1To chart fertility over a long period, demographers and historians have often used synthetic indicators such as completed fertility or total fertility rate. These indicators describe an average number of children per woman for a cohort or a period of time, but do not track changes in family size. Has the percentage of childless women evolved over time? What is the share of families with four children? Through a meticulous comparison of the questions and data from eight population censuses and family surveys, Sandra Brée describes the changes in the family size of women born between the 1850s and the 1960s, spanning more than a century of cohorts. She shows that large families began to decline in the late nineteenth century, that levels of childlessness varied over the period, and that the share of families with two children began to increase with the cohorts born in the 1920s. This study highlights the benefits of associating indicators of mean fertility with measures of parity.
2The history of French fertility since the nineteenth century is now well known. However, the distribution of women’s lifetime number of children, in other words, their family size or parity, is rarely mentioned. Fertility tends to be approached through the mean number of children born to each woman (completed fertility for cohorts, and total fertility rate for period analysis) and through age-specific fertility rates, but these averages are rarely broken down to describe how many women had no children, one child, two children, etc. An analysis of averages conceals the specific trend of each family size. Is the decline in completed fertility atributable to the increasing rarity of large families or to a higher frequency of small families? And what is the impact of childlessness on completed fertility?
3Authors who have analysed the fertility of French cohorts over the long term (Daguet, 2002; Festy, 1979; INED, 1991; Sardon, 1990; Toulemon, 2001) have mainly worked with aggregate civil registration data (using the number of births as the numerator) and census results (using the number of women as the denominator). These aggregate data can be used to describe fertility trends, but not to determine the exact number of children born to women by the end of their reproductive lives.  However, as Patrick Festy stresses (1979, p. 96), “changes in family size are central to the explanation”. Only Laurent Toulemon (1995, 2001) has used family surveys to estimate the fertility of the cohorts born around 1930 and to calculate family sizes. Furthermore, these studies rarely distinguish between ever-married and never-married women.
4Consequently, there is no comparative research on cohort fertility going back to the cohorts of the 1850s that uses a retrospective method and differentiates between the fertility of ever-married women (including widows and divorcees)  and that of women who were never-married at age 45. This article begins with a review of the available sources and data, and of the methods used to analyse the fertility of the cohorts born in France between 1850 and 1965. It then examines fertility changes over a long period based on family size and then on parity progression ratios. The simultaneous study of changes in family size and completed fertility improves our understanding of fertility trends and, more specifically, the impact of each family size on the pattern of overall fertility.
I – Sources and method used to measure family size
5The retrospective method for estimating fertility uses data on the reported number of children born to women aged 45 and over during their reproductive lives. Questions on the number of children have not always been asked in censuses in France or elsewhere (Eggerickx and Begeot, 1993); France is rather unique in that regard – the censuses of 1906, 1931 and 1946 all included the question “How many children have you had?”.  In 1931 and 1946, the question on family size was asked of married, widowed and divorced women and of widowed men (and of household heads in 1906). This information is valuable for analysing fertility. From 1954 onwards, the question was included in a separate “Family survey” (Enquête famille, also called Étude de l’histoire familiale in 1999 and Enquête famille et logements in 2011), as recommended by Louis Henry (1953). Henry suggested that detailed analysis of fertility should be based on “the entire history of the family [collected by means of] specific surveys” (Henry, 1953, pp. 489-490), but that the question “How many children do you have?” should be kept in the five-year population censuses to provide statistics on families receiving family allowances based on their number of dependent children.
1 – What data are available?
6Since 1954, the question on the number of children born has been moved from the census to the Family surveys associated with the census (Table 1) that only cover a sample of the population.  The first family survey was conducted simultaneously with the census of 1954, and subsequent surveys were conducted with the censuses of 1962, 1975, 1982, 1990, 1999 and 2011 (Table 1). In the first three surveys, the question was only asked of ever-married women.  In subsequent ones, it was asked of all women, irrespective of their marital status, and in the last two surveys it was also asked of men.
