Final Conference of the RTN Network Economic of Education and Education Policy in Europe (EEEPE)

Hosted by the Centre for Economic Performance, LSE Organized by Jo Blanden (CEP, London School of Economics; University of Surrey), Steve Machin (University College London, CEP, London School of Economics and CEPR) and Sandra McNally (CEP, London School of Economics) London, 13-14 March 2009

If You Pay Peanuts do You Get Monkeys? A Cross Country Comparison of Teacher Pay and Pupil Performance
*Peter Dolton (Royal Holloway, University of London), Oscar Marcenaro-Gutierrez (University of Malaga)

This conference is supported by: FP6 Marie Curie RTN (Contract no. MRTM-CT-2003-504796) The views expressed in this paper are those of the author(s) and not those of the funding organization(s) or of CEPR, which takes no institutional policy positions.

PRELIMINARY FIRST DRAFT: DO NOT QUOTE

If You Pay Peanuts do You Get Monkeys? A Cross Country Comparison of Teacher Pay and Pupil Performance.
Peter Dolton 1 Oscar D. Marcenaro-Gutierrez 2

Abstract Why are teachers paid up to four times as much in some countries compared to others and does it matter? Specifically, is the quality of teachers likely to be higher if they are paid higher up the income distribution in their own country, and are pupil outcomes influenced by how well their teachers are paid? The policy relevance of these questions deserves attention. This paper considers the determinants of teachers salaries across countries and examines the relationship between the real (and relative) level of teacher remuneration and the (internationally) comparable measured performance of secondary school pupils. We use aggregate panel data on 39 countries published by the OECD to model this association. The results confirm the importance of market supply forces in the determination of teacher pay and suggest that relative (and absolute levels) of teacher salaries exert a powerful influence on pupil performance.

Address for correspondence: Prof. Peter Dolton Department of Economics Royal Holloway College University of London Egham Surrey, TW20 0EX United Kingdom E-mail: Peter.Dolton@rhul.ac.uk

Dr. Oscar D. Marcenaro-Gutierrez Department of Applied Economics (Statistics and Econometrics) University of Malaga Malaga 29071 E-mail: odmarcenaro@uma.es

Acknowledgements:
The authors wish to thank seminar participants at LSE and Royal Holloway for comments, especially Victor Lavy and Josh Angrist.

1 2

Royal Holloway University & Centre for Economic Performance, LSE. E-mail: Peter.Dolton@rhul.ac.uk University of Malaga. E-mail: odmarcenaro@uma.es. 1

If You Pay Peanuts do You Get Monkeys? A Cross Country Comparison of Teacher Pay and Pupil Performance. 1. Introduction Why do teachers in Holland earn four times what those in Israel and why are teachers in Korea paid at the 80th percentile in their countrys wage distribution whereas those in Norway are paid at only the 45th percentile? Do these massive variations in the way different countries treat their teachers matter for the educational outcomes of their children? Specifically, is the quality of teachers likely to be higher if they are paid higher up the income distribution in their own country, and are pupil outcomes influenced by how well their teachers are paid? The answers to these questions are of central policy importance and it is possible that we can learn a lot about the relationship between teacher quality in pupil outcomes from cross national comparisons that we cannot learn by looking at individual countries. This paper considers the determinants of teachers salaries across countries and examines the relationship between the real (and relative) level of teacher remuneration and the (internationally) comparable measured performance of secondary school pupils. We use aggregate panel data on 39 countries published by the OECD to model this association. In seeking to model the relationship between teachers pay and pupil outcomes one seeks a potential causal mechanism to link them. There are basically two potential explanations as to why they may be causally linked. The first is that higher pay will induce the more graduates into the profession making teaching more competitive. Once recruited, higher relative pay will provide teachers with more incentive 3 for them to make more effort to improve the educational outcomes of the children they teach. The second mechanism is more subtle namely that improving the pay of teachers improves their standing in the countrys wage distribution and hence the status of teaching as a profession in that country. As a result of this higher status more kids in that country will want to be teachers in the future and as a consequence this will make more teaching as a profession more selective and hence recruit the more able individuals. We will not be able to distinguish between these different causal mechanisms but essentially this does not matter. The central proposition is that better pay in teaching induces a higher quality of graduate into the job and that this will improve student performance. We seek to establish whether aggregate country data supports this hypothesis, and if so, can we retrieve an estimate of the relative size of this effect. A final and

In some countries this could possibly operate directly through incentive or performance related pay. For example the UK has had a performance pay threshold for classroom teachers since 2000. See Dolton et al (2003). 2

more difficult question is whether we can interpret the effect we find as a causal effect or simply a correlation. We return to this issue at some length. A well-educated labour force is vital for every economy. The human capital and endogenous growth models suggest that the level and/or quality of education in an economy directly affect a countrys growth 4. Hence one clear motivation for this study is to identify factors affecting the stock of human capital in the economy. Despite research interest in this topic, there is not much evidence regarding the quality of formal education and its link to human capital. In developed economies the demand for schooling has been rising and there has not been a sustained rise in the real pay of teachers 5. In the countries where teacher pay is not rising this is likely to have resulted in a more intense utilization of teachers. It may be affecting teacher recruitment and motivation, contributing, at least partly, to a lower quality of education provision. However, if the quality of education is affected by relatively worse teachers salary conditions, this could impact - in the long run - on economic growth by adversely affecting the process of human capital accumulation. The main emphasis of the literature on the relationship between school quality and pupil outcomes has been focused on class size. This means that other significant dimensions of resource input policy decisions have largely been ignored. That is the case of the relationship between some teachers attributes, such as a teachers highest degree attained or experience, and student achievement, which is not well-known 6. This seems to be shifting to some extent, as the knowledge and skills required by the teachers is a matter of increasing concern from a policy perspective in most countries. The impact of teacher quality is far larger than any other quantifiable schooling input 7 (Rivkin et al, 2001, and Goldhaber, 2002). In fact, Goldhaber suggests that it is key to attracting and retaining high quality teachers, because of the link between teacher salaries and student outcomes 8.

This argument is strongly supported by the Human Capital Theory (HCT), which relies on the assumption that more educated workers enjoy greater productivity, in that manner contributing to the countrys economic growth. 5 This is verified by looking at Figure 2 and considering the following countries: USA, UK, Australia, Denmark, Netherlands, New Zealand, Austria, Ireland, Sweden, Switzerland, Norway, Finland, Italy, Belgium, and Germany. 6 This being partly due to the difficulty to conduct a cost-benefit analysis in the education system, as in any notfor-profit sectors. 7 Hanushek (2003) provides a wide and critical review of the evidence around the world on the effectiveness of input-based schooling policies. 8 Hanushek et al (1998) found that the best teachers can produce an additional year's worth of learning out of their students compared to the least effective teachers. 3

Another important factor which has been well documented, particularly for the US (See Goldin 2006 and Perlmann and Margo 2001) is that teaching used to be a more or less exclusively feminine profession as there was relatively few other occupations open to women. This means that many of the high ability women were drawn into teaching. In more recent years as other professional labour markets opened up to women this has increasingly the high ability women have been drawn into alternative occupations. As a result their places have been taken by lower ability males quite possibly to the detriment of pupil performance. Comparative international analysis could be a useful instrument to illuminate this less explored strand of the ongoing educational debate, particularly in a context in which the national policies do not seem well orientated in that they are not sensitive enough to substantiate the link between good teachers and effective student learning. Within a country it is difficult to get a satisfactory measure of teacher quality as most teachers are paid (conditional on experience) very similarly on the same pay scale. This is not true if we consider cross country data as Figures 1a, 1b and 1c show how variable teachers pay is across countries. We propose to use this variability to try and identify the effect of teachers salaries on pupil educational outcomes. Our argument is that within a country the relative wage in teaching will, on average, determine the ability and hence the quality of teachers on gets. Clearly if teachers are paid in the top 20% of the earnings distribution of a country then one would expect that this profession would attract some of the most able graduates in the country (provided that the compensating wage differential of poor working conditions was not too large). Likewise, if teaching is really poorly paid in relative terms then one would expect that only the less able would end up in the job. (The exception of course is that some individuals may be highly motivated by teaching as a vocation and not care about the remuneration but we assume that this will not be a totally dominating effect.) So, our story is that with cross country data we expect that each country will get the quality of teacher it deserves by setting the relative wage in teaching. As a result we suggest that in cross country data the relative wage of teachers is a sufficient statistic for their average quality. The low quality of education and especially basic education is a severe problem in most developing countries. Although this problem does not exclusively affect poor countries, (see Hanushek and Luque, 2003), it is obvious that there is a clear positive relationship between spending per student and GDP per capita. Poorer countries tend to spend less per student than richer countries. Nevertheless, there are several countries, e.g. Japan and the
4

Netherlands, which have comparatively moderate expenditure per student and have the highest level of students attainment 9. In the context of pressures to reduce overall government spending, there is a growing public concern about the allocation of resources to educational institutions. This has direct implications in terms of education and teachers, to the extent that most countries spend a large proportion of their government budgets on education and a very high fraction of that spending goes for teachers' salaries. Hanushek et al (1998) estimated that variations in teacher quality account for at least 7.5 percent of the total variation in student achievement. This is a much larger share than any other school characteristic. Furthermore, Hanushek and Kimko (2000) demonstrate that observed differences in measures of labour-force quality based on student cognitive performance have a dramatic impact on productivity and national growth rates. The relationship between teacher supply and economic growth is of interest (see Tamura (2001)). We present some evidence of how economic growth and teachers supply conditions have interacted with each other in the OECD (and WEI) countries economy. To be precise, we have used teachers salary level as a proxy measure of teacher supply conditions. In this paper we will use two different ways of measuring teachers salaries to pick up the power of teachers salaries to purchase goods and their power to attract individuals to the teacher profession, respectively. This paper is organised as follows: in section two we provide a overview of the main contributions to the literature on teachers salaries followed. In section three we provide a description of the cross country data to be used in our econometric analysis. Section four sets up an educational production function framework for our analysis and the econometric identification issues in our analysis are discussed in section 5. The econometric estimation results relating to the factors affecting teachers salaries relative to GDP per head and average teachers' salaries per hour are evaluated in section six as are the results which show a link between teachers salaries and educational performance. Finally, in section seven we discuss possible policy implications and summarise the main conclusions.

2.

Review of International Literature on Teacher Quality and Pupil Attainment Competitive salaries and benefits for teachers are crucial to attracting and retaining

high-quality teachers (Murnane et al, 1991, Ballou and Podgursky, 1997). Research has shown that levels of compensation and criteria for awarding salary increases affect who goes

Using the performance in mathematics as proxy for educational achievement. 5

into teaching, who stays in teaching and for how long (Odden & Kelley, 1997; Dolton, 1990, 2006, Dolton and van der Klaauw 1999). Without motivated and qualified teachers the positive impact of teaching on student learning may be constrained. As in any other occupation, quality can only be demanded and motivation can only be expected if working conditions, including salary, are attractive. Of course good working conditions alone do not guarantee high-quality education but poor pay in teaching is unlikely to attract high quality teachers and secure favourable pupil outcomes. Figlio and Stone (1997), Figlio and Lucas (2000) and Ferguson (1991) find that higher salaries are associated with better-qualified teachers. Advanced economies have faced an increasing demand for skill outside teaching, which has pushed up its price. This reduces teacher quality and may raise teacher quantity, which becomes relatively cheaper than quality. This helps explain why advanced countries experience a growth in primary school teacher-student ratio, while experiencing reductions in relative teacher quality (Lakdawalla, 2007). The low quality of education and especially basic education is a severe problem in most developing countries (Glewwe, 1999). Although this problem does not exclusively affect poor countries, Hanushek and Luque (2003), show that there is a clear positive relationship between spending per student and GDP per capita. Poorer countries tend to spend less per student than richer countries, nevertheless, there are several countries, e.g. Japan and the Netherlands, which have comparatively moderate expenditure per student and still remain among the countries with the highest level of students attainment. The debate on different measures of school quality has predominantly focused on class size, sidelining other significant dimensions of input policy decisions, like teacher remuneration. Until recently the relationship between teachers attributes, such as a teachers qualifications or experience, and student achievement, was not well-known. This seems to be shifting, as the knowledge and skills required by teachers is becoming a matter of increasing concern from a policy perspective in most countries. Recent studies suggest the impact of teacher quality is far larger than any other quantifiable schooling input (Hanushek et al, 2005, and Goldhaber, 2002). Hanushek (2003) provides a wide and critical review of the evidence around the world on the effectiveness of input-based schooling policies. In fact, Goldhaber asserts that it is key to attract and retain high quality teachers, because of the link between teacher salaries and student outcome. Hanushek et al (1998) found that the best teachers can achieve an additional year's worth of learning from their students compared to the least effective teachers. Further, they estimated that variations in teacher quality account for at least 7.5 percent of the total variation in student achievement. This is a much larger share than any other school characteristic. Much of the evidence cited above uses data from
6

specific US states like Texas. There is relatively little research which takes a US wide perspective (Flyer and Rosen 1997 being one dated exception and Temin 2002 being a qualitative overview). A well-educated labour force is vital for every economy. The human capital and endogenous growth models suggest that the level and/or quality of education in an economy directly affects a nations growth. (see Hanushek and Woessman (2007) for a summary of recent evidence.) Hanushek and Kimko (2000) demonstrate that observed differences in measures of labour-force quality based on student cognitive performance have a dramatic impact on productivity and national growth rates. There is also a considerable literature focused on the factors affecting the stock of human capital in the economy. However, there is not much evidence regarding the quality of formal education as a relevant dimension of human capital. Specifically, the relationship between teacher supply and economic growth is of interest (see Tamura, 2001). This research will examine the evidence of how economic growth and teacher supply conditions have interacted with each other in the OECD countries.

