Does public expenditure management matter for education outcomes?

Abstract

This paper examines the significance of public expenditure management for primary education outcomes in public schools in two South African provinces (Gauteng and North West). Using cross-sectional data from 175 public primary schools and 13 local education offices, the analysis finds that while misappropriation of education funds (leakages) is not strongly associated with poor education outcomes, delays on the part of the government in disbursing funds to schools are correlated with Grade 5 dropout rates. The paper finds no evidence that public expenditure and total resource wealth (including public and private contributions) are significantly associated with education outcomes. Increased spending on learning and teaching support materials is associated strongly with lower Grade 1 and Grade 7 repetition rates. The paper also finds that repetition rates are driven strongly by poverty indicators at the district level, while dropout rates are driven strongly by district and school inefficiency.

Keywords:

• economics of education
• education finance
• education outcomes
• efficiency
• public expenditure management
• public expenditure

JEL classifications:

• I21
• I22
• I28

1. Introduction

Primary education has long been established as a cornerstone of poverty alleviation and socio-economic development (Psacharopoulos, 1985, 1994, 2004; Birdsall, 2001, 2006; Duflo, 2001; Su 2006; Hanushek & Wossman, 2007a). In most national development strategies in Africa, relatively large proportions of development resources have been allocated to education. Public spending on primary education in sub-Saharan Africa rose by 29% in real terms between 2000 and 2005 (UNESCO, 2010). The reasons for sizable investments in education are numerous: young people with at least a primary education are less likely to contract HIV than are those with no schooling; greater investment in female education leads to more productive farming and significant decreases in malnutrition; an additional year of schooling results in higher wages; and education supports democratic growth and political stability (Birdsall, 2006). Nevertheless, there is a paradox. While large proportions of gross domestic product are allocated consistently to the education sector, the desired outcomes (such as high pass and completion rates, strong academic achievement, absorption into a skilled labour force and competitive wages) are severely lacking in many African countries. In fact, in about one-half of the countries in sub-Saharan Africa, education enrolments and quality have deteriorated since 1990 (Al-Samarrai, 2003).

This decline in education outcomes despite increased expenditure, coupled with the ambiguity surrounding the determinants of education outcomes, motivates this paper. While not discounting the importance of expenditure in education, I argue here that a focus on public spending alone is misleading and that the management of public expenditure plays a potentially important role in fostering good education outcomes. South Africa is an appropriate case study as a country that spends about 5.5% of its gross domestic product on education and yet is burdened with levels of academic achievement that are even worse than other developing countries who spend only a fraction of what South Africa spends on education. Using original data, I assess the relationship between public expenditure management (PEM) and primary education outcomes in South Africa. Specifically, I investigate the following five research questions:

1. Is there a significant relationship between leakage of funds (misappropriation) and education outcomes?

2. Is there a significant relationship between delays in the disbursement of funds and education outcomes?

3. Is there a significant relationship between levels of public expenditure (government subsidies) per capita and education outcomes?

4. Is there a significant relationship between a school's total resource wealth per capita and education outcomes?

5. Is there a significant relationship between learning, teaching and support material (LTSM) expenditure per capita and education outcomes?

In what follows, I review the literature in Section 2, provide the South African contextual background in Section 3, present the data and methodological approach in Section 4, discuss findings from the cross-section regression analyses in Section 5 and conclude with policy implications in Section 6.

2. Review of the economics of education literature

The importance of education for development is rooted in the early human capital theory, where emphasis was placed on education as an input to economic production rather than as an outcome of economic production (Becker, 1964; Mincer, 1970). Education economists have continued to raise important policy questions on the role of education in improving the welfare of the five billion people living in developing countries. Education has been found to be instrumental in the adoption of new agricultural technologies (Foster & Rosenzweig, 1996) and as a means to improve health and reduce fertility (Straus & Thomas, 1995). Sen (1999) describes education as an intrinsic good in itself. Primary education, in particular, has long been argued to be a cornerstone of poverty reduction and socio-economic development. For example, Psacharopoulos (1994), in his review of empirical studies, showed that both private and social rates of return to investment in education were highest at the primary level, regardless of the number of students. Thus, the importance of education for development has been firmly established in the literature, but there remains a great deal of ambiguity as to the precise inputs that yield good education outcomes. This section reviews the literature on both input and outcome variables.