Characteristics of the Family surveys
Characteristics of the Family surveys
7Fertility estimates based on retrospective questions can be skewed by various biases due to recall errors (especially for the oldest women at the time of the survey), selection effects due to deaths of children and their mother or to migration of mothers, and non-response. Analyses of French and Belgian data show that the risks of bias stemming from recall errors and selection effects are very low (Brée et al., 2016b), which confirms Neels’ (2010) and Van Bavel’s (2014) findings for Belgium, and those of Andersson and Sobolev (2013) for Sweden. Women’s completed fertility calculated from censuses and surveys is consistent with completed fertility from civil registration data. Moreover, concordant results are obtained for the fertility of women from the same cohort observed across different censuses or surveys. Survey data nevertheless appear to be less reliable than data from exhaustive censuses, especially for analyses at regional level, where the numbers may be quite small (Brée et al., 2016b). 
8The problem of non-response bias was raised by Paul Vincent as early as 1946. He identified two types of non-respondents to the question “How many live-born children have you had?”, namely “careless” respondents, who simply forgot to answer the question (or who did not wish to, although he does not really consider that possibility), and respondents who thought that, by leaving the space blank, they were indicating “none”. After analysing the non-responses, Vincent decided that the careless group accounted for only 3% of the total (in 1931) and assumed that the remaining 97% were childless. He distributed the first 3% of non-responses proportionately across the other family sizes. Louis Henry questioned this distribution method (Henry, 1953), but conceded that while the solution of considering all non-responders as childless was not quite satisfactory, it was doubtless impossible to do better. The researchers in charge of family statistics derived from the 1946 census also considered that all the women who did not answer the question “How many live-born children have you had?” were childless. While that may seem like a strong assumption, the differences between the 1931 and 1946 censuses are small (Figure 2). The level of total childlessness (all women) measured in this way (26% in France) is consistent with that measured in other western countries: 32% for Belgium (Brée et al., 2017), 26% in Germany (Dorbritz and Schwartz, 1996), 23% in the Netherlands (Rowland, 2007), 24% in the United States (Morgan, 1991) and 31% in Australia (Rowland, 2007).
9In the Family surveys, although the non-response rates vary considerably from one survey to another  (Brée et al., 2016b), consistent estimates are obtained for the different parities and for completed fertility by birth cohort.
2 – Measuring completed fertility and family size by women’s marital status
10Figure 1 shows estimates of completed fertility of women ever-married by age 45 calculated from the censuses of 1931 and 1946 on the one hand, and from the surveys of 1954, 1975, 1982, 1990, 1999 and 2011 on the other. 
11The estimates based on the two types of data source are highly consistent, with the exception of the last family survey (Famille et logements, 2011), which leads to much lower completed fertility than the three previous surveys (particularly 1999). There seems to have been an under-reporting of children born before the current union, in particular when those children were not living with the couple (Mazuy and Toulemon, 2013).
Completed fertility of ever-married women by cohort according to different sources (survey data smoothed over three years)
Completed fertility of ever-married women by cohort according to different sources (survey data smoothed over three years)Interpretation: Average number of live-born children born per married, widowed or divorced woman in each cohort.
12But what if we look more specifically at the parity distribution (Figure 2)? With regard to childlessness among women who had been married at least once, the 1946 census is highly consistent with the data from the 1954 and 1975 surveys, and with the 1999 survey for the cohorts born after 1926. Compared with those data, the 1982, 1990 and 2011 surveys overestimate childlessness. The data from the different sources are concordant for the “one child” and “two children” categories (except the data from the 1999 survey for the cohorts of women born before 1920 who reported having had two children). The data are slightly less homogeneous for women who reported having had three children, although they are still fairly consistent. However, the 2011 survey underestimates the percentage of mothers of four or more children compared with the data from the other surveys.
