Human capital as a factor of growth and promotion of employment at the regional level: the case of France and Germany

Transcription

1 QG EN-C Rue Belliard 101, B-1040 Brussels Tel. (32-2) Fax (32-2) Human capital as a factor of growth and promotion of employment at the regional level: the case of France and Germany EN

2 Human capital as a factor of growth and promotion of employment at the regional level: the case of France and Germany CoR Studies E 2/2005 Brussels, July 2005

3 This study was produced by the National Centre for Scientific Research (CNRS), Delegation Paris A, France, and the Centre for European Economic Research (ZEW), Germany. It does not represent the official views of the Committee of the Regions.

4 EUROPEAN UNION Committee of the Regions Foreword Since 2000, a large part of Community intervention has been structural in character and has aimed at strengthening areas such as productive activities, infrastructure and entrepreneurial environment. However, as already recognised by the European Commission and underlined by the Committee of the Regions over recent years, human capital plays a fundamental role in a knowledge-based economy. Moreover, in the context of the revised Lisbon strategy, it also constitutes a key element for reducing the gap between European regions and improving regional competitiveness and labour force quality. This study assesses the contribution of human capital to growth and employment in France and Germany. It describes in depth, and through a solid empirical analysis, the relations existing between a set of relevant indicators for improving human capital and its output in terms of productivity, wages and employment. It provides an estimation of private and social returns from schooling and training. Finally, it provides insights into the links between education and public finances and expenditures. High quality in human capital produces positive effects on economic growth and employment. It directly affects both individual (in terms of salary, participation in the labour market and employability) and regional positions. On the whole, regions with higher levels of human capital present higher outputs and income levels, and are more competitive, as the French and German cases analysed in the study show. Differences in human capital endowment can then partly explain the disparities between regions in terms of productivity. Schooling and training are worthy investments for society as a whole. Higher and better education levels can then be considered as a necessary and suitable objective to be pursued as well as a relevant instrument of regional development policies. Regional policy-makers should then take into account this important factor in the planning of regional strategies for the high returns it could provide. Promoting human resources at regional and local levels is a fundamental step towards regional convergence, growth and employment in an increasingly competitive Europe. Peter Straub President of the Committee of the Regions

5

6 Section I The case for France Marc Gurgand, CNRS-PSE and Crest, Paris David Margolis, CNRS-TEAM and Crest, Paris Mathieu Perona, ENS, Paris 1. Introduction General introduction French NUTS 2 and Objective 1 regions Educational expenditure, enrolment and attainment Regional expenditure Educational attainment Educational flows Econometric estimates of the effects of schooling Schooling and wages Participation and employment Schooling and aggregate productivity Public policies, returns to schooling and incentives Methodology Observed and raw returns Decomposing returns A typology of regions Conclusion Main findings Policy recommendations References List of abbreviations for region names... 52

7 Section II The case for Germany Andreas Ammermüller, ZEW Anja Kuckulenz, ZEW Charlotte Lauer, ZEW Thomas Zwick, ZEW 1. Executive summary Introduction Educational expenditure and attainment in the German states Regional expenditure on secondary and university education School attainment and other educational indicators The evolution of regional school attainment levels Attainment by cohort and the likely evolution of educational disparities Other indicators of educational performance Econometric estimates of the effects of schooling Schooling and wages Schooling, labour force participation and employment Schooling and aggregate productivity Private returns to schooling Methodology Average returns across all educational levels Results by educational level The long-term impact of schooling expenditure on public finances Methodology Results The social rate of return to schooling Data and sources Parameter values Results The relative returns to investment in schooling and in physical capital A typology of regions References Appendix The German education system Data sources

8 Section I The case for France 1. Introduction 1.1. General introduction Using the Structural Funds, mainly the European Regional Development Funds and the European Social Fund, the European Union leads an ambitious policy aiming at the reduction of imbalances between regions. Until recently, investment in infrastructure has been the main tool of this policy. However, the European Commission has, for several years, underlined the crucial role of human capital in a knowledge-based economy. Along the lines of the Lisbon strategy to narrow the gap in human capital between regions, the European Commission has launched a series of surveys to evaluate the contribution of human capital discrepancies to the gap in employment and growth between regions. For decades, the international research on education economics has argued that schooling is a major input to sustainable growth and employment, based both on theoretical and empirical arguments. Education is believed not only to increase the level of output that can be attained in a given state of the technology (Solow, 1957; Mankiw et al., 1992), but also to foster technological innovation and adoption, thereby increasing the potential for long-run growth (Romer, 1990; Lucas, 1992; Benhabib and Spiegel, 1994). This becomes a major issue in a knowledge-based economy, not only for growth per se, but also for employment, as emphasised in the Lisbon strategy. At the country level, a report by Angel de la Fuente (2003) underlined that the strong returns to human capital investment made it a tool of choice for reducing economic imbalances between regions. However, data used for cross-country comparisons often lack homogeneity and present significant measurement error issues, which may lead researchers to conclude erroneously that human capital has only a weak influence on growth. In the absence of a sufficiently robust existing data set, it becomes important to construct new data for reconsideration of the initial study. For this reason, the research programme has been extended to the regional level, first in Spain and then in other member countries. The main finding of the already-completed studies is that human capital investment returns are better than the return to infrastructure investment in poorer regions. This finding hinges on the integration of the yields in terms of increased employment probabilities of individual human capital investment. Taking into account this effect leads to private, social and even fiscal returns significantly above those found by former studies at the European level. In France, the regional differences in GDP per capita have increased since the mid-1980s (Davezies, 2001). Also, France has one of the highest levels of regional wage dispersion in Europe (OECD, 2000). Understanding the sources of these discrepancies and their dynamics is thus a major issue for public policy.

