South Africa's economics of education: A stocktaking and an agenda for the way forward

Abstract

This paper reviews some of the existing economics of education literature from the perspective of South Africa's education policymaking needs. It also puts forward a suggested research agenda for future work. The review is arranged according to five areas of research: rates of return, production functions, teacher incentives, benefit incidence analysis and cross-country comparisons. Production functions, especially if translated to cost-effectiveness models, can point to important policy solutions. Teacher incentives is a policy area that is in need of a better theoretical and empirical basis. Rates of return are difficult for policymakers to interpret, but suggest a need for a qualification below the Grade 12 level. While benefit incidence analysis can demonstrate large improvements in the equity of public financing, cross-country comparisons reveal that not only is the distribution of schooling outcomes particularly unequal, on average it is well below what the country's level of development would predict.

Keywords:

• economics of education
• rates of return
• production functions
• teacher incentives
• benefit incidence analysis

JEL classification:

• H52
• I21
• I28

1. Introduction

Psacharopoulos, arguably one of the founders of the current economics of education tradition, observes that ‘[i]n the field of education, perhaps more than in any other sector of the economy, politics are substituted for analysis’ (1996:343). This problem in the education sector is conceivably brought about by three factors: an absence of relevant analysis, analysts who are unsuccessful in communicating their findings to the policymakers, or policymakers who resist paying attention to the analysts. This paper examines the first two factors in the South African context.

The paper takes stock of the economics of education literature that is influencing, or should influence, South Africa's education policymaking, through reference to a few key texts, though by no means all the available literature. Gaps in the literature are identified on the basis of assumptions about what policymakers need. The bias is towards a utilitarian view of the literature: it should inform policymaking and development in rather explicit ways. Less policy-oriented and more academic pursuits in the economics of education field are undoubtedly important, but they are not the subject of this paper. The discussion of the literature is organised according to five models or areas of research: rates of return, production functions, teacher incentives, benefit incidence analysis and cross-country comparisons. The paper concludes with a tentative research agenda for the economics of education in South Africa.

2. Rates of return

The unconditional relationship between earnings and years of schooling in South Africa points to an average increase in earnings of around 22% for every additional year of schooling completed in the range of two to 11 years of schooling, and a large increase of around 125% associated with the difference between 11 and 12 years of schooling, in other words with having attained Grade 12. (This finding is based on the authors' analysis of the Statistics South Africa 2005 Income and Expenditure Survey data, focusing on anyone who reported earning an income.) This kind of unconditional analysis suffers from two weaknesses. Firstly, the net benefits are not clear, because the cost, both private and social, of completing more years of schooling is not taken into account. Secondly, other factors such as years of experience, gender and (in particular in the case of South Africa) race, which may play a separate role in determining income, are ignored. Two distinct methods are commonly used to overcome these two weaknesses, though it is rare to find both weaknesses addressed within the same analysis. Herein lies some of the confusion that surrounds rates of return to education. A further problem is the fact that the policy implications of rates of return analyses are often not explored, or they are explored in a manner that is too rudimentary to be helpful to policymakers (Glewwe, 1996:283).

The first of the two methods, which has been called the ‘elaborate method’ (Psacharopoulos, 1981:322), uses the same internal rate of return calculation that would be used to calculate the return on a non-education investment. This method considers both income benefits associated with more education, and the private and public costs of education. Psacharopoulos & Patrinos (2004) argue that a cross-country comparison of annual rates of return, where these rates are based on the elaborate method, reveals two clear patterns. Firstly, primary schooling yields better returns than secondary schooling, which in turn yields better returns than tertiary education. Secondly, rates of return are higher the less developed a country is. Education rates of return for South Africa published by Psacharopoulos & Patrinos (2004:125) using the elaborate method are too old and unrepresentative to be useful for South African education policymakers. They are based on 1980 data for African residents of Durban (Psacharopoulos, 1993:41). It would seem that no subsequent rates of return estimates using the elaborate method have been published for South Africa.

The second method, generally referred to as the Mincerian approach, uses an earnings function to examine the relative effects of years of schooling and years of experience on earnings (Mincer, 1974:130). It considers the cost of formal schooling only in terms of the opportunity cost of income forfeited, not in terms of the direct private and public costs of schooling.