13Based on these different sources, we estimated women’s fertility using the weightings shown in Table 2. The data from the censuses of 1931 and 1946 were used for the cohorts born before 1906. For the 1912-1916 cohort onwards,  the family surveys were used, but not the 1999 and 2011 surveys for the oldest cohorts (aged over 75 at the time of those surveys) because the numbers were very low (Appendix Figure A.1). We chose to use these simple estimates because they make it easier to distinguish between the fertility of ever-married women and never-married women while maintaining data consistency for all women. Moreover, these estimates can be updated with data from future family surveys. Estimates of parity distribution and completed fertility are shown in Appendix Table A.1.
Distribution of ever-married women at the time of the survey or census by family size (%) and by five-year cohort according to different data sources (data smoothed over three years)
Distribution of ever-married women at the time of the survey or census by family size (%) and by five-year cohort according to different data sources (data smoothed over three years)Interpretation: For each figure: proportion of ever-married women at the time of the survey who had x live-born children.
Weightings used for the estimates based on censuses (C) and family surveys (S)
Weightings used for the estimates based on censuses (C) and family surveys (S)
14Regarding women never married at age 45, the 1982, 1990, 1999 and 2011 surveys yield consistent results on parities and completed fertility despite the small numbers (Appendix Figure A.2). For the cohorts born before 1915, the fertility of never-married women could not be measured directly. To estimate general fertility from the 1946 census (which does not provide data on never-married women), Festy calculated that never-married women contributed to general fertility by 0.06 first births and 0.02 second births, i.e. a total of 0.08 births for completed fertility of 2 to 2.6 births for all women (Festy 1979; INED, 1966, 1991). This is equivalent to 3% to 4% of births outside wedlock, a calculation consistent with Nizard and Maksud’s (1977) data on illegitimate births. However, this means, for the 1891-1895 cohort for example, that 52% of women who were never married at age 45 were childless, that 32% had one child and 16% had two children. The idea that almost half of the women who were never-married at age 45 had at least one child seems excessive, especially given that, among women born in 1912-1916 (the first cohorts for which parities of never-married women are at levels consistent with those of subsequent cohorts), the 1999 survey found that 84% of never-married women were childless, 10% were mothers of one child and 6% mothers of two children, i.e. just 16% of never-married women had at least one child. We therefore decided to apply these percentages for the earlier cohorts, at the risk of under-estimating slightly the contribution of never-married women to general fertility if illegitimate births decreased over those cohorts.  These estimates of the fertility of women never-married at age 45 are therefore lower than Festy’s, as is also the case, therefore, for my estimates of general fertility.
15The parity distribution of all women and their completed fertility is estimated using the following formula:
17where Pxg = overall fertility at birth-order x
18Pxm = fertility at birth-order x of women ever-married at age 45
19Pxnm = fertility at birth-order x of women never-married at age 45
20Cp = Permanent celibacy 
21The estimates for the cohorts born from the 1920s onwards  are highly consistent with those based directly on the survey data (Appendix Figure A.3). The estimates for the previous cohorts differ from those derived from the 1946 census since the data available in that census for all women assumed that all never-married women were childless (Appendix Figure A.4). They also differ from those that can be derived from the data in Depoid (1941) or Festy (1979) based on a diagonal reading  of age-specific fertility rates (Appendix Figure A.4). These estimates, for the cohorts born before 1890, are higher than those based an assumption that all never-married women had at least one child, which is clearly improbable: either completed fertility calculated retrospectively underestimates fertility, or a diagonal reading of age-specific fertility overestimates it.
22Under the first hypothesis, it would mean that the women who died or migrated had higher fertility than the others. Yet, the women lost to observation generally had slightly lower fertility than the survivors (Andersson and Sobolev, 2013; Brée et al., 2016b). As suggested above, assuming that all of the women who did not respond were childless may also be excessive. It is also possible that when they answered the question at age 45 or more, some women may have omitted children who were born alive but who died a few hours or a few days after birth. 
23Under the second hypothesis, completed fertility could be overestimated by a diagonal reading “because the fragments of the history of the various cohorts represented by each of these rates in a given year depend on the past history of these same cohorts” (Sardon, 1990, p. 17).
