Population (english edition)
I.N.E.D
I.S.B.N.sans
200 pages
p. 9 à 48
doi: en cours
Veille sur la revue
Veille sur l'auteur
I.N.E.D
I.S.B.N.sans
200 pages
p. 9 à 48
doi: en cours
Veille sur la revue
Veille sur l'auteur
Vous consultez
Volume 59 2004/1
The Returns to Education and Experience : Trends in France over the Last Thirty-Five Years
Marion Selz [*] Marion Selz, CNRS, Laboratoire d’Analyse Secondaire et de Méthode Appliquée à la Sociologie – Institut du Longitudinal (LASMAS – IdL) Claude Thélot [**] Claude Thélot, Cour des Comptes.
A single model applied to data from a large number of surveys covering thirty-five years provides highly empirical but comparable measurements of the economic returns to education and to experience, in the most general sense of that term, and of their evolution in France. While this model does not form the “true” model of these returns, it supplies an interpretative framework of relatively clear synthetic rates. The relative fragility of the approach in respect of conceptual choices (which concepts of education and experience ?) and empirical choices (what level of data precision, which estimates ?) raises doubts about the potential contribution of excessive sophistication in models and econometric methods. The results point up the explanatory power of the model and indicate that the returns to education exceed those to experience. The income returns to education fell gradually between 1962 and 1985, since when they have remained stable. This overall trend in fact masks a rise in the returns to very short periods of schooling, a clear decline at around thirteen years of schooling (corresponding to the baccalauréat), and virtual stability for the longest periods.
L’application d’un modèle unique aux données d’un grand nombre d’enquêtes étalées sur trente-cinq ans fournit des mesures très empiriques, mais comparables, du rendement salarial de la formation initiale et de l’expérience, au sens le plus global du terme, et de son évolution en France. Même si ce modèle ne constitue pas le « vrai » modèle de ce rendement, il donne une « grille de lecture » d’indices synthétiques assez clairs. La relative fragilité de la démarche vis-à-vis de choix conceptuels (quels concepts de formation et d’expérience ?) ou empiriques (quelle précision des données, quelles estimations ?) conduit à rester dubitatif quant à l’apport éventuel d’une sophistication excessive des modèles ou des techniques économétriques. Les résultats mettent en évidence la puissance explicative du modèle et indiquent que la formation est plus rentable que l’expérience. La rentabilité salariale de la formation a baissé progressivement de 1962 à 1985 et elle est restée stable ensuite. Cette évolution globale recouvre en fait une hausse de la rentabilité des formations très courtes, une baisse nette autour de 13 ans de formation (ce qui correspond au bac) et une quasi-stabilité pour les formations les plus longues.
La aplicación de un modelo único a los datos de un gran número de encuestas llevadas a cabo durante los últimos 35 años ofrece medidas muy empíricas, pero comparables, del rendimiento salarial de la formación inicial y de la experiencia, en el sentido más amplio del tér-mino, y de su evolución en Francia. Aun si tal modelo no constituye el “verdadero” modelo de tal rendimiento, ofrece una matriz interpretativa de índices sintéticos bastante claros. La relativa fragilidad del método en la elección de conceptos (¿cómo definir formación y expe-riencia ?) y en las cuestiones empíricas (¿cuán precisos son los datos, y qué estimaciones hay que escoger ?) sugieren escepticismo ante una sofisticación excesiva de los modelos o técnicas econométricas. Los resultados ponen en evidencia el poder explicativo del modelo e indican que la formación es más rentable que la experiencia. La rentabilidad salarial de la formación disminuyó progresivamente entre 1962 y 1985 y permaneció estable a continuación. Esta evo-lución global esconde un aumento de la rentabilidad de las formaciones muy cortas, una clara disminución alrededor de los 13 años de formación (duración correspondiente al bachillerato) y una estabilidad de las formaciones más largas.
Over and above the issue of measuring the economic returns to education in the labour market, this article byMarion Selzand Claude Thélotis original for its analysis of changes in these returns since the 1960s. Based on a classic model with which the situation in France can be related to that in other countries, they evaluate the impact over time on earnings of the length of schooling and of experience. A noteworthy feature of the article is its use of successive comparable surveys. The authors review in detail the problems that are inherent in this type of study, introducing elements of comparison between men and women, and between the situations in the public and private sectors, which is a dimension of particular interest in the French context. The results of this solid empirical analysis are also situated in the broader context of the trends in earnings inequality in France.
People who have received a long education have more success in the labour market than others. To mention just two of their principal advantages in this field, they have a lower risk of unemployment and they receive higher salaries. That education produces “general economic returns” is a universal finding ; what varies, sometimes considerably depending on period and country, are their size and modalities.
In the present article, these returns will be considered solely in terms of individual earnings. As a rough illustration of the importance of the phenomenon, Figure A1 (presented in the Appendix) shows the strength of the association between length of schooling and salary for France, and Table 1 presents the association between the average net salary of highly educated individuals (20 years of schooling) and that of individuals with little education (4 years of schooling)
[1]. The association is very strong even though it has weakened particularly since the 1960s : over thirty-five years, the ratio between these two average salaries has fallen from approximately 3.5 to 2.5, for both men and women (even though it is on the whole slightly lower for the latter).
Table 1
Ratio between the average net salary of highly educated individuals (20 years of schooling) and of individuals with little education (4 years of schooling)
Descriptive statistics of this kind, highly valuable in its own right, illustrates the question that this article seeks to address : is the decline in the ratio of average earnings a result or not of a decline in the returns to educational investment in France over the last thirty-five years and, if so, how large is this decrease and how is it to be interpreted ?
The apparent association between earnings and education could be misleading, if the most highly educated employees earn more for reasons completely unrelated to their education, for example because they are more likely to work for large companies or in major urban centres, or because they are more “intelligent” or more “productive” etc. Discussion must therefore proceed from a model. Following the fundamental article by Mincer (1958), analysis of the relationship between education and earnings has traditionally involved relating the logarithm of earnings to three groups of variables : variables describing education, those describing work experience (and seniority), and a third heterogeneous group taking into account other factors that affect earnings (individual characteristics such as sex, nationality, occupation, etc ; collective characteristics such as activity sector, company size, or even about the company itself : its profitability, unionization, location etc). As this article aims to characterize trends occurring over thirty-five years, the model that we use is less sophisticated and includes only initial schooling and experience ; but the same model is used over the whole period, i.e. it is applied to all the surveys employed.