3.

Data Sources: Teacher Salaries and Internationally Comparable Pupil The single most comprehensive source of comparative information about teachers in

Attainment Scores. different countries around the world is in OECD (2001b). Together with the Education at a Glance publications from the OECD (1996-2007) it is possible to build up a picture of what has happened with teachers in different countries. These publications provide information on teacher salaries both starting salaries and salaries at the top of the profession as well as after 15 years of teaching experience. The same sources also provide country by country details of: educational expenditure, the percentage of teachers who are women, teacher work hours, educational personnel as a percentage for the labour force and instruction time by subject. In addition much of this information is available separately across primary, lower secondary and upper secondary school sectors. Data of the kind examined for the USA and the UK by Dolton et al (2003) relating to pupil numbers, teacher numbers, teacher wages and other factors relating to teacher supply are difficult to obtain on a consistent basis for most other countries for any reasonable time series. Hence, in order to understand what is happening across countries we examine evidence by pooling OECD (and WEI program) countries data. In this paper, we examine the

relationship between teachers salaries and some economic and educational variables for most of the OECD countries. 10 The data which are available from the OECD Education at a Glance publications allows us to construct a panel data set relating to 1995, 1996, 1998-2005 11. For most of these years up to 39 countries are observed. At the maximum this gives us a panel data set of 1170 observations (i.e. 39 countries, times 3 education sectors times 10 years). However, for some variables not all years are observed. This means that our resulting panel data is unbalanced. In this data basic characteristics of the educational system are observed (or derived). Our task is to explain the variation in teachers salaries across different countries by relative supply and demand factors having controlled for the basic cross country heterogeneity which can be observed. It should be noted that a major difficulty in any study of this sort is the existence of heterogeneity in the educational systems of the different countries that cannot be easily observed 12. A major concern is that there is considerable variation in teacher recruitment procedures in different countries this makes any comparison at international level more difficult. This is an inevitable shortcoming in the dataset that we have used here. Nevertheless an examination of this nature, i.e. using a cross-country set of data would be interesting. In such an analysis, each country would be at different points in the economic cycle and hence, any significant relationship between the variables representing teacher supply and the economic cycle in this data set would be evidence of a link between these two components. In this paper we have used two different ways of measuring teacher salaries: 1) the power of teacher salaries to purchase goods and services measured by converting salaries to U.S. dollars using purchasing power parities (PPPs), and secondly taking this level of teacher pay in real terms and normalising it to an hourly rate by dividing by the number of working hours to facilitate homogenous comparisons across countries. Teacher salaries converted using PPPs represent the sacrifice of other goods and services that a country gives up to support a teacher. In other words, teacher salaries expressed in U.S. dollars indicate the cost of teachers relative to the general market basket of goods.

The data is available for countries which participated in the WEI program. This program was carried out by the OECD and UNESCO, with the support of The World Bank. 11 There are unfortunately no data for 1997 published by the OECD. 12 An additional difficulty is based on the changes in policy and institutional settings in each country that have contributed to growth aside from the changes in inputs of physical and human capital (e.g. lower variability in inflation). 8

10

Another measure of the investment in teaching is the statutory teacher salary relative to the number of working hours per year. It provides a rough estimate of the cost of an hour of teacher instruction time across countries. The average statutory salary per teaching hour after 15 years of experience was, in 2001, $37 in primary education, $45 in lower secondary education, and $52 in upper secondary (general) education across OECD countries (OECD, 2003). For primary education, Philippines and Hungary have relatively low costs per hour of instruction; by contrast, costs are relatively high in Germany ($59, $69 and $77, respectively) and Japan ($68, $77 and $90). Salary and teaching time are only two components of teacher working conditions. The amount of professional development time supported by a school or district, student behaviour, participation in school decision making, class size, quality of facilities, and adequacy of resources are examples of conditions that could also influence a teachers desire to teach or not teach at a particular school. Many of these conditions, however, are very difficult to measure. As trends in teacher salaries have been affected by demographic shifts in the distribution of teachers ages, we decided to use all the salary data available to us. We therefore use the data on beginning teachers, teachers with 15 years of experience, and teachers on the top of their salary scale. This makes use of the maximum amount of information for each country for each year. In using all this data it is worth pointing out that this will then embody all the structural differences in the determination of teachers pay in each country. Specifically we will then have the both flat, declining and very progressive wage structures as well as data from countrys where teachers wages have been rising, constant or falling. We measure the nature of the countrys investment in education by the level of educational expenditure as a fraction of GDP and we control for the rate at which a country is growing, since clearly this will constraint its choice set of educational investment possibilities. The relationship between educational and total spending repeats itself in nearly all developing countries and across all continents. In Latin America, for example, countries, such as Uruguay, that experienced high economic growth generally increased public spending as a percentage of GDP, which led to hikes in educational spending of 5% and 4.3% per annum, respectively. Across OECD countries there is a broad range of expenditure per student. At the primary level, expenditure ranges from US$863 in Mexico to US$6713 in Denmark. Differences between countries are even greater at the secondary level. To get a clearer picture of the relative importance of the educational sector, in the WEI countries at least 4% of all employed persons work in the education system, but in some
9

cases (e.g. Tunisia) this figure rises to 7% and represents a high proportion of the labour force that have a tertiary qualification (OECD, 2001a). A quick look at the data allows us to check that educational expenditure on instructional services, below the tertiary level, account for the highest proportion of educational spending. What is more, changes in enrolments do not seem to have been the main factor driving changes in expenditure per primary and secondary student 13. The relative supply of teachers is measured by the number of teachers (and other educational staff) as a fraction of the labour force and the Pupil/Teacher ratio in the education system. An additional supply factor relates to the composition of the teacher stock in terms of the proportion of teachers in the country who are women. We also control for the number of teacher hours supplied in the country, as obviously fewer teachers can be compensated for by a lower number of teachers working more hours. The changing nature of the demand for teacher services is proxied for in this data by the demographic growth in the size of the population of school age. In addition we are able to control for the salary differences in the three education sectors: primary, lower secondary and upper secondary schools. The second part of our analysis is to examine the relationship between teacher remuneration and educational attainment. To this end we collected data to measure the quality of educational output by using results on the PISA (2000, 2003 and 2006) and TIMSS (1995, 1999 and 2003) tests across all countries for all available years 14 . The suggestion here is that since there is a relative shortage of teachers in these subjects in most schools then this might show up in the relative earnings of teachers if the fraction of time devoted to science and mathematics was higher. In the event neither of these two variables were significant in our regressions and since they reduced our sample size somewhat further we have omitted these results. The basic form of the salary data can be seen in Figures 1a, 1b and 1c which respectively examine the pattern of secondary school teacher salaries after 15 years in 2007 in PPP$ in Figure 1a, in relative terms to GDP per head in Figure 1b and relative to the position in the earnings distribution in Figure 1c. Although there is some concordance in the ordering of the countries in the three measures there is also some notable differences. Specifically whilst the USA is in the middle of graph in Figure 1a and 1c it is very low in Figure 1b. Likewise, whilst Israel is nearly at the bottom of actual salary in Figure 1a it is near the top in

But in the case of Spain where 10 per cent drop in enrolment has led to a significant increase in spending per student. 14 Several approaches were undertaken with PIRLS (2001, 2006) data, but we do not report results on that to the extent they only assess 4th grade students. 10

13

terms of relative salary in Figure 1b. Such differences will make consistent results found on the relationship between educational outcomes and teacher salaries logically more robust. Figure 2 provides a visual picture of how teachers salaries and their structure vary across countries and years. Specifically this figure graphs the percentile position of teachers wages in the income distribution of the country. In some countries it is clear that teacher salaries have been rising whilst in others they have been falling. The figure also makes clear how in some countries there is a large difference in teacher salaries at the beginning of their career to the end of their career (e.g. Thailand, Austria, Korea, and Hungary) whilst in other countries the teacher wage profile is very flat (e.g. Peru, Sweden, Finland, Denmark, Norway and Brazil.) This figure also nicely illustrates the variability of the data we have at our disposal by considering countries in a panel over years from 1995-2005. An indication of the data available to us from PISA and TIMSS is captured in Figures 3a and 3b. Here we order the standardised scores and their variability by country. Looking at Figure 3a and 3b we see there is a remarkable difference between the orderings. It seems that the countries with the highest standardised mean score are slightly more likely to have a low variability and vice versa. This impression is bourn out in Figure 3c where we can see a loose negative association between standardised score and the variability of the score. This relationship is worthy of some further investigation in our regression analysis. Figure 4 is the figure at the heart of our investigation. It plots the relationship between standardised scores and average teacher salaries by country. Here we can see quite a good positive association between these two variables of interest. Naturally this relationship needs careful analysis including all the conditioning information which is relevant but this basic scatter plots provides the rationale for a thorough investigation. Figure 5 provides us with a basic idea of the weak relationship that exists between class size (as measured by Pupil/Teacher ratios) and pupil performance. Its relatively weak association is reflected in the diverse literature which finds the evidence on the effect of class size and student performance is contradictory 15.

See, e.g. Akerhielm (1995), Becker and Powers (2001), Hedges et al. (1994a, 1994b), Hoxby (2000) and Krueger (1999). 11

15

4.

An Economic Model of Educational Outcomes. The most straightforward way to conceptualize an economic model of educational

outcomes is to consider the production of educational outcomes as a function of inputs and resources. Lee and Barro (2001) have suggested such a simple formulation:

Q Q( F , R)

(1)

Where Q denotes schooling outcomes, F family resources, and R resources used by schools. In this context the educational outcomes can be measured by the achievement of pupils or students on international tests. Cross national comparative analysis is then possible using the assessed learning achievement in specific areas like Maths and Reading and Science. Typically the effect of family factors on educational outcomes can only be estimated using individual child level data and even then it is hard to identify a causal impact of these family factors separately from resource factors. Using cross country data the family factor cannot even be considered. The bulk of the literature which uses international cross country data has focussed on the use of resources to try and explain the variation in pupil outcomes. We seek to extend this literature and specifically focus on the role of teacher remuneration. In this context, teacher pay or the amount spent on teacher pay, is just one of the resources used in the production of pupil outcomes. Alternative resource inputs to this production process are total teaching hours, and pupil teacher ratios. This is the case because clearly the educational spending authorities can choose to trade off: higher teacher pay for larger classes or indeed fewer teachers and make them teach more hours. Hence teacher pay as a resource input into teacher has a clear relationship to pupil teacher ratios and teacher hours. Inferring the optimal combination of teacher hours, teacher pay and pupil teacher ratios in these trade offs is exactly what educational authorities must seek to do if they wish to maximise the outcome of pupil attainment. The logic of teacher pay as an input into the production of pupil outcomes is that higher pay will facilitate the recruitment of higher ability graduates into the profession. Clearly if teacher pay is low relative to other professions then the quality of new recruits will be lower than in those alternative professions. As pointed out by Zabalza (1979), who developed an occupational choice model for teachers in UK, the decision process for entrants and leavers into the teacher labour market is the earning profile, measured as its present level and its rate of progress over time. The individual will select teaching or other alternative, depending on which of them presents the higher expected return. He concluded that the occupational decisions of men resulted more
12

influenced by career prospects while women were more sensitive to changes in immediate earnings. Thus, the path to teaching careers can be interpreted as a series of sequential decisions affected by ability and gender. A slightly more sophisticated model of the determination of aggregate pupil outcomes at the country level would consider the following simultaneous relationship:
T T ( S , D, H , P, G , E ) Q Q ( H , P, T )

(2) (3)

where
T is teacher wages S is relative teacher supply (as measured by the proportion of teachers in the labour

force)
D is relative changing demand (as measured by changing pupil numbers) H is total teaching hours G is aggregate country economic growth E is aggregate educational spending P is the Pupil/Teacher ratio Q is average pupil performance.

In this context we are interested in modelling the outcome of pupil performance (Q) across countries explicitly acknowledging the alternative resource options open to each country vis--vis the mix of teaching contact hours (H), Pupil/Teacher ratios (P) and teacher salaries (T) within the overall spending and resource envelope of the country. So clearly a country could decide to reduce its teacher salary budget by having fewer teachers teach to larger classes or indeed decide to give teachers higher pay in return for teaching larger classes. These are all explicit options open to the educational decision makers. We assume that these decision makers will attempt to maximise Q subject to their available envelope of educational resources. The educational spending envelope will be determined by the aggregate level of economic growth (G) and the fraction of a countrys GDP that is devoted to educational expenditure (E). To a large extent E and G from the perspective of the educational administrator - will be pre-determined as they will not be able to change E or G to effect an increase in teacher pay they must work within the framework of the resources given to them by the political system within the country. In most countries the Pupil/Teacher ratios and teaching contact hours are chosen by educational authorities based on administrative conventions and political directives from
13

government and to this extent, at the level of the individual country (in the short run at least), they are largely exogenously determined. Teacher pay on the other hand is variable and can be changed in the short run subject to the supply and demand pressures specifically the number of trained teachers who are available (S), and the demographically determined number of pupils to be educated (D). The initial approach will be to estimate the two equations (2) and (3) separately to attempt to explain, firstly, the variation in teacher pay across country, and then the variation in pupil performance across countries. Inevitably the potential problem is one of identification. One can consider that this is a problem of the same unobservables influencing teacher pay as influence pupil outcomes. It could be that there are unmeasurable, inherent characteristics of certain good educational systems which make pupils learn effectively and reward teachers well. This may itself be determined by the importance attached to education by politicians, and voters. Alternatively one can consider this as a simultaneity issue i.e. how can we be sure that it is not the prospect of teaching well performing kids that tempts recruits into teaching rather than it being well paid teachers who facilitate good pupil performance?

5.

Econometric Model Identification

Our econometric analysis seeks to identify the relationship between teachers salaries and its determinants and average pupil attainments and its determinants. Since we are fortunate to have panel data for our estimation, identification of the parameters of interest relies on i) cross country variation in the dependent variables across countries over time, ii) that unobserved heterogeneity for each country is constant across time or that such unobserved heterogeneity, if it varies across time, is orthogonal to the explanatory variables in our regressions. Under the above conditions, fixed effect or random effect estimation will provide consistent estimates of the marginal relative importance of the determinants of average salaries and average pupil attainment. The scope for there being considerable country specific unobserved heterogeneity which changes across time is strictly limited. This is because all the data we use is measured at an average level within the country so that many idiosyncratic observations within a country (for example by school) will be averaged out. There is also limited scope for changing unobserved heterogeneity within a country (or state) over time since most of the political and educational institutions of each country (or state) will remain constant and relatively unchanged over the 10 year period of our data.
14

Examination of Figure 2 and Figure 3a shows us that there was considerable variation in both average teacher salaries and average pupil attainment across countries. This gives us some confidence that we are seeking to estimate models on changes in the underlying earnings structure on the changes in the educational performance of children when such variation does exist. Our attempt to identify the influence of teacher wages on pupil outcomes will be fairly eclectic. Firstly we estimate equation (3) as a reduced form equation relying on the panel data variation of countries across time to attempt to identify the teacher wage effect. Secondly we seek to model equation (2) and (3) simultaneously via 2SLS IV. In this model our assumption will be that supply and demand for teachers determine teacher salaries but that they play no role in the determination of pupil outcomes other than their indirect effect through changing teacher pay. These exclusion restrictions facilitate the estimation of the model in (2). Our third approach will be to consider how the growth in teacher wages (within a country) over the life cycle affects pupil outcomes where this is akin to a generalised model of differences in that we will be averaging over teacher wages at the beginning of a career, 15 years into a career and at the end of a career. Finally, since we actually have a variety of pupil outcome measures in different subjects (at each point in time) we can also model these outcomes jointly by Seemingly Unrelated Regressions (SUR). Whilst with observational data it is impossible to prove there is a causal effect of teacher wages on pupil outcomes we suggest that if our estimated marginal effect stands up to all of our different estimation methods it is quite possible that there is some systematic relationship between teacher pay and pupil outcomes which may be more that a simple statistical correlation. Notwithstanding this logic it must be recognised that the estimated relationships do not identify causal parameters, but rather reveal the strength and significance of statistical association between the variables of interest in our underlying reduced form relationships.

6.

Empirical Results

The Determination of Teacher Pay.

In the first part of the empirical analysis we seek to explain the determination of teacher wage across all our countries. As explained in section 3 there are a variety of ways of measuring a teachers salary and we seek to examine the determinants of each proposed measure of salary. We measure the nature of the countrys investment in education by the level of educational expenditure as a fraction of GDP and we control for the rate at which a country is growing, since clearly this will constraint its choice set of educational investment
15

possibilities. The relative supply of teachers is measured by the number of teachers (and other educational staff) as a fraction of the labour force and the student/teacher ratio in the education system. An additional supply factor relates to the composition of the teacher stock in terms of the proportion of teachers in the country who are women. We also control for the number of teacher hours supplied in the country, as obviously fewer teachers can compensated for by a lower number of teachers working more hours. The changing nature of the demand for teacher services is proxied for in this data by the demographic growth in the size of the population of school age. In addition we are able to control for the salary differences in the three education sectors: primary, lower secondary and secondary schools. Table 1 presents the regression estimates of the aggregate factors that have an impact on the teachers salaries measure in two ways; Firstly, as average teachers' salaries per hour, secondly as salary expressed in PPP. Separately we examine in Table 2 relative pay of the teacher expressed as the percentile that the person is paid at in the earnings distribution of the country in question 16. In these regressions we make use of all the data on teachers earnings on starting in the career, after 15 years and also at the top on the pay scale. In each case we use data on teachers salary in primary education, lower secondary education and upper secondary education. Table 1 shows the results of the two different specifications for each of the first two dependent variables, those obtained by using OLS (pool regressions) on the whole data and fixed and random effects when treating the data as a panel (variable definitions are found in the Data Appendix A). It should be borne in mind that this table represents quite a diversity of results as there are two different dependent variables and three different estimation procedures reported in this table. As a result we would not expect there to be full agreement across the rows of this table with respect to the sign and significance of particular marginal effects of specific regressors on different dependent variables using different estimation methods. Table 2 then presents the same results for the teachers wage expressed as a percentile in the earnings distribution. Looking at both Table 1 and Table 2 there is an indication that the relative supply of teachers, as measured by the stock of teachers in the labour force, has a clear effect on teacher salaries in the appropriate way that it to say the greater the potential supply, the lower will be teachers earnings. In Table 1 we find this effect clearly when OLS is used but somewhat dissipated when Fixed Effects (FE) is used in the salary specification i.e. column 4. When we consider teacher relative pay in Table 2 the effect is even clearer. This finding accords with numerous studies of individual countries (for example Dolton et al 2007) which

16

Separately in the Appendix B we also consider teachers pay measured relative to GDP per head. 16

show that the size of the stock of inactive teachers (mainly qualified women teachers on career or family breaks) will affect the salary earnt by those serving in the profession. Our educational model suggested that one way of teaching more pupils is to increase the pupil/teacher ratio. To compensate teachers for this deterioration in their working conditions and induce them to accept this state of affairs one would expect to have to pay them more. Therefore one would expect that as the pupil/teacher ratio rose we would expect salaries to rise. This is what we see in the FE and RE estimates for teacher pay in panel 2 of Table 1. We also find in Table 2 that this effect is strongly significant for beginning teachers on the lowest wage, teachers with 15 years experience and those at the top of their profession. Another way to increase resources into educational production is to simply put more hours into the school year. There is some evidence that this will reduce the hourly wage of teachers this is logical as if we increase the denominator of the hourly pay computation then as hours rise the hourly wage rate will fall. If however we look at the absolute wage in real terms at any point the teaching career (Table 1 panel 2) or we look teachers wage percentile (Table 2) then we see that there is no effect on pay of having a longer school year. However if we consider pay per hour , i.e. Table 1 panel 1 then this falls as the Pupil/Teacher ratio rises. Further evidence of the influence of supply factors is present in the significance of the percentage of teachers who are women for both of the dependent variables under consideration. Worldwide (in our sample) this fraction has not been falling and shows a slight rising trend up to the most recent figures of around 67%. This variable has a negatively significant impact on teachers salaries when OLS is used and for teachers salaries per hour when FE or RE is used. This may result from the possibility of gender wage discrimination or from the occupational segregation which takes place in most countries where teaching is still regarded as predominantly a female occupation. Alternatively, it may be a consequence of the different career promotion prospects faced by the male and female teachers in the various countries that we examine. On the demand side it appears that as the stock of school age children grows then this demand push will translate lower wages and lower wages per hour (Table 1, panel 1) and lower percentile relative wages at every point on the teacher pay scale (Table 2). One explanation of these findings is that as the demographic size of the cohort rises then, in the short run, there is more pressure for teachers to add extra hours to their teaching day for no extra pay or that the more experienced teachers bear this burden 17.
17

The result is the opposite if we use teacher wages relative to GDP per head (see Table A5). One explanation of this finding is that, it may simply be that as the demographic burden of increasing pupils to teach rises then teachers wage relative to GDP per head rises accordingly in the longer run.

17

With our two different dependent variables we must be careful how we control for the relative wealth of a country and the effect of economic growth. However in the teachers wage per hour equation we must control for this absolute wealth effect, although this will limit the specification to exclude growth in GDP. This specification strategy will hopefully allow us to assess the importance of wealth in explaining teachers salaries in the different countries. The results relating GDP growth to teachers relative salaries compared to GDP per head indicate that there exists a positive relationship between the changing wealth of a country and their teachers salaries. In both the teacher pay and the teacher pay per hour results we find that the wealthier a country is (as measured by GDP per head) the more likely they are to pay their teachers more per hour. This is logically related to the relative wealth of the country with more developed economies having the potential for larger public sectors. As expected, any increase in the expenditure on educational institutions (as a percentage of GDP) has a significant and positive effect on teachers salaries when we consider OLS estimations. This occurs because a large part of any countrys educational budget is taken up with salary costs as so logically as a country chooses to spend more on education it is more or less obviously spending more, in proportionate terms, on teacher remuneration. This effect is mostly a result of absolute values of the fraction of GDP per head spent on education as FE and RE do not reveal that changes in this variable affect teachers salaries. In the equations reported in Table 1 we use two dummy variables to control for whether the teacher pay variable is for those with 15 years experience or those at the top of the teacher pay scale (relative to those just starting the job) and two dummy variables are used to measure the differences among teachers salaries in the lower secondary, upper secondary and primary education levels (the latter is the reference group). These variables have positively significant coefficients when the panel models are estimated suggesting that when cross country heterogeneity is accounted for then there is evidence that teachers in lower or upper secondary schools are paid slightly more than their primary school counterparts. Likewise, the dummy variables being a teacher with 15 years experience or at the top of the scale show a reasonable marked return to experience. The nature of these coefficients are that they give average marginal effects which may disguise a plethora of salary reward patterns in different countries. Considering again the evidence in Figure 2 we see that in some countries the pay schedule by experience is fairly flat - ( Sweden, Norway,

18

Peru, Germany, Tunisia, Spain and Turkey) but that in others there is a marked difference between wages at the beginning of the career and after 15 or more years - notably in Portugal, UK, USA, Austria, Netherlands, France, Thailand, Korea, Belgium, Uruguay, Malaysia, Hungary, Denmark, Argentina and the Czech Republic.) The overall goodness of fit of these estimated equations is reasonable with around 4968% of the variation in teachers wages relative to GDP per head explained and 63-65% of the variation in teachers wages per hour explained. This indicates that the included variables have a reasonable capacity to explain the variance observed in teachers salaries in the countries sampled. Moreover, the results of the F-test indicate that the model estimated is significant at the 1 percent confidence level (for both specifications). In the first panel of Table 5 we report the Hausman test statistics which report the test on the null hypothesis that the coefficients estimated by the efficient random effects estimator are the same as the ones estimated by the consistent fixed effects estimator. Since all of the test statistics indicate that there is a significant P-value, this suggests that the FE should be preferred.

The Determination of Pupil Performance.

Having examined the determination of teacher remuneration we now turn to the question of whether the relative pay that a teacher receives in any given country has an impact on pupil performance. Logically one would expect that where teachers are paid in the wage distribution of a country is a measure of how teachers are regarded in that country. Or more specifically we can take this relative pay of teachers to be an indication of the relative quality of teachers. With our data we know this relative quality by country across time. Hence we know from what part of the wage distribution teachers in each country are paid at over time. We would expect this to potentially be related to the possible relative performance of the pupils. Naturally in this investigation it is important to have standardised tests of school children which are the same across countries. Such tests are provided by TIMSS and PISA 18, (and PIRLS). We use the scores on these tests suitably standardised to examine how the performance of pupils in different countries over time is affected by the resources that have been used to teach them in terms of Teaching hours, Pupil/Teacher ratios and other factors.

18

We used PIRLS in our earlier estimation but these results are less clear. 19

Conditioning on these factors we then address the question of whether there is any relation between teachers relative pay and pupil performance. Our main results are tabulated in Table 3a. Here we use two different possible regressors to examine the effect of pay Teachers salaries after 15 years experience and percentile in the wage distribution. Our main result is that we find that pupil scores rise significantly as teacher salaries rise (in both real terms and relatively). Our results suggest that a $5000 dollar rise in teacher pay (or a 15% increase in real pay) increases pupil standardised performance by .2 of a standard deviation in the standardised score which is equivalent to around an 8% increase in pupil performance. Likewise a 5% increase in the relative position of teachers in the salary distribution increases pupil performance by around 8%. There are also other interesting effects of what determines pupil performance apart from teacher salaries. The most important effects found in Table 3a are that in the FE and OLS estimates Pupil/Teacher ratios have a 4-10% negative effect on pupil performance i.e. that as the pupil teacher ratio rises this will affect pupil performance by a sizeable factor. Likewise we find that an increase of pupil contact hours of 100 hours will increase pupil performance by between 6-9% depending on whether FE, RE or OLS is used 19. Since average teaching contact hours is around 750 hours this constitutes a rise of teaching hours of around 13%. The scaling of our conclusion is not made any clearer by the Hausman Test of FE versus RE since when we perform this test (see Table 5) we see that FE is to be preferred in thes estimations. In the Specification I this means that we have a lower marginal effect (of 0.0296) whereas when we consider the relative percentile position then we are to prefer the FE marginal effect (of 4.0123) which is the largest. One important caveat of the results in Table 3a are that as panel estimates they depend crucially on the assumption that the unobserved heterogeneity does not vary across time. Whilst this seems not to be a very strong assumption with respect to the unobserved characteristics of educational systems over a 10 year period this is something which we should investigate. One way to investigate the robustness of our results is to estimate the model using 2SLS IV to capture the possibility that teacher wage determination is endogenous. Hence we use our reduced form estimation of teacher salary determination (i.e. equation (2)) to estimate a first stage regression which we use to compute an IV for teachers wage.
19

Here our exclusion restriction is that our supply and demand forces of teacher

When we add a dummy for PISA (vs TIMSS) and/or a dummy for science, read (vs Math) , their coefficients are not significant and do not change the rest of coefficients.

20

numbers in the economy and growth in pupil numbers (respectively) do not directly influence pupil test scores. The results of the 2SLS estimations are reported in column 4 and column 8 of Table 3c for the two specifications we estimated by panel estimation methods. Reassuringly when we estimate our pupil performance equation using this 2SLS IV we see that the results of our estimation are directly comparable to those we found with OLS, FE and RE. Specifically we find that the effect of teachers wages on pupils outcomes is still positive and statistically significant and with a marginal effect which is directly comparable to our earlier results. A second form of robustness check we can perform is to consider the growth in teachers earnings over the life cycle rather than simply the level of teachers wages. This we can do as we have teachers earnings at the beginning, after 15 years in the job and at the end of the job. Hence, with this data we can approximate the slope of the age earnings profile over the whole life cycle and use this as our regressor instead of the level of teachers earnings per se as we are concerned about the endogeneity of the former variable. Hence our suggested regressor is an average difference (of wages over the lifetime) and this, it could be argued, will not be endogenous. Table 3c reports our estimation results when we use this growth in earnings variable. Here we see that using FE and RE we again have an earnings effect on pupil outcomes. The implication of this result is that a positive wage profile in terms of a steady advancement of earnings over the life cycle will give able teachers the appropriate incentive to stay in the job. As a result of this we are suggesting that pupil test scores will benefit directly. A final form of robustness check is to consider the separate pupil tests scores we have in Maths, Reading and Science as separate regressands as up to now we have averaged them. This then facilitates a Seemingly Unrelated Regression (SUR) estimation of our model. The advantage of this is that we are then estimating them model as a system exploiting the possibility that the errors in the separate equations are related. This will lead to more efficient estimation. Since this technique has already been used in this context by Lee and Barro (2001) we adopt exactly their specification to estimate this model. Our results are reported in Table C1. Despite the fact that their model uses quite different regressors namely Expected Years of Education and Educational Expenditure we continue to get a clear effect of teachers earnings on childrens educational outcomes. Indeed our estimation results may be encouragingly compred to those of Lee and Barro (2001) since we get clear effects of Educational spending, Expected years of education and Pupil/Teacher ratios whilst the Lee and Barro paper (2001) do not get clear results on the presence of teacher wage effects.
21

One logical question which arises with this result linking average student performance with average teacher salaries is whether this relationship is affected if we consider the effect on the distribution of pupil attainment. In other words, does rewarding teachers better (or recruiting a higher quality teacher) drive up average performance of pupils at the expense of narrowing the distribution of performance. This can be investigated in two ways. Firstly to see if our original result is robust to controlling for the standard deviation (SD) of pupil performance within the country and secondly by using the measure of the variance of performance as a dependent variable to see what is driving its variation across countries. In Table 4 we report that the link between teachers salary and pupil performance is not changed much by conditioning on the SD of pupil performance i.e. that a $1000 rise in teacher pay would increase pupil performance by around .25 of a standard deviation or around 5%. However the effect of the relative pay of teachers, in terms of percentile of teachers pay is not robust to adding the SD of pupil performance as a control. Adding the SD of pupil performance to this regression shows that there is a negative effect on pupil performance i.e. that there is some trade off resulting in higher average performance of pupils may result in a lower SD of performance in which case there is some evidence that putting resources into pushing up mean pupil performance may result in there being a smaller variance in this performance. Such a result is not surprising as initiatives which teach to the test such as those in the UK can result in bringing up the performance of the less able children whilst curtailing the performance of the most able children and hence reducing the variance of pupil performance overall. Continuing the line of investigation relating to what determines the variability of student performance we explore the statistical correlates of this variable in Table 4. Here we see that as the teacher quality rises (in terms of percentile position) then the variability of student performance is reduced but as teacher pay rises in PPP terms then the variability of pupil performance also rises. It is not immediately clear that there is a logical explanation for this finding. Perhaps the most interesting determinant of the variability of pupil performance is that it rises as the number of teaching hours rises. This is a fundamental relationship which suggests that as teaching hours rises by 100 per year then the SD of pupil performance will rise by around .17-.25 of a SD for a standard normal variate. At the mean of such a variate this is equivalent to around an 8% rise in the SD. Such a finding has a logical plausibility as it suggest that as more hours are taught then the brighter students will learn more and increase the gap of achievement between themselves and the weaker students.

22

7.

Conclusion and Policy Implications

Our starting point for this investigation was the proposition that teacher quality is not easy to observe within a country but may be measured by teacher relative wages when we consider cross country data. This is a reasonable hypothesis on the grounds that where a teacher is paid in a countrys income distribution will result in drawing out candidates for teaching of the appropriate quality. The idea is that each country gets the teachers it wants and deserves. With this idea in mind we set out to investigate whether the variable quality of teachers in different countries in the world could help to explain the variation in the performance of 15 year olds in cross country national tests in maths reading and science. Since our measure of teacher quality was the wage teachers are paid in a country we sought to model the determination of teacher salaries and pupil outcomes. We first considered reduced for estimations for a teacher salary equation and a pupil outcome equation. Using panel data for 39 countries for 10 years we attempted to establish the relationship between teacher salaries and pupil outcomes. The advantage of panel data in this context is that is country specific unobserved heterogeneity does not vary over time then we may be able to establish whether the link between teacher wages and pupil outcomes is anything most that a correlation. We also sought to test our findings by estimating a variety of other models as robustness checks. Specifically we estimated a 2SLS IV model in which we suggested that movements in the supply and demand for teachers (in terms of the number of trained teachers in the economy and the changing population of school children) were exogenous to pupil test scores. We also investigated models of: the growth in teacher earnings, the distributional effects of teacher quality on the dispersion of test scores and the separate performance on subject by subject specific tests using SUR estimation. In each case we found that there was still a highly significant and positive effect of teacher wages on pupil test scores. To provide some idea of the scale of the effects we find our coefficients suggest that a 15% percent increase in teacher pay would give rise to around a 6-8% increase in pupil performance. Likewise a 5% increase in the relative position of teachers in the salary distribution would increase pupil performance by around 6-8%. These effects are significant and robust to the estimation procedure we use and the different identification assumptions we make to facilitate each estimation technique. Whilst we cannot, of course, claim unequivocally that these effects are causal it seems plausible to suggest that there robust
23

nature across different estimation strategies might lead us to suggest that the relationship between teacher salary and pupil outcomes is more than a simple correlation. The policy implications of our findings are relevant to the recruitment of teachers and the improvement of educational standards for young people. The link we have found between high educational standards and teachers quality has logical implications for any governments commitment to recruit, retain and reward good teachers. In this regard, it seems that increasing teachers salaries, and the speed at which they can reach higher pay levels within a particular pay structure, will help schools recruit and retain teachers that schools need to offer all pupils a high-quality education. At a wider policy level the improvement in human capital appears to be a common factor behind economic growth in recent decades in all OECD countries, especially in the so called Mediterranean countries20, where the increase in human capital accounted for more than half an extra percentage point of growth in the 1990s compared with the previous decade. Therefore it would seem that one clear way to improve the human capital stock is to invest in higher quality teachers.

20

Spain, Italy and Greece. 24

References

Akerhielm, K. (1995). Does Class Size Matter?. The Economics of Education Review, vol. 14 pp. 229-241. Angrist, J. and Lavy, V. (2001). Does Teacher Training Affect Pupil Learning? Evidence from Matched Comparisons in Jerusalem Public Schools. Journal of Labor Economics 19(2), pp. 343-369. Ballou, D. and Podgursky, M. (1997). Teacher Pay and Teacher Quality, W.E. Upjohn Institute for Employment Research, Kalamazoo, Michigan. Barro, R. and Lee, J-W. (1993) International Comparisons of Educational Attainment', Journal of Monetary Economics, vol.32, pp363-394. Becker, W. E. and Powers, J. R. (2001). Student performance, attrition and class size given missing student data. Economics of Education Review, vol. 20, pp. 377-388. Black, P. A. and Hosking, S. (1997). The teacher crisis in South Africa: quitting, shirking and `inferior substitution'. South African Journal of Economics, 65(4), pp. 500-509. Bonesronning, H., Falch, T., and Strom, B. (2005). Teacher Sorting, Teacher Quality, and Student Composition. European Economic Review, 49(2), pp. 457-483. Borghans, L. (1991). Occupational choice: the market for primary school teachers. ROARM-1991/3E, Maastricht. Card, D. and Krueger, A. (1992) Does school quality matter? Returns to education and the characteristics of public schools in the United States, Journal of Political Economy, vol.100, pp.1-40. Chapman, D. W. and Mhlck, L. O. (Eds.). (1993). From Data to Action: Information Systems in Educational Planning. Paris: UNESCO, International Institute for Educational Planning and London: Pergamon. Dolton, P. (1990) 'The Economics of UK Teacher Supply', Economic Journal, vol.100, pp.91-104. Dolton, P. (2006). 'Teacher Supply' Chapter 19 Handbook of the Economics of Education vol.2, ed Eric Hanushek and Finis Welch. Dolton, P Makepeace, G. and van der Klaauw, W. (1999) Occupational choice and earnings determination: The Role of Sample Selection and Non-Pecuniary Factors, Oxford Economic Papers, vol.41, pp.573-594. Dolton, P and van der Klaauw, W. (1999) ' The Turnover of Teachers: A Competing Risks Explanation', Review of Economics and Statistics, vol.81(3), pp.543-550. Ferguson, R. F. (1991). Paying for public education: New evidence on how and why money matters. Harvard Journal on Legislation, vol. 28, pp. 465-498. Figlio, D. N. and Stone, J. A. (1997). School Choice and Student Performance: Are Private Schools Really Better?, Institute for Research on Poverty Working Paper 1141-97. Figlio, D. N. and Lucas, M. E. (2000). Do high grading standards affect student performance. NBER Working Paper 7985. Flyer, F. and Rosen, R. (1997) The New Economics of Teachers and Education, Journal of Labor Economics, vol.15(1), p.S104-S138. Goldhaber, D. (2002). The Mystery of Good Teaching. Education Next, vol. 2(1), Hoover Institute. Glewwe, P. W. (1999). The Economics of School Quality Investments in Developing Countries: An Empirical Study of Ghana. MacMillan Press, U.K. Hanushek, E., Kain, J. and Rivkin, S. (1998). Teachers, Schools and Academic Achievement. Econometrica. Hanushek, E., Kain, J. and Rivkin, S. (2002). Inferring Program Effects For Special Populations: Does Special Education Raise Achievement For Students With Disabilities?. Review of Economics and Statistics, vol. 84(4), pp. 584-599.
25

Hanushek, E., Kain, J. and Rivkin, S. (2005). 'Teachers, Schools and Academic Achievement', Econometrica, 73(2), p.417-458. Hanushek, E. (2007) Handbook. Hanushek, E. and Kimko, D. D. (2000). Schooling, Labor Force Quality, and the Growth of Nations. American Economic Review, pp. 1184-1208. Hanushek, E. and Luque, J. A. (2003). Efficiency and Equity in schools around the world. Economics of Education Review, vol. 22(5), pages 481-502. Hanushek, E. and Woessman, L. (2007) The Role of Education Quality in Economic Growth forthcoming in the Journal of Economic Literature. Hedges, L., Laine, R. and Greenwald R. (1994a). Does Money Matter: A Meta Analysis of Studies of the Effects of Differential School Inputs on Student Outcomes. Educational Researcher, vol. 23(3), pp. 5-14. Hedges, L., Laine, R. and Greenwald R. (1994b). Money does matter somewhere: a reply to Hanushek. Educational Researcher, vol. 23(4), pp. 9-10. Hoxby, C. (2000). The effects of class size on student achievement: new evidence from population variation. Quarterly Journal of Economics, vol 115(4), pp. 1239-1285. Jones, C, I. (2002). Introduction to Economic Growth. Norton and Company. Krueger, A. B. (1999). Experimental estimates of education production functions. Quarterly Journal of Economics, vol. 114, pp. 497-532. Lakdawalla, D. (2002). The declining relative quality of teachers. RAND Working Paper, Santa Monica. Lakdawalla, D. (2006). The Economics of Teacher Quality, Journal of Law and Economics, vol.XLIX, pp.285-329. Lavy, V. (2001) 'Evaluating the Effect of Teacher's Performance Incentives on Pupil Achievement' mimeo. Lee, J-W. and Barro, R.J. (2001) 'Schooling Quality in a Cross-Section of Countries', Economica, vol.68, pp.465-488. Lewis, P. and Norris, K. (1992). Demand, Supply and Adjustment in the Teachers LaborMarket. Australian Journal of Education, vol. 36(3), pp. 260-277. Loeb, S. and Page, M. (2000) 'Examining the Link between Teacher Wages and Student Outcomes: The Importance of Alternative Labor Market Opportunities and NonPecuniary Variation', Review of Economics and Statistics, vol.82, no.3, pp.393-408. Murnane, R. J., J. D. Singer, J. B. Willett, J. J. Kemple and R.J. Olson (1991). Who Will Teach? Policies That Matter. Cambridge, MA: Harvard University Press. OECD (1996, 1997, 1998, 1999, 2000, 2001a, 2002, 2003, 2004, 2005, 2006, 2007). Education at Glance. Paris. OECD, 2001b. Teachers for Tomorrows Schools. Paris. OECD, 2002. Teacher Demand and Supply: Improving Teaching Quality and Addressing Teacher Shortages, Education Working Paper, Paris. Odden, A., and C. Kelley (1997). Paying Teachers for What They Know and Do: New and Smarter Compensation Strategies To Improve Schools. ED 404 312. Thousand Oaks, CA: Corwin Press. Perlmann, J and Margo, R. (2001) 'Women's Work: American Schoolteachers 1650-1920', University of Chicago Press, Chicago. Rivkin, S.G., Hanushek, E.A. and Kain, J.F. (2001). Teachers, schools, and academic achievement, NBER Working Paper No. 6691. Lee, J-W, and Barro, R. (2001) 'Schooling Quality in a Cross-Section of Countires', Economica, vol.68, pp.465-488. Lopez-Acevedo, C. and Salinas, A. (2001). The Distribution of Mexico's Public Spending on Education. WP-2404, The World Bank. Savedoff, W. and Piras, C. (1998). How much do teachers earn? Office of the Chief Economist. Inter-American Development Bank, mimeo.
26