9 8 SECTION I THE CASE FOR FRANCE Consistent with this objective, this study analyses the contribution of human capital to regional economic development in France. This instrument is partly managed at the State level through nationwide educational institutions but is increasingly dependant on local financing. Its geographical distribution also results from individual mobility in part driven by regional attractiveness. We have three main objectives: (1) to provide coherent series on education, employment and production at the regional level and describe correlations between them in all French regions; (2) to compute individual and social marginal rates of return to the investment in human capital, contrast them by regions and study the incentives inherent in existing public policies with respect to education; (3) to evaluate the potential gain from educational policy for Objective 1 regions (Guadeloupe, Martinique, RØunion, Guyane) and the transitory Objective 1 region (Corse) and more generally for regional convergence. Transitory Objective 1 districts Valenciennes, Avesnes and Douai will only be considered when data allow for analysis of such small areas. For comparability, the study follows closely the methodology used in De la Fuente et al. (2003). Although this report focuses on formal schooling, training over the life cycle is also important to ensure that ageing workers keep pace with technological progress. There exists some research on the question in France (Gauron, 2000; Foug re et al., 2001), but this remains an important subject for future study ( 1 ) French NUTS 2 and Objective 1 regions France is divided into 26 NUTS 2 regions. Figure 1.1 shows a map of metropolitan France with each of its NUTS 2 regions, as well as darker lines distinguishing its NUTS 1 regions. Figure 1.2 shows a world map with the location of France s døpartements d outre-mer (DOMs) and territoires d outre-mer (TOMs). Each of France s DOMs makes up its own NUTS 2 region, and each of these is an Objective 1 region. Corse (Corsica, in English) makes up another NUTS 2 region by itself, and is a transitory Objective 1 region. The remaining transitory Objective 1 regions (Valenciennes, Avesnes and Douai) are all part of the NUTS 2 region Nord-Pas-de-Calais. However, the presence of the dynamic city of Lille in the same NUTS 2 region implies that Nord-Pas-de-Calais is a very heterogeneous region, and therefore we do not devote much attention to this region when we focus on Objective 1 regions. The neighbouring region of Picardie has an economy more similar to that of Valenciennes, Avesnes and Douai, and thus we will occasionally comment on this region as a proxy for what may be happening in the southern part of Nord-Pas-de-Calais. There is an important amount of variability across regions in France in the level of economic activity. At one end of the scale is the le-de-france region, which includes Paris and the surrounding døpartements. As Figure 1.3 shows, GDP per person in le- ( 1 ) The authors would like to thank the following people for their help and contributions to the preparation of this report: FranÅoise Martinez and Mich le Jacquot (DEP, Minist re de l Øducation nationale), Patrick Aubert and GØrard Forgeot (INSEE), Angel De la Fuente and Vittorio Campanelli.

10 INTRODUCTION 9 de-france is over 50 % higher than the all-france average. In addition, le-de-france is by far the largest region in terms of population, comprising over 18 % of France s population. These two factors combine to draw GDP per person upwards to such a degree that only three regions le-de-france, Alsace and Rhône-Alpes (the region where Lyon is situated) have a GDP per person above the population-weighted average. Figure 1.1: NUTS 2 regions in Metropolitan France Source:

11 10 SECTION I THE CASE FOR FRANCE Figure 1.2: DØpartements and territoires d outre-mer Source:

