______1
Louis Chauvel
Cellule de Sociologie,
Observatoire Français des Conjonctures Economiques,
Fondation Nationale des Sciences Politiques,
69, Quai d'Orsay
75007 Paris.
fax: 33.1.44.18.54.54.
E-mail:
Cohort changes in Education, Social Stratification andMobility, the Case of France (1964-1995)[*]
Résumé
Few research works on social stratification and mobility are directly concerned with the cohorts. Most frequently, the observed changes, such as the expansion of education, the growth of the part of skilled positions, the upgrading and the increase of upward mobility, are intuitively considered to be continuous trends, concerning equally any age and any cohort. Actually, in France, these trends are not similarly distributed between cohorts: some of them benefited from a fast change when the others knew stagnations relatively to the prior cohorts. The level of education, the opportunities to have access to the middle or higher categories (EGP-II et EGP-I), the social value of academic titles in terms of access probabilities to the most prestigious categories of the society, but also in terms of upward mobility, are criteria which underline the specificity of cohorts born during the ’40, which appear to have benefited from the fast social change opportunities. The later experience a clear pause relatively to these trends.
The results are: (1) the expansion of the level of education is not linear; (2) globally, when they are estimated by cohort, the opportunities to enter various social groups, notably the highest on the social hierarchy, evolve by steps; (3) the ‘social value’ of a given level of education, in terms of opportunity to enter the highest social strata, is not stable nor linearly evolving from a cohort to the other; (4) even if social fluidity is more or less stable from a cohort to the other, the opportunities to experience an upward or downward mobility differ considerably from a cohort to the other. Thus, it might be useful, for the research of essential social changes in stratification, to insist more on cohort analysis: the stability of fluidity might hide some changes due to the evolution of the margins of the mobility tables, and important evolutions of inflows and outflows. When that cohort analysis is avoided, the risk is to miss the rhythm, the pace and the timing of social change.
The research analyses on the changes of social stratification avoid frequently the question of birth cohorts[1]. Generally, the cohort is considered to be neutral. The analysis of the expansion of education, of middle and higher social strata of society (upgrading), as well as many phenomena concerning the social stratification on various aspects, seem to be rarely interested in cohorts. Marchand et Thélot (1997), in their book on the long-term evolution of the French social structure, present changes (tertiarisation, development of skilled employment, of middle and higher categories, of education, etc.) which are similarly observed in most industrial countries, and never refer to the distribution of these changes between birth cohorts. This is not a specific case: when one consider the great trends of social changes, the implicit idea is that they are smoothly distributed by birth cohort, the later progressively benefiting from changes that the former did not knew. The trends might have regular and proportional consequences on cohorts, without substantial acceleration nor braking. Thus, the cohort might be negligible. That hypothesis should be evaluated.
Soma classical empirical research works in social mobility since World War II have had other intuitions and suspicions: the members of the successive cohorts can have met a destiny more or less favourable than former or later; History sometimes implies non linearities. Glass and Hall (1954), for example, asked these questions: the succession of economical periods, the crisis and the emergence of public interventionism — for education, research, but also for health, social services, etc. — could have had an impact on the destiny of consecutive cohorts, and the authors tempted to illustrate that hypothesis; nevertheless, the survey used by Glass, which was simply one cross-section sample, could not be sufficient to separate age and cohort effects. In USA, the comparison of life cycles and intergenerational mobility by Jaffe and Carleton (1954) showed the specificity of cohorts born near the year 1910, who experienced the 1929 crisis at their entrance into the labour market and who met a social destiny less favourable than successors and predecessors. Blau and Duncan (1967, pp. 81-113, 177-188) located similar effects for that 1929 generation, whose first five or ten years in the labour market were more difficult than for the elder and the younger, implying a comparative handicap which characterised them during their whole life. More recently, Erikson and Goldthorpe (1992, mainly pp. 72-99) observed inflow and outflow rates by cohort, but here like elsewhere having only one survey for each country they could not separate the cohort effects from the age effects.
Linearity or fracture of stratification change by cohort
As the 1929 crisis, the economical braking which begun 25 years ago in many industrial countries could have caused a rupture in the trend of social upgrading, in the development of skilled employment, and even partly in the public educational investment, and might have implied irreversible marks in the social destiny of the cohorts which entered into the labour market after 1975.
More generally, analysing the evolution of social positions of successive cohorts on a rather long term, and the comparison of their social positions at the same age in various periods, might be interesting for the point of view of an evaluation of the cohort dynamics linearity and stability. Thus, such an analysis might provide an evaluation of the participation of various cohort to global social change. I will describe the opportunities to enjoy longer education, to have access to better social positions, to see the social value — in terms of probabilites to reach such or such social position — of a given academic title, and, last but not least, to benefit from an upward intergenerational mobility.