2.1 Review of education input variables

Education inputs may be defined as the resources provided by parents, school management and government authorities to facilitate learning in schools. These include qualified and well-paid teachers, classrooms, learning materials, facilities, organisation, curriculum, tuition fees, government subsidies and technical support. Their importance with regard to education outcomes may seem obvious. Nonetheless, their significance has been the subject of intense debate among economists, and the findings have been largely contradictory. In both the international and the South African literature, socio-economic status seems to matter fairly consistently (Coleman et al., 1966; Heyneman, 1979; Thomas, 1996; Case & Deaton, 1999; Lam, 1999; Van der Berg, 2008). Case & Deaton (1999) found that pupils in better-off black households in South Africa do better in their education. In Van der Berg's (2008) study, socio-economic differentials at the school level played a major role in educational outcomes at the primary school level in South Africa. The literature on class size typically makes reference to Eric Hanushek (1981, 1986, 1995, 1998, 1999), who has long argued that school inputs, including class size, have no systemic impact on student performance. Conversely, Akerhielm (1995), Crouch & Mabogoane (1998), Krueger (2002), Lee & Barro (2001) and several other researchers have all found smaller classes to be associated with improved achievement.

The evidence produced in the literature regarding the impact of teacher characteristics on education outcomes is equally mixed. Hanushek (1995) summarises 96 studies on the estimated effects of resources on education in developing countries; the majority of studies that focused on teacher education showed a statistically significant and positive relationship between teacher education and outcomes, but the majority of studies on teacher experience and salaries showed a statistically insignificant relationship. In his view, differences in school quality are better explained by teacher ‘skills’ – an indicator that captures background and training, as well as choice of presentation styles and curricular materials (Hanushek, 1981). In contrast to Hanushek's findings, Lee & Barro's (2001) cross-country study of 58 countries found that teacher salaries had a positive impact on reading scores. In the South African literature, the research on teacher characteristics has also produced mixed results. Van der Berg & Louw (2006) find teacher absenteeism and teacher quality (measured by a dummy variable indicating that the mathematics teacher has a degree and another indicating that the teacher has teaching training) to have a large effect on test scores. Crouch & Mabogoane (1998) found little evidence that teacher experience mattered for education outcomes although additional teacher training does matter.

School-based management – the devolution of authority from central government to principals, teachers, parents, communities and sometimes students – is gaining widespread acceptance globally. The theoretical appeal is strong: through participative decision making and autonomy, schools under school-based management can be more efficient in the use of resources and can respond better to local needs (Santibanez, 2006), and schools and parents who are interested in maximising their children's learning outcomes have incentives to make decisions that will impact positively on the learning process (Hanushek & Wossman, 2007b). It is from these arguments and some limited empirical evidence (Wossman, 2003; Barera-Osorio, 2009) that school-based management interventions in South Africa are incorporated in this study as potential explanatory variables.

The role of public expenditure in education outcomes continues to be among the most heated areas in the education literature. A cross-country study by Al-Samarrai (2002) showed a weak relationship between public spending and education outcomes. In a study of 50 developing and transition countries, Gupta et al. (1999) found that total education spending hardly affects outcomes, but that the proportion of primary and secondary spending in total spending has a positive and significant impact. In their survey of developing countries, Pritchett & Filmer (1999) found that resources only marginally affect education quality; however, they also found that there could be significant efficiency and productivity gains by allocating expenditure to areas of high marginal productivity, such as learning materials (textbooks and other types of instructional aids). In summarising various multiple regression results that try to explain education outcomes, Fiske (2000) arrived at the conclusion that increasing specific education inputs tends to matter more for low-income countries than for higher-income ones, and therefore that resources matter when they are added to a low resource base; then, at a certain level, one observes diminishing returns. This aligns with general interpretations of production functions.