12 INTRODUCTION 11

13 12 SECTION I THE CASE FOR FRANCE At the other end of the spectrum are the Objective 1 regions (the DOMs and Corse). They have the lowest GDP per person in France and are also the smallest in terms of population. GDP per person in the poorest region (RØunion) is only 53 % of the average for the whole of France, although (as noted above) this average is drawn upwards by the preponderance of le-de-france in the statistics. Even the richest of the DOMs (Martinique) only has a GDP per person amounting to 63 % of the average for all of France, although Corse is significantly better off with a GDP per person of 76 % of the overall average, only 2 percentage points behind that of Languedoc-Roussillon and 3 percentage points behind Nord-Pas-de-Calais. The presence of Lille in this last region suggests that the other transitory Objective 1 regions (Valenciennes, Avesnes and Douai) may be worse off than Corse, although data are not available at a level of detail fine enough to test this idea. 2. Educational expenditure, enrolment and attainment 2.1. Regional expenditure French decentralisation policies have progressively transferred the management of school buildings to the local authorities. Municipalities are responsible for primary schools, døpartements manage lower secondary schools (coll ges) and regions are in charge of upper secondary schools (lycøes). The State remains responsible for the general organisation and supervision of teaching and curricula and employs the staff. It is also in charge of most of the public higher education system, although the local authorities were involved in the development of universities during the 1990s (Plan UniversitØ 2000). One of the dangers of local initiatives at this level is the proliferation of small-scale universities in medium-sized towns. Because public spending at the regional level is not available in a unified and homogenous source over time, it is difficult to analyse the evolution of the disparities in, and the share of, local and State expenditures in total spending. The data presented in Table 2.1 are for 1998 only and were computed by the Ministry of Education, with the exception of the contribution by the municipalities, which had to be built separately ( 2 ). Private spending is not available at the regional level either. We computed it from a household expenditure survey (EnquÞte Budget des familles ). This data source contains very detailed spending and income information. Unfortunately, the version of these data that is publicly available only contains NUTS 1 identifiers, whereas the rest of this report is at the NUTS 2 level. In addition, overseas døpartements are not present. ( 2 ) At the municipal level, the only information we have is average per student spending for five classes of town size. We merged this information with a file containing the town size and number of children in primary school for every French municipality. We allocated a spending to each of them, based on its size and its number of children in school. Because the level of per student spending varies relatively little across the five size groups and the distribution of municipalities by size is not very contrasted across regions, the resulting regional average spending is very homogenous.

14 EDUCATIONAL EXPENDITURE, ENROLMENT AND ATTAINMENT 13 Table 2.1 presents the disparities in public per student spending at four schooling levels and the relative contributions of State and local spending ( 3 ). It is well known that, in the French system, spending at the secondary level is high relative to spending on higher education. To a large extent, this is explained by the cost of vocational and professional training, the availability of a large spectrum of courses and the very high teacher pupil ratio in the system of classes that prepare students for the entrance examinations to elite higher education institutions (classes prøparatoires aux grandes Øcoles). It is often claimed that a substantial effort in favour of higher education is warranted if France is to be an active contributor to innovation at the international level (e.g. Aghion and Cohen, 2004). Table 2.1: Normalised public spending per student, 1998 Primary Lower secondary Municipalities (%) State (%) Total DØpartements (%) State (%) Total le-de-france Rhône-Alpes Alsace Haute-Normandie Champagne-Ardenne Bourgogne Provence-Alpes-Côte d'azur Aquitaine Pays de la Loire Centre Franche-ComtØ Midi-PyrØnØes Bretagne Auvergne Basse-Normandie Lorraine Limousin Poitou-Charentes Picardie Nord-Pas-de-Calais Languedoc-Roussillon Corse DOMs France (%) France spending (EUR, 2004) Coefficient of variation (%) ( 3 ) Unless stated, all tables are presented by decreasing regional GDP per capita.

15 14 SECTION I THE CASE FOR FRANCE Upper secondary Tertiary Regions (%) State (%) Total State le-de-france Rhône-Alpes Alsace Haute-Normandie Champagne-Ardenne Bourgogne Provence-Alpes-Côte d'azur Aquitaine Pays de la Loire Centre Franche-ComtØ Midi-PyrØnØes Bretagne Auvergne Basse-Normandie Lorraine Limousin Poitou-Charentes Picardie Nord-Pas-de-Calais Languedoc-Roussillon Corse DOMs France (%) France spending (EUR, 2004) Coefficient of variation (%) Source: GØographie de l'ecole (2001), Minist re de l'øducation nationale and authors' calculations. The share of spending on primary schools coming from municipalities is high, at least twice as large as that of døpartements and regions. Because of the strong variation of local spending, there is more disparity in total per pupil spending at the secondary levels than there is at the primary and tertiary levels. However, inequality is contained by the strong share of State spending on secondary education. The DOMs and, to a lesser extent, Corse are strong outliers, with high per pupil spending at almost all levels. Table 2.2 presents private (household) spending in 2000, gross and net of transfers ( 4 ). Amounts are much lower than public spending, because the private sector is not very developed in higher education and is strongly subsidised at the primary and secondary levels. As a result, net spending is often negative. There is also much more regional variation than for public spending: as expected, spending is higher in richer areas. ( 4 ) Only scholarships are considered. Transfers through the fiscal system are potentially important, and known to be regressive (CERC, 2003), but they are not computed here.