Strictly speaking, only the elaborate method yields proper rates of return values, though Mincerian beta coefficients for years of schooling are commonly also described as rates of return – Psacharopoulos & Patrinos prefer the term ‘wage effects’ (2004:116). The argument that the Mincerian approach produces rate of return-like statistics is sound, but it is important to explain to policymakers that this approach produces values that are often much lower than those produced by the elaborate approach.

There have been numerous rates of return analyses applying the Mincerian approach to South African data. Keswell & Poswell (2004) present their own data analysis, plus a meta-analysis of analyses carried out in previously published South African texts. (Though Keswell & Poswell refer to some of their models as ‘non-Mincerian’ [2004:841], they are essentially non-linear versions of the Mincerian approach.) Authors using the Mincerian approach typically add demographic variables not included in Mincer's original model (1974). Keswell & Poswell include race, gender and rurality in their 2004 analysis. Their main finding is that the returns to schooling increase with years of schooling, but only beyond 11 years, and that before Grade 12 each additional year of schooling yields almost no income returns. Put differently, the rate of return curve viewed from the years of schooling axis is convex. This is in stark contrast to the finding by Psacharopoulos & Patrinos (2004) that returns are highest at the primary level and, implicitly, that their shape is concave. It is important to view this not as a fundamental dispute between economists over the effects of education, but rather as a natural outcome of using different methods and attaching different meanings to the term ‘rate of return’. Crucially, the Mincerian approach does not take into account the high direct cost to the household of tertiary education.

From an education planning perspective, it is useful to view the typical Mincerian analysis in the light of labour market signals produced by education qualifications. The very sharp increase in the returns to schooling at the point where 12 years of schooling are completed, as seen in Keswell & Poswell's (2004) sharp ‘take-off’ in the rate of return at or one year after Grade 12, is to a large degree associated with the possession of a Grade 12 certificate. This certificate, which is the only standardised qualification issued in the South African schooling system, may put pressure on the system to improve the skills and knowledge of pupils to an exceptional degree in the one or two grades preceding Grade 12, but at least part of the income advantage of having successfully completed Grade 12 (as opposed to Grade 11) must flow from one's possession of a crucial and widely recognised means for signalling to employers the value of one's human capital. Assuming that qualifications do influence earnings in a manner independent of actual education, an obvious question for the policymaker is how changing the system of qualifications, for instance by introducing a Grade 9 certificate (something that has been on the South African policy agenda for a while), might influence the profile of returns to years of schooling. Figure 1 illustrates how crucial a policy question this is. It shows that the recent trend has been for about 60% of young South African adults to have no qualification at all, and that there is no evidence (at least not in the graph) of a downward trend in this statistic.

Figure 1: Highest qualification held by age (2007)

There is a challenge, not just in South Africa, to make the policy implications of patterns seen in rates of return analyses clearer. One part of this challenge is to examine how the structure of the country's qualifications system affects decisions and earnings in the labour market. Glewwe's study of data from Ghana (1996), using a relatively rare combination of variables covering years of schooling, innate ability, knowledge acquired through schooling and qualifications held, approximates the type of analysis that policymakers ideally require. Here, the separate effect of qualifications is measured. The analysis also highlights what should be obvious, but is often not, namely that what pupils actually learn, rather than years spent at school, is what lies behind productivity and improvements in the earnings of individuals. Du Rand et al. (2011) reach similar conclusions in a study that is constrained by missing data. Inadequate data in South Africa on the competencies of individuals in the labour market remains a hurdle for better human resources planning in the country.

Finally, rates of return analyses can also be used to compare income returns to different types of education at the same level. Psacharopoulos (1993:48) provides a comparison of the rates of return for academic and vocational secondary schooling, and finds that when the higher costs of vocational schooling are taken into account, the returns to academic schooling are higher. What might the situation be in South Africa, where recent years have seen substantial growth in the budgets and enrolments of pre-tertiary vocational schooling (the ‘FET colleges’),1 and the introduction of a new vocational curriculum, but also concerns that students enter vocational colleges when they are too old (partly due to the absence of a lower level basic schooling qualification)? No published rates of return analysis seem to exist, though unpublished and exploratory analysis conducted by one of the authors of this paper suggests that when controlling for race and gender, and taking into account private and public costs, historically vocational training has yielded better rates of return than ordinary schooling. This suggests that South Africa, like many other countries (Bennell, 1996), does not conform to the global pattern described by Psacharopoulos (1993).