The management of public expenditure and how this affects outcomes has largely been neglected in the literature. However, an important body of work using public expenditure tracking surveys, which has been led by the World Bank, reveals gross inefficiencies in the use of public funds to deliver education services For example, in Uganda only 13% of the annual capitation grant from the central government reached schools between 1991 and 1995 (Reinikka & Svensson, 2004). Therefore, district officials captured 87% for purposes unrelated to education. In Zambia, when funds were channelled to schools at the discretion of district authorities, less than 20% of schools received their funding (Das et al., 2004). Aside from leakages, delays and bottlenecks in the allocation of resources through public administrations have been identified in tracking surveys as a potentially serious problem affecting quality of services. In Tanzania, there was evidence of delays in disbursing school subsidies and the processing of non-wage funds, ranging from 6 to 42 days at the Treasury (Gauthier, 2006). In Rwanda, delays in the payment of capitation grants occurred between the submission of a funding release request by the line ministry and the authorisation by the Ministry of Finance and most of the money arrived in the fourth quarter of the budget year (Fofack et al., 2003). The relationship between these inefficiencies and learning outcomes has not been fully explored. This paper contributes to the literature by examining this relationship.

2.2 Review of education outcome variables

Education outcomes are typically grouped into completion rates, learning achievement (e.g. test scores), average years of education of the adult population and labour force, and earnings and productivity (Boissiere, 2004). Vaizey (1971) argues that assertions about the global output of education are not meaningful if they are intended as a guide to the efficiency with which resources in education as a whole are used compared with the way resources are used in other parts of the economy. He illustrates this challenge with the example of a school that remains unchanged in every respect except for an additional teacher or an extra $100 to spend on equipment. A reasonable assumption can be made that changes with respect to pupils could be attributed to the extra teacher or extra $100. The reality is that, in the course of the year, pupils and teachers become a year older and there are withdrawals and additions to the student body and teaching staff. Thus, changes in the cultural ambiance of the school and even within pupils' families can be explanatory factors. Education production function studies try to control for such external factors, but Vaizey's concern remains an important caveat that must be borne in mind in the analytical discussions. Indeed, the cumulative nature of learning makes it difficult to isolate indicators of education outcomes.

Several studies (for example, Colclough & Lewin, 1993; Schultz, 1995; Case & Deaton, 1999; McMahon, 1999) have used enrolment rates or attainment as their measures of outcomes. Jones et al. (2008) used school attendance – an output variable influenced by enrolment and the number of schools in a district. Rajkumar & Swaroop (2008) used failure to complete an adequate level of primary schooling. Other notable studies (for example, Lee & Barro, 1997; Gupta et al., 1999; Rumberger & Thomas, 2000; Al Samarrai, 2003) have used survival, repetition and dropout rates in their analyses. The most recent of these studies, however, have used test scores (Duflo et al., 2008; Holmlund et al., 2008; Van der Berg, 2008; Glewwe et al., 2010).

This study uses pass, repetition and dropout rates as measures of outcomes. Test scores were not incorporated into the study, because not all schools selected in the sample would have participated in the international, standardised tests available at primary level in South Africa – the Progress in International Reading Literacy Study (PIRLS), the Southern and East African Consortium for Monitoring Education Quality (SACMEQ), and the national Systemic Evaluations, which assess a sample of primary school students in South Africa in literacy and numeracy. Pass (or completion), repetition and dropout rates are used because they conform to the literature; however, they are not without limitations.

Pass rates, while indicative of achievement, suffer from a lack of comparability due to the arbitrary setting of examinations by primary schools. Repetition rates may reflect weak learning abilities but, in the South African context, these rates could be inflated at the Grade 1 level by underage enrolment. Conversely, artificially low repetition rates at the Grade 7 level could be due to policy restrictions on the number of repetitions allowed and restrictions on average repetition. Dropout rates, on the other hand, do not distinguish between pupils who leave the school system never to return and those who leave a school to transfer to another school. In fact, dropout rates have been more controversial as an indicator of outcomes also because students may drop out of schools not because they are weak students but for several other reasons such as dissatisfaction with school quality (Hanushek et al., 2006) or socio-economic conditions (Levy, 1971; UNESCO, 1984) – reasons that are independent of ability.

At the same time, the literature recognises that repetition rates are symptomatic of learning difficulties and therefore are associated with poor outcomes (Alexander et al., 1994; Jimmerson, 2001; Corman, 2003; Brophy, 2006). The measure's appropriateness in South Africa is underlined by an influential report on learner retention that acknowledges both dropout and repetition rates as symptomatic of learning abilities and internal inefficiencies within schools (Ministerial Committee, 2008). Moreover, available quantitative evidence shows that test scores in primary schools in South Africa are correlated with repetition and dropout rates, thus making the latter variables valid indicators of learning outcomes.