16 EDUCATIONAL EXPENDITURE, ENROLMENT AND ATTAINMENT 15 In contrast, with the exception of the high income outlier ( le-de-france) that has average spending, the public financial effort seems to compensate for regional income disparities: Figures 2.1a d( 5 ) show a negative correlation for most of the sample, suggesting that public intervention in schooling is redistributive at the regional level ( 6 ). The case is particularly clear for DOMs on the left of the figures, at all levels except for tertiary education. Table 2.2: Normalised private spending per student, 2000 Total spending Direct transfers Net spending GDP per capita le-de-france Bassin parisien Ÿ Nord Ÿ Est Ÿ Ouest Sud-Ouest Centre-Est MØditerranØe France France, expenditure (EUR) Source: EnquÞte Budget des familles, 2000, authors' calculations. ( 5 ) The list of abbreviations for region names used in the figures is provided on page 52. ( 6 ) The correlation coefficients between relative GDP per capita and relative public spending at the primary, lower secondary and upper secondary levels are 0.46, 0.45 and 0.42, respectively. When excluding le-de-france, these correlations increase (in absolute value) to 0.75, 0.68 and 0.61, respectively. The correlation coefficient between relative GDP per capita and relative public spending on tertiary education is much smaller (in absolute value), being 0.07 with le-de-france included and 0.05 when le-de-france is omitted from the calculations.

17 16 SECTION I THE CASE FOR FRANCE

18 EDUCATIONAL EXPENDITURE, ENROLMENT AND ATTAINMENT 17

19 18 SECTION I THE CASE FOR FRANCE 2.2. Educational attainment This section presents the regional dispersion of educational attainment and its evolution over time. We consider (1) the years of schooling (including possible class repetition) of the population over 25 and not undertaking formal education at survey time and (2) the completed levels of education (described in Table 2.3). In order to avoid measurement errors, we used data from the 1968, 1975, 1982, 1990 and 1999 censuses as much as possible ( 7 ). But, for some years and for the DOMs, we had to refer to EnquÞte emploi which has a 1/300 sampling rate ( 8 ). Table 2.3: Educational level coding Code Level French equivalent Cumulative duration L1 Primary or no schooling Enseignement ØlØmentaire 5 L2.1 Lower secondary BEPC 9 L2.2 Vocational/professional secondary CAP/BEP 11 L2.3 Upper secondary BaccalaurØat 12 L3.1 Lower tertiary DEUG 14 L3.2 Upper tertiary Licence, maîtrise, 17 DESS/DEA, th se, grande Øcole (median) NB: The CAP/BEP and BaccalaurØat represent parallel tracks in the French educational system, both of which start after the end of coll ge. Table 2.4 shows normalised averaged years of schooling for the censuses. The regional dispersion is not very strong, with coefficients of variation to the order of 3 4 % and the dispersion is decreasing slightly. le-de-france is a strong positive outlier, while the DOMs stand at the bottom end of the distribution albeit without clearly appearing as outliers (with the possible exception of RØunion). Nord-Pas-de- Calais, the region that contains the transitory Objective 1 districts, also has low relative levels. Over time, the relative positions are stable, but normalised levels increase somewhat in Bretagne, Limousin and Poitou-Charentes, contributing to a reduction in overall dispersion. Figure 2.2 shows that educational attainment is very clearly correlated with regional income per capita ( 9 ). le-de-france and the DOMs have very high and very low GDP respectively and draw much of the correlation. ( 7 ) The public use files of the 1968, 1982 and 1990 censuses are $ samples. The 1975 public use census file is a & sample. Although there exists a $ sample of the 1999 census, we used the!0 sample version due to limits in file transfer sizes. ( 8 ) Completed levels of education are available in all censuses, but years of schooling are only available in the 1968, 1975 and 1982 censuses. For 1990 and 1999, we computed average years of schooling by level, sex and cohort based on EnquÞtes emploi in the same years and then reweighed by the distribution of educational level, sex and cohort found in the censuses. In the absence of census data in the DOMs, we used directly the EnquÞtes emploi that are available in 1993 and 1998 and contain about individuals. ( 9 ) The correlation coefficient is 0.64.

