of that, the efficiency measures are sensitive to the sampling variations underpinning the estimate of the frontier. A way to assess this sensitivity is using bootstrapping. Bootstrapping is a computer-based statistical method that generates an artificial, new random data set obtained by sampling with replacement from a given data set. This new data set could be used to calculate some statistics called “replicates.” The procedure is repeated many times, each generating new replicates to build a sample. Based on this sample conclusions can be drawn on the distribution of the original data under the assumption that it mimics the distribution of the bootstrapped sample.13

Bootstrapping allows bias in the efficiency scores to be corrected and it calculates corresponding confidence intervals. As mentioned, as the DEA frontier is based on best-performing units, it would capture only the lower bound of the true frontier. This, by definition, generates an upward bias in the estimated efficiency scores. If it is assumed that the distribution of the difference between the estimated and the bootstrapped efficiency scores mimics the distribution of the difference between the true and the estimated efficiency scores, the bias can be estimated, the efficiency scores corrected, and their confidence intervals can be found.

Annex 7.2

Annex Table 7.2.1.

Point Estimates and Confidence Intervals. Input Oriented— Primary Education

article image
Source: Authors’ calculations.
Annex Table 7.2.2.

Point Estimates and Confidence Intervals. Output Oriented—Primary Education

article image
Source: Authors’ calculations.
Annex Table 7.2.3.

Point Estimates and Confidence Intervals. Input Oriented— Secondary Education

article image
Source: Authors’ calculations.
Annex Table 7.2.4.

Point Estimates and Confidence Intervals. Output Oriented—Secondary Education

article image
Source: Authors’ calculations.
Annex Table 7.2.5.

Point Estimates and Confidence Intervals. Input Oriented—Health

article image
Source: Authors’ calculations.
Annex Table 7.2.6.

Point Estimates and Confidence Intervals. Output Oriented—Health

article image
Source: Authors’ calculations.