3. The state of primary education in South Africa

After more than 300 years of colonial and apartheid rule, South Africa consolidated democracy in 1994 and put an end to a long history of oppression of black South Africans by a white minority. The new democratic administration proceeded to establish and entrench a system of democratic governance with a strong emphasis on individual and collective rights. However, apartheid left South Africa with a legacy of severe socio-economic problems. Many of these problems are rooted in the history of education in South Africa, because the former apartheid government had entrenched an inequitable system under which black South Africans received a sub-standard level of education known as Bantu education. Bantu education was designed and imposed on black pupils by the apartheid government as ‘a specialised education system that would meet the special needs of the African people and assist them to develop, along their own lines, towards their own separate national destinies’ (Murphy, 1973:233). In reality, the architects of apartheid aspired to develop a black labour policy that would encourage white capital accumulation without endangering the legitimacy and stability of white political domination (Terreblanche, 2002). Inevitably, the disparities in education between white and black South Africans were vast. The new democratic government that emerged in 1994 was faced with the daunting task of correcting this historical injustice by providing equal educational opportunities to all South Africans.

This was done mainly by means of fiscal measures in terms of which resources were allocated to schools according to their poverty profile, also known as poverty quintile. Poverty quintiles are categories into which schools are classified on the basis of rates of unemployment, income and illiteracy within the schools' catchment areas. Therefore, schools ranked within a lower quintile (between quintiles one and three) would receive a higher subsidy than schools ranked in the fourth or fifth quintiles. Many schools, particularly those in the higher quintiles, are not dependent on provincial allocations; rather, by charging school fees, as provided for in the South Africa Schools Act 84 of 1996, some schools are able to operate self-sufficiently. The decision to allow fees to be set in public schools remains controversial. Some feel that it perpetuates inequalities, while others see it as allowing for a necessary supplement to scarce resources and as an incentive for middle-class families to keep their children in the public school system. Moreover, on the basis of their poverty ranking, certain schools have been categorised as ‘no-fee schools’ as of 2007. Essentially, these schools have been barred from charging tuition fees and entirely depend on government funds.

A critical question is whether the extensive redistributive measures have been able to address the equity concerns. The findings are mixed. Earlier studies, such as Fiske & Ladd (2004) and Jansen & Taylor (2003), found that significant progress has been made towards a fairer distribution of resources across provinces and across schools. The authors, however, maintained that much less progress has been made in ensuring that schools have sufficient resources to meet optimal educational standards. Similarly, Van der Berg & Burger (2002) argued that racial inequality in primary education is still a concern in post-apartheid South Africa, and although the massive resource shift to black schools was successful in narrowing the pupil-to-teacher ratio gap, matriculation results remained substandard and had even deteriorated in the post-apartheid era.

South Africa has participated in three comparative international education surveys: the PIRLS, in 2006 for Grade 4 pupils; the Trends in International Mathematics and Science Study, in 2003 for Grade 8 pupils; and the SACMEQ, in 2000 and 2007 for Grade 6 pupils. The results from all three comparative tests are clear. South African students lag behind internationally and regionally. Moreover, within the country, there are regional disparities in performance.

4. Data and methodology

Given that there is no existing dataset capturing both the education outcomes and PEM variables, I use original data I collected in 2007 using two micro-level surveys: a public expenditure tracking survey and a quantitative service delivery survey. Both are the ideal instruments for this study because of the precision with which they ‘track’ the flow of resources from the source to the end user. The combined use of the public expenditure tracking survey and the quantitative service delivery survey goes beyond normal public expenditure reviews, which examine the composition of spending and sectoral allocations. They are tools that quantitatively measure provider incentives and behaviours, and they allow lessons to be learned about how spending is transformed into services (Dehn et al., 2003). In the survey, questions were asked about characteristics of the school such as: geographic location; number of students; poverty quintile ranking; pass, repetition and dropout rates; teacher qualifications and salaries; school facilities; characteristics of the governing body; and sources of funding and expenditure (see Appendix A for more detail).