20 EDUCATIONAL EXPENDITURE, ENROLMENT AND ATTAINMENT 19 Table 2.4: Normalised average years of schooling le-de-france Alsace Rhône-Alpes Champagne-Ardenne Bourgogne Haute-Normandie Centre Aquitaine Provence-Alpes-Côte d'azur Pays de la Loire Midi-PyrØnØes Franche-ComtØ Auvergne Bretagne Basse-Normandie Picardie Lorraine Poitou-Charentes Limousin Nord-Pas-de-Calais Languedoc-Roussillon Corse Martinique Guadeloupe Guyane RØunion France France, years Coefficient of variation (%) Source: Census (1/20) and EnquÞte emploi for 1990, 1999 and DOMs, and authors' calculations; 1993 and 1998 DOM data are presented with 1990 and 1999 respectively. Sample restricted to individuals aged years old not undertaking formal education.

21 20 SECTION I THE CASE FOR FRANCE The sources of dispersion are detailed in Table 2.5 for the year le-de-france has lower primary (L1) attainment and much higher upper-tertiary level (L3.2) than most regions. The reverse is true for DOMs, where a strikingly large stock of residents have only a primary education. There is less dispersion overall at the intermediary levels. One should be aware that the age structure of the resident population almost certainly contributes significantly to these contrasted outcomes, but the population in the DOMs is rather young. Table 2.6 provides some evidence about the past dynamics of education levels. Average schooling is computed separately for 10-year cohorts using the 1999 census and EnquÞte emploi for the DOMs. Looking at coefficients of variation, it appears that there has been less dispersion between regions over time, except for the youngest cohort. One must be careful however, since this disappears when the DOMs are excluded given that the cohort data for the DOMs are not very reliable because of small initial sample size (about observations). Overall, the negative relation between cohort age and the coefficient of variation implies that convergence is taking place, but the orders of magnitude are small because initial disparities between regions are limited. Looking more closely into convergence, Figure 2.3 plots the change of the relative position between the and cohorts (first two rows of Table 2.6) against the initial position of cohort This answers the question: did the regions that had lower schooling levels increase their performance more strongly than other regions between the two generations? The answer seems to be yes, because the slope of the

22 EDUCATIONAL EXPENDITURE, ENROLMENT AND ATTAINMENT 21 figure is slightly negative ( 10 ). Many regions that had initial levels between 95 and 98 % of the average improved their relative position, whereas almost all above average regions decreased their relative position. The same conclusion arises from Figure 2.4 for tertiary-level education. It plots the share of the cohort that completed some higher education against the increase in that share between the and cohorts. Again, progress was faster for lagging regions ( 11 ). The extreme outlier is lede-france, which concentrates more and more of the higher educated population. Table 2.5: Educational levels in 1999 (%) L1 L2.1 L2.2 L2.3 L3.1 L3.2 Total le-de-france Alsace Rhône-Alpes Champagne-Ardenne Bourgogne Haute-Normandie Centre Aquitaine Provence-Alpes-Côte d'azur Pays de la Loire Midi-PyrØnØes Franche-ComtØ Auvergne Bretagne Basse-Normandie Picardie Lorraine Poitou-Charentes Limousin Nord-Pas-de-Calais Languedoc-Roussillon Corse Martinique Guadeloupe Guyane RØunion France Source: Census (1/20) and EnquÞte emploi 1998 for DOMs, and authors' calculations. Sample restricted to persons aged years old not undertaking formal education. ( 10 ) The correlation coefficient is ( 11 ) The correlation coefficient, excluding le-de-france, is 0.38.

23 22 SECTION I THE CASE FOR FRANCE Table 2.6: Normalised average years of schooling by cohort, le-de-france Alsace Rhône-Alpes Champagne-Ardenne Bourgogne Haute-Normandie Centre Aquitaine Provence-Alpes-Côte d'azur Pays de la Loire Midi-PyrØnØes Franche-ComtØ Auvergne Bretagne Basse-Normandie Picardie Lorraine Poitou-Charentes Limousin Nord Pas-de-Calais Languedoc-Roussillon Corse Martinique Guadeloupe Guyane RØunion France France, years Coefficient of variation (%) Source: Census (1/20) and EnquÞte emploi 1998 for DOMs, and authors' calculations. Sample restricted to those not undertaking formal education.