24The fertility of never-married women and of all women born before 1890 probably falls somewhere between Festy’s estimates (probably overestimated) and ours (probably underestimated). This analysis of the sources and the various estimates derived from those sources suggests that the family size estimates for married women are the most robust, in particular for the oldest cohorts.
II – Changing family size in cohorts born between 1850 and 1966
1 – Trends in family size
25Figure 3 shows the trend in family size and completed fertility of ever-married women in the cohorts born in France between 1850 and 1966. We can see that similar fertility levels are not always the result of the same parity components (Toulemon, 2001). The 1897-1901 and 1957-1961 cohorts exhibit the same low completed fertility (2.22 children per woman), but for the earlier cohorts it can be attributed mainly to high percentages of childless women (17%) and mothers of only one child (26%), whereas for the later cohorts it results from a predominance of two-children families (41%). 
26Five groups of cohorts emerge from Figure 3. First, the cohorts born before 1890, for whom the dominant model (more than 25%) was large families of four or more children, and whose share began falling steadily at the end of the nineteenth century (De Luca-Barrusse, 2008); second, the cohorts born between 1891 and 1911 who had their children between the two World Wars and for whom the dominant model was one child; third, the cohorts born between 1912 and 1931 who had most of their children during the baby boom, among whom we see an increase in the number of families of three or more children; fourth, the 1932-1946 cohorts among whom the share of families of four or more children declined and the two-children family began to dominate (more than 30% of families); last, the cohorts born in and after 1947, among whom we observe a general convergence of fertility behaviours (Daguet, 2002; Prioux, 2002) and a stability (Toulemon et al., 2008), which seems to persist to this day (Mazuy et al., 2013).
Parity distribution of ever-married women (at age 45) by cohort*
Parity distribution of ever-married women (at age 45) by cohort** The mean age at childbirth was 28 years for the oldest cohorts, and 27 for the more recent cohorts (INSEE, population estimates and civil records). For ease of interpretation, an average age of 28 years was added to the birth cohorts to give an idea of the reproductive period.
27We thus observe a “cyclical” pattern, where each cohort seems to react to the previous one by adopting the opposite behaviour. The daughters of the women who had large families (the cohorts born before 1890) themselves had fewer children (the cohorts of 1891-1911); in turn, their daughters had large families (1912-1931 cohorts), but their granddaughters returned to lower fertility (1932-1951 cohorts). However, the most recent cohorts (1952-1966) seem to have maintained a steady model of fertility mainly characterized by a majority of two-children families, although it is too early to say whether this is a long-term trend or not.
28The fertility decline at the end of the nineteenth century has been interpreted as a sign of the change in parent-child relationships when affection became central (Ariès, 1960, 1980) and the child became “precious” (Praz, 2005). Controlling fertility enabled parents to invest, emotionally and financially, in the “quality” of their children (Alter, 1992; Becker, 1965; Caldwell, 1976). An analysis of family sizes reveals that it is mainly large families (four or more children) that have become less and less common: their share fell by more than half between the cohorts of 1850 and 1895. Anne-Marie Sohn (1996, p. 809) has called the fourth child a “threshold number”, beyond which “parents decide to abort”. However, the decrease in families of four or more children was not offset by an increase in three-children families but rather by an increase in the percentages of childless women and mothers of only one child among the women who had children in the inter-war period. In other words, the fertility decline that began with the cohorts born around 1850 has not been linear across all family sizes, with a progressive downsizing from four children to three, and from three children to two, and so forth, but represents a shift from a regime where more than one-third of families had four or more children to a regime where large families accounted for fewer than one-fifth of the total.