Before presenting our study a brief summary can be given of the conclusions reached by earlier studies and that are the most useful for our purpose. They can be appraised from the following two tables. Table 2 concerns France and was synthesized by Marchand and Thélot (1997), while Table 3, produced by Bils and Klenow (2000), shows the situation in 52 countries.
Table 2
Rate of return to one year of schooling (%) : selected results from previous studies in France
Table 3
Results of the simple Mincer equation for 52 countries, in the late 1980s (with a few exceptions)
Regarding France, the various authors mentioned in Table 2
[2] do not all use the same equation, thus making comparison difficult. Three conclusions do nonetheless emerge, which we will attempt to refine.
First, the returns to education appear to have declined over time. In the mid-1960s an extra year of education would have brought 10-12% additional earnings, compared with 7-9% in the early 1990s. This conclusion is tenuous due to the diversity of the models, but the trend, which needs to be specified, is clear to see.
Next, the returns to an extra year of schooling seem to depend on when it occurs. The use by some authors of a Mincer equation of the second or third degree therefore appears justified for the length of schooling.
Finally, taking other variables into account, particularly those of activity sector, job, and even company greatly reduces the “pure” returns to education. When attention is limited only to experience, we cannot claim to have taken into account all the different factors that influence earnings, even if it is the most important.
The results presented in Table 3 are all based on the same model : a simple Mincer equation, but in which length of schooling is present only at the first degree — the return to an extra year of schooling is thus assumed to be constant. Because the same model is used, the conditions for cross-national comparisons appear to be satisfied here (except for the problem of measuring the variables). The rate of return to education appears to differ greatly between countries : in some it is very high, around 15% (Brazil, Colombia, Guatemala, Honduras, etc), even 20% or over (Côte d’Ivoire, Jamaica), or on the contrary very low, between 3% and 5% (Austria, Canada, Greece, Italy, etc). These are large differences and, if all these surveys are reliable (some of the samples used are tiny, the variables are perhaps incorrectly measured, etc), the results are disturbing, not as regards the existence of the relationship between education and earnings but as regards its size. It is true that the content of education of any given duration differs greatly between countries, which can influence productivity, and hence earnings, in widely varying ways. On the whole, returns to education seem higher in developing countries and lower in developed countries — though for contradictory conclusions, based on a more exhaustive survey in Madagascar, see Arestoff (2001).
In what follows, we present the model and the data used in this article with the reasons for the choices made, then we present the results obtained. A discussion of the empirical robustness of these results can be found in the Appendix.
I. The model and the data
We employed the following model where S denotes the salary, d the duration of schooling and e the duration of experience :
The marginal earnings return, 
, i.e. the percentage of additional earnings procured by an extra year of schooling or of experience, is therefore :
, i.e. the percentage of additional earnings procured by an extra year of schooling or of experience, is therefore :- For an extra year of schooling :
- For an extra year of experience :
This percentage therefore depends both on the duration of schooling (d) and on the duration of experience (e) relative to which the increase is assumed to occur.
By comparison with the most frequently used models, that used here prompts three remarks.
First, it uses a third degree polynomial for each of the variables, whereas usual practice — though see Baudelot and Glaude (1989), and Goux and Maurin (1994), who had already adopted this approach — is to use a first or second degree polynomial for the length of schooling and a second degree polynomial for experience. The present specification makes it possible to estimate the second degree effects of each of these variables on the earnings return, which can, depending on the value of the coefficients, first increase and then decrease. This profile is a priori attractive and consistent with the results already suggested, namely that from the start of schooling (and respectively of adult life) up to a certain point, the return to an extra year of schooling (and respectively of experience) can reasonably be expected to increase ; and beyond a certain point to decrease. From this is derived the idea of a parabolic return (and we therefore expect a2 and b2 to be positive and a3 and b3 to be negative).
Such a hypothesis is a priori richer than those usually made (the marginal return is constant with a first degree polynomial, while with a second degree polynomial it rises or falls continuously depending on the sign of a2 or b2) ; the coefficients a2, a3, b2 and b3 will of course be tested for the null hypothesis, which if accepted would lead back to the standard formulations. This model will be applied to all the major surveys conducted in France by INSEE (French National Institute of Statistics and Economic Studies) over the past thirty-five years to gather data on the earnings, education, and experience of employees, in order to measure the changes in the returns to education.
The second remark is that, complementary to this succession of temporal estimates, the model incorporates an interaction between education and experience that reflects the hypothesis according to which the returns to (the duration of) education — and thus its “economic earnings value” — may have varied over time. If education was remunerated more “in the past” than it is today, the oldest employees — and therefore, because of the definition of experience given below, the most experienced — should receive, depending on their education, higher earnings than the younger — and thus more recent — employees, which will be the case if c is positive. The c sign can also, when it is negative, translate another phenomenon : the possible economic obsolescence of education acquired in the past. In this case, when the same length of time is spent in education to acquire the content adapted to the skills needed for working life by today’s employees, and that have become obsolete for the oldest employees, the latter will earn less than their younger counterparts.
The third observation is that education and experience are the only variables present in the model, something which, as seen above, is common but not systematic in the literature. This choice is deliberate and was motivated by the fact that our model is applied to different surveys where the other conceivable factors are identified using classifications that are variable and hard to compare across successive surveys. In addition, the originality of the present work lies in using one and the same model — however basic it may be — over time, and in adopting a descriptive perspective, as opposed to examining a single survey using a very precise and sophisticated model with apparently exhaustive explanatory powers. Moreover, the remarks concerning the two factors selected will go in the same direction : we will use this model as an interpretative framework intended to provide summary indicators and the trends in the economic returns to education, but we do not suggest that it is the or even a “true” model.
1. Education
Education is identified by length of schooling, from age 6 (i.e. without taking pre-elementary schooling into account). Measuring education by length of schooling is the rule in existing studies, but presents two well-known disadvantages.
First, the content of the education is unknown, whereas an approach based on qualifications rather than length of schooling would allow this to be apprehended more closely. In reality, in France, and with two exceptions — admittedly major ones — concerning third level education, the existing research shows that using just the length of schooling is a satisfactory approximation(
[3]).
Explicitly including qualifications could raise another question, one that is important and well-known in this research field. Is it really the content of what is learned at school that determines economic returns, or is it the fact of having been selected — one of the functions of the educational system and qualifications — which informs the employer about the potential productivity of the possible future employee : in other words, the theory of human capital stricto-sensu versus the theory of signalling ? However, when attention is limited, as here, to length of schooling, this issue cannot be resolved.