3. Production functions

While rates of return analyses are crucial in attempting to explain the external efficiency of education systems, production functions assume the same role with respect to the internal efficiency of education institutions. Production functions essentially aim to identify which education inputs, such as teacher qualifications, teaching materials, teaching time, and so on, have the largest effect on outcomes as measured by pupils' test scores. Economists have shown keen interest in exploring this model though, for a number of reasons discussed below, the reception by policymakers has been mixed.

Perhaps the most important reason to be sceptical about the use of a single production function study to inform policy is that the data used for the analysis were in most cases not compiled specifically for this kind of analysis. For this reason, Hanushek, a prolific analyst in this area, laments the fact that most production function analysis is ‘opportunistic’ (2002:12). In particular, the ideal of test scores from two points in time, allowing for a value-added approach, and data at the level of pupils (including socioeconomic data) and not just the school, are often not realised. Despite these problems, identifying what inputs emerge as important across several production function studies through meta-analyses, such as that produced by Hanushek (2002) for the US, has come to be regarded as a valuable process that can indeed inform policy. Of course, this solution presupposes the existence of a critical mass of studies from the country concerned.

For South Africa, a pioneering production function study is that conducted by Case and Deaton (1999:1078), who use pupil-level data from the 1991 to 1993 period. More recently, a number of studies have used pupil-level data to examine the production of learning results in primary schools, including Gustafsson (2007), Van der Berg (2008), Taylor & Yu (2009) and Spaull (2011). Studies focusing on secondary schooling have relied on school-level data: Crouch & Mabogoane (1998b), Van der Berg & Burger (2003), and Bhorat & Oosthuizen (2006). Curiously, no one appears to have used the opportunity offered by the 2003 TIMSS data (Mullis et al., 2004) to produce a pupil-level production function at the secondary level. A similar opportunity will arise when the 2007 TIMSS data are released. The South African findings regarding school inputs have been fairly intuitive ones. For instance, libraries, teacher housing in rural areas, and better teacher knowledge all advance pupil performance. The South African work has also confirmed what studies from elsewhere have found, namely that socioeconomic status, in particular the level of education of the pupil's parents, plays a large role, in fact much larger than is commonly believed. In most countries, socioeconomic status has a somewhat greater effect on the performance differences between schools than all school resource factors combined (OECD, 2007:171).

Arguably the biggest influence that production function studies around the world have had on policy relates to one factor that they have consistently said does not play a role in improving pupil performance, namely class size. This finding, obviously an important one from a budgetary perspective, has been taken seriously by policymakers. The question of class size illustrates the importance of distinguishing statistical significance from policy significance in an analysis. Certain models do in fact find class size to be a statistically significant predictor of pupil performance (see for instance Taylor & Yu, 2009), but even then, when coefficients are translated into financial values, it is nearly always found that reducing class sizes would be among the most costly of all the available policy interventions implied by the model, relative to the desired performance improvement. This translation of a production function into a cost-effectiveness model, by bringing in actual unit costs, is seldom done in the academic literature and is arguably one important reason why policymakers find production functions difficult to interpret. Pritchett & Filmer (1997) and Mingat et al. (2003:56) explain the methodology required for this translation.

On the matter of class size, there is an important South African caveat. South Africa's class sizes are exceptionally large even by developing country standards. To illustrate, 16% of the country's Grade 8 pupils in 2008 were in classes exceeding 55 pupils (Gustafsson & Patel, 2008:25). In countries such as Botswana, Malaysia and Egypt the figure is considerably lower. Data from other years and other grades confirm this pattern. One wonders whether the orthodoxy on class size, partly resulting from production function findings, has perhaps created a blind spot in education policymaking, considering how little policy attention has been devoted to excessively large classes. Case & Deaton (1999) do in fact argue strongly that in South Africa, at least in the early 1990s, over-sized classes were a noteworthy predictor of poor pupil performance. However, Crouch & Mabogoane's criticism (1998a) with respect to an earlier draft of the article still applies to Case & Deaton (1999): the results are insufficiently robust, with respect to R ² values and t statistics, to support the policy conclusions. More focused attention on the effects of extremely large class sizes, with good data, in the South African production function work may reveal that South Africa's exceptional situation allows for the identification of critical thresholds beyond which pupil performance is affected by class size.