29The interwar period was highly specific because it simultaneously represented the end of the first transition (or the beginning of the stabilization phase if we view the post-war baby boom as simply an “accident”; Desplanques, 1988; Knibiehler, 1997) and a time of major socioeconomic change (the demographic, social and economic impact of the First World War and the economic crisis of the 1930s).  Despite public policies to boost fertility through repressive measures  and incentives  aimed at rebuilding the population and making up for the deficit in births, the women who had their children between 1920 and 1940 were among the least fertile of the twentieth century. There was clearly a mismatch between legislative action and behaviour (Cova, 1997), since these laws had no apparent impact on the birth rate (McLaren, 1990). While the first child was usually desired, subsequent births were often precluded: “maternal instinct” was often referred to in the joyful testimonies recorded by Rebreyend (2003, p. 214), but much less so for successive, closely-spaced pregnancies, with unwanted pregnancies being described as a trap. One-child families seem to have been truly characteristic of the interwar period (Brée et al., 2016a, 2017). Some researchers attribute the high prevalence of small families to the specific events of the period rather than to personal preference (Rowland, 2007). For others, the rise in childlessness and in the proportion of one-child families indicates a growing acceptance of behaviours (especially being married and voluntarily childless) that had been unthinkable earlier (Anderson, 1998; Morgan, 1991). These behaviours are sometimes described as pioneering precursors of future change (Anderson, 1998; Van Bavel and Kok, 2010). It is hard to fully comprehend the reasons behind the low fertility levels of the interwar period, although, in addition to the continued fertility decline linked to the end of the transition, most theories emphasize the aggravating effect of context (Brée et al., 2016a).
30How should we interpret the fertility upturn that began with the 1895 cohort (Festy, 1979)? The French baby boom was characterized by a decline in childlessness and one-child families, a slight increase in two-children families, but above all by a very sharp increase in the share of families of three or more children. If we simulate a linear decline in families of four or more children between the cohorts of 1897 and 1937, instead of the observed increase, and proportionally redistribute the other family sizes, completed fertility levels off in accordance with the theoretical model of the demographic transition (Appendix Figure A.5). This result supports the interpretation of the baby boom as a temporary deviation (Coleman, 2004) and establishes the interwar period as the end of the demographic transition, or at least as the beginning of a long phase of stagnation of average fertility, because family sizes subsequently changed again. Demographers continue to argue about the factors that led the cohorts of only children to have so many large families (Calot and Sardon, 1998; Van Bavel and Reher, 2013). The baby boom has been seen as a consequence of the optimism triggered by the end of the Second World War. However, fertility began to rise before the end of the war, so that hypothesis has been rejected (Calot and Sardon, 1998; Van Bavel and Reher, 2013). Economic explanations have also often been advanced but do not find much empirical support (outside studies of the United States; Doliger, 2008). According to Van Bavel and Reher (2013), the baby boom can be attributed largely to a boom in marriage, a decrease in the mean age at marriage and a consequent decrease in the mean age at first birth. Longer marriages and the use of still traditional means of contraception (mainly coitus interruptus) led to an increase in family sizes. In the specific case of France, according to Calot and Sardon (1998) and to Chesnais (2006), France’s pro-family policies may also have created a “sense of urgency” to avoid population decline.
31In the cohorts born in the early 1930s, fertility fell again. In France the baby bust was characterized by an increase in families of two children and a decline in families of four or more. The bust is usually attributed to the contraceptive revolution (Westoff and Ryder, 1977), even though it predated by a few years the distribution and use of the contraceptive pill (but not necessarily its appearance; Caldwell, 2001). According to Leridon (1987), the baby bust was the combined result of a decrease in the desired number of children and more effective family planning. Van de Kaa and Lesthaghe (Lesthaghe, 1995; Lesthaghe and Van de Kaa, 1986; Van de Kaa, 1987) have proposed the concept of “second demographic transition” to describe the changes in the last third of the twentieth century that led to the decline in fertility. According to these authors, the emergence of an individualistic family model, in which the family is built upon the couple relationship, has affected both the family formation process (delayed marriage and greater propensity to divorce) and reproductive behaviour, with a pronounced decline in fertility. Coleman (2004), along with others, has contested this theory, claiming that the fertility downturn is simply a phase in the overall process of fertility transition that began at the turn of the twentieth century or even earlier.