Use of such a general variable — the total length of schooling — has another disadvantage, that of precluding any differentiation of the return on years according to their nature. What is the value, for example, of repeated years (années de redoublement) or years that do not result in a qualification ? It has been shown, at least for males during the 1990s, that repeated years yield nothing, and that the income return to years that do not lead to a qualification is positive but only half what it is to years that do lead to a qualification (see Goux and Maurin, 1994 ; Hanchane and Moullet, 1997). Ignoring these aspects is nevertheless acceptable in the perspective adopted here : to bring to light effects that are general, rather than specific, for the purpose of synthetic description.
2. Experience
Experience is measured by the number of years separating the start of working life from the point where the earnings under analysis are received. Our study is concerned therefore with individuals in employee jobs at the time of the survey, though we know nothing of their “past career”. It is for this reason that we refer not to “work experience” but simply to “experience” (even though this term can be slightly misleading) : periods of unemployment, national service, and time out to raise children certainly contribute to experience and no doubt count as useful experience for future employment, but they are not “work experience” in the true sense of the term. Furthermore, in these years of experience, the division between work experience and “other” experience varies between men and women and by period — as have varied the level and continuity of female employment, on the one hand, and those of unemployment, on the other hand, over the last thirty-five years. The result is a certain heterogeneity in the content of this experience, between men and women, between years of observation, and between birth cohorts.
In this context, the interaction between d and e is likely to have a meaning, if work experience and “other” experience have different impacts on earnings. If c is positive, it could be interpreted as a greater role for the former, since the continuity of work careers — both through access to employment and through not experiencing unemployment — is greater for the individuals with the most education.
3. Extensions and interpretations
In the literature, the basic Mincer model, independently of the precise variables that are incorporated into it, has been the subject of criticisms and extensions, which though we will not adopt them are useful to have in mind at the start of this study
[4]. Two in particular can be mentioned.
The first concerns the interpretation of the model’s coefficients. From the outset, analysts have discussed the possibility of a selection bias ; see, for example, Griliches (1977). If it is people possessing certain qualities who at the same time spend longest in education and receive the highest earnings, the coefficients for the length of schooling cannot be interpreted as measuring the “pure” effect of education. In an extreme case this “pure” effect might not even exist. This question has generated numerous articles proposing widely varying solutions for controlling this potential bias, ranging from introducing IQ measurement variables into the equation, to estimating the model on identical twins — for example Rouse (1998), and Bound and Solon (1998) — and including two-stage estimation, with instrumental variables, or the addition of an explanatory equation for education. From these studies it emerges that the coefficients of the “simple” Mincer equation give a possible overestimation of the pure, real effect of education but that this appears to be compensated by a possible underestimation due to measurement errors. In other cases, Hanchane and Moullet (1997) or Boumahdi and Plassard (1992), treating education as endogenous leads to an overestimation of the rate of return. On balance, however, according to Ashenfelter and Rouse’s synthesis (1999), the usual simple estimators of the Mincer equation could well be the most accurate. Yet, according to Card (2001), no study can at present conclude decisively about the size of the bias introduced by the least squares estimator of the Mincer equation.
The second question results from the possible behaviour of individuals and employers, particularly in a country like France where income distribution is affected by the existence of a national minimum wage, the SMIC (salaire minimum interprofessionnel de croissance). But even without the SMIC, the issue would arise : for if education influences productivity and, through this, earnings, then the influence is on the potential salary of individuals, that which they could expect to receive, rather than on the salary they actually receive, which is the result of many other factors. Technically, instead of a single equation, two are needed : one that models the fact of being in paid employment or not, and another that explains the earnings of these employees. This formalization would correspond to a classic Tobit model. Moreover, such a formalization would be necessary if we wanted to estimate the “total economic returns” to schooling, i.e. in terms of earnings returns and of unemployment avoided. For it is known that not only has the risk of being unemployed always been lower for individuals with qualifications, but that over the last twenty-five years (since the onset of the economic “crisis”) in France this effect has increased : qualifications now provide relatively more protection against unemployment than they did in the early 1970s, and from this standpoint they have gained in value. Conversely, because attention is limited here to employees, our study also fails to include the upward careers of employees who become self-employed : yet these itineraries are not necessarily unrelated to the length of education.
As the objective here is only to measure the evolution over time of earnings returns, the approximation involving use of a simple Mincer equation seems acceptable.
4. Choice of dataset and population
Our model is applied to the different surveys conducted by INSEE over the past thirty-five years and that supply information on the three basic topics of earnings, education, and experience. Specifically we use the FQP (Formation et qualification professionnelle – Education and work qualification) surveys of 1964, 1970, 1977, 1985 and 1993, and the annual Labour Force surveys (Enquêtes emploi), of 1991, 1993, 1995 and 1998
[5].
We wish to analyse the salary received by full-time employees. In the FQP surveys, individuals are asked about their salary in the year prior to the survey. In these surveys, therefore, we have had to restrict attention to individuals who worked full-time for the whole of the year preceding the survey (we thus avoid any entries and exits occurring in the course of the year, and careers that are starting or ending, when earnings are more uncertain and variable). In the Labour Force surveys, the same data are not available. To define a comparable population, we selected employees working full-time at the time of the survey (in March) and one year previously.
The salary recorded is the net salary, including all bonuses. For the FQP surveys, there is no conceptual ambiguity even though questions remain over empirical accuracy (because earnings are self-reported, but also because of the coding, since an upper limit for salary is set in some surveys). For the Labour Force surveys, the earnings analysed are in theory those received at the time of the survey. But the fact of interviewing the household member present (on behalf of all the wage earners in the house hold), and the need to report any bonuses (explicitly asked for), leads to thinking that over and above the inaccuracy that results, the respondent sometimes relies on the income tax declaration, in which case the earnings and bonuses he or she reports will be those of the previous year — as is the case in the FQP surveys (which is in fact more reliable regarding annual bonuses or the “thirteenth month”, for example).
We have limited ourselves to employees in work at the time of the survey, aged 59 years or less on 31 December of the year for which the earnings are reported. In this way can be avoided, at least for the last two decades, the selection effects produced by retirement, since those employees who continue to work after age 60 have specific characteristics(
[6]).
Within the aggregate thus formed, we estimate a Mincer equation for several sub-populations, which raises some problems of principle or of empirical delimitation depending on the case :
- for both men and women : this is unusual since as a rule attention is limited to males on the grounds that for them experience is a good approximation of work experience. But this argument carries less weight today than it did twenty or thirty years ago ; and besides, in our perspective of synthetic description, it is worthwhile estimating the returns to schooling and experience for women also ;
- only for “French nationals by birth” : in addition to its specific interest, a study of this population approximates to that of employees who have been educated in France, which is a useful population if we want to interpret the estimates in terms of the effectiveness of the French educational system ;
- for private sector employees on the one hand, and for those of the public sector (central government, local authorities, and hospitals) on the other [7]. The theories from which the Mincer equation was developed belong to the classical economic approach that assumes a competitive labour market, and are thus strictly speaking adapted to the private sector. However, this intellectual origin does not prevent the equation from also being applied to the public sector (still with the view of obtaining synthetic indicators that describe reality) ; this also makes possible comparisons of the respective roles of education and experience in each sector, which is obviously very useful. The difficulty lies in clearly defining the two sectors, both within the different surveys and to allow for the transformation of the private sector in France over the last thirty-five years. The definition used is that present in the various surveys ; no attempt was made to define a common boundary, which has two consequences : a) this traces the boundary specific to the period, which is therefore variable over time ; and b) unfortunately, it is drawn differently depending on the type of survey, as can be seen from the example of 1993, a year in which both an FQP survey and a Labour Force survey were conducted.
Finally, at the intersection between education and experience stands the indentured apprenticeship. This period, when it can be identified, can be considered, and with equal legitimacy — it is characteristic of any form of part-time training, but the question arises concretely only for indentured apprenticeship — as a period either of schooling or of employment. This gives a choice of three variants :
- First variant : indentured apprenticeship is not considered as education ; in this case, the length of schooling is measured excluding any apprenticeship, and this length of apprenticeship is counted as an integral part of experience.
- Second variant : indentured apprenticeship is considered both as a period of education and of employment ; in this case it is counted both in the length of schooling and in the length of experience, which therefore overlap.
- Third variant : indentured apprenticeship is not considered as experience ; in this case, the length of schooling includes this period, and experience only begins at the end of the apprenticeship ; this variant is the opposite of the first.
At the conceptual level, the second variant is the most useful and sensible. The third variant is certainly the least likely, while the first is perhaps a lesser evil : better than the third, less satisfactory than the second. This is the one that we have preferred to use here, because although, concretely, all three variants can be used with the FQP surveys (except for the 1964 FQP), they cannot be used with the Labour Force surveys. The results of the other two variants are presented in the Appendix and compared with the first one, which is the central variant (Table A7).
II. The main results
The Mincer equation has been estimated separately for each survey using ordinary least squares and after weighting the numbers by the coefficient for making the survey representative at the national level. It is essential to estimate the model using weighted data (on this subject see “Empirical robustness” in the Appendix)(
[8]).
We examine successively the goodness of fit of the equation being tested, the returns to education and to experience, and the value of the interaction between education and experience.
1. The fit of the model
The model fits the data quite well, and in particular the fit is quite good across the different surveys. For the employees as a whole, the adjusted R2 stands at around 0.33 at the beginning of the period, then decreases slightly and stands at 0.28 in 1998 (Table A1). Thus the explanation by length of schooling and of experience is quite good — the seven variables in the model explain just under a third of the variance — but the goodness of fit does not improve over time. On the contrary, it deteriorates slightly, which signifies a slight increase in the role of other factors in earnings, at the expense of the two major factors from human capital theory. It is also possible that experience is a less good “approximation” of work experience at the end of the period, which would give it a weaker explanatory role, at least if we consider that it is strictly work experience that influences earnings.
Note also that the adjusted R2 is slightly higher for men than for women, perhaps once again because of the different role of experience. It is also slightly higher for employees in the public sector than for those in private sector companies, which probably reflects the tighter control of earnings by level of qualification in the public sector.
All the coefficients are estimated quite precisely since the samples used are large (from several thousands to tens of thousands of individuals). They are significantly different from zero, except occasionally for the interaction between education and experience (see below). The discussion that follows is concerned not with the coefficients considered separately but with combinations of coefficients that can be used to describe the form and size of the earnings returns to schooling and experience.
2. The returns to education
The marginal rates of return depend on the length of schooling and experience. We have calculated them in two ways. First, by average lengths of schooling and experience estimated by survey and by sub-population, i.e. at the average points of the samples and sub-samples : males, females, private sector employees, etc. (Table A2). But these average points differ widely between surveys because the average lengths of schooling and experience — the former in particular — have increased greatly over time (Table 4), so that rates calculated at these average points necessarily incorporate this general movement. Hence the second calculation, in which the returns to education and experience are measured at a “fixed point” that is the same for all periods and all sub-populations. Taking into account the previous averages, we selected a length of schooling of 10 years (i.e. a completion of education at age 16) and a length of experience — or of “adult life” — of 20 years, which corresponds roughly to mid-career point for uninterrupted work careers. The marginal rates of return measured at this fixed point are given in Table A3.
Table 4
Average lengths of schooling and experience (years)
The model reveals a very high marginal rate of return to education, but which is lower today than thirty-five years ago. At the average point, the marginal rate stood at 11.1% of additional earnings for an extra year of schooling in the mid-1960s (FQP survey 1964), and 8.8% at the end of the century (Labour Force survey 1998). At the fixed point corresponding to 10 years of schooling and 20 years of experience, the return from an extra year of schooling was 12% in 1964 and 8% in 1998. However, the rate of return declined exclusively between the early 1960s and the mid-1980s (between the 1964 and 1985 FQP surveys, which concern earnings received in 1962 and 1984, respectively), and has remained stable since then. This represents a drop of between 1 and 1.7 points every 9 years (depending on the method of calculation) during the first 22 years of the period
[9]. The stability we observe since the mid-1980s and that emerges when the most recent surveys are included is a new and important result, which we will attempt to interpret below.
The marginal return to education is quite similar for both men and women, but a closer examination shows that it was slightly higher for men until 1993 and may well be slightly lower as of this date at the fixed point. However, the differences are very small.
The marginal return to education is also very close for men and women when attention is limited to French nationals by birth, perhaps very slightly higher for males, though barely so.
On the other hand, the difference between the private sector and the public sector is considerable. The earnings return to education calculated at the fixed point is far higher in private companies. At the start of the period, the difference in favour of the private sector was 4 points (9.5% in the public sector against 13.5% in the private sector) and still stood at 1.4 points at the end of the period (6.6% against 8%)
[10]. This means that the reduction in the marginal return to education was much greater in the private sector. In both private and public sectors, this reduction occurred mainly up to 1985, though for men in the private sector it continued for slightly longer.