A very practical application of the production function technique is demonstrated by Crouch & Mabogoane (1998a). In preparing a list of top performing schools, with respect to the Grade 12 examinations, for the Sunday Times newspaper they used both a traditional approach of simply taking the best results, and a more socioeconomically sensitive approach where they compare actual results to the expected results emerging from a production function. In the list using the first approach, only one historically black public school appeared in a list of the 10 best performing schools. In the list using the second approach, nine of the 10 best performing schools were historically black public schools. The second approach clearly helps the education administration to identify and praise schools that succeed in overcoming contextual and historical difficulties, and it helps to identify schools that should be the subject of more qualitative case studies aimed at finding out what makes these schools so successful and efficient. (In 2009 the Sunday Times again published a list of top schools, but this time only the traditional approach was used, resulting in only one historically black public school appearing in the top 10.)

A discussion of production functions can probably not avoid including a reference to multi-level modelling, also known as hierarchical linear modelling. This method, used for instance by Gustafsson (2007), has become popular, but has arguably made explaining production function findings even more difficult than when using the ordinary least squares method. Johnes (2004:647) finds multi-level modelling ‘computationally intractable’.

Much of the challenge with respect to production function analysis lies in working towards data collections from schools that are better suited to this kind of analysis. Crouch & Mabogoane (1998a) emphasise the need for better variables on school management in order to reduce the residual, or unexplained, part of South African production functions. One variable that is rare yet of great potential importance in a production function is a direct measure of teacher knowledge. Such a variable has become available, for the first time in South Africa, with the release of the 2007 SACMEQ (Southern and Eastern Africa Consortium for Monitoring Educational Quality) dataset, which includes teachers' scores in subject knowledge tests (Spaull, 2011). Two recent data collections with test scores from two points in time, the National School Effectiveness Study and a cross-country study covering just over 100 schools on either side of the South Africa–Botswana border, are currently being analysed and are likely to add immense value to the South African policy discourse. The advantage that the two-point data has over cross-sectional analysis is that it overcomes some of the endogeneity problems of the omitted variable variety – for example, the problem that test scores from just one point in time reflect pupils' innate ability and learning acquired from other sources, for instance other teachers. The additional information results in better estimates of the magnitude of the effect of the various inputs in the schooling process.

What is striking in South Africa is that the government's sample-based Systemic Evaluation testing programme, which produces a dataset that is exceptionally well suited for production function analysis focusing on South African policy questions, has barely been used for this purpose because researchers have not been able to access the dataset. This is unlike the situation in Brazil or the US, whose equivalent SAEB (System for the Evaluation of Basic Education) and NAEP (National Assessment of Educational Progress) datasets, respectively, are widely available to research institutions. The institutional problems that make data inaccessible in South Africa should be resolved in the interests of a greater volume and variety of analysis that can inform difficult education policy decisions.

Lastly, a word of warning by Hoenack (1996:332) on the influence of production function work on education policymaking deserves mention. Underlying this work is a quest for the right ‘recipe’ for effective schooling, so that this can be expressed in the right policies and budgets. This whole approach obviously reflects a rather top-down paradigm. An alternative paradigm states that the education authorities should simply insist on good learning outcomes, and monitor such outcomes, and let schools themselves find the right recipes. Both arguments have merit, depending partly on the kind of school one is focusing on. Clearly, production functions should not reinforce an overly top-down policy agenda.

4. Teacher incentives

Recent developments in the policy on teacher incentives in South Africa have been turbulent. A much publicised wage agreement with unions in 2008 stipulated the introduction of salary notch increases every second year for outstanding teachers, where the evaluation of teachers would be based on a mix of schools-based and external inputs. This element of the agreement was dropped in 2009 due to insufficient support from unions.

Economic analysis of teacher incentives has been conducted in two key areas. The first looks at the core teacher salary as an incentive for joining the teaching profession and staying there. Because teacher pay is to a large degree determined through a political process, and not in an open market, research into how it is determined becomes especially important. In economic terms, the determination of teacher pay in South Africa displays elements of both monopoly (there is in a sense one supplier, the teaching force represented by unions) and monopsony (there is largely just one buyer of teaching services, the state). The political nature of the wage negotiation process makes the risk high that teacher pay will be substantially higher or lower than it would have been if it had been more market-driven. In order to determine whether it is too high or too low, economists typically perform a conditional wage comparison between teachers and other professionals in the economy using household data. The analysis is subject to a number of complexities, including how to define the group of professionals against which teachers are compared, how ‘teacher’ should be defined and how teacher productivity should be dealt with. Teacher productivity relative to that of other professionals is virtually impossible to assess empirically using the household data that are typically used for this kind of work. Yet it is important to at least acknowledge this dynamic.