If these trends are not linear by cohort, their cohort decomposition could be important to understand whether they are the result of a global dynamic shared by any cohort, or the consequence of the replacement of former cohorts by new ones, arriving with a new social structure, which does not evolve no more.
Methods
Many methods exist for the decomposition of age-period-cohort effects. The methodological literature is important, and now classical (see Chauvel, 1997a and 1997b for a quick survey on this question). The basic instrument is the Lexis — or «Lexis-Becker-Verweij-Pressat»[2] — diagram, of which the objective is the simultaneous representation the three chronological dimensions which are time (or period), age (life cycle position), and birth cohort (diagram 1). Fundamentally, this representation does not need any panel data, but simply series of cross-sectional surveys (Deaton, 1985). The diagram crosses[3] years of observation (periods) horizontally, age vertically, and thus birth cohorts appear diagonally. It is the modern standard presentation. Thus, for periodic surveys, a new period will be added with a column on the right. That representation allows a simple reading of the three dimensions which are linearly linked: when somebody is a a-years-old person during year t, the next year, in t+1, she will be (a+1)-years-old; she was born in c=t-a. The perfect colinearity between the three dimensions is the source of many methodological difficulties. When the evolution is linear for one dimension, the mathematical decompositions are infinite, but when we have a rupture in any simple chronological dimension, its location is perfectly clear.
1- Lexis-Becker-Verweij-Pressat diagram (Pressat style)
The Lexis diagram and its homologues are not a scientific revolution, but a very clever easy way to present the interaction of the three chronological dimensions: in line one can read the destiny of successive cohorts at the same age, in column, the «apparent life cycle»[4] for a given year (isochronon), and the «actual life cycle», that a cohort will follow during its life, in diagonal (line of life). Because of the linear link between the three dimensions (age, period and cohort), the sociological interpretation of social times might be as dangerous as fecund (Riley, Foner, Waring, 1988, pp. 244-268).
2- Examples of age period and cohort effects - and life cycle recomposition
On some typical examples, phenomena properly linked to one of these dimensions are easily identifiable (diagram 2). A period effect shared by the whole population when it has a given age will typically be characterised by an horizontal rupture; it might be the case of the access to majority — if law is not modified («age effect»). A phenomenon known by the whole population after a given year («period effect») will be marked by a vertical rupture on the diagram; it might be the case of a disease without sequels that a vaccine will stop. A phenomenon characterising some cohorts, and not others, whatever age or period, will provoke a diagonal rupture («cohort effect»); for example, «those who lived through World War-II». Thus, some examples exist for which the diagram and any method of decomposition are appropriate.
A forth situation is, where the interaction between age and period does not follow vertical, horizontal nor diagonal line. That is a «life cycle recomposition» (Chauvel, 1997a), when the actual life cycle of the cohorts changes with no clear age nor period nor cohort effect. Here is a complex interaction between two dimensions. A modification of the legal age is a typical example. Clearly, The evolution of the proportion of «cadreset professions intermédiaires»[5] in the French population is a cohort effect (diagram 3).
3- Proportion of «cadreset professions intermédiaires» at work (Lexis diagram)
Source: compilation FQP-Emploi. (females and males)
The question is to offer an easy to read representation of that diagram. The simple bi-dimensional figures are of six types corresponding to the number of couples between the three chronological dimensions. The most classical consists in the couple (age, period) representing the «apparent life cycles» at various periods (figure 4): the access to the category of «cadres et professions intermédiaires» was more frequent for young than for olders during the ’70 and, after, the progressive growth for elders appears clearly.
4- Proportion of «cadreset professions intermédiaires» at work
Lexis age / periodLexis age / cohort
Source: compilation FQP-Emploi (females and males).
Here, to locate a cohort effect is specifically difficult. Another classical figure is funded on the couple (age, cohort), which presents the actual life cycle, longitudinally, of each cohort: each curve materialise the destiny of each cohort at successive ages.
A third diagram seems to be more interesting (figure 5): it uses the couple (cohort, age). That cohort diagram permits the comparison, at various ages, the position of the cohorts. If the curves at age 30, 35, 40, etc. presents similar shapes, and identical fractures revealing the same cohort cleavages at different ages, the cohort effect might be there. For the «cadres et professions intermédiaires», if the figure locates an age effect (the curves for the elder are above those of younger), the position of any cohort at 30 years old seems to condition the later positions. The interest of the cohort effect is clearly more important when it does not reveal a linear and continuous progression, but fractures, accelerations or brakings which characterise similarly any age. Here, the great step up of the ’40 cohorts is systematic at any age.
5- Proportion of «cadreset professions intermédiaires» at work: Lexis (Cohorte /Age) (cohort diagram)
Source: compilation FQP-Emploi. (females and males)
Cohort diagram: theoretical examples
The cohort diagram consists in the representation, for successive cohorts (horizontally), of the value of a given variable — percentages of housing property, suicide rates, income, or the proportion of EGP-I etc. — (vertically) at different ages materialised by the curves which allows to follow the situation of the cohorts at the same age. On series of hypothetical examples, we may understand the logic of that diagram.