Attention must be drawn to a major difference between men and women. While the marginal returns to education for men and women are close at the level of the French economy as a whole, this is an effect of averages, since it is not true in each sphere, private sector and public sector, taken separately. In reality, education yields higher returns for men than for women in the private sector while the opposite tends to be observed in the public sector. In both cases, the disparities are smaller at the end of the period, but they have not disappeared.
The decline over time in the marginal return to education is the most marked for men, particularly in the private sector where it has continued throughout the period. For women on the other hand, the trend follows the pattern already described, i.e. a reduction up to 1985 followed by a levelling off. Thus in each sphere the disparities in returns between men and women have lessened.
These results can be considered from another point of view, that of individuals. For women, the returns to an extra year of schooling are roughly the same in the private sector as in the public sector. At the fixed point, they have declined respectively from around 11.5% and 11% to 8.5% and 8% respectively. But this has not been a steady decline : it occurred between 1964 and 1985.
For men, on the other hand, this return is much higher in the private sector over the entire period, even if the gap has narrowed : at lengths of experience and schooling fixed at 20 and 10 years respectively, 4.7 points higher in the early 1960s (14.3% against 9.6%) and still 2.2 points higher at the end of the period (8.7% against 6.5%). This closing of the gap occurred because in the private sector the reduction in the marginal return to education was marked and, above all, more regular, continuing beyond 1985
[11].
Overall, the convergence is appreciable : the marginal return to education is much less disparate and unequal in the French economy today than thirty-five years ago, both between men and women and between sectors of the economy (even though it remains higher for men in the private sector). A gradual homogenization of the labour market is occurring along this dimension
[12].
3. The parabolic profile of the returns to education
In every case, the estimated coefficients are positive for a2 and negative for a3. In other words, the parabolic profile of the returns to education emerges clearly : initially rising for the early lengths of schooling — with average experience, between 1% and 2% additional earnings per extra year of schooling for the early lengths of schooling up to 13% extra at 12-14 years of schooling (in the 1960s) or 9% (at the end of the 1990s) — then falling to 5-6% for 20 years of schooling.
A more detailed examination of these estimated parameters and the resulting parabolas (Figure 1 and Tables A4 and A5) leads to the following conclusions :
- As already noted, the estimated parameters are significantly different from zero at the 5% confidence level [13] for all the surveys and all the sub-populations. This shows the pertinence of the hypothesis of a marginal return that initially rises then falls ; the analysis would be poorer if, as has often been the case, only a constant or decreasing return was admitted.
- The curvature of the parabolas changes from one survey to another. The parabolas obtained from recent surveys are flatter than those from older surveys, which means that the variation in the return to an extra year of schooling as a function of the period in which it occurs is smaller at the end of the century than thirty-five years earlier. The index of curvature (or “width” of the parabola [14] increases from 17 years in 1964 and 17.8 years in 1970 to 22.8 in 1995 and 21.7 in 1998. We should note, however, that because most of the FQP surveys predate the Labour Force surveys it is difficult to distinguish in this trend between a survey effect and a period effect. This reduction in the curvature is clearer in the results for women, even within each series of surveys, which suggests, at least for women, the existence of a period effect.
- Next, the maximum marginal return is obtained for around 13 years of schooling, with little change over time and with little difference between men and women, or between the public and private sectors ; for FQP 1964 the optimal level is closer to 12 years.
- This stability in the optimal length of schooling, both over time and across sub-populations, is remarkable : the highest return to an extra year of schooling is invariably obtained for education completed at around age 19, i.e. shortly after the baccalauréat (the general, technological, or scientific qualification marking the end of secondary schooling), at the possible entry to third level education.
- Finally, the return by the different lengths of schooling evolves diversely over time :
- the maximum marginal return decreases following the familiar profile : decline up to the 1980s followed by stability. The movement is very broad : for example, for all employees the maximum rate in mid-career (for 20 years of experience) decreased from 12.9% in 1964 to 8.9% in 1998 (Table A4) ; for zero experience (Table A5), at the start of the career therefore, it decreased from 11.9% in 1964 to 9.0% in 1977 and 9.1% in 1985, which is roughly its current level (9.1% in 1995 ; 8.4% in 1998) ;
- on the other hand, the marginal rates of return to short periods of schooling have tended to increase over time : with 20 years of experience the return to a fifth year of schooling was between 1% and 2% up to 1985, compared with between 2% and 3% at present. The movement is even more marked for women than for men [15] ;
- marginal rates of return to long periods of schooling seem to have been fairly stable for thirty-five years : for 20 years of schooling, which corresponds to what is practically the maximum length of education (measured from age 6), the rate is around 5% with 20 years experience as with zero experience (Tables A4 and A5), except for the 1964 FQP which again appears to be an exception. For 18 years of schooling, the rate is also almost constant at around 7.5% [16]. It is slightly higher for men than for women, but stable in both cases.
Figure 1
Rate of return to an extra year of schooling by length of schooling for 20 years of experience (%) – Weighted data
Thus, around the average movement — reduction up to 1985 and stability since then — the change over time in the marginal rates of return is very different depending on the length of schooling. It is tending to increase for short or virtually non-existent lengths of schooling, clearly decreasing at that of maximum return (13 years, which corresponds to the baccalauréat or shortly afterwards), and more or less stable over the last thirty-five years for lengths of schooling that while long are still realistic (in the order of between 18 and 20 years, which corresponds to the baccalauréat + 6 years or baccalauréat + 8 years).
4. “Pure” earnings hierarchy by length of schooling
Attention now turns to the trend in the earnings distribution by length of schooling. Table 1 gave an initial impression of this by showing the relationship between the average earnings of individuals with 20 years of schooling and those of individuals with 4 years of schooling. But in what follows we seek to calculate the “pure” effect of the length of schooling, as formalized by the model used in this article, by calculating the ÅœM⁄Åœm ratio where the Åœ are the estimators of maximum and minimum earnings obtained from the model’s parameters.
Given the use of a cubic formulation and the sign of the estimated coefficients, the curve which represents the variation in salary (S) as a function of the length of schooling (d) presents a minimum Åœmfor a value dmof the length of schooling, which can be negative or positive, and then increases to a maximum value of ÅœMfor a positive value of dM. We calculated this ratio at mid-career (20 years experience) and at the beginning of working life (zero experience) for the purposes of comparison.