24 EDUCATIONAL EXPENDITURE, ENROLMENT AND ATTAINMENT 23

25 24 SECTION I THE CASE FOR FRANCE 2.3. Educational flows In France, the main break in one s academic progression occurs between the upper secondary and the lower tertiary levels as a result of the national school-leaving examination, the baccalaurøat, successful passage of which is compulsory in order to attend almost any tertiary level educational institution. There is also a high drop-out rate during the first years of tertiary schooling in France. As shown in Table 2.7, educational enrolment rates fall from nearly 84 % at ages to 35 % at ages Table 2.7: Enrolment rates at and 20 24, (%) years old years old Women Men Total Women Men Total le-de-france Rhône-Alpes Alsace Haute-Normandie Champagne-Ardenne Bourgogne Poitou-Charentes Aquitaine Provence-Alpes-Côte d'azur Centre Franche-ComtØ Midi-PyrØnØes Bretagne Auvergne Basse-Normandie Lorraine Limousin Picardie Pays de la Loire Nord-Pas-de-Calais Languedoc-Roussillon Corse Martinique Guadeloupe RØunion Guyane France Source: GØographie de l'ecole (2001), Minist re de l'øducation nationale. Schooling rates at ages are highly and positively correlated with rates at ages 20 24, with the bulk of French regions concentrated around the national average. There are two clear outliers, le-de-france, due to the concentration of tertiary schooling institutions in the capital, and French Guiana, at the other end, which is significantly below national average on both accounts (and has only a technical university at the L3.2 level). Objective 1 regions are among the worst performers, with only Martinique being comparable with non-objective 1 regions.

26 EDUCATIONAL EXPENDITURE, ENROLMENT AND ATTAINMENT 25 Table 2.8: Outcomes after secondary education (%) No degree (L2.1) Vocational (L2.2) BaccalaurØat (L2.3) Pays de la Loire Bretagne Midi-PyrØnØes Limousin Auvergne Aquitaine Poitou-Charentes Rhône-Alpes Provence-Alpes-Côte d'azur Bourgogne Languedoc-Roussillon Nord-Pas-de-Calais Lorraine Alsace Franche-ComtØ Basse-Normandie le-de-france Haute-Normandie Centre Champagne-Ardenne Picardie Corse Metropolitan France Source: GØographie de l'ecole (2001), Minist re de l'øducation nationale. Table 2.8 describes outcomes of pupils after lower secondary education. The figures in this table cover all people at the end of their lower secondary education (public, private and apprenticeship) in metropolitan France. A significant minority ends up with no degree at all. Vocational education receives on average 25.5 % of pupils, with no discernable correlation with the region s overall educational performance. Once again, Corse is a strong outlier, with a much higher proportion of pupils leaving the schooling system without any qualification than any other region. This table also underlines the difficult situation of former industrial regions, such as Picardie, which never completely recovered from the fall of mining and traditional heavy industries. The areas of Valenciennes, Douai and Avesnes, in the Nord-Pas-de-Calais region, are in a similar situation but do not appear clearly here because of the importance of the urban area of Lille (also in this region) and its dynamic performance. It is among those enrolling in a tertiary degree programme that inter-regional mobility becomes an issue. The baccalaurøat is required to access almost any tertiary education, and the bulk of this education is provided by universities. In principle, a person leaving secondary schooling and going to university is required to stay in the same region, unless he or she can justify a change (parental move, speciality not provided by the local university, etc.). However, much of inter-regional mobility at this level is formed

27 26 SECTION I THE CASE FOR FRANCE of pupils with good academic records or looking for specific specialisations. Table 2.9 describes such moves. The percentages are computed as follows: l Continue denotes the percentage of baccalaurøat recipients in the region joining higher education; l Stay denotes the percentage, among the Continue group, of those going to a university in the same region; l Arrivals denotes the number of people coming from another region, as a percentage of the total number of baccalaurøat recipients going to university in this region; l Departures denotes the percentage of baccalaurøat recipients going to another region as a percentage of the number of laureates of the sending region going to university. Corse and the DOMs appear much less attractive than other regions, with a lower percentage of baccalaurøat recipients continuing on to university in the region and almost none moving to it. Although not much different from the average in terms of the share of baccalaurøat recipients who continue on to tertiary education (with Corse being the region that sends the highest share of its baccalaurøat recipients on to tertiary education), this mobility implies that the better-qualified pupils leave the region at this level, thus contributing to the maintenance of a low stock of human capital. This is probably compensated in part by labour migration, but data are lacking for a careful analysis of the phenomenon and we are not aware of any general analysis of inter-regional migration by education. Table 2.9: Inter-regional movements at entry into tertiary education (%) Continue Stay Departures Arrivals le-de-france Rhône-Alpes Alsace Haute-Normandie Champagne-Ardenne Bourgogne Poitou-Charentes Aquitaine Provence-Alpes-Côte d'azur Centre Franche-ComtØ Midi-PyrØnØes Bretagne Auvergne Basse-Normandie Lorraine Limousin Picardie Pays de la Loire Nord-Pas-de-Calais Languedoc-Roussillon Corse DOM France Source: GØographie de l'ecole (2001), Minist re de l'øducation nationale.