4.1 Sample design

The study aimed for a sample size of 8.5% of all public primary schools in the North West and Gauteng provinces. The 8.5% value was guided by the fact that nationally and internationally comparable studies on education have used samples ranging from 3 to 18%. Firstly, all schools were stratified according to district or region. This was to ensure that each school in a particular region or district had an equal chance of being selected. The stratification was followed by a simple random sampling method, also referred to as ‘epsem’ (equal probability of selection method) sampling, a tool that is used widely in applied educational research because it usually leads to self-weighting, in which a simple arithmetic mean obtained from the sample data is an unbiased estimation of the sample mean (Ross, 2005). SPSS software was used to randomly select 8.5% of schools from each of the strata, which would total 207 schools. Responses were received from 175 schools. Of these, 96 were schools located in Gauteng and 79 were located in the North West. This represented 7.3% and 7.1% respectively with minor variations, percentage-wise, at district levels. Table 1 presents the sample allocation by region and district. It is important to emphasise that the sampling procedure did not take school size into account, and therefore, in the random selection, some schools would have more pupils than others. This does not bias the study, because the unit of analysis is at the school level and not st the pupil level. Conclusions must also be interpreted as such.

Table 1: Sample allocation over strata, Gauteng and North West Province

	Total number of schools (N)	Sample size (n)
District (Gauteng)
Erkurhuleni North	103	8
Erkurhuleni South	117	9
Gauteng East	109	9
Gauteng North	33	2
Gauteng West	83	5
Johannesburg Central	157	12
Johannesburg East	78	5
Johannesburg North	102	8
Johannesburg South	64	5
Johannesburg West	92	7
Sedibeng East	53	4
Sedibeng West	98	6
Tshwane North	59	2
Tshwane South	125	10
Tshwane West	45	4
Total	1 318	96
Region (North West)
Bophirima	172	11
Southern	218	14
Central	348	31
Bojanala	376	23
Total	1 114	79

The sample of 175 schools mirrored the actual population in the reflection of their poverty quintiles. There were about 20% of schools in quintile one; 15% in quintile two; 36% in quintile three; 17% in quintile four; and 12% in quintile five. Therefore, the majority of the schools fell into the low to medium poverty spectrum (quintiles one to three). All of the schools were mixed-gender schools. The student population ranged from 33 to 1644 students, with a median school population of 562 and an average of 633 students. A table of means for the outcome variables is presented in Table 2.

Table 2: Summary of statistics for pass, repetition and dropout rates

Province	Variable	Observations	Mean	Standard deviation	Minimum	Maximum
Gauteng	Pass	83	97.3	7.3	50	100
	Gr_1_rep	91	5.4	5.8	0	24
	Gr_7_rep	84	1.0	1.8	0	9
	Gr_1_dropout	85	4.1	8.3	0	54
	Gr_5_dropout	85	3.3	4.3	0	18
North West	Pass	53	95.6	7.6	58	100
	Gr_1_rep	72	12.0	13.4	0	77
	Gr_7_rep	60	3.0	5.2	0	21
	Gr_1_dropout	71	4.4	9.5	0	72
	Gr_5_dropout	68	3.9	4.9	0	22
Gauteng and North West	Pass	136	96.7	7.4	50	100
	Gr_1_rep	163	8.3	10.4	0	77
	Gr_7_rep	144	1.8	3.7	0	21
	Gr_1_dropout	156	4.2	8.7	0	72
	Gr_5_dropout	153	3.6	4.6	0	22

4.2 Data analysis

Before assessing the significance of the PEM variables while controlling for other inputs, I undertake bivariate regressions to determine whether there is a correlation between the PEM variables and each of the outcome variables. The purpose is to discount any PEM variable that is not correlated with education outcomes from the next stage of testing, when control variables are factored in. I find that leakages have no significant relationship with outcomes, even when outliers are removed. Delays in fund disbursements are associated significantly with dropout rates. Total resource wealth per capita and LTSM spending per capita are also significantly associated with several of the education outcome variables. I find no significant correlation between the levels of public spending and outcomes. This remains true even when the redistributive component (the disproportional allocation of funds to disadvantaged schools) is taken into account.

Having established which PEM variables to incorporate in the model for the next stage of testing, I then undertake a series of ordinary least squares (OLS) regressions to establish the importance of PEM while conditioning for other factors.