In the totality of surveys and sub-populations observed, dmis usually positive, at most between 3 and 4 years : at 20 years experience, dmranges on average between 1.9 and 3.8 years (between 1.3 and 3.2 years for men and between – 0.5 and 4.5 years for women) ; at zero experience, the values of dmare very similar. These values of dmcorrespond to very little or practically no schooling(
[17].
At 20 years of experience, the lengths of schooling for which earnings are maximum, dM, stand on average between 21 and 24.7 years (between 20.9 and 24.3 years for men and between 19.8 and 24.1 years for women). Once again, the result is very close at zero experience. The lower end of this range corresponds to plausible maximum lengths of schooling, while the upper end does not. These values exceed what is actually plausible, though only by a little. This means that over the “useful” interval, [2-3 years – 20 years], the model does indeed reflect a growth in earnings, with the curve decreasing outside the interval [dm, dM]. Even if dMis often slightly too high to be plausible, it is useful to calculate the theoretical “pure” earnings hierarchy given by the model, i.e. the ratio between the two extremes Åœm and ÅœMthat correspond to these theoretical durations.
Table 5 highlights again one of the main results of the “raw” ratio presented at the start of this article. At 20 years of experience the “pure” earnings hierarchy as a function of length of schooling has declined over the last thirty-five years ; the fall is fairly steady for men, although the movement is stabilized or perhaps even reversed at the end of the period, while for women the trend is more erratic. But overall, for men in mid-career, the ratio goes from 4.5 in the early 1960s to around 3.7 at the end of the century. The result is similar for men at the beginning of their career, but is not confirmed for women, for whom, whether because of a survey effect or not, the “pure” earnings hierarchy by length of education at zero experience was higher in the late 1990s than during the 1960s and 1970s. Here again the decrease is observed mainly up to the mid-1980s.
Table 5
Ratio of maximum and minimum estimated salaries and the average rate of return to one year of schooling for 20 years and 0 years of experience
From this ratio between minimum and maximum earnings we obtain an average rate of return
[18] (Table 5) — rather than a marginal rate as in the previous section — to an extra year of schooling, either with 20 years experience or with zero experience. The result of this calculation can be considered as the best measure of the earnings return to education (at fixed experience and in the framework of the model). It completes the previous measures that are more partial : the average rate, with 20 years experience, falls from around 9% or 9.5% in the mid-1960s to 6% or 6.5%. This fall is amplified by a small survey effect, since in 1993, the rate of return is 0.3 points higher in the FQP survey than in the Labour Force survey. At the beginning of the period, the return may have been slightly higher for women, slightly lower at the middle of the period, and today is practically identical. Does this trend reflect a convergence in salaries between men and women caused by women’s increased participation in the labour market ?
These results confirm that the return to education has not declined uniformly : it fell considerably between 1964 and 1985 but has been stable since 1985.
5. Changes in the returns to education and in the earnings hierarchy
These results need to be compared with the overall trend in earnings inequality in France. The long-term series recently published by INSEE (Casaccia and Seroussi, 2000) have been used to produce Table 6, which is limited to differences of salaries in the private and semi-public sectors. Between 1965 and 1997, the interdecile salaries ratio decreased both for men (from 3.9 to 3.29) and for women (from 3.4 to 2.70) and once again, the decrease occurred in the first twenty years, from 1965 to 1985. Since 1985 stability has prevailed.
Table 6
Indicators of salary inequalities in France
The trend in the returns to education is thus consistent with that in overall earnings inequality. This is understandable, since a policy of reducing income dispersion, whether through large increases in the minimum wage (SMIC) and low incomes or through a downward pressure on high salaries, necessarily results in lower returns to education. There is therefore a degree of contradiction between these two objectives : that of reducing inequalities in earnings and that of maintaining (or increasing ?) the returns to education.
6. The returns to experience
Unlike the returns to education, the returns to experience have been quite stable over time : between 1% and 2% per year of experience depending on the year, irrespective of the method of calculation (i.e. either at the average point of schooling or for 10 years of schooling).
On the whole, returns to experience are slightly higher for men than for women, in both the private sector and the public sector, but the differences are small. In view of their small size, it could be said that experience gained outside the labour market yields nearly as much as work experience strictly defined, since experience, here in an “undifferentiated” form, yields little more for men than for women, and yields little more for women today than it did in the past, even though women’s strictly work careers are now more frequent, longer and less interrupted, than was the case thirty-five years ago.
However, although these differences are small, they are nearly all in the same direction (the returns to experience are higher for men than for women and they are higher for women today than in the past). We can therefore also conclude for a slight rise in the value of work experience compared with non-work experience. What makes this second conclusion more plausible is that if the analysis is limited to single people, i.e. to women for whom experience approximates closely to work experience, we no longer observe higher returns to experience for men compared with women.
Therefore, the slightly lower returns to experience for women would owe more to the fact that in their case the “undifferentiated” experience used here encompasses more time out of the labour market, which is rewarded less in terms of earnings. But the difference seems slight in comparison with the widely held view that what is learned on-the-job, in working life, is extremely important. This confirms another result, obtained using a different method and other sources, by Alain Bayet (1996) : a negative effect is associated with career “gaps” but it is not systematic, and in any case “women accumulate non-work experiences which are in part counted for them”.
The parameters estimated by the model for the returns to experience are either such that e2 is negative and e3 positive, or such that one or the other is not significant. When these coefficients are significant they have values such that the vertex of the parabola, which is then a minimum, corresponds to enormous values for experience that are empirically implausible. In other words, the only part of the parabola that really corresponds to our data is its “descending branch”, before its lowest point, and this is the case for all the surveys studied. The conclusion to be drawn from this is clear : in any period, the marginal return to experience decreases over the entire life cycle. Depending on the survey, an extra year of experience is associated with 3-6% of additional earnings at the beginning of working life (5 years after leaving school), with 1-2% when this extra year is at mid-life or mid-career (after 20 years), and has practically no effect — and so “contributes” almost nothing extra — after 30 years (Table A6).
The emergence of this configuration of a constantly decreasing return to experience over the life cycle, means that we could, to reduce the number of parameters and because they are not always significant, merely use an equation where experience appears only at degree 2 :
In this case, the hypothesis of a fall in the return to an extra year of experience can be adopted a priori (since b2 is negative, which is expected). The estimates are very close to those given by the full model, which for reasons of homogeneity is the one that will be used.