There have been at least four South African conditional wage comparisons dealing with teachers: Crouch (2001), Gustafsson & Patel (2008), Armstrong (2009) and Van der Berg & Burger (2010). The findings of the four are broadly similar. Crouch (2001) finds teacher pay to be more or less comparable to that of other professions, and argues that if there is an under-supply of teachers in South Africa, this is due more to an insufficient provisioning of teacher training than to a problem of poor pay dissuading prospective teachers. However, the pay scales are said to be insufficiently generous for older teachers, increasing the possibility that older teachers will leave the profession. Gustafsson & Patel (2008:18) explain that the administered nature of teacher pay tends to result in a situation where the actual spread of pay over years of experience does not match the official salary scales, because when the official scales change, they are not implemented retroactively, meaning that teachers' actual pay will to a large degree be a reflection of previously existing official scales. Like Crouch (2001) and Van der Berg & Burger (2010), they find that older teachers are under-paid, but conclude that more generous increments linked to years of experience introduced in 2008 will over the years eliminate this problem (2009 policy developments have, however, removed some but not all of this deferred generosity). Gustafsson & Patel (2008) emphasise that the employer needs to signal the existence of future incentives to prospective teachers in an active manner, since they are not visible if the prospective teacher simply looks at what older teachers are currently earning.

The second key area of analysis looks at incentives paid to those who perform exceptionally well. Despite such incentives being a highly topical policy issue, there has been little work of an academic nature in South Africa on this subject. There are arguably two things that should inform the policy process. One is better data on teachers' perceptions, for instance with respect to their pay, their sense of professional identity and how to improve pupil performance. To some extent such data are collected in programmes such as the Systemic Evaluation, which includes teacher questionnaires and allows for linking to pupil performance. But for a more in-depth understanding of the necessary performance incentives for teachers, the other thing that is probably required is a dedicated and nationally representative teacher opinion survey, perhaps along the lines of the OECD's TALIS programme (OECD, n.d.) In the absence of such data, the policy process ends up relying too heavily on the assumption that teacher unions are able to represent adequately what incentivises teachers, an assumption that will clearly not always be a sound basis for policy.

Furthermore, the policymaking process would benefit from a systematic stocktaking of lessons from abroad, and an understanding of the relevant theory, in order to counteract fairly pervasive misperceptions in the policy debates. One misperception which has arguably brought with it large costs in recent years is the notion that performance-linked incentives for individual teachers are easily realisable. The international evidence suggests that the nature of schools requires incentives to be largely group-based if they are to be practical. The notion that teachers will respond in predictable ways to incentives, even well-designed and group-based ones, must also be interrogated.

An important trend within economics of education has been the increasing emphasis on better understanding of cause and effect in policy areas, such as teacher incentives, where misunderstandings are common (63% of articles with the word ‘causality’ in the whole text in the Economics of Education Review are from 2005 or later, against 28% for the term ‘rate[s] of return’). We can see the beginnings of a critical stock of literature on the effects of performance-linked teacher incentives in developing countries, which tend to experience effects that differ from those in developed countries (see literature review provided by Bruns et al., 2011). Some of this literature draws from randomised controlled trials of different teacher incentive approaches at a local level. This method stands out for the degree of certainty it can offer with regard to how teachers will react to incentives. Even here, however, the policymaker should exercise caution. As Woessman (2011) argues, small-scale experiments cannot gauge the effect of new incentive programmes on long-term teacher recruitment and retention trends.

5. Benefit incidence analysis

A society where it is widely felt that people's opportunities in life are unfairly distributed cannot be a healthy society. In the economics literature, recent empirical evidence indicates that in developing countries where inequality is extreme, economic growth is retarded (Barro, 2000). As the foregoing discussion has suggested, social and income inequalities are perpetuated by unsound education policies that fail to educate the poor. Part of the solution lies in ensuring that sufficient public spending goes towards schooling for the poor, while another part lies in ensuring that improved spending on the poor is translated into better learning outcomes. Benefit incidence analysis focuses on the first of these two challenges.