Assuming that variation between pupils is fixed, the model tries to predict the percentage change in education outcomes given a unit change in total resource wealth, LTSM spending and delays. I estimate the relationship as follows:

where Educ represents the five education outcome variables (examination pass, Grade 1 and Grade 7 repetition rates, and Grade 1 and Grade 5 dropout rates); Delay represents a dummy variable for no/minimal delay or extensive delay in the disbursement of subsidies to schools; LTSM represents spending on learning and teaching materials per student; and Totalres represents a sum of government subsidies, household contributions in terms of tuition fees and donor contributions per student. I control for school inputs (schinp) and non-school inputs (nonschinp) of theoretical importance as established in the literature. The error term is represented by μ.

It is important that the model addresses issues of endogeneity in order to minimise doubt as to whether the relationship between two variables is being driven by another unobservable variable. I identify two sources of potential endogeneity, selective migration and purposive programme placement, as these are common problems affecting education production function studies. The former is a problem of self-selection and refers to instances where parents choose the school their children attend on the basis of its resources; consequently, schools with more resources end up with pupils from more favourable family backgrounds. Similarly, better-qualified teachers may choose to work in more affluent schools. In this instance, self-selection can lead to overestimates of the effect of resources on outcomes.

The first problem is minimised to a large extent in the context of South Africa because, in the public school system, schools are obliged to give admission preference to children living within a ‘school feeder zone’ as stipulated in Article 34 of the Admission Policy for Public Ordinary Schools (Department of Education, 1998). Other children outside this area are taken on a first-come, first-served basis or are placed on a waiting list. Self-selection is therefore unlikely to occur on a large scale in the case of parental selection of schools. The preference of better-qualified teachers for affluent schools, however, remains an endogenous problem that I have to address in the model. I do this by controlling for teacher salary as a proxy for teacher qualification. The reason for this is that teachers in South Africa are paid based on their qualifications. The second source of endogeneity pertains to the targeting of educational resources at weaker/disadvantaged schools from a policy standpoint. This is true in the case of South Africa, as the government provides more non-salary resources per capita to poorer schools according to quintile rankings. The model addresses this problem by controlling for poverty quintiles, which would be the factor determining how resources are allocated.

Finally, I acknowledge that OLS regressions do not address aggregate biases and nesting of data, which are more appropriately addressed by hierarchal linear models. This is not a concern for this study because the research collects and analyses data at the school level and also draws conclusions at the school level and not the pupil level. This is an important caveat to bear in mind when interpreting the results. OLS regressions can also be limiting when the parameters of the dependent variables take on relatively few values, including zeros and when they are non-normal with heteroskedastic distributions, as is the case with repetition and dropout rates. It is worth mentioning that a second model (negative binomial regressions) is also undertaken alongside the OLS regressions. However, the results of the negative binomial regressions emerge to be fairly similar to the OLS results and therefore the latter are reported and discussed below.

5. The effect of PEM on education outcomes: Regression results

Table 3 presents the regression results. There are two sets of regressions for each outcome variable. The first model only controls for socio-economic variables (non-school inputs). The second model controls for both socio-economic and theoretically relevant school inputs, as established in the literature, such as class size, teacher characteristics and school-based governance.