7. The interaction between length of schooling and length of experience
In the model, coefficient c measures the effect on earnings of the interaction between education and experience. Table 7 summarizes the number of cases where the estimator of coefficient c is positive, negative, or not significantly different from zero, the count being made on the basis of each sub-population and each survey :
Table 7
Sign of the coefficient c of the interaction between the length of schooling and the length of experience
When the coefficient is significant it is hardly ever negative ; with roughly equal frequency it is either positive and significant or not significantly different from zero, with a slight predominance for the former, at least in the FQP Surveys.
This coefficient is significantly positive in half the cases. This reflects the fact that the return to an extra year of schooling is greater for the more experienced — and hence older — employees, which fits well with the theses about the decline in the effectiveness of education. The cases where it cannot be considered different from zero, indicating that the return to one year of schooling is the same regardless of experience — i.e. regardless of the age of the employee — occur also in the recent period (covered by the Labour Force surveys) during which the decline in returns, observed over successive surveys, has stopped. Hence there is a certain coherence between the two methods of following changes in the returns to one year of schooling.
Conclusion
As in previous studies, the simple model relating earnings to the lengths of schooling and of experience has proved its worth. It has good explanatory qualities and makes possible a synthetic description of the relationship between these two components of human capital and earnings. Experience, which procures an increase in earnings of 1-2% per extra year, is not reduced here to work experience : experience acquired outside the labour market, “in real life”, is also remunerated, albeit certainly less so. Education is remunerated much more than experience : on average by between 9% and 9.5% per extra year in the mid-1960s, by between 6% and 6.5% today, and this return is of the same order for men and women, at the middle of their working lives. However, it has to be accepted that the estimates of the returns to education are imprecise : those obtained here for the return to one year of schooling vary by between 0.5 and 1 point depending on empirical choices that are in part determined by convention (see in Appendix), which is a very high degree of imprecision relative to results in the region of 6% and 9%.
In France, the returns to education, measured in salaries, declined over twenty years (1965-1985) and have been stable for the past fifteen years. Compared with thirty-five years ago, the gaps are smaller, particularly between private sector and public sector employees. The returns to education are nonetheless higher in private sector companies for men, whereas for women they are comparable in both the public and private sectors.
APPENDIX
Figure A1
Average annual net salary (in current-rate francs) by length of schooling – Weighted data
Source : INSEE.
Robustness analysis
We have conducted an in-depth analysis of empirical robustness, but for reasons of space only the main conclusions are presented here.
Robustness by treatment of apprenticeship
As was mentioned in the introductory discussion, indentured apprenticeship occupies an intermediate position between school and work, and can be counted either as time in education, or as experience, or as both. By taking it into account in these three ways we can evaluate the “conceptual robustness” of the marginal rates of return to education and experience (Table A7).
Regarding returns to education, the orders of magnitude are quite similar and the movement of decline in returns is qualitatively well described in all three variants.
Counting indentured apprenticeship as training and not only as experience (Variant 2) leads, first, to lower returns at the beginning of the period studied here, and second to a decline in these returns over time that is also smaller.
As concerns the returns to experience, the differences are very small. We can therefore consider that the convention in respect of apprenticeship does not greatly affect estimates of the return to one year of experience.
Empirical robustness
The influence of specifically empirical choices on the results does not receive the attention it deserves. And yet, as we shall see, this influence is sometimes large, while the empirical choices themselves are not always obvious.
Sensitivity of the estimates to the population observed
On occasions the age at completion of education given for a person is under 6 years, or a specific code indicates “no education” (this represents between 0.1% and 1% of the population observed). In such cases we attributed value 0 to the length of schooling. Likewise when the indicated age at first employment is under 10 years, as is the case for between 0.1% and 1.5% of the members of the population used, the start of working life is by convention set at 10 years. But these choices are in part arbitrary. By removing the individuals involved from the data we were able to measure the effect of these choices ; it is not always negligible (Table A8). In the case of a return calculated at the “fixed point”, the greatest difference was among male private sector employees in the 1998 Labour Force survey, where the marginal rate of return to education decreased from 8.70% (Table A3) to 7.33% (i.e. a difference of 1.4 points), even though the number of individuals involved is quite small (191 persons excluded out of 19,068, i.e. in weighted figures 64,994 out of 6,274,538). The return calculated for the average durations presents a much smaller difference : 10.02% instead of 9.78% for male private sector employees in 1998.
Likewise the estimators are appreciably different according to whether we consider the whole population or restrict analysis to the employees whose salary is such that ln(S) is in the interval [Average (ln(S)) ± 3σ] (Table A9). The effect is at times very large and is quite disturbing : the estimators can — although it is not systematic — depend appreciably on whether or not a small number of individuals are included in the analysis, the decision to include them or not being to a large extent conventional. This kind of purely empirical sensitivity must preclude detailed discussion of variations below a certain size, including when they relate to basic issues (conceptual and econometric).
“Survey” effect evaluated by comparing the estimates obtained from the 1993 FQP and Labour Force surveys
The variables used to select the individuals of the population are not the same in both series of surveys (FQP surveys and Labour Force surveys). A survey was conducted in each series in 1993, so these two surveys were used as a basis to adjust the populations. The differences in returns calculated for the two surveys in 1993 on 1992 salaries are not negligible, reaching nearly 1 point for men in the private sector, with no clear explanation as to the reason. In all probability the origin of these disparities must be attributed to intrinsic differences in the two series of surveys on which this study is based. This survey effect seems large and cannot be corrected for in the present study. This is why both estimates are presented for 1993.
Sensitivity to weighting
In the FQP surveys, unlike in the Labour Force surveys, the sample is stratified. Because the weights differ widely from one individual to another (for example, they vary from 200 to 2,000 for the 1964 FQP and from 659 to 15,062 for the 1993 FQP), the estimates and calculations must be carried out with weighting so that results derived from the FQP surveys apply to the entire French population. The difference can exceed 2 points for certain sub-populations, as is seen for private sector employees in the 1964 FQP : the earnings return to an extra year of schooling at the average point is 11.62% with weighting (Table A2) as against 14.33% without (Table A10). This is easily understood given that only the FQP surveys are conducted on stratified samples and that, within this stratification, the higher social categories are sampled much more than the lower social categories. Weighting is necessary if the sample used is to be representative of the parent population, and the authors of earlier studies who did not weight the results of FQP surveys before performing their econometric analyses should have done so. The measurement of sensitivity performed in this case is thus for information only, and the conclusion it yields does not have the same meaning at all as the previous ones.
All the results presented and discussed in this article have therefore been obtained from weighted calculations (including for the Labour Force surveys, for which, however, weighting is less necessary).