Benefit incidence analysis typically uses a combination of unit cost figures from public accounts and data on the use of public services from household surveys to draw conclusions such as that 35% of public spending goes towards the poorest 20% of the population, and that public spending is thus pro-poor. Comparisons between countries using the same approach can reveal important patterns. For instance, Davoodi et al. (2003:21) argue that sub-Saharan Africa has been particularly unsuccessful at targeting public education spending towards the poor. For example, at the secondary level on average only 7.4% of spending goes towards the poorest 20% of the population.

In South Africa, Van der Berg (2005, 2009) has calculated benefit incidence patterns for a range of public services, including education. This analysis indicates that public spending in 2006 on primary and secondary schooling was clearly pro-poor, and well targeted by international standards (Van der Berg, 2009:13). It also reveals that between 1993 and 2006 there was a clear trend towards better targeting of the poor (Van der Berg, 2005:8, 2009:14). The public funding of tertiary education reveals a different pattern. As in virtually all countries, in South Africa this funding is pro-rich, though in South Africa it is even more so than elsewhere. Van der Berg (2009) explains that this finding could be influenced by an important data problem, namely that in the household data poor students living away from their families are likely to appear artificially better off than their families of origin actually are.

Van der Berg (2009) points to a few common areas of misunderstanding. The standard benefit incidence approach of examining the breakdown of public education spending by quintile of the population means that the age pyramid of different segments of the population will influence the results. Above all, if a larger proportion of the poor are young, which is the case in most countries, then a pro-poor pattern will emerge even if the state spends an equal amount on each pupil. In fact, if just public spending per enrolled pupil is considered, then public spending is slightly pro-rich, and not pro-poor as seen in the typical benefit incidence analysis (Gustafsson & Patel, 2006). It is obviously important for education policymakers to recognise the differences between the two approaches.

The finding that publicly funded education inputs, at least at school level, are more or less equitably distributed is important information for policymakers as it provides evidence that ambitious post-1994 policies to correct the grossly distorted spending patterns of the apartheid era have paid off. There is thus ample empirical justification for the current and rather strong policy shift away from education inputs and towards education outcomes. Clearly there are education input issues that must still be resolved, but devoting the bulk of the policy attention to outcomes appears completely justified and necessary.

Substantial public funding for the poor in South Africa would suggest that private inputs are low. The evidence suggests that this is the case. The analysis by Gustafsson & Patel indicates that overall around 8% of the funding of public schools was private in 2005, though for the poorest two quintiles it was around 2% (2006:71). Kattan & Burnett (2004) find that private inputs into public primary schooling in many other developing countries are at a substantially higher level: 21% in China, 30% in Ghana and 43% in India. They moreover find that policy analysis of private inputs into public primary schooling is often confounded by definitional problems. Many countries that claim to have abolished ‘fees’ in fact still demand that parents pay, for instance for textbooks, because textbooks are not regarded as a part of ‘fees’. In South Africa, at least anecdotally, the newly declared ‘no fee’ schools for the poor have in many cases simply renamed what were previously ‘fees’ as ‘voluntary donations’. Though private contributions to schooling in South Africa may not be large, at least by international standards, they receive considerable policy attention and warrant better analysis. For this, Statistics South Africa's Income and Expenditure Survey data would be useful (Stats SA, 2005). But apart from clarifying the numbers, it is necessary to gain a better idea of the reasons why even in poor communities parents often encourage private contributions to the school fund. It is unlikely that insufficient public funding is the only reason; there are probably also reasons relating to the way schools are viewed and used by their communities.

6. Cross-country comparisons

Even simple cross-country comparisons can be very informative for policymakers. Often these comparisons can correct misperceptions or at least nuance existing perceptions within one country. In the case of South Africa, they have indicated that although the top decile of pupils, in terms of test results, perform exceedingly well compared to the remainder of pupils, they in fact do not perform well compared to the top decile of other middle income countries (see for instance Mullis et al., 2004:34). Despite the common perception in South Africa that too many pupils drop out of school before completing 12 years, a cross-country analysis reveals that in terms of secondary school completion South Africa is slightly above the average for similarly developed countries (Gustafsson & Morduchowicz, 2008:28).

The increasing availability of internationally comparable data on quality of education has allowed for new insights into education and country development. Above all, these data have permitted economists to adapt traditional cross-country growth models so that they now include educational quality, and not just years of schooling. This adaptation has profound implications for education policy. In particular, it implies that focusing exclusively on increasing enrolments in developing countries, without explicitly considering improvements in pupil performance, can be misguided (Hanushek & Woessman, 2009).

Two areas of exploration with respect to cross-country analysis should interest education policymakers. Firstly, as summarised recently by Stiglitz et al. (2009), there are important debates about what country development indicators policymakers should be focusing on. The traditional economic growth or GDP per capita indicators are inadequate on their own and excessive focus on them can be dangerous, for instance if income inequality is ignored. Development models using non-traditional dependent variables such as happiness or freedom from poverty are still rare, partly due to data availability problems. One potentially valuable data source that has received little attention in this regard is the World Values Survey (WVSA, n.d.), in which South Africa has participated for three waves in the last two decades.

The second area of exploration is cross-country models designed to predict not a development indicator such as GDP growth, but educational quality as measured by standardised test scores. In such models critical explanatory variables would cover education policy choices made by particular countries. What is currently clear from the literature is that a policy intervention that does not appear to be strongly associated with better educational quality is higher spending per pupil (Hanushek & Woessman, 2007:60). In what is likely to become an important strand in the literature, Bishop (1997) points to the importance of having standardised examinations, while more recently Woessman (2011) finds evidence in the cross-country data of the value of performance-linked teacher incentives for improving learning outcomes. In this type of analysis the direction of causality must be a concern. For instance, do standardised examinations produce good educational performance, or is it rather that countries that perform well educationally are willing to have more standardised examinations? Here the policymaker should look for evidence of a proper examination of cause and effect of the kind provided by Hanushek & Woessman (2009), who use an instrumental variables approach for testing causation.

7. Conclusion

The five areas of research considered here do not of course cover the whole economics of education field. Two other important areas could have been included, had space permitted. One is the estimation of ideal unit costs in, say, primary schooling for countries at a particular level of development. Psacharopoulos (1996), in a paper which, like this one, explores a possible economics of education agenda, emphasises the need for work in this area. Another is teacher supply and demand analysis. Here too, little work has been done in South Africa. Crouch (2001) suggests a simple model for South Africa but, as he himself admits, it is merely exploratory.

An economics of education research agenda for South Africa, based on the discussion in the preceding sections, might look as follows.

Rates of return: Here a clearer sense of what the typical rates of return analyses mean for policymakers is important. Perhaps some atypical rates of return analyses that took into account the qualifications that people have obtained would throw new light on what should be done with the qualifications structure of the schooling system. Rates of return comparisons of general and vocational education could be an important empirical input when considering the current policy shift towards higher enrolments in FET colleges.

Production functions: Considering that these models are most informative for policymakers when there are many of them, it is important to build on the stock of existing local literature and to perform periodic meta-analyses. Policymakers are best served by production function analyses that incorporate a final cost-effectiveness step so that the meaning of the coefficients can be understood by non-economists. Class size thresholds are a matter that seems to deserve more focused attention. Ensuring that data are structured in a way that facilitates good production function analysis should be a priority.

Teacher incentives: Periodic comparisons of teacher pay with other professional pay will always be necessary. With regard to performance-linked teacher incentives, there is much work to do. What can be done without new data is to apply the knowledge that has been gained in other countries, with respect to theory, empirical evidence and policy design, to the South African context in order to clarify key issues around which there is still too much confusion. New data on the behaviour and preferences of South Africa's teachers would greatly assist in determining what kinds of incentives will work best.

Benefit incidence analysis: Here it is necessary to build on the existing stock of standard benefit incidence analyses to monitor the progressivity of public spending into the future. Further interrogation of the socioeconomic status of tertiary students might produce revisions of the country's very regressive distribution of public expenditure on tertiary education. Using the existing data, the incidence of private spending on education could be better investigated.

Cross-country comparisons: Exploring the relationship between education and non-traditional country development indicators (other than growth or per capita income) represents an exciting research frontier. The same can be said of cross-country models that explore how the education policy choices that countries make can affect education outcomes.

Acknowledgements

Both authors are lecturers on the economics of education course offered to education planners studying for the Professional Certificate in Education Finance, Economics and Planning at the University of the Witwatersrand. This initiative, funded by GIZ (German Society for International Cooperation), is targeted at sub-Saharan Africa. This article is partly informed by what has appeared important to practitioners attending this course. An earlier working paper version of this article can be found online: http://ideas.repec.org/p/sza/wpaper/wpapers105.html

Notes

¹FET colleges are institutions that may accept school leavers in the Grades 9 to 12 range. They were previously known as technical colleges.