Table 3: OLS regressions of PEM and education outcomes

	Pass rates		Grade 1 repetition		Grade 7 repetition		Grade 1 dropout		Grade 5 dropout
Variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
LTSM expenditure per capita	0.007 (0.005)	0.01 (0.006)	−0.016* (0.01)	−0.012 (0.01)	−0.005* (0.003)	−0.007 (0.004)	0.001 (0.009)	−0.003 (0.01)	−0.000 (0.004)	0.001 (0.005)
Total resource wealth per capita	−0.000 (0.000)	−0.000 (.0001)	−0.000 (0.001)	0.000 (0.001)	0.000 (0.000)	−0.000 (0.000)	−0.000 (0.001)	−0.000 (0.001)	−0.000 (0.000)	0.000 (0.000)
Delay dummy	–1.623 (1.186)	–1.897 (1.422)	1.267 (2.256)	2.58 (2.144)	1.132 (0.758)	0.897 (0.932)	–1.368 (2.072)	–1.603 (2.017)	1.111 (0.956)	2.233** (1.067)
Quintile one	–4.903 (2.979)	–1.187 (4.17)	5.597 (5.473)	7.678 (5.879)	2.388 (1.806)	−0.607 (2.473)	3.759 (5.208)	3.421 (6.031)	−0.592 (2.374)	0.268 (3.15)
Quintile two	–5.625** (2.609)	–4.043 (3.658)	6.427 (5.473)	4.908 (5.407)	3.91** (1.612)	3.263 (2.14)	1.82 (4.867)	−0.598 (5.692)	1.735 (2.21)	1.632 (2.942)
Quintile three	–1.867 (2.475)	0.105 (3.514)	6.913 (4.878)	3.165 (5.152)	1.242 (1.505)	0.112 (2.008)	1.793 (4.699)	1.891 (5.607)	2.106 (2.119)	3.548 (2.831)
Quintile four	–3.261 (2.439)	–3.546 (3.601)	2.243 (4.878)	–2.124 (5.312)	0.592 (1.522)	0.259 (2.089)	6.818 (4.579)	3.084 (5.622)	4.14* (2.096)	6.367** (2.896)
District average household size	4.831** (1.728)	3.267 (3.601)	0.336 (3.067)	–5.207 (3.177)	–2.168** (0.061)	−0.848 (1.381)	0.813 (2.793)	0.902 (3.077)	1.549 (1.323)	2.668 (1.673)
District access to electricity	0.349** (0.141)	0.411** (0.17)	−0.176 (0.277)	−0.119 (0.267)	−0.182** (0.091)	−0.163 (0.113)	0.034 (0.253)	0.07 (0.251)	0.018 (0.116)	−0.057 (0.132)
Class size		0.012 (0.075)		0.057 (0.103)		0.054 (0.042)		−0.301** (0.107)		−0.079 (0.05)
SGB parental representation		0.059 (0.525)		1.157 (0.816)		−0.333 (0.338)		–1.441* (0.78)		−0.068 (0.394)
Teacher absenteeism		–2.53* (1.524)		−0.782 (2.24)		1.207 (0.984)		−0.3316* (2.083)		0.17 (1.12)
Teacher salary		−0.000 (0.000)		0.000 (0.000)		−0.000 (0.000)		0.000 (0.000)		0.000 (0.000)
Observations	94	77	114	95	101	83	109	90	108	89
R-squared	0.212	0.253	0.096	0.192	0.202	0.255	0.063	0.252	0.107	0.239

Notes: Data presented as regression estimates (standard errors). Quintile five is omitted due to collinearity with other quintiles. *Significant at the 0.1 significance level. **Significant at the 0.05 significance level.

For the PEM variables, the regressions produce insignificant results for total resource wealth in all the outcome variables. This finding aligns with many empirical studies which argue that resources alone do not matter. The lack of significance remains even when total resource wealth is interacted with socio-economic variables to check whether resource wealth interacts with other variables and whether it is only through the interaction that impact is made on education outcomes.

For the delay variable, there is one significant result with the expected positive sign, indicating that increased delays in disbursing funds are associated with a remarkable 2.2% increase in dropout rates at Grade 5 level. This significant result comes from a model that controls for both poverty variables and other sources of school and district inefficiencies. For LTSM spending, two models yielded significant results suggesting that an increase of R1 per student in LTSM spending is associated with a 0.016% decrease in Grade 1 repetition rates and a 0.005% decrease in Grade 7 repetition rates. Put differently, an increase of R100 (approximately US$10) per child is related with a 1.6% decrease in Grade 1 repetition rates and a 0.5% decrease in Grade 7 repetition rates. The two models that yielded statistically significant results are ones that only controlled for poverty indicators. This may suggest that inclusion of certain school inputs captures some of the effects of LTSM spending.

With regard to the robustness of this finding, the Durban–Wu–Hausman post-estimation test (see Cameron & Trivedi, 2009) is undertaken to assess whether LTSM is indeed endogenous. The null hypothesis for the test is that the variables are exogenous. The p values emerge as insignificant and therefore the hypothesis cannot be rejected. This demonstrates that the exogeneity of the LTSM variable cannot be rejected. If the LTSM variable is indeed exogenous, the regression outputs should be efficient. The OLS results showing LTSM spending to be significantly associated with outcomes, specifically Grade 1 and Grade 7 repetition rates, thus remains a valid finding.

Another interesting finding from Table 3 is that the factors which influence repetition rates are largely different from factors that influence dropout rates. Repetition rates tend to be influenced by certain socio-economic factors such as low poverty quintiles and access to electricity at the district levels. Dropout rates, on the other hand, tend to be more influenced by governance and school-related efficiencies such as teacher absenteeism, proportion of parental membership on school governing bodies and class size.