28 ECONOMETRIC ESTIMATES OF THE EFFECTS OF SCHOOLING Econometric estimates of the effects of schooling In this section, we estimate the effects of schooling on several economic outcomes. We first evaluate the impact of schooling on wages, participation and employment, and contrast the strength of the effects between regions. We then estimate the correlation between regional productivity and average level of education. This last estimation needs to be treated with care due to a lack of data for certain variables and important measurement error problems for others Schooling and wages Using standard econometric procedure, we estimate, separately by region, the percentage increase in hourly wage that can be attributed to an additional year of schooling, holding professional experience constant. The specification has the form: lnw i ¼ c þ us i þ a potexp i þ b potexp 2 i þ d sex þ u i (1) where W i is the hourly wage of worker i (net of employee payroll taxes but gross of income taxes), c is a constant and u i is a random disturbance. The explanatory variables are the number of years of schooling (S), potential experience (potexp, defined as the time elapsed since the completion of education) and its square and sex. The second specification is identical to equation (1) except that the years of schooling variable (S) is replaced by a set of indicator variables, one for each of the educational levels listed in Table 2.3. In the latter case, we then obtain yearly returns by dividing the level effect by the theoretical number of years in each level. Because wages are not observed for every individual and those who receive a wage are not a random sample of the population, the coefficient of the education variable could be biased unless we also specify and estimate the selectivity rule (Heckman, 1979) ( 12 ). The results presented in Table 3.1 use data from three years ( ) of the labour force survey, EnquÞte emploi, in order to increase sample size ( 13 ). Depending on the region, sample sizes are between and It should be noted that Corse only provides 519 observations, thus implying that the estimates may not be very reliable for this region. Table 3.1: Wage returns to education, (%) All levels L2.1 L2.2 L2.3 L3.1 L3.2 le-de-france Ÿ Rhône-Alpes Ÿ Alsace Ÿ Haute-Normandie Ÿ Champagne-Ardenne Ÿ ( 12 ) Selectivity correction is obtained by simultaneous estimation of a probit model with demographic variables to explain participation and employment. This is done by maximum likelihood. ( 13 ) Given that the EnquÞte emploi is not administered annually in the DOMs, we use data from the 1998 wave.

29 28 SECTION I THE CASE FOR FRANCE All levels L2.1 L2.2 L2.3 L3.1 L3.2 Bourgogne Ÿ Provence-Alpes-Côte d'azur Ÿ Aquitaine Ÿ Pays de la Loire Ÿ Centre Ÿ Franche-ComtØ Ÿ Midi-PyrØnØes Ÿ Bretagne Ÿ Auvergne Ÿ Basse-Normandie Ÿ Lorraine Ÿ Limousin Ÿ Poitou-Charentes Ÿ Picardie Ÿ Nord-Pas-de-Calais Ÿ Languedoc-Roussillon Ÿ Corse Martinique Ÿ Guadeloupe Ÿ Guyane Ÿ RØunion Ÿ France Ÿ Source: EnquÞtes emploi , and authors' calculations. (%)

30 ECONOMETRIC ESTIMATES OF THE EFFECTS OF SCHOOLING 29 The marginal return of an additional year of education varies between 4.2 % and 9.7 %, but most coefficients are between 6.8 % and 8.9 % which is in the range of the returns routinely found for developed countries in the literature. In international comparisons, it is often observed that higher returns are present in less-developed countries that have also lower attainment levels (Psacharopoulos, 1994). The picture is also valid at the regional level, as illustrated by Figures 3.1 and 3.2. Returns decrease with both GDP per capita and average attainment over most regions ( 14 ). If these coefficients reflect the return on educational investment, this implies that it is most warranted in poorer areas. Table 3.1 also reports returns by educational level. Notice that vocational education is associated with lower wages than lower secondary education, although it corresponds to two more years of schooling. Tertiary education (L3.1 and L3.2) has the highest average marginal returns and the baccalaurøat (upper secondary, L2.3) has the lowest average marginal returns. It must be emphasised that dummy variables measure completed (and not attended) levels, so that returns may be lower for dropouts. This general pattern is present in most regions and there is no schooling level that has a higher relative return in poor or rich regions. ( 14 ) The correlation coefficient between GDP per capita and returns to schooling (Figure 3.1) is The correlation coefficient between educational attainment and returns to schooling (Figure 3.2) is 0.46.