The first one (1) represents a society where the reproduction is perfect, with no collective progress: the successive cohorts knows the same position at the same age, identically, with the same age effect: 15% of EGP-I at 30 years old for the cohort born in 1935 as in 1965. Of a similar manner, whatever the birth cohort, the proportion of EGP-I goes from 15% at age 30 to 25% at age 50.
The second example (2) presents the case of a regular and linear trend of social progression, equally distributed by any cohort: from a previous cohort to the later, at the same age, the part of EGP-I is increasing; for the 1935 cohort, 15% of EGP-I at age 30 and 21% for the 1965 cohort. Clearly, in a linear equally shared by cohort trend, any new cohort should experience a better destiny than the previous ones at the same age.
The third figure (3) reveals a very different evolution: the progress is, but it is entirely concentrated on one cohort: the 1945 cohort. It is a step progression. The cohorts born before knows a first model of society with 20 % of EGP-I at age 40; those born after a second model, with 25% at the same age. Evidently, the interpretation could be ambiguous: if the 1960 cohort benefited from the progress that the 1945 cohort initiated, the privilege of the 1945 cohort was to be the first to benefit from an higher proportion of EGP-I in a society where elders had a lower proportion of EGP-I. People born in 1960 have exactly the same position as their close elders at the same age. Here, a long term social progress exists: with the replacement of old cohorts by new cohorts, the proportion of EGP-I increases. But that progress is not regularly distributed between cohorts. To be member of the 1945 cohort is the best, but for the 1940 cohort, the situation is distressing. For the 1960 cohort, clearly, it is not possible to understand, by one’s personal experience, the signification of that social progress of which the elders of the 1945 cohort talk about.
The cohort diagram: six theoretical cases
Example 1Example 2
Example 3Example 4
Example 5Example 6
The fourth figure (4) is a composition of the two previous situations: progressive growth, plus one step for the 1945 cohort. Here, the 1960 cohort continues to enjoy some progress, even if it is less spectacular than those of the 1945 cohort.
The fifth case (5) is a situation of stop and of contraction for the post-1945 cohorts: for the global mean, from the arrival in the labour market of the 1945 cohort to its maturity, the global mean proportion of EGP-I in the society will grow, but more and more slowly, and the fell down of the trend is inscribed in that cohort dynamic.
The sixth diagram (6) is inspired from the fifth, but is more complex : the ages deviate gradually form preceding cohort to following one: the 1930 cohort seems homogenous, at least from age 35 to 50. The following cohorts know a progressive divergence: the life cycle is recomposing, and youth and maturity are less and less similar.
It is possible to use other models of separation of age period and cohort effects, inspired of the (APC) model of Mason, Mason, Winsborough et Poole (1973), and other types, discussed in Chauvel (1997a) . The colinearity between the three chronological dimension which are age, period and cohort remains the difficulty which imply the impossibility to clearly separate them when the evolutions are perfectly linear (a linear upward cohort effect formally corresponding to the combination of an upward period effect and of a downward age effect). But when the effects are not linear, and are not a complex interaction between two chronological dimensions, the separation is possible. Other models such as loglinear models using variables such as age, period and cohort (see infra) might be useful.
Data
Sources
The data used here comes from an original source: a compilation of the surveys «FQP» (Formation-qualification-professionnelle: 1964-1970-1977) and «Emploi» (1983-1989-1995), which describe the French labour force, and the civilian population. These surveys were obtained from LASMAS-IDL-IRESCO (CNRS), and extracted with the help of Irène Fournier. These large surveys of the French official statistical office (INSEE) are huge[6], and permit, with more or less continuous definitions, to follow the French population since 30 years on questions such as the level of education, social position and origins, and on many other aspects. The results presented here come from a file consisting in a normalisation of the six surveys: FQP 1964-1970-1977 et Emploi 1983-1989-1995, named here «compilation FQP-Emploi».
Having only one measure point each six or seven years might imply important constraints, particularly for the Lexis diagrams. To establish it, one should have regular time intervals, for example any 5 years, to cross periods with age classes of the same amplitude. To do so, a possible interpolation method is about to simulate the potential annual series of surveys presenting a credible image of the years between two successive surveys, providing thus estimations of years 1965, 1970, 1975, etc., 1995. Consider first table Aa,p constituted with the 6 surveys of the «compilation FQP-Emploi» for the 6 periods p = 1964, 1970, 1977, 1983, 1989, 1995 and presenting a specific measure (for example, the proportion of EGP-I) for each age a between 20 et 65 years old. To simulate the complete table for each age a and for each period p between 1964 et 1995, we may: