The mystery surrounding the French demographic transition and the reversal of the marriage pattern can be dispelled by regional and spatial investigations. Nineteenth-century France is known for being highly heterogeneous. It is therefore an ideal candidate for exploratory methods. The exploratory methods used in this section aimed to reveal the existence of clusters. I sequentially used two multivariate statistical techniques. First, I ran the principal component analysis needed to reduce the number of variables. Second, I relied on a hierarchical cluster analysis to identify clusters and establish a typology of French counties based on their demographic and socio-economic characteristics in the 1850s.
Principal component analysis
Principal component analysis (PCA) enables us to condense information from interdependent variables to a smaller set of factors (Jolliffe 2002). Before running PCA, the dataset is checked for appropriateness. I used the Kaiser-Maier-Olkin measure (KMO) of sampling adequacy to test whether the variables were largely independent or correlated very closely. The numbers were all above 0.5, which justified the use of principal component analysis.
The purpose of this exploratory method was to synthetize the information contained in a dataset by reducing the number of dimensions of the dataset, in order to identify and uncover latent patterns. PCA uses orthogonal transformations to convert a set of possibly correlated variables into a set of linearly uncorrelated variables (called principal components).Footnote 10 To construct a typology of fertility and marriage patterns, I reviewed the major influences on these things, selecting 25 quantitative variables. The variables capture a set of socio-economic, demographic, and cultural characteristics in the PCA. Table 1 presents the descriptive statistics of the variables used in the analysis. Most of the variables come from the Statistique Générale de la France. These variables are available for 85 counties (corresponding to NUTS-level 3) and concern the 1850s.
Table 1 Descriptive statistics—French counties, 1850s. Sources: Data from Statistique Générale de la France (see Appendix A for a description of the variables) A crude birth rate of 30 children per one thousand individuals is a level from which we can distinguish counties using birth control from those that do not. Above this level, it is likely that only a very small segment of the population uses fertility control. A crude birth rate below 20 children per one thousand individuals, however, suggests that a large segment of the population practices birth control (Chesnais 1992). Hence counties experiencing a fertility transition should exhibit intermediate crude birth rates ranging between 30 and 20 per one thousand individuals. As shown in Table 1, the average crude birth rate in mid-nineteenth century France was 26, ranging from a minimum of 17.7 (in Calvados) to a maximum of 33.5 (in Cher). Although the fertility transition had affected France about half a century before, fertility behaviors remained significantly heterogeneous across counties.
The demographic variables used in the analysis captured fertility behaviors (crude birth rates and marital fertility), marriage patterns (age at marriage, celibacy, share of married women, share of early marriagesFootnote 11), infant mortality, and out-of-wedlock births. As education variables, I used the literacy rates and the enrollment in public primary schools. Among the economic variables used in the analysis were the level of urbanization (density and agglomerated population), the occupational structure (employment in industry and in agriculture), landownership inequality measured by the share of landowners,Footnote 12 and the wealth captured by the GDP per capita. With regard to the cultural variables, religious practices are captured by the share of Protestants and by the share of juring priest in 1791 (Civil Constitution of the Clergy). The variables when accurate account for the gender dimension.Footnote 13
Seven principal components present eigenvalues greater than 1, as shown by the scree plot presented in Fig. 5. These 7 components explain 82% of the total variability. To avoid redundancy, I present and integrate in the remaining part of the analysis only the five major components out of the seven displaying eigenvalues greater than 1. These 5 components explain 72% of the total variability. In order to simplify the interpretation, the principal components are identified using orthogonal rotation (varimax–Kaiser) that allow us to put a smaller number of highly correlated variables under each component. Table 2 presents the key characteristics of each component. From each column, it is possible to define the variable with which each component is most closely associated. Only correlations above 0.3 are presented. The first principal component of a set of variables is the linear index that captures the largest amount of information common to our variables (Filmer and Pritchett 2001).
Table 2 Five Components of the PCA—orthogonal varimax rotation The first component (PC1), which explains 25% of the variance, is positively correlated with the literacy and enrollment rates in public primary schools for both genders. PC1 represents the overall level of human capital. The second component (PC2) is almost as important as the first and explains 20% of variance. This component is related to the marriage pattern (modern/natural vs. traditional). It is closely correlated with the share of married women ( +), the female and male ages at marriage (-), the proportion of women who married early (below 25 years old) (-), and definitive celibacy ( +). Additionally, this component correlates closely with the share of juring priests (i.e., those who swore an oath of loyalty to the Civil Constitution of the Clergy ( +), which captures the low weight of religious practice and traditions in the counties exhibiting a high share of juring priests (see below). The third component (PC3), which explains 12% of variance, is closely correlated with the density and agglomerated population. It represents the urban vs. rural structure of the economy. Figure 6 provides an illustration of the positioning of French counties along the factorial axes PC1 and PC2 (Fig. 6a) and PC1 and PC3 (Fig. 6b).Footnote 14 The French counties appear in blue and the variables used in the analysis are represented by red vectors (Fig. 6).
The fourth (PC4) and fifth (PC5) components both explain 7% of the variance. PC4 is positively related to fertility and is referred to as reproductive behavior. PC5 is positively correlated with men and women working in the industrial sector. It represents the industrial sector (as opposed to the primary and tertiary sectors).
Hierarchical cluster analysis
PCA enabled us to extract essential information from the dataset and express latent structures within a few variables (principal components). Combining a hierarchical cluster analysis (HCA) with our PCA allowed us to group our counties in relevant clusters based on their (socio-economic, demographic, and cultural) characteristics.
HCA is a method which explores the organization of samples in groups and among groups presenting a hierarchy (Lee and Yang 2009). From a set of 85 individuals (in our case, counties), the HCA spreads the individuals into a number of heterogeneous groups within which individuals share homogeneous characteristics. The first five factors of the PCA are used for hierarchical clustering with Euclidean Distance as distance measure and Ward’s computation method as agglomerative clustering. The dendogramFootnote 15 resulting from this method illustrates the sequence in which the counties were partitioned into clusters (Fig. 7).
A choice of six clusters was retained. This number of clusters appeared the most appropriate and presented the most meaningful association of counties.Footnote 16 The centroids of each class are presented in Table 3. Class 6, comprising one single county, is very different from the rest of the sample and appears as a clear outlier. Its density, agglomerated population, and GDP per capita are all considerably larger than any other county. The remaining five classes present a pertinent typology.
Characteristics of the clusters:
-
Class 1—Counties within class 1 are characterized by endowments in human capital below the national average. These counties present the characteristics of the Eastern EMP, without any clear and direct evidence of fertility control. Fertility rates are close to the national average. The labor force is mostly agrarian.
-
Class 2—Counties belonging to class 2 present a high endowment of human capital for both men/boys and women/girls. These counties present the characteristics of the Eastern EMP. Men and women marry earlier than anywhere else in France. The share of women who marry below the age of 25 is the highest. The share of married women is also the highest and the share of definitive celibacy is the lowest. Yet these counties present the lowest fertility rates reported in France. The population living in these counties is highly educated and exerts fertility control within marriage. This population is sparse and rural but benefits from a dynamic industrial sector.
-
Class 3—Like those in class1, the counties belonging to class 3 present endowments in human capital below the national average. But unlike class 1 counties, these present the characteristics of the Western EMP and exhibit high fertility rates. The great majority of the population living in these counties make their living from agricultural pursuits.
-
Class 4—Class 4 is the most closely marked by the industrial sector. Its population is dense and agglomerated and its fertility is above average. Endowments in human capital and marriage patterns vary substantially from one county to another; they cannot be used to classify the counties belonging to this class because they occur in an indiscriminate way.
-
Class 5—Like those in class 2, the counties belonging to class 4 reveal high endowments of human capital. Yet in major opposition, the counties belonging to class 5 present the characteristics of the Western EMP. Women marry later than in any other parts of France. The share of married women is the lowest. The share of women marrying below the age of 25 is twice as low as that in class 2. The share of definitive celibacy is (with class 3) the highest. Counties from class 5 present fertility rates that are below average. The population living in these counties is highly educated and exerts fertility control through marriage. This population is sparse and rural but, unlike that of class 2, does not benefit from a particularly dynamic industrial sector.
-
Class 6—Seine appears a unique county that does not match any of the other five classes. Seine displays the highest endowments in human capital for both men and women. Yet enrollment rates in education are the lowest (together with class 3). Marital fertility is very low but overall fertility is far above national average. Illegitimate births are twice as high as the national average. The density is 30 times higher than the national average and the GDP per capita is by far the highest in France.
The geographical distribution of the classes (Fig. 8) appears astonishingly coherent. Although 25 variables were used to construct the classes, among which many variables appear at first glance to be highly unevenly distributed across the country, very clear geographic classes emerge on the map that visually confirm the importance of the role played by the norms and the culture. The operating forces explaining the location of coherent geographic classes share socio-economic, demographic, and cultural characteristics that are specific to these geographic areas.
Figure 9 presents the position of the French counties along the two main discriminatory dimensions identified by PCA: endowments in human capital (the horizontal axis) and marriage patterns (the vertical axis). Counties displaying a high endowment of human capital are located on the right side of the zero vertical axis; those displaying low endowment on the left side. Counties presenting the characteristics of Eastern EMP are located on the upper side of the horizontal axis; those presenting the characteristics of Western EMP on the lower part of this axis. A clear division appears between our classes 1, 2, 3, and 5 (as summarized in Table 4). On the one hand, highly educated counties (classes 2 and 5) can be identified in contrast to poorly educated counties (classes 1 and 3). On the other, some counties characterized by free marriage practices (classes 1 and 2) may be set against counties controlling the incidence of marriage (classes 3 and 5). Counties from class 4 are not so clearly divided between our two main dimensions. Counties are on average more educated and present the characteristics of the Western EMP, but certain counties composing class 4 do not strictly follow this pattern.Footnote 17
Table 4 Classes on the two main dimensions Counties with strong traditions/authoritarian (classes 3 and 5) rely on ‘traditional’ marriage practices, as described by Hajnal. These counties controlling marriage present a low share of married women, late age at marriage, and high celibacy. These counties exhibit the highest marital fertility rates. Class 3 also exhibits some of the highest crude birth rates, in contrast to class 5 in which some of the lowest rates are recorded. Progressive/libertarian counties (class 2), where low importance is placed on religion and traditions, control fertility within marriage and do not exert any control on marriage. They display high shares of married women, a high proportion of early marriage, and low celibacy. Yet these counties—which I call progressive—control fertility within marriage.Footnote 18 Class 2 exhibits the lowest marital fertility and crude birth rates. Class 1 exhibits a similar marriage pattern to that of class 2. However, counties composing class 1 do not control their fertility within marriage as much as the counties of class 2 do and crude birth rates remain among the highest. Class 1 is not as economically advanced and successful as class 2.
In the traditional family economic system, marriage was considered the ultimate control mechanism of fertility (Van de Walle 1986). For Malthus (and contemporaneous authors), the share of single and the age at marriage formed the perfect mechanism for allowing individuals to adjust their fertility (‘adaptation mechanisms’). What our findings suggest is that early forms of birth control were used in certain regions to control fertility, in parallel to the traditional ‘Malthusian’ regulation used in others.
Counties with similar economic structures can display very different demographic patterns. The analysis conducted above shows that the weight of tradition and religious cult triggers different effects in similar (socio-economic) environments. Counties once educated used different means (ways) to attain a similar goal. Counties followed different trajectories, resulting in different timings. Eventually, all counties experienced the demographic transition, but they did not all follow the same trajectory. Culture, norms, traditions, matter profoundly. Accounting for the weight of traditions enables us to better understand the mechanisms (and the puzzles) behind the observed long-run economic and demographic process.
France can be divided into 2 main groups of counties on the basis of their marriage pattern (conservative vs. progressive) and can be divided into two groups on the basis of their fertility (‘natural’ vs. controlled). These two different divisions point to different situations and outcomes: fertility control within marriage, fertility limitation through marriage control, and natural fertility. If we want to understand the dynamic of the demographic transition, we need to more systematically account for the role played by tradition, culture, norms, and values. Economic development does not inevitably trigger clear-cut changes in a society. The path followed by counties in the process of development is not unique, nor linear. The stickiness of culture and norms explains why it took longer for certain areas in France to experience their fertility transition, despite conducive economic circumstances. The diffusion and spread of fertility control required cultural changes.
Cultural beliefs, family structure, and development
Our analysis shows that the marriage-fertility patterns are more complex than originally described by Hajnal (1965). We have observed the existence of two main types of fertility behavior in the strategy adopted by individuals and households. Fertility regulation can be the result of traditional means of control such as sexual abstinence, delaying first marriage, celibacy, age at first birth. But it can also be the result of ‘modern’ behaviors consisting in a direct control of the number of births within marriage through spacing out the intervals between births or stopping child-bearing at a certain age (Knodel and van de Walle 1979).
According to the innovation diffusion hypothesis (Carlsson 1966), the fertility transition is the consequence of new behaviors and new knowledge, changes in culture or attitude toward fertility. For Alter (1992), the fertility transition can be interpreted as ‘a shift in the mechanism of population control from restriction of marriage to limitation of childbearing within marriage.’ But what triggers this shift? What accounts for the persistence of Malthusian behavior in some counties and the shift to birth control in other counties?
Family structure and inheritance practices
Table 5 presents the correlations between our classes and the variables used in our PCA. Significant coefficients at the 5% level appear in bold. One qualitative variable—family structures—is integrated to our PCA to see how it connects with our typology. France is characterized by the coexistence of diverse family systems and inheritance practices (Berkner and Mendels 1978; Todd 1983, 1990, 2011; Le Bras and Todd 2013). Four main types of family emerge from their libertarian versus authoritarian structure (different parents-children relationships) and their equal versus unequal division of property (between siblings). The types of nuclear family (egalitarian and absolute) are characterized by a liberal relationship between parents and children, while the types of complex family (communitarian/cooperative and stem) are characterized by an authoritarian relationship between the generations.
Table 5 Pearson correlations In the nuclear egalitarian family, the distribution of inheritance between the children is equal. The equal division of properties induces, a priori, a decline in living standards (or the need to combine working with making a living from other activities). Already, in the seventeenth century, the peasants of the Parisian Basin practiced deeply egalitarian sharing—such as no legislation required (De Brandt 1901). In counties where peasants were dividing their land equally between their children, young people married earlier and established their own family sooner than in regions where lands were held undivided. In contrast to the nuclear egalitarian family, the nuclear absolute family is indifferent when dividing the inheritance to the principle of equality between children. The assets and properties are distributed by testament or will and usually go to one single individual, often the eldest or only son.
The communitarian family allows several households to live in the same house. All the sons can marry and bring their wives to the family home. All the brothers inherit equally from their parents. The success of this system depends on the age at which the parents die. In the stem-family system, one child only (usually the eldest) inherits the assets and property of the family (and preserves the lineage). Other children have to leave the family home when they get married but may stay if they remain single. The cooperative family (patrilocal egalitarian) shares strong similarities with the communitarian family. The main difference rests in the temporary nature of the co-residence.
The types of family are projected on the two main dimensions of the PCA in Fig. 9. It illustrates how the family systems relate to the classes. The coefficients of correlation are presented in Table 5. Classes 1 and 3 are positively and significantly correlated with extended types of family. In particular, class 1 correlates with communitarian families and class 3 with stem families. Classes 2 and 5, however, are positively and significantly correlated with the nuclear types (one household per house).Footnote 19 In particular, Class 2 correlates with the nuclear egalitarian type and class 5 with the nuclear absolute type.Footnote 20
Religious practice
For the Church, marriage is sacred. Contraception is forbidden and sex is not allowed outside marriage. Yet taking into account the difficulty of life for peasant families and their limited resources (Le Bras and Todd 2013), the Church softened the demographic pressure by allowing women to delay marriage. There is a demographic ideal type of Roman Catholic family characterized by a late age at marriage, high fertility and few illegitimate births. Hence, the rise in illegitimate births together with the decline in the ‘traditional’ marriage pattern marked a break in religious practices.
After the French Revolution, the government required all clergy to swear an oath of loyalty to the Civil Constitution of the Clergy. Constitutional priests chose to accept the Civil Constitution and to become State workers. This measure aimed at removing Christianity from everyday life in France. As shown in Table 5, Classes 1 and 2 on one side, are significantly correlated with our measure of religious practice proxied by the share of juring priests in 1791, while on the other side classes 3 and 5 are negatively and significantly correlated.Footnote 21 The collapse of clerical institutions in some parts of the country led to the disappearance of the traditional framework of religious life.
Religion matters in understanding how people controlled their fertility. The main difference observed among those who did so lies between individuals controlling their marriage and individuals controlling births (within their marriage). Our findings contradict the linear view that the dechristianization (only) triggered the demographic transition (see Blanc 2020). People living in regions characterized by strong religious practices controlled their fertility as well, but in a more traditional way. Such practices are hidden by the use of variables measuring marital fertility, such as the Ig index developed by Coale and Watkins (1986), as previously argued by Wetherell (2001).
Gender equality and women’s agency
Classes characterized by greater gender equality and higher investments in girls’ (and boys’) education display lower fertility. The pattern observed in class 2 is in line with Mendels’ (1984) argument that proto-industrialization (and more generally the combination of industrial and agricultural activities) in the countryside may have weakened the Malthusian preventive checks, characterized by late marriage. Yet some nuance should be introduced alongside this argument, as evidenced by the situation of class 5. A major difference between classes 2 and 5 is the level of religious practice. Culture and norms matter (as already argued in the literature, e.g., Bisin and Verdier 2000; Baudin 2010; Alesina et al. 2013; among others).Footnote 22 It can bring different outcomes and explain the differences observed between classes 2 and 5 with regard to their way of controlling fertility: birth control within marriage (class 2) versus control of marriage (class 5).
The literature on gender equality commonly argues that girls’ age at marriage is a good measure of the subordination of women in a society. The classic argument is that the increase in human capital delays marriage and gives greater autonomy to girls/women.Footnote 23 Our analysis shows that this is indeed the case in traditional societies (where religious belief and pressure are strong). In such conditions, greater gender equality occurs in parallel to low nuptiality, late marriage, and a high proportion of individuals remaining single. Yet what our analysis reveals is that in a progressive society (where the pressure of traditions is low), greater gender equality is practiced in environments characterized by active nuptiality, early marriage, and low celibacy.
The characteristics of class 2, combining high education for girls, active nuptiality, early marriage, and low fertility, reflect women’s greater decision-making power within the household.Footnote 24 What the characteristics of class 5 reveal is that outside of marriage women’s autonomy is possible, but, once married, women lose their autonomy. Late marriage, low nuptiality, and still relatively large marital fertility reflect a lower degree of women’s agency within marriage.
Gender equality, as observed in nuclear egalitarian families, fosters economic development. It is not the marriage pattern itself that fosters economic development. It is the level of gender equality, which can take different forms depending on the traditional versus progressive values of the territory where individuals live. Class 2 presents a combination of unexpected characteristics with regard to the conventional literature about marriage patterns and economic development (i.e., sparse and rural population, active and early nuptiality, recourse to birth control).
Different ways of controlling fertility co-existed in mid-nineteenth century France. The different patterns observed between regions cannot be explained as merely a process of diffusion with early and late adopters of fertility control behaviors. They reflect the existence of different strategies of adaptation used by individuals in response to their degree of autonomy/dependence on cultural conventions and norms, and of their degree of adaptation to socio-economic changes. Cultural beliefs and traditional values are (by definition) sticky (Alesina et al. 2013). Changes might ensue more easily/faster in regions that share certain values of change (Rocher 1973) and are more in favor of accepting changes.Footnote 25 The homogenization of behaviors over time certainly offers greater autonomy for people to decide for themselves and keep their distance from cultural conventions.
Class 2 enjoyed the optimal conditions for becoming the cradle of fertility transition (Fig. 10). The long-run trends presented in Sect. 3 suggest that the modern practice of birth control (within marriage) started as early as the eighteenth century and (slowly) diffused itself across France in the course of the nineteenth century. Rural people tended to be more inclined than urban people to choose to restrict their offspring and were thus ahead in terms of birth control. In certain places with high agglomerated population, the growth of industry stimulated fertility. In other places, such as the Parisian Basin, dynamic economic activities and rising living standards went hand-in-hand with birth control. Evidence suggests that these are not stable equilibriums but different forms of evolution with their own repulsive areas and attractive areas (Garden and Le Bras 1988). Ultimately, modern behaviors of fertility control spread to both urbanized and rural areas and led to the unification of the territory. Hence, two clear phases can be identified in the French demographic transition. Before the mid-nineteenth century, it consisted mainly in a rural phenomenon led by more prosperous districts, with no apparent link between industrialization and urbanization—Paris being an exception (Van de Walle 1986). After 1876, France witnessed a second wave of fertility limitation with the diffusion of modern behaviors to the rest of the country.Footnote 26
The complexity of marriage patterns is likely to explain why Dennison and Ogilvie (2014) found no correlation between female age at first marriage, celibacy rates, household complexity, and economic development.Footnote 27 Our analysis emphasizes the need to improve our knowledge of the long-run evolution of marriage patterns and to more systematically account for changes from a dynamic perspective. The analysis conducted in Sect. 4 shows that greater gender equality and women’s agency can occur in the context of active and ‘early’ marriage. The characteristics of the marriage pattern seem to be highly dependent on cultural beliefs and traditional norms. My analysis suggests that it is not the EMP itself but the greater level of gender equality and women’s autonomy (within the household, as long as there is low pressure from religious beliefs) that is at the heart of the positive feedback loop that triggered greater female agency, encouraged female participation in the labor force, increased human capital investment, reduced fertility, and fueled economic growth (Diebolt and Perrin 2013, 2019). The arguments developed by De Moor and Van Zanden (2010) and the elements raised by Dennison and Ogilvie (2014)—that triggered a stimulating discussion—could then be reconciled by (better) accounting for the role played by cultural factors—producing different effects and types of adaptations in different contexts.Footnote 28 In the French context, it seems highly unlikely that the reversal of the marriage pattern—occurring at the same time as the French Revolution and the weakening of the Catholic Church—happened to be a coincidence.