31 30 SECTION I THE CASE FOR FRANCE 3.2. Participation and employment In order to evaluate the impact of education on individual participation decisions and on the probability of being employed when participating, we estimate model: P* = Zb P +u E* = Xb E +v where P * and E * are latent variables for participation and employment respectively. A person participates if P * ;0 and she is employed if E * ;0. Z and X contain several observed determinants including the schooling level. If u and v, the unobserved determinants, are correlated, the employment model delivers parameters that are biased by selectivity into the participant population. As with the wage equation, the two events are estimated jointly using the EnquÞtes emploi ( 15 ). We used an individual s declaration of participation (employed or unemployed by the ILO definition) as our measure of participation, although this has been the matter of some debate in the labour economics literature (Jones and Riddell, 1999). Table 3.2 presents the marginal effects of schooling on the probability of participation ( 16 ). Generally, the effects are rather small: an additional year of education increases participation by about 1 percentage point (which means by less than 1 %). Vocational education has a stronger effect on participation, probably because it is more often undertaken by those who do want to be active. Table 3.3 presents the corresponding effects for the probability of being employed, given that the person is in the active population. Again, employment effects are much smaller than wage returns, typically between 1 and 3 percentage points. As illustrated in Figure 3.3, a higher impact of the level of education on employment is found in poorer regions. Direct inspection of Table 3.2 (that is ordered by GDP per capita) reveals that this is also true for participation. It is therefore a general feature. Table 3.2: Marginal effect of schooling on the probability of participation, All levels L2.1 L2.2 L2.3 L3.1 L3.2 le-de-france 0.73 Ÿ Rhône-Alpes 0.58 Ÿ Alsace 0.73 Ÿ Ÿ 0.36 Haute-Normandie Ÿ 0.36 Champagne-Ardenne 1.06 Ÿ ( 15 ) Identification requires that some determinants of participation are not direct determinants of the probability of being employed, given that one participates. Employment is assumed to depend on education, experience, a year dummy and sex (the variables in X); participation is further influenced by marital status, the presence of children and the presence of children below 6 (i.e. Z=X plus these additional variables). The residuals are assumed to be distributed as a bivariate normal and the model is estimated by maximum likelihood. ( 16 ) More precisely, because the marginal effect of schooling is different for every individual in a non-linear model like this one, we present the average effect in the population.

32 ECONOMETRIC ESTIMATES OF THE EFFECTS OF SCHOOLING 31 All levels L2.1 L2.2 L2.3 L3.1 L3.2 Bourgogne 1.03 Ÿ Provence-Alpes-Côte d'azur Ÿ 0.73 Aquitaine 0.59 Ÿ Ÿ 0.97 Pays de la Loire 0.89 Ÿ Centre Franche-ComtØ Midi-PyrØnØes 0.74 Ÿ Bretagne 0.75 Ÿ Ÿ 1.31 Auvergne 0.93 Ÿ Ÿ 1.33 Basse-Normandie 0.74 Ÿ Ÿ 0.05 Lorraine Ÿ 0.15 Limousin 0.78 Ÿ Poitou-Charentes 0.86 Ÿ Picardie Nord Pas-de-Calais Languedoc-Roussillon Corse 1.53 Ÿ Ÿ Martinique Guadeloupe Guyane Ÿ 0.77 RØunion France 1.16 Ÿ Source: EnquÞtes emploi , and authors' calculations. Coefficients are interpreted as percentage point changes for an additional year of schooling. Table 3.3: Marginal effect of schooling on the probability of employment, All levels L2.1 L2.2 L2.3 L3.1 L3.2 le-de-france Ÿ 0.33 Rhône-Alpes Alsace Ÿ 0.06 Haute-Normandie Champagne-Ardenne Ÿ Ÿ 3.65 Bourgogne Provence-Alpes-Côte d'azur Aquitaine Ÿ 0.59 Pays de la Loire Centre Ÿ 1.91 Franche-ComtØ Ÿ 0.74 Midi-PyrØnØes Ÿ 0.49 Bretagne 1.00 Ÿ Auvergne Ÿ Basse-Normandie Ÿ 0.48 Lorraine Limousin Poitou-Charentes Ÿ 0.01 Picardie Ÿ 0.71 Nord-Pas-de-Calais Languedoc-Roussillon

33 32 SECTION I THE CASE FOR FRANCE All levels L2.1 L2.2 L2.3 L3.1 L3.2 Corse Ÿ Ÿ 4.35 Martinique Ÿ Guadeloupe Guyane RØunion Ÿ France Source: EnquÞtes emploi , and authors' calculations. Coefficients are interpreted as percentage point changes for an additional year of schooling Schooling and aggregate productivity Empirical estimation of the relationship between schooling and aggregate productivity has a long tradition in the economic literature (Krueger and Lindhal, 2001). It is usually based on the comparison of GDP and average schooling levels in different countries at various dates, conditioning on the stock of physical capital and sometimes other variables. This approach should allow one to demonstrate that education has a favourable impact on production and growth. Results have sometimes been disappointing and they seem to be very sensitive to data quality (e.g. De la Fuente et al., 2002; Cohen and Soto, 2001). In this respect, we start from a disadvantaged position. This is mainly because physical capital data are not available at the regional level in France. As a result, we were forced to build them from firm data accessible at the French statistical institute, INSEE. However, capital data are only reported at the firm level and are assigned to the geographical location of the firm s headquarters, instead of having their geographical distribution over establish-