Tamura, R. (2001). Teachers, Growth and Convergence. Journal of Political Economy, vol. 109( 5), pp. 1021-1059. Temin, P. (2002) Teacher Quality and the Future of America, NBER Working Paper 8898. Tremblay, A. (1997). Are we Headed Toward a Teacher Surplus or a Teacher Shortage? Education Quarterly Review 4(1), pp. 53-85. Zabalza, A. (1979). From Shortage to Surplus: The Case of School Teachers. Applied Economics, vol. 11, pp. 55-7

27

28

Table 1. The Determination of Teachers Salaries Across Countries 1995-2005. Pool of Salaries, controlling for the type of salary, i.e. starting (reference group), after 15 and top.
Teachers salaries/hour Fixed Random OLS Effects Effects 10.2077*** 10.1606*** 10.1625*** (0.7899) (0.5777) (0.5816) 19.2087*** 19.1950*** 19.1973*** (0.7919) (0.5797) (0.5836) -10.2478*** -2.3312** -3.8791*** (0.5976) (1.1406) (1.0257) -3.8842*** -3.2947*** -3.3823*** (0.2769) (0.3548) (0.3435) -0.2281*** -0.1257*** -0.1371*** (0.0299) (0.0374) (0.0364) 1.2181 3.6384*** 3.4562*** (0.9277) (0.8556) (0.8438) 4.8027*** 5.5408*** 5.2085*** (1.3460) (1.4404) (1.4010) 2.1485*** 0.0229 0.0497 (0.5444) (0.5065) (0.4997) 0.0095 0.1190 0.1188 (0.0717) (0.0866) (0.0841) -0.0845*** -0.1098*** -0.1184*** (0.0174) (0.0301) (0.0280) 1.1637*** 0.7243** 1.0530*** (0.0516) (0.2888) (0.1441) 75.5462*** 47.9141*** 48.2036*** (3.7521) (7.3166) (5.9407) 1243 1243 1243 167.72*** 89.60*** 0.59 0.59 0.75 0.76 0.72 0.66 0.68 Teachers salaries (1000 $) Fixed Random OLS Effects Effects 6.9708*** 6.9334*** 6.9342*** (0.4863) (0.3206) (0.3226) 12.9288*** 12.9209*** 12.9212*** (0.4876) (0.3217) (0.3237) -6.7324*** -1.0378 -1.6545*** (0.3680) (0.6329) (0.6005) 0.0931 0.3595* 0.3118 (0.1705) (0.1969) (0.1942) -0.0820*** -0.0088 -0.0153 (0.0184) (0.0208) (0.0205) 0.4942 1.9963*** 1.9009*** (0.5712) (0.4748) (0.4727) 2.3421*** 3.8758*** 3.6028*** (0.8288) (0.7993) (0.7898) 0.8379** 0.1460 0.1077 (0.3352) (0.2811) (0.2798) -0.0445 0.1035** 0.0983** (0.0441) (0.0481) (0.0475) -0.0300*** -0.0049 -0.0150 (0.0107) (0.0167) (0.0161) 0.9224*** 0.4152*** 0.7497*** (0.0318) (0.1602) (0.1010) 26.0377*** 7.2279* 4.1083 (2.3102) (4.0600) (3.6175) 1243 1243 1243 137.25*** 93.42*** 0.60 0.60 0.60 0.54 0.68 0.49 0.59

Dummy for wages after 15 years of experience Dummy for wages at the top of the earnings distribution Teachers and educational staff as a fraction of the labour force Teaching hours per year (100s) Women fraction of teaching staff (%) Lower secondary dummy Upper secondary dummy Expenditure on educational institutions as a percentage of GDP Pupil/Teacher ratio Growth in the size of the population at the age of primary/lower secondary and upper secondary GDP per head (1000 $) Year dummies Constant Observations F-statistic R-squared Within R-squared Between R-squared Overall

29

Table 2: Percentile position of teachers salaries in the earnings distribution (OLS)

Percentile position of teachers salaries Starting Salaries after 15 years Salaries at the top salaries of experience of the scale -0.0617*** -0.0582*** -0.0663*** (0.0116) (0.0116) (0.0105) 0.0061 0.0072 0.0007 (0.0052) (0.0052) (0.0047) 0.0010* -0.0003 -0.0004 (0.0006) (0.0006) (0.0005) 0.0691*** 0.0524*** 0.0438*** (0.0176) (0.0176) (0.0160) 0.1742*** 0.1419*** 0.1106*** (0.0252) (0.0252) (0.0230) 0.0522*** 0.0471*** 0.0333*** (0.0094) (0.0094) (0.0085) 0.0041*** 0.0040*** 0.0037*** (0.0014) (0.0014) (0.0012) -0.0012*** -0.0011*** -0.0012*** (0.0003) (0.0003) (0.0003) -0.0051** -0.0021 -0.0010 (0.0024) (0.0024) (0.0021) -0.0104 (0.0334) -0.0363 (0.0293) -0.0298 (0.0301) -0.0497* (0.0287) -0.0571** (0.0290) -0.0351 (0.0301) 0.3220*** (0.0680) 314 13.75*** 0.41 0.0037 (0.0334) -0.0314 (0.0293) -0.0263 (0.0301) -0.0345 (0.0287) -0.0368 (0.0290) -0.0054 (0.0302) 0.5032*** (0.0681) 314 12.66*** 0.39 0.0050 (0.0315) -0.0277 (0.0279) -0.0380 (0.0285) -0.0370 (0.0274) -0.0687** (0.0277) -0.0250 (0.0286) 0.6990*** (0.0620) 308 15.39*** 0.44

Teachers and educational staff as a fraction of the labour force Teaching hours per year (100s) Women fraction of teaching staff (%) Lower secondary dummy Upper secondary dummy Expenditure on educational institutions as a percentage of GDP Pupil/Teacher ratio Growth in the size of the population at the age of primary/lower GDP growth (%) Year dummies: (reference year 1995) 1998 1999 2000 2001 2002 2003 Constant

Observations F-statistic R-squared Note: Standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%

30

Table 3.a. Determinants of Standardised Pupil Scores across Countries 1995-2005. (Each type of Assessment for 8th grade students.)
OLS -0.0270 (0.0304) -0.0655*** (0.0093) 0.0037 (0.0048) 0.0518*** (0.0057) Specification I Fixed Effects Random Effects 0.0933 -0.0225 (0.0664) (0.0514) 0.0367* -0.0432*** (0.0215) (0.0139) 0.0141** 0.0014 (0.0063) (0.0054) 0.0296** 0.0532*** (0.0147) (0.0094) 2SLSa -0.0938** (0.0381) -0.0574*** (0.0098) 0.0043 (0.0074) 0.0436*** (0.0112) OLS 0.1014** (0.0412) -0.1029*** (0.0157) -0.0230*** (0.0068) Specification II Fixed Effects Random Effects 0.2144** 0.0691 (0.1077) (0.0838) 0.0293 -0.0845*** (0.0361) (0.0247) 0.0151 -0.0187 (0.0215) (0.0128) 2SLS -0.0268 (0.0328) -0.0742*** (0.0103) -0.0095 (0.0062)

Teaching hours per year (hundreds) Pupil/Teacher ratio Women fraction of teaching staff (%) Teachers' salaries after 15 years in 1000$ Percentile position of teachers (after 15 years) GDP growth (%) Year dummies: (reference year 1995) 1999 2000 2003 2005 Constant

0.1448*** (0.0235) 1.4149*** (0.3309) 0.7098** (0.3186) 1.1066*** (0.3160) 0.8921*** (0.3260) -2.0626*** (0.4750) 205 26.20***

-0.0617** (0.0249) 1.6836*** (0.1718) 1.2427*** (0.1742) 1.2384*** (0.1826) 1.3256*** (0.1968) -3.4701*** (0.7228) 205 18.15*** 0.50 0.19 0.01

-0.0050 (0.0231) 1.5767*** (0.1742) 1.1282*** (0.1738) 1.1902*** (0.1746) 1.2681*** (0.1883) -1.9778*** (0.5343) 205 0.44 0.68 0.44

0.1839*** (0.0265) 0.5979** (0.2841) -0.1532 (0.2480) 0.5649** (0.2572) 0.1090 (0.2656) -0.9665 (0.6034) 192 0.490

1.0803 (0.7961) 0.1237*** (0.0347) 1.3475*** (0.4114) 1.0499*** (0.3956) 1.5328*** (0.3940)

4.0132*** (1.3151) -0.1305*** (0.0484) 1.7414*** (0.2029) 1.3713*** (0.2630) 1.1548*** (0.2437)

2.9723*** (1.1385) -0.0055 (0.0405) 1.6726*** (0.2016) 1.2449*** (0.2293) 1.3770*** (0.2127)

1.9714** (0.8876) 0.1720*** (0.0275) 0.4640 (0.2899) 0.0272 (0.2545) 0.8386*** (0.2570) 0.5476** (0.2568) -0.5000 (0.7200) 192 0.462

Observations F-statistic R-squared Within R-squared Between 0.54 R-squared Overall Note: Standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%.

-0.4621 (0.6959) 141 13.76***

0.44

0.45

-5.5432*** (1.5000) 141 17.02*** 0.56 0.46 0.10

-1.4101 (0.9280) 141 0.46 0.46 0.31

0.49

31

Table 3.b. Determinants of Standardised Pupil Scores across Countries 19952005. (Using Overall Teachers Wage Growth on Each type of Assessment for 8th grade students.)
Specification I Fixed Effects 0.0455 (0.0674) 0.0374* (0.0217) 0.0102 (0.0066) 4.5595** (2.1250)

OLS Teaching hours per year (hundreds) Pupil/Teacher ratio Women fraction of teaching staff (%) Overall growth in teachers wage average Percentile position of teachers (after 15 years) GDP growth (%) Year dummies: (reference year 1995) 1999 2000 2003 2005 Constant Observations F-statistic R-squared Within R-squared Between R-squared Overall 0.1134*** (0.0279) 1.1859*** (0.3964) 1.0072*** (0.3803) 1.5751*** (0.3747) 1.6089*** (0.3834) -0.5700 (0.8821) 203 0.0763** (0.0371) -0.0668*** (0.0111) -0.0212*** (0.0048) 0.3940 (0.8788)

Random Effects -0.0692 (0.0565) -0.0352** (0.0160) -0.0031 (0.0058) 5.2076*** (1.6116)

-0.0604** (0.0249) 1.7469*** (0.1684) 1.3952*** (0.1718) 1.4325*** (0.1661) 1.5560*** (0.1865) -6.4451*** (1.7915) 203 0.50 0.39 0.04

-0.0139 (0.0239) 1.6906*** (0.1753) 1.3529*** (0.1772) 1.5024*** (0.1703) 1.6345*** (0.1888) -5.0771*** (1.4193) 203 . 0.45 0.31 0.13

0.36

Note: Bootstrapped standard errors in parentheses. significant at 10%; ** significant at 5%; *** significant at 1%.

32

Table 3.c. OLS Estimates explaining the Standardised Scores for each type of Assessment, 8th grade students (controlling for standard deviation).
Specific. I -0.1145** (0.0491) -0.0121 (0.0320) -0.0331*** (0.0125) 0.0086* (0.0046) 0.0485*** (0.0052) Specific. II -0.1355* (0.0757) 0.0979** (0.0471) -0.0736*** (0.0218) -0.0165** (0.0070)

Standard Deviation for each type of Score Teaching hours per year (hundreds) Pupil/Teacher ratio Women fraction of teaching staff (%) Teachers' salaries after 15 years in 1000$ (deflated) Percentile position of teachers salaries (after 15 years) GDP growth (%) Year dummies: 1999 2000 2003 2005 Constant

0.1028*** (0.0238) 1.2202*** (0.3259) 0.7523** (0.3158) 0.9847*** (0.3131) 0.8097** (0.3175) -2.0903*** (0.4961) 208 0.45

0.9317 (0.8045) 0.0881** (0.0367) 0.8474** (0.4089) 0.9645** (0.4007) 1.2943*** (0.4067)

Observations R-squared adjusted Note: Standard errors in parentheses * significant at 10%; ** significant at 5%; *** significant at 1%.

-0.2814 (0.7299) 138 0.28

33

Table 4. Estimates explaining the Standard Deviation of the Standardised Scores for each type of Assessment, 8th grade students.

OLS Teaching hours per year (hundreds) Pupil/Teacher ratio Women fraction of teaching staff (%) Teachers' wages after 15 years in 1000$ Percentile position of teachers (after 15 years) GDP growth (%) Year dummies: 1999 2000 2003 2005 Constant -0.0529 (0.0342) -1.2263*** (0.4654) -1.0348** (0.4525) -1.3215*** (0.4448) -0.8332* (0.4560) 3.2858*** (0.6821) 210 11.94*** 0.2472*** (0.0430) 0.0145 (0.0181) -0.0054 (0.0067) -0.0138* (0.0074)

Specification I Fixed Effects 0.0637 (0.0846) -0.1813*** (0.0527) -0.0108 (0.0149) 0.1287*** (0.0358)

Random Effects 0.1712*** (0.0611) 0.0127 (0.0237) 0.0052 (0.0090) -0.0059 (0.0107)

Specification II Fixed Random OLS Effects Effects 0.2447*** 0.0960 0.2051*** (0.0505) (0.1083) (0.0694) -0.0022 -0.2002** -0.0115 (0.0253) (0.0899) (0.0314) 0.0086 -0.1339** 0.0121 (0.0081) (0.0541) (0.0109)

0.0264 (0.0607) -1.1689** (0.4792) -1.4019*** (0.4090) -2.1269*** (0.4267) -1.7999*** (0.4543) 3.6356** (1.4074) 210

-0.0507 (0.0428) -0.9721** (0.4456) -0.9622** (0.4091) -1.3795*** (0.4000) -1.0123** (0.4196) 3.1345*** (0.7946) 210 0.08 0.64 0.33

-0.1013** (0.0417) 0.5095 (0.9352) -1.0205** (0.4673) -0.8758* (0.4600) -1.4097*** (0.4568)

-0.2214* (0.1206) 7.6671** (3.1788) -0.3492 (0.5754) 0.1538 (0.5973) -0.9123 (0.5563)

-0.1479*** (0.0536) 1.7174 (1.2616) -0.9315** (0.4721) -0.6747 (0.4448) -1.4648*** (0.4359)

Observations F-statistic 0.19 R-squared Within 0.35 R-squared Between 0.35 0.02 R-squared Overall Note: Standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%.

2.3881*** (0.8230) 138 13.09***

8.9486*** (3.1342) 138 0.23 0.17 0.02

2.0799** (0.9667) 138 0.12 0.79 0.43

0.45

34

Table 5. Hausman test Fixed vs Random effects (Ho: difference in coefficients not systematic)
2(15)=26.67*** 2(15)=29.93** Table 1 Prob.> 2=0.0453 Prob.> 2=0.0189 Table 3a Prob.> 2= 0.0000 Prob.> 2=0.0000 Table 4 Prob 2(9)=0.0449 Prob 2(8)=0.0025

Columns 2 vs 3 Columns 5 vs 6

Columns 2 vs 3 Columns 5 vs 6

2(9)=143.63*** 2(8)=31.37***

Columns 2 vs 3 Columns 5 vs 6

2(9)=17.26*** 2(8)=23.76***

35

Figure 1.a. Actual and fitted Upper Secondary school teachers salaries after 15 years experience in 2007 $ PPP
Teachers salaries after 15 years Is 10000 25000 40000 55000 70000 H rae un l Cz g ec T ary h urk Re e pu y b Ic lic el a G nd re Po ece rtu ga Ita l Fr ly a N Sw nce ew e Ze den al N and or w A ay us tri U a S Sp A Fi ain nl an d A UK us tra Ire lia D lan en d m ar Ja k pa Be lg n iu K m G ore N erm a et a h Sw erla ny itz nd er s la nd Actual Teachers salaries after 15 years Fitted values

Upper Secondary education

Note: Fitted value based on a regression controlling for teachers and educational staff as a fraction of the labour force, teaching hours per year (100s), women fraction of teaching staff(%), expenditure on educational institutions as a percentage of GDP, students/Teacher ratio and GDP annual growth (%).

Figure 1.b. Actual and fitted Upper Secondary school teachers salaries after 15 years experience/GDP per head in 2007 $ PPP
Teachers salaries after 15 years/GDP in 2005 Ic .5 1 1.5 2 2.5 el N and or w ay U SA G r H eec un e g Sw ary ed en Ita Ire ly la Fr nd Cz an ec A ce h us Re tri pu a bl ic U Fi K n D lan en d A mar us k tr N Po alia ew rtu Ze gal a Be land lg iu m Sp ai Ja n p N I an et sr he ae r l G land Sw erm s itz an er y la n K d or Tu ea rk ey

Upper Secondary education

Actual Teachers salaries/GDP per head

Fitted values

Note: Fitted value based on a regression controlling for teachers and educational staff as a fraction of the labour force, teaching hours per year (100s), women fraction of teaching staff(%), expenditure on educational institutions as a percentage of GDP and students/Teacher ratio.

36

Figure 1.c. Actual and fitted Upper Secondary school teachers salaries after 15 years experience relative to the earnings distribution of the whole population (2003)
Teachers salaries relative position (percentile) .4 .5 .6 .7 .8 Ch N i or le w A P ay rg e en ru A tina us Fr tria a Ire nce la n Ita d Sw l U ed y ru en g Cz Hun uay g ec h Uary Re S p A Be ubl A lgiuic u N stra m ew G lia Zereec al e an M Ud al K ay s D Sp ia en ai m n Fi ar Ponlank N et rtu d he g rla al n G Japds er a m n Tu an n y Tu isia rk B ey SwTha razi i il l Ph tzer and ili lan pp d i K nes or ea

Upper Secondary education

Actual Teachers salaries (percentile)

Fitted values

Note: Fitted value based on a regression controlling for teachers and educational staff as a fraction of the labour force, teaching hours per year (100s), women fraction of teaching staff(%), expenditure on educational institutions as a percentage of GDP, students/Teacher ratio and GDP annual growth (%).

37

Figure 2. Relative position (percentile) of teachers salaries in the earnings distribution of the whole working population (Upper secondary education)
Arge ntina
0.2 .5.7 5 51

Australia

Austria

B elgium

B razil

C hile

C zech R epublic
0.2 5.5.7 51

Denm ark

Finland

Franc e

Germ any

Greec e

Hungary
0.2 5.5.7 51

Indonesia

Ireland

Italy

Japan

Kore a

Percentile

M alaysia
0.2 .7 5.5 51

Nethe rlands

New Zealand

Norway

Peru

Philippines

Portugal
0.2 5.5.7 51

Spain

Sweden

Switz erland

Thaila nd

Tunisia

19 95

19 96 19 97

19 98

19 99

20 00

20 01

20 02

20 03

20 04

20 05

19 95 19 96

19 97

19 98

19 99

20 00

20 01

20 02

20 03

20 04

Turkey
0.2 5.5.7 51

UK

USA

Uruguay

19 95

19 96

19 97

19 98

19 99

20 00 20 01

20 02

20 03

20 04

20 05

19 95

19 96

19 97

19 98

19 99 20 00

20 01

20 02

20 03

20 04

20 05

19 95

19 96

19 97

19 98 19 99

20 00

20 01

20 02

20 03

20 04

20 05

19 95

19 96

19 97 19 98

19 99

20 00

20 01

20 02

20 03

20 04

Percentile (starting wages)

Percentile (after 15 years)

20 05

Percentile (top wages)

Graphs by Country

20 05

38

-2 0 1

-1

Source: Authors own calculations.

Figure 3.a. Standardised Average Scores (8th grade students) by country (2006)

Source: Authors own calculations.

Figure 3.b. Average Scores Standard Deviation by country (2006).

Sl o I cv en F el ia A inland us an tra d lia Th ItUK a N ew aila ly ZeSpand N S ala in et w n he ed d Berlanen Mlgi ds U exum Hruguico D ung ay e Sw Nnmary a itzorw rk er ay Jo lan r d Brdan I a Porelazil r n G t ug d re a l Ja ece Fr pa Cz a n ec K nce h Tu ore Re n a pu isia GAusblic er tr m ia a U ny I S TusraA e In Crkeyl do hi A n l rg e e en sia tin a
Tunisia Argentina Brazil Indonesia Mexico Thailand Uruguay Turkey Chile Israel Greece Italy Portugal Spain USA Norway Hungary France Iceland Denmark UK Czech Republic Austria Sweden Germany Ireland Belgium Switzerland Slovenia Japan Australia Netherlands New Zealand Korea Finland

39

Figure 3.c. Standardised Scores against average standard deviation of the scores (2006)
2
Australia Finland New Zealand Sweden Netherlands Denm ark Norway Ireland Hungary Belgium Greece Portugal Switzerland France Korea Germ any USA Austria

Standardised Scores -1 0

Iceland

UK

Italy

Spain

Uruguay

Czech Republic

Japan

Thailand

M exico

Turkey

4 Average Score's Standard Deviation

Tunisia

Brazil

-2

Indonesia

Source: Authors own calculations.

Figure 4. Scores percentile at 8th grade students as a function of teachers salaries after 15 years experience.
100
Finland New Zealand Australia Japan Belgium Ireland Germany Austria UK Denmark Korea Netherlands Switzerland

Score's Percentile 8th grade students 20 40 60 80

Sweden Czech Republic Iceland Hungary France

Israel Turkey

Portugal Italy Greece

Norway USA Spain

0 10000

20000

30000 40000 Teachers' salaries (2007 USA$ PPP)

50000

60000

Source: Authors own calculations.

40

-2
Portugal Spain Greece

Standardised Scores 8th grade students -1 0 1 2

10
Italy Israel

Belgium France Switzerland Iceland Austria UK ark Denm Hungary Czech Republic New Japan Zealand Sweden Germ any Slovenia USA

15
Turkey Brazil

Source: Authors own calculations.

Netherlands Korea Australia

Figure 5. Standardised Scores as a function of Pupil/Teacher ratio.

20 Pupils/Teacher ratio 25 30

Ireland

Finland

M exico Chile

41

Data Appendix A: Definition of variables 21: Teachers and educational staff as a fraction of the labour force (%), based on head counts (1999 and 2001) 22. (Source: Education at a Glance, OECD (2001 & 2003), table D2.4). Teachers' salaries 23: - Teachers wage/GDP per head: Teachers weekly salaries relative to per capita Gross Domestic Product (GDP) in equivalent US dollars converted using PPPs; years 1995-2005. (Source: Education at a Glance, OECD (1997-2007), table D1.1 & D3.1). - Teachers wage per hour: This variable refers to the teachers average salaries per statutory hour of teaching in primary education, lower secondary education and upper secondary education; 1995-2005. (Source: Education at a Glance, OECD (1997-2007), table D1.1 continued & D3.1 continued).

- Teachers annual statutory salary at starting of their careers/after 15 years/at the top of the scale: Annual statutory teachers' salaries in public institutions at starting salary, after 15 years of experience and at the top of the scale by level of education, in equivalent US dollars converted using PPPs; 1995-2005 24. (Source: Education at a Glance, OECD (1997-2007), table D1.1 continued & D3.1 continued).
Teaching hours per year (1996-2005): This refers to the statutory number of teaching hours per year (1996-2002), i.e. net contact time in hours per year in primary, lower secondary and upper secondary education 25. (Source: Education at a Glance, OECD (1997-2007), table D3.1 continued & & D4.1). Fraction of teaching staff who are women (%); 1996-2005: Percentage of women among teaching staff in public and private institutions, in primary, lower secondary and upper secondary education. 26

For the year 1999 Education at a Glance was not published, which means that we have missing data for 1997. 22 For the period 1995-1998 we impute the figures from 1999 and for the rest of the period we impute the 2001 figures. 23 For Uruguay, after 1999, teachers salaries were reported for working 20 hours per week. Most teachers hold two positions, therefore the figures have been multiplied by 2 to make then comparable with the rest of countries. 24 Teachers' salaries at the top of the scale for Thailand in 2002 have been dropped, as they do not make sense. 25 Information for 1995 is not available. To overcome this we computed the ratio teachers salaries after 15 years experience/wage (after 15 years experience) per teaching hour to obtain a proxy for teaching hours in 1995. 26 Teachers' salaries at the top of the scale for Thailand in 2002 have been dropped out, as they do not make sense. 42

21

(Source: Education at a Glance, OECD (1997-2007), table D2.3 continued & table D7.2 & table D6.2).
Lower secondary: This is a dummy variable that is equal to 1 if the observation refers to teachers of lower secondary education (primary education teachers is the reference group). Upper secondary: This dummy variable is equal 1 if the observation refers to teachers of upper secondary education (primary education teachers is the reference group). GDP growth (%): This measures the GDP (per capita, in equivalent US dollars converted using a Purchasing Power Parity Price Index) growth per year in each OECD and WEI program country (%) between 1995 and 2005. (Source: http://www.worldbank.org/data/onlinedatabases.html)). GDP per head (1995-2005): The absolute GDP per capita (in equivalent US dollars converted using a Purchasing Power Parity Price index) in the OECD and WEI countries. (Source: http://www.worldbank.org/data/onlinedatabases.html). Expenditure on educational institutions as a percentage of GDP (1995-2005): The direct and indirect expenditures on educational institutions, in primary and secondary education expressed as a percentage of the Gross Domestic Product 27.

(Source: Education at a Glance, OECD (1997-2007), table B2.1 & B2.2).

Pupil/Teacher ratio (1995-2005): Ratio of student numbers to teaching staff for public and private institutions in primary, lower secondary and upper secondary education.

(Source: Education at a Glance, OECD (1997-2007), table D2.2 continued).

Growth in the size of the population at the age of primary/lower secondary and upper secondary education (%); 1995-2005: This is calculated as a proxy of the growth in demand for education, using the relative change in the population at the age of primary/lower secondary and upper secondary education between 1995 and 2005. (Source: Education at a Glance, OECD (1997-2007), table A21).
Intended instruction time (1995-2005) 28 : Average number of hours per year of total compulsory instruction time.

(Source: Education at a Glance, OECD (1997-2007), table D4.1 (1995-1996) & D1.1 (1998 onwards).
GINI index: Taken from the World Income Inequality Database, developed by United Nations University (UNU)-World Institute for Development Economics Research (WIDER).

27 28

The data for 2002 has been imputed using information from 2001. For the years before 2000 it is an average for the three educational levels. There are no data available for 1995. 43

Imputation of Gini indices: For the whole set of countries 2005 Gini index=2004. On top of that we have interpolated the data for the following countries and years:
Australia Austria Belgium Brazil Chile Czeck Republic Denmark Greece Hungary Indonesia Ireland Italy Korea Malaysia Netherlands New Zealand Peru Philippines Portugal Switzerland Thailand Tunisia Turkey UK Uruguay Japan Slovenia 1999,2003 2002 2002 2000 2001,2002,2004 2004 1996 2001 2004 1998, 2000, 2001, 2003, 2004 2002 2003 1999-2004 1996-1998; 2000-2004 2004, 2005 1998-2005 1995, 1996; 2004, 2004 1995-1999; 2004 2002,2003 1995,1996 1995; 2002-2004 1995-1999; 2001-2004 1995-1999;2004 2004 1999 1996;1999-2004 2004

PISA: Is the Programme for International Student Assessment. It has been conducted in 2000, 2003 and 2006, assessing math, reading and science skills of 8th grade students. Source: http://www.oecd.org/document/43.html PIRLS: Is a program to evaluate the Progress in International Reading Literacy Study. It undertook an International Student Achievement in Reading (years 2001 and 2006, with 26 and 40 countries assessed respectively): 4th grade students. Source: PIRLS 2006. Assessment Framework and Specifications. TIMSS: Trends in International Mathematics and Science Study: 4th grade (1995, 1999 and 2003) and 8th grade assessments (1995, 1999 and 2003). In 2003, some 46 countries participated in TIMSS, at either the fourth- or eighthgrade level, or both. Source: NCES (2004). Highlights from the Trends in International mathematics and Science Study (TIMSS) 2003.

44

Educational outcomes at different point is time (8th grade Students):

1995 1999 2000 PISA- math PISA - reading PISA - science 2003 PISA - math PISA - reading PISA - science TIMSS - math TIMSS - science 2006 PISA- math PISA - reading PISA - science

TIMSS - math TIMSS - science

TIMSS - math TIMSS - science

Countries included in each estmate:

Table 3a Table 3b Table 4 Spec. 1 Spec. II Spec. 1 Spec. II Spec. 1 Spec. II Argentina Argentina Austria Austria Austria Austria Austria Austria Austria Austria Belgium Belgium Belgium Belgium Belgium Belgium Belgium Belgium Brazil Brazil Brazil Brazil Chile Chile Chile Chile Chile Chile Chile Chile Czech Republic Czech Republic Czech Republic Czech Republic Czech Republic Czech Republic Czech Republic Czech Republic Denmark Denmark Denmark Denmark Denmark Denmark Denmark Denmark Finland Finland Finland Finland Finland Finland Finland Finland France France France France France France France France Germany Germany Germany Germany Germany Germany Germany Germany Greece Greece Greece Greece Greece Hungary Hungary Hungary Hungary Hungary Hungary Hungary Hungary Iceland Iceland Iceland Iceland Indonesia Indonesia Indonesia Indonesia Indonesia Indonesia Indonesia Indonesia Ireland Ireland Ireland Ireland Ireland Ireland Ireland Ireland Israel Israel Israel Italy Italy Italy Italy Italy Italy Italy Italy Japan Japan Japan Japan Japan Japan Japan Japan Jordan Jordan Jordan Korea Korea Korea Korea Korea Korea Korea Korea Malaysia Malaysia Mexico Netherlands Netherlands Netherlands Netherlands Netherlands Netherlands Netherlands Netherlands New Zealand New Zealand New Zealand New Zealand New Zealand New Zealand New Zealand New Zealand Norway Norway Norway Norway Norway Norway Norway Norway Philippines Philippines Philippines Philippines Philippines Philippines Philippines Philippines Portugal Portugal Portugal Portugal Portugal Portugal Spain Spain Spain Spain Spain Spain Spain Spain Sweden Sweden USA UK USA Tunisia Turkey USA Tunisia USA Tunisia Turkey USA Tunisia USA Tunisia Turkey USA Tunisia USA Table 1 Table 2

45

Table A1. Descriptive Statistics for all countries by year.

Teachers starting salaries Teachers salaries after 15 years experience Teachers salaries at the top of the pay scale Teachers salaries/hour Teachers salaries/GDP per head Teachers percentile position in the earnings distribution Teaching hours per year Pupils/teachers ratio Women fraction among teachers staff Teachers as a fraction of the labour force Expenditure on education/GDP (%) GDP growth (%) GDP per head Standardised Scores 8th grade students N. Obs. 51 51 51 46 51 51 86 53 38 54 117 195 22 1995 Mean Std. Dev. N. Obs. 20122.35 5646.14 86 27403.53 7979.89 86 33843.33 11278.64 83 36.57 11.57 84 1.81 0.58 86 0.57 0.11 83 732.06 162.07 83 15.69 6.29 80 58.00 14.65 72 2.94 0.73 54 53 5.19 3.97 117 15274.71 7689.79 195 2.6e-7 1.02 1996 Mean Std. Dev. N. Obs. 16819.98 8263.61 96 23386.01 11810.25 96 29270.20 15093.32 94 29.68 20.48 64 1.98 0.93 96 0.56 0.13 91 739.61 158.84 61 18.34 7.75 90 63.19 15.43 78 2.95 0.74 57 2.11 0.72 82 5.07 2.09 117 15539.96 7755.08 195 1998 Mean Std. Dev. N. Obs. 16774.14 8165.54 108 23411.26 11386.71 108 31289.82 13879.45 105 40.35 15.49 101 1.87 0.78 108 0.57 0.13 99 687.39 154.80 98 18.72 7.15 104 63.21 16.16 84 2.97 0.73 99 2.08 1.10 102 3.11 3.65 117 16188.07 8228.63 195 30 1999 Mean Std. Dev. N. Obs. 16276.72 7902.36 110 21774.62 10864.02 110 27527.63 13567.49 107 30.16 17.50 97 1.87 0.88 110 0.57 0.13 102 749.02 158.36 98 18.43 7.26 89 63.20 16.06 76 3.24 0.82 99 1.99 0.91 99 3.58 3.03 117 16486.04 8458.33 195 7.4e-8 1.01 74 2000 Mean Std. Dev. 16470.20 7990.71 22401.31 11286.90 28575.91 14001.16 32.32 18.15 1.76 0.75 0.56 0.13 751.05 170.80 18.46 7.57 64.07 15.62 3.24 0.82 1.94 0.87 5.84 1.95 16883.17 8664.93 -2.1e-4 0.99

Teachers starting salaries Teachers salaries after 15 years experience Teachers salaries at the top of the pay scale Teachers salaries/hour Teachers salaries/GDP per head Teachers percentile position in the earnings distribution Teaching hours per year Students/teachers ratio Women fraction among teachers staff Teachers as a fraction of the labour force Expenditure on education/GDP (%) GDP growth (%) GDP per head Standardised Scores 8th grade students

2001 2002 2003 2004 2005 N. Obs. Mean Std. Dev. N. Obs. Mean Std. Dev. N. Obs. Mean Std. Dev. N. Obs. Mean Std. Dev. N. Obs. Mean Std. Dev. 107 17037.30 8239.39 110 16845.34 8302.97 110 17468.02 8488.14 84 20637.28 6787.04 78 21611.96 6407.24 107 23191.54 11386.94 110 23094.50 11705.49 110 23800.25 11871.67 80 28634.64 9325.19 80 29020.10 8931.89 107 28873.35 13788.28 101 29252.65 14689.59 110 29240.67 14241.85 81 33671.10 12051.85 78 34877.31 11579.63 90 33.33 17.36 68 40.17 15.60 63 40.84 13.03 60 41.67 13.16 60 42.01 12.52 107 1.69 0.67 110 1.66 0.63 110 1.60 0.55 80 1.44 0.46 80 1.40 0.44 102 98 107 93 99 105 117 195 0.56 749.15 17.43 64.13 3.18 2.05 3.21 16995.61 0.13 169.70 7.14 16.44 0.88 0.94 2.71 8776.10 102 95 111 90 99 102 117 195 0.56 753.79 17.20 63.72 3.21 2.05 2.81 17187.50 0.13 174.06 6.76 16.69 0.83 0.91 3.61 8883.30 102 87 114 72 99 90 117 195 128 0.57 762.15 16.67 64.54 3.18 2.03 4.13 17417.61 -1.7e-8 0.13 178.77 6.43 15.82 0.85 0.85 1.97 8909.32 1.0 81 87 78 99 75 117 195 749.35 15.38 66.19 3.13 2.09 6.75 18008.70 170.31 6.03 14.93 0.83 0.77 2.12 9102.54 75 87 87 102 75 117 195 98 743.11 15.05 66.69 3.25 2.09 6.50 18451.75 3.4e-8 175.49 5.23 15.14 0.83 0.77 2.51 9226.89 1.00

46

Table A2 Descriptive Statistics for all years by country.

(The rows for each country denote : number of observations/mean/standard deviation)
Starting salaries Argentina Australia Austria Belgium Brazil Chile Czech Republic Denmark Finland France Germany Greece Hungary Iceland Indonesia Ireland Israel Italy Japan Jordan Korea Malaysia Mexico 21 9338.4 2567.7 27 23352.5 1737.2 30 20679.0 874.8 30 22826.9 2957.6 21 8894.2 3364.1 24 10350.0 1427.8 29 9435.4 3174.4 30 25760.1 1570.4 29 21279.7 3781.7 30 20108.1 1211.9 30 31602.3 2648.1 30 17525.8 2366.5 27 7022.5 2221.0 21 17486.2 2078.4 20 1595.8 963.9 30 21234.0 1067.0 9 10891.9 553.6 30 19661.3 1224.0 21 20163.0 696.5 15 6835.5 1024.8 29 22451.2 989.1 21 10388.5 3174.0 19 11619.8 1473.9 Salaries After 15 21 13028.9 3566.4 26 34366.8 569.2 30 27288.8 2019.5 30 31652.1 4622.2 21 10829.9 3133.3 24 12144.3 1565.1 29 12221.4 4034.1 30 32001.2 3501.2 30 27315.4 3817.4 30 26572.0 1302.2 30 38722.2 3104.1 30 21293.4 2590.8 27 9796.1 2892.3 21 20259.5 3034.4 20 2461.2 1304.7 30 33721.8 1609.7 6 13309.7 853.6 30 24142.2 1725.2 21 37574.4 520.0 15 9522.8 1610.6 29 38268.2 2039.8 21 16817.0 5064.1 17 14652.0 1960.8 Salaries at top 21 15300.7 4494.9 27 34394.6 546.4 30 42868.5 2858.3 30 38848.3 5780.1 18 13158.2 3121.9 24 15762.8 2532.8 29 15501.4 4853.3 30 32369.0 3847.9 30 28285.5 3706.9 30 38135.9 1528.0 30 41817.0 3240.6 30 25756.9 2942.0 27 12954.0 3797.5 21 22768.2 3870.5 20 4133.5 2245.5 30 38354.0 1955.4 9 18598.3 946.8 30 29782.0 2274.6 21 48299.5 1061.7 15 24465.6 1240.6 29 61506.2 3042.0 21 22434.8 6991.7 19 24840.8 3277.9 Salaries Wage Teaching Pupils/ Women fraction /hour Perc. hours teacher 9 10.3 6.3 23 41.5 2.0 30 41.7 6.1 30 44.4 9.6 9 8.7 5.8 9 15.3 1.6 29 18.6 6.4 30 53.1 12.6 24 46.9 9.7 30 38.3 7.8 30 53.0 6.5 30 33.4 7.4 27 16.5 5.2 21 34.4 8.7 12 2.8 1.5 30 42.9 5.1 0 . . 30 37.4 5.7 9 69.7 9.3 9 13.2 1.5 27 65.1 14.5 12 22.7 7.6 8 15.6 1.1 21 0.5 0.1 20 0.6 0.0 24 0.5 0.0 24 0.5 0.1 21 0.6 0.1 21 0.5 0.0 23 0.4 0.1 24 0.6 0.0 24 0.6 0.0 24 0.5 0.0 24 0.7 0.0 24 0.6 0.0 21 0.4 0.1 0 . . 20 0.3 0.2 24 0.6 0.1 0 . . 24 0.5 0.0 15 0.7 0.0 0 . . 23 0.8 0.0 21 0.6 0.1 0 . . 27 846.9 53.9 24 830.3 36.5 30 659.6 58.6 30 731.4 62.9 24 766.8 58.8 24 863.8 5.6 30 662.3 66.4 30 612.5 46.0 27 596.1 75.9 30 718.6 134.1 30 735.4 41.5 30 671.6 80.4 28 598.2 104.7 21 602.0 67.6 15 912.0 254.7 30 796.0 88.0 0 . . 30 654.5 68.8 18 540.5 72.9 21 792.8 85.2 28 596.4 127.7 18 771.7 7.6 20 919.9 158.1 21 18.0 4.8 23 15.2 2.3 27 11.2 2.1 19 11.5 2.0 27 27.2 8.0 27 31.0 3.9 30 15.5 3.3 30 11.5 1.3 25 14.5 3.0 27 16.0 4.2 30 16.4 2.8 30 11.2 2.1 30 11.1 0.8 22 11.9 1.7 18 20.0 3.1 22 18.2 3.1 15 14.8 3.0 30 10.5 0.5 24 16.8 2.9 19 19.0 2.0 30 24.2 5.4 21 19.1 1.6 21 28.9 3.7 18 75.2 10.4 0 . . 27 68.6 16.0 27 64.5 10.3 21 84.7 8.5 24 65.4 12.2 27 75.1 12.6 30 56.4 13.0 29 67.9 6.3 26 65.5 11.5 30 58.9 17.6 12 56.8 6.4 27 77.8 12.1 22 65.2 16.0 18 44.4 7.2 23 68.4 12.6 15 76.2 8.0 30 76.4 14.4 15 42.9 17.6 14 50.2 14.1 30 53.5 16.6 16 61.1 2.6 21 51.8 10.7 Teachers Expen. GDF GDP/ /labour /GDP growth head force 21 18 30 21 4.4 2.0 2.5 1.3 0.0 0.9 7.1 0.4 21 27 30 26 3.1 2.4 4.6 1.5 0.0 1.1 0.6 0.3 30 27 30 30 3.2 2.2 4.0 1.2 0.0 0.6 1.0 0.2 9 24 30 30 4.1 1.8 4.0 1.4 0.0 0.6 1.4 0.3 21 21 30 21 3.1 1.8 3.1 1.8 0.0 1.0 2.3 0.5 21 27 30 24 2.1 2.4 5.5 1.6 0.0 0.8 3.1 0.4 21 27 30 29 2.7 1.7 5.6 0.9 0.0 0.4 2.5 0.1 30 27 30 30 3.4 2.6 3.9 1.4 0.1 1.5 1.4 0.3 21 27 30 30 3.0 2.1 5.3 1.3 0.0 0.6 1.1 0.3 30 27 30 30 3.2 2.3 3.8 1.3 0.1 0.6 1.0 0.3 30 27 30 30 2.3 1.8 3.3 1.8 0.1 0.4 1.0 0.3 12 21 30 30 2.9 1.2 5.5 1.5 0.2 0.2 1.5 0.3 30 27 30 27 3.6 1.7 6.5 0.8 0.8 0.4 1.3 0.2 24 3 30 21 3.6 2.2 5.2 0.8 1.0 0.8 2.5 0.2 21 15 30 20 2.8 0.7 4.3 1.0 0.5 0.3 6.5 0.6 30 27 30 30 3.7 1.8 8.1 1.5 0.2 0.8 2.2 0.5 0 24 30 9 . 2.5 3.2 0.6 . 0.2 3.8 0.1 30 24 30 30 3.7 1.8 3.3 1.2 0.1 0.4 1.2 0.2 30 27 30 21 1.9 1.7 2.9 1.6 0.0 0.6 3.7 0.2 0 12 30 15 . 2.6 4.6 3.1 . 1.5 2.9 0.6 30 27 30 29 1.8 2.4 6.5 2.7 0.0 0.7 4.8 0.5 21 12 30 21 3.5 2.1 4.8 2.5 0.0 0.4 5.2 0.8 30 24 30 16 3.4 2.4 4.2 1.8 0.3 1.1 4.1 0.3 Standardised 8th Grade Scores 2 -2.0 0.1 13 0.6 0.4 9 0.4 0.2 15 0.6 0.3 9 -2.4 0.7 6 -1.2 0.4 9 0.4 0.3 9 0.2 0.4 9 1.3 0.3 9 0.4 0.2 9 0.2 0.4 9 -0.5 0.4 15 0.3 0.4 9 0.3 0.2 10 -1.6 0.4 9 0.6 0.3 5 -0.2 0.3 13 -0.2 0.3 15 1.1 0.4 4 -0.7 0.3 15 1.3 0.3 4 0.3 0.2 9 -1.8 0.4

47

Netherlands New Zealand Norway Peru Philippines Portugal Slovenia Spain Sweden Switzerland Thailand Tunisia Turkey UK USA Uruguay

30 24307.4 1673.0 30 15495.5 1127.8 30 21616.4 2921.0 15 4500.3 369.1 21 8733.6 1417.2 26 16170.1 629.6 0 . . 29 25689.4 2041.5 29 18923.3 1959.3 29 36396.5 5055.9 18 5260.8 356.3 15 14214.9 2653.9 21 10443.6 2297.0 26 21184.3 1928.2 30 24352.7 1037.9 18 8912.7 1562.5

30 35125.6 6688.4 30 28103.6 2780.9 30 25241.5 3091.1 15 4500.3 369.1 21 9448.2 1692.1 29 25249.1 1335.9 0 . . 29 29983.6 2301.1 29 23188.4 1718.5 29 47996.6 6422.1 18 12928.9 876.3 15 14842.7 2810.1 21 11900.2 2244.3 26 32724.2 1861.3 30 33530.9 2196.4 18 10639.8 1845.1

30 41402.8 7510.4 30 28104.3 2779.3 30 26841.6 3461.1 15 4629.9 203.3 21 10442.0 1856.8 29 42941.9 2195.6 0 . . 29 37567.7 2713.7 17 26556.4 2489.7 29 56135.2 6974.1 15 27329.8 6614.0 15 16657.8 3122.0 21 13806.8 2409.5 26 34327.7 5381.0 24 41732.9 1731.8 15 21902.5 7637.1

29 40.7 10.9 30 30.1 1.9 30 41.6 7.8 9 7.9 0.7 10 9.6 1.4 29 38.7 6.9 0 . . 29 45.3 11.9 5 35.7 5.1 14 60.0 17.4 12 14.7 7.8 11 26.8 14.7 20 19.7 3.9 4 42.9 2.4 24 32.6 1.2 9 14.4 10.1

24 0.6 0.1 24 0.6 0.0 24 0.4 0.0 15 0.4 0.0 21 0.8 0.0 23 0.7 0.0 0 . . 23 0.6 0.0 23 0.5 0.0 23 0.7 0.0 18 0.8 0.0 15 0.7 0.1 15 0.7 0.0 20 0.6 0.0 24 0.5 0.0 18 0.5 0.1

30 869.6 67.8 27 963.4 23.9 30 628.9 83.6 12 677.5 71.6 27 1176.0 0.0 30 666.2 122.8 0 . . 30 697.1 151.9 6 576.0 42.9 18 802.8 95.3 18 650.9 73.0 21 609.1 90.2 3 384.0 41.6 6 760.0 31.0 30 1049.6 81.0 9 662.9 75.4

21 16.6 1.0 30 17.8 3.8 27 10.2 1.4 14 24.2 5.5 21 34.4 6.0 19 10.0 1.8 3 13.6 2.1 30 13.7 2.9 30 13.3 1.3 23 13.2 2.2 15 21.9 2.4 15 21.6 2.8 28 23.5 9.4 30 17.3 3.3 30 15.8 1.0 18 19.4 5.8

23 55.3 18.0 29 67.3 12.0 21 62.1 13.6 6 51.2 10.2 18 80.2 5.4 15 73.0 6.3 3 82.3 13.6 26 59.3 8.9 30 63.8 13.6 12 50.5 17.4 3 62.3 1.2 15 44.4 3.9 9 42.8 1.9 26 66.4 11.3 30 67.7 14.8 0 . .

21 3.4 0.0 30 3.1 0.5 21 4.4 0.0 21 3.8 0.0 21 2.3 0.0 0 . . 0 . . 30 3.7 0.1 30 3.4 0.0 21 2.8 0.0 21 2.1 0.0 21 6.0 0.0 30 2.7 0.3 30 3.0 0.1 30 2.7 0.2 21 2.8 0.0

27 1.9 0.8 21 2.6 0.8 21 2.4 0.8 12 1.9 1.1 15 2.6 1.9 24 2.3 0.8 0 . . 11 1.6 0.5 27 2.5 0.8 27 2.4 0.5 18 1.7 1.0 0 . . 18 1.5 0.6 24 1.8 0.7 0 . . 15 1.2 0.6

30 4.1 1.4 30 4.5 1.6 30 4.4 1.5 30 4.1 3.2 30 4.1 2.4 30 3.8 1.6 30 6.0 0.8 30 4.9 1.1 30 4.9 1.4 30 3.3 1.5 30 5.2 5.7 30 5.7 1.9 30 4.5 5.6 30 4.6 0.7 30 4.0 1.2 30 3.3 7.7

30 1.6 0.4 30 1.6 0.3 30 0.9 0.1 16 1.4 1.0 21 3.0 0.6 29 1.8 0.3 3 1.1 0.0 29 1.8 0.4 29 1.1 0.2 29 1.9 0.5 21 2.2 0.9 17 2.5 1.1 24 1.8 0.6 26 1.5 0.4 30 1.2 0.2 21 1.3 0.5

12 0.8 0.3 15 0.48 0.7 13 -0.1 0.5 4 -2.1 0.2 8 -0.3 0.3 5 -0.1 0.9 9 -0.1 0.6 13 0.6 0.3 9 0.5 0.6 6 -1.2 0.1 10 -1.7 0.6 6 -1.0 0.2 6 0.6 0.2 14 0.02 0.5 6 -1.1 0.1

48

APPENDIX B. Teacher Pay relative to GDP per Head.

The power to attract individuals to become teachers as measured by the ratio of teacher salaries to gross domestic product (GDP) per capita. Teacher salaries relative to per capita gross domestic product (GDP) are an indication of the extent to which a country invests in teaching resources relative to the financial ability to fund educational expenditures 29. A high salary relative to per capita GDP suggests that a country is making more of an effort to invest its financial resources in teachers. Relative to per capita GDP, teacher salaries are relatively low in the Czech Republic, Hungary, and Indonesia and relatively high in Philippines and Denmark. Wealthier countries do not necessarily spend a greater share of their wealth on educational resources, however. Although the Czech Republic and Hungary have both relatively low GDP per capita and low teacher salaries, other countries with GDP per capita below the OECD average, including Philippines and Tunisia, have comparatively high teacher salaries. In contrast, Norway and France, two countries with relatively high GDP per capita, have below-average teacher salaries. When using the teachers wage relative to GDP per head we can clearly only control for economic growth and not the absolute size of the wealth in the country. The results relating GDP growth to teachers relative salaries compared to GDP per head indicate that there exists a negative relationship between the changing wealth of a country and their teachers salaries. This may be due to the rate at which an economy grows is largely determined by the productivity of the private sector. Those countries which have private sectors which are growing more rapidly are more likely to be leaving their public sector workers behind in relative pay terms.

29

The mean public funding of education is 81 % in the OECD countries. 49

Table B1. Estimation explaining the variation in teachers' salaries/GDP per head from 1995-2005.
Teachers' salaries/GDP per head Fixed Random OLS Effects Effects -0.4312*** -0.0630 -0.1320*** (0.0389) (0.0424) (0.0402) 0.0539*** 0.0153 0.0198 (0.0184) (0.0132) (0.0135) 0.0027 0.0030** 0.0032** (0.0019) (0.0014) (0.0014) 0.2006*** 0.1914*** 0.2012*** (0.0608) (0.0317) (0.0331) 0.5801*** 0.3550*** 0.3813*** (0.0887) (0.0534) (0.0550) 0.1334*** 0.0104 0.0167 (0.0355) (0.0190) (0.0198) 0.0099** 0.0107*** 0.0123*** (0.0046) (0.0032) (0.0032) 0.0008 (0.0011) -0.0216** (0.0090) 1.6342*** (0.2429) 415 18.74*** -0.0007 -0.0006

Teachers and educational staff as a fraction of the labour force Teaching hours per year (100s) Women fraction of teaching staff (%) Lower secondary dummy Upper secondary dummy Expenditure on educational institutions as a percentage of GDP Pupil/Teacher ratio Growth in the size of the population at the age of primary/lower secondary and upper secondary GDP growth (%)

(0.0011) (0.0011) -0.0101* -0.0110** (0.0054) (0.0056) Year dummies 1.0353*** 1.1520*** Constant (0.2126) (0.2175) 415 415 Observations 7.95*** F-statistic 0.27 0.26 R-squared Within 0.35 0.39 R-squared Between 0.45 0.30 0.38 R-squared Overall Note: Standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%.

50

51

Table C1. PISA Standardised Score Determinants. Barro & Lee (2001), (2000, 2003, 2006).
Specification I Log (GDP per capita) Pupil/Teacher ratio Log (Teachers salary after 15 years experience) Log (Educational expenditure) Log (Teaching hours)
Expected years of education Asia Pacific

SUR estimates Specification II Specification III

Specification IV

-0.7555*** (0.1127) 0.0263*** (0.0070) 0.8213*** (0.0721) 0.1598*** (0.0475) 0.0418 (0.1085)

-0.8087*** (0.1040) 0.0291*** (0.0067) 0.8080*** (0.0676) 0.1685*** (0.0439) -

Constant Observations R2 (for each equation)

-8.8703*** (1.1268) 147 0.2517 0.1545 0.2995 0.2952 0.0719 0.2749 0.2472 0.1878 0.2325

-8.4765*** (0.7209) 159 0.2541 0.1465 0.2886 0.2864 0.0751 0.2865 0.2538 0.1828 0.2309

-0.908*** (0.139) 0.006 (0.008) 0.673*** (0.085) 0.283** (0.112) -0.929*** (0.176) 0.044** (0.022) 1.044*** (0.075) -1.762 (1.341) 125 0.579 0.527 0.582 0.659 0.399 0.527 0.522 0.392 0.569

-0.900*** (0.150) 0.010 (0.008) 0.572*** (0.091) 0.516*** (0.113) 0.072*** (0.021) 0.771*** (0.080) -7.238*** (0.835) 136 0.520 0.436 0.542 0.599 0.354 0.481 0.446 0.390 0.459

Standard errors in parentheses * significant at 10%; ** significant at 5%; *** significant at 1%

52