6. Conclusion and policy recommendations

These results suggest that both technical and allocative efficiency matter for primary education outcomes. The study generates a number of policy implications. First among these is the need for better management of resources at the sub-national level. My research reveals that even though the misappropriation of funds (leakage) is not correlated with outcomes, it occurs in 13% of the sampled schools in Gauteng province and 28% of schools in the North West province. These are potentially funds that could be allocated towards learning and teaching materials.

Secondly, it is important that pupils and teachers receive adequate learning and teaching support materials. The results have shown that LTSM spending is significantly related to repetition rates at early and later stages of primary education. A valid concern from a public finance perspective is whether expenditures of R100 per child to attain 0.016% and 0.005% decreases in Grade 1 and Grade 7 repetition rates, respectively, demonstrate good value for money.

Thirdly, delays in the disbursing of funds to schools warrant attention. Although the vast majority of schools received their allocations between two weeks and two months after the start of the school year, I found evidence that delays in remitting funds were negatively associated with Grade 5 dropout rates. The results suggest that eliminating delays in fund disbursements can result in a 2.2% decrease in Grade 5 dropout rates.

Fourthly, the study shows that while repetition rates are strongly influenced by the levels of poverty at the district level, school and district inefficiencies are associated with pupils dropping out. From a policy standpoint, school and district authorities can make quick wins in reducing dropout rates by improving efficiency and the quality of services. A reduction in repetition rates requires a more systemic change to improve socio-economic conditions at the district level, and therefore takes considerable time.

It is clear from the results that PEM matters for education outcomes. This study corroborates other empirical findings, which argue that merely increasing expenditure does not improve outcomes. Good management of public funds, specifically allocating funds towards priority needs such as learning and teaching materials, and disbursing funds on time matter for education outcomes.

Appendix A. The public expenditure tracking survey and the quantitative service delivery survey instrument

The public expenditure tracking survey and the quantitative service delivery survey were applied at two levels: primary schools and district education offices. There were therefore two distinct questionnaires used in this study. For the primary schools survey, the head teacher was the main respondent. A deputy head teacher was an alternative respondent as these two individuals are best placed to have a global view of the affairs of the school. In the rare event that these two educators were not available, either a long-standing teacher or a senior administrator was interviewed. The district survey was designed to have the District Director as the main respondent, with allowance for the heads of various units to complete relevant parts of the questionnaires. Each of these two surveys are briefly elaborated upon below.

A.1 Primary schools survey

There were 10 sections in this survey. The first part dealt with identification. Here, data were collected on basic information about the school; for example, its geographic location, number of students, whether it was a no-fee or fee-charging school and the school's poverty quintile ranking. Section 2 dealt with student outcome indicators such as enrolment, pass, repetition and dropout rates. Section 3 pertained only to schools that charged fees. There were questions on the amount of fees and percentages of schools exempted either fully or partially.

Section 4 asked about the experience and qualifications of principals and teachers as well as additional information about teachers such as their salary and the degrees of absenteeism. School facilities such as adequacy of school furniture, toilets and drinking water were covered in Section 5. Proximity of the schools to clinics and public transportation were covered in Section 6. In Section 7, questions were asked about the existence of school governing bodies and parent teacher associations; the top issues discussed at these meetings, the frequency of the meetings and the extent of parental participation in these bodies. Section 8 dealt with supervision of schools by district education inspectors and accountability of schools to district authorities in terms of submitting academic reports and audited financial reports. The last two sections dealt with finance, where sources of funding were requested in addition to the timeliness of the disbursements and finally what the school spent its money on and how much.

A.2 District Education Office surveys

The survey collected a lot of information on the characteristics of the District Education Office; human resource capacity; supervisory and advisory roles; financial disbursements; among others. Some of the questions were designed as cross-triangulation to verify or supplement information received from the schools. For instance, there were sections that dealt with delays in receiving funding and also on leakage (the difference between the amount schools were supposed to receive and the amount actually disbursed to schools).

A.3 Interviews

Both surveys were designed for face-to-face interviews, although a few of the school surveys were done telephonically. All interviews at the district education offices in the two provinces were conducted face to face by the author. For the primary school surveys, five research assistants in Gauteng and five in North West – a total of 10 – were recruited and trained to conduct the survey of 175 schools.