Sensitivity for full or restricted models
The variables for which we obtain estimated parameters non-significant at the 5% level are usually the interaction (de) and those associated with experience only (e2 or e3). Among the FQP surveys, that of 1993 had a considerably smaller sample (see Table A12) and non-significant parameters are more frequently obtained. To take into account the parameters non-significant at the 5% level, regressions were performed with an option whereby the variables with the least significant parameters were eliminated one by one, until all of the remaining variables were estimated as significantly different from zero at this level of confidence (Table A11). The average returns obtained using the parameters estimated in this “restricted model” are always very close to those obtained using the initial full model. Reference is therefore to the latter throughout this article.
Table A1
Trend in the adjusted determination coefficient (r2) in the earnings return model
Table A2
Rate of return to an extra year of schooling at the average point of the sub-populations (%) – Weighted data
Table A3
Rate of return to an extra year of schooling at the fixed point (d = 10 years, e = 20 years) (%) – Weighted data
Table A4
Rate of return to an extra year of schooling by length of schooling for 20 years of experience (%) – Weighted data
Table A5
Rate of return to an extra year of education by length of schooling for zero experience (%) – Weighted data
Table A6
Rate of return to an extra year of experience by length of experience for a length of schooling of 10 years (%) – Weighted data
Table A7
Rate of return to an extra year of schooling for an average length of experience (%) : comparison of the 3 variants – Weighted data
Table A8
Rate of return to an extra year of schooling at the fixed point (d = 10 years, e = 20 years), excluding from the population those “with no education” and those “starting employment before age 10” (%) – Weighted data
Table A9
Rate of return to an extra year of schooling of sub-populations at the average point, retaining only those observations for which ln(S) is in the interval [Average (ln(S))±3σ] (%) – Weighted data
Table A10
Rate of return to an extra year of schooling of sub-populations at the average point, with no weighting of the data (%)
Table A11
Rate of return to an extra year of schooling of sub-populations at the average point calculated only with parameters significant at the 5% level in the Mincer equation (%) – Weighted data
Table A12
Numbers observed and weighted in the samples used
Table A13
Average weighted annual net salaries (in current-rate francs)
Acknowledgements
We would like to thank the two anonymous readers whose valuable comments helped to improve this article.
BIBLIOGRAPHIE
·
Arestoff F., 2001, “Taux de rendement de l’éducation sur le marché du travail d’un pays en développement. Une analyse micro-économétrique”, Revue économique, 52(3), pp. 705-716.
·
Ashenfelter O., Rouse C., 1999, Schooling, Intelligence and Income in America : Cracks in the Bell Curve, NBER Working paper No. 6902, National Bureau of Economic Research.
·
Baudelot C., Glaude M., 1989,“Les diplômes se dévaluent-ils en se multipliant ?”, Économie et statistique, 225, INSEE.
·
Bayet A., 1996, “Carrières continues, carrières incomplètes et salaires”, Économie et statistique, 299, INSEE.
·
Bils M., Klenow P.J., 2000, “Does schooling cause growth ?”, The American Economic Review, 90(5), pp. 1160-1183.
·
Boumahdi R., Plassard J.-M., 1992, “Note à propos du caractère endogène de la variable éducation dans la fonction de gains”, Revue économique, 43(1).
·
Bound J., Solon G., 1999, “Double trouble : on the value of twins-based estimation of the return to schooling”, Economics of Education Review, 18(2), pp. 169-182.
·
Card D., 2001, “Estimating the return to schooling : progress on some persistent econometric problems”, Econometrica, 69(5), pp. 1127-1160.
·
Casaccia M., Seroussi G., 2000, “Séries longues sur les salaires”, Insee-Résultats, 735.
·
Degenne An., Lebeaux M.-O., Vallet L.-A., 1998, Les données de l’enquête Formation et qualification professionnelle de 1964 sont de nouveau disponibles, Lasmas/Institut du Longitudinal.
·
Goux D., Maurin E., 1994,“Éducation, expérience et salaire”, Économie et prévision, 116, Ministère de l’Économie et des Finances, direction de la prévision.
·
Griliches Z., 1977, “Estimating the returns to schooling : some econometric problems”, Econometrica, 45(1), pp. 1-22.
·
Hanchane S., Moullet S., 1997, “Mesure et analyse des rendements éducatifs : le cas français”, document de travail du Greqam, No. 97B07.
·
Hanchane S., Moullet S., 1999, “Accumulation du capital humain et relation éducation salaire”, document de travail du Greqam, No. 99C02.
·
Jarousse J.-P., Mingat A., 1986, “Un réexamen du modèle de gains de Mincer”, Revue économique, 6, pp. 999-1031.
·
Lévy-Garboua L., 1972, Une analyse économique de la distribution des revenus individuels, doctoral thesis, Université de Paris I.
·
Marchand O., Thélot C., 1997, Formation de la main-d’œuvre et capital humain en France depuis deux siècles, Les dossiers d’Éducation et Formations, No. 80, ministère de l’Éducation Nationale.
·
Mincer J., 1958, “Investments in human capital and personal income distribution”, The Journal of Political Economy, 66(4), pp. 281-302.
·
Mincer J., 1974, Schooling, Experience and Earnings, New York, Columbia University Press for NBER.
·
Plassard J.-M., Tahar G., 1990, “Théorie du salaire d’efficience et disparités non compensatrices : évaluation à partir de l’enquête FQP”, Économie et prévision, No. 92-93, Ministère de l’Économie et des Finances, Direction de la Prévision.
·
Riboud M., 1977, An Analysis of Earnings Distribution in France, PhD thesis, University of Chicago.
·
Riboud M., 1978, Accumulation du capital humain, Paris, Economica.
·
Rouse C.E., 1999, “Further estimates of the economic return to schooling from a new sample of twins”, Economics of Education Review, 18(2), pp. 149-157.
NOTES
[*]
CNRS, Laboratoire d’Analyse Secondaire et de Méthode Appliquée à la Sociologie – Institut du Longitudinal (LASMAS – IdL).
CNRS, Laboratoire d’Analyse Secondaire et de Méthode Appliquée à la Sociologie – Institut du Longitudinal (LASMAS – IdL).
[**]Cour des Comptes. Translated by Accenta Ltd.
Cour des Comptes. Translated by Accenta Ltd.
[1]Those with only four years of schooling are employees of older ages or who were educated abroad
Those with only four years of schooling are employees of older ages or who were educated abroad