32In the most recent cohorts (women born in and after the 1950s), fertility reached very low levels, characterized by strong polarization around two-children families and a smaller percentage of large families with four or more children. However, even if the very low levels of fertility observed since the late 1970s and especially in the 1980s are one of the characteristic features of the “second demographic transition”, current fertility levels are not exceptional and are similar to those observed before the baby boom (Brée et al., 2017). The predominance of two-children families is, conversely, specific to the women born since the 1940s. However, Frejka (2008) suggests that this predominance may be challenged by a possible future increase in childless and one-child families. In France, childlessness remains at fairly low levels compared with other European countries, but one-child families are quite common and their percentage has remained relatively stable since the cohorts born in the 1940s. This stability is remarkable given that the expected impact of the increase in the mother’s age at first birth and the higher frequency of separation among couples with one child would be to raise the probability of having only one child (Breton and Prioux, 2009).
33While there are relatively few never-married women in the cohorts studied (7-12%), their fertility behaviour warrants analysis, especially as their proportion has increased in the most recent cohorts, and hence the share of their births in overall fertility. 
34While the women born before the First World War had very few children out of wedlock – illegitimacy was still socially stigmatized (Fauve-Chamoux and Brunet, 2014) – behaviours changed after the Second World War. A growing proportion of never-married women born in the late 1920s and early 1930s had at least one child (Figure 4), and the phenomenon amplified in the subsequent cohorts. While explanations for the baby boom often revolve around the increase in legitimate fertility and age at marriage (Van Bavel and Reher, 2013), it appears that never-married women also contributed to the increase in fertility, at least in France.
Parity distribution of never-married women at age 45, by cohort*
Parity distribution of never-married women at age 45, by cohort** The mean age at childbirth was 28 years for the oldest cohorts, and 27 years for the more recent ones (INSEE, population estimates and civil records). For ease of interpretation, an average age of 28 was added to the birth cohorts to give an idea of the reproductive period.
Note: For the oldest cohorts (up to the 1912-1916 cohort), the distribution is the same because we applied the distribution of the 1912-1916 cohort to them. Even if the shift in the distribution between the 1855 and 1910 cohorts is probably not as constant, the limited change observed between the 1912-1916 and 1922-1926 cohorts seems to validate our methodological choice.
35Starting in the 1970s, with the increase in cohabitation before marriage (Roussel, 1978) and non-marital unions (Guibert-Lantoine et al., 1994; Rault and Régnier-Loilier, 2015) marriage was no longer a pre-requisite for family formation. Extra-marital births became more common (Muñoz-Perez and Prioux, 1999) even if childbearing was generally preceded by marriage. By 2007, however, the percentage of children born outside marriage exceeded that of children born to married couples (INSEE, civil records). Among never-married women, a large number have nevertheless been in a relationship, have lived with a partner and even officialized their union (through a civil partnership, introduced in 1999; Rault and Régnier-Loilier, 2015), but never-married women have always had fewer children than their ever-married counterparts. In the cohorts born in 1962-1966, a much higher percentage of never-married women are childless than ever-married women (40% versus 6%). However, among women who have had at least one child, the behaviours are much more similar. Never-married mothers more frequently have only one child than ever-married mothers (39% versus 19%), but they almost as often have two children (40% versus 44%).
36These differences in fertility can be attributed to the fact that some never-married women do not have a partner, or at least not during their reproductive years, but also to the fact that women who do not want children are much less likely to marry than those who do.
37The family sizes of all women (Figure 5) closely match those of ever-married women, since the latter represent the overwhelming majority (88% to 93%) of women. The main variations are observed for parities 0 and 1, since these are the parities most affected by the behaviour of never-married women. The changes in completed fertility appear to be determined by large families, even more so than for married women alone.
38Childlessness should therefore always be analysed with care, since levels vary strongly depending on whether all women or only ever-married women are taken into account. In France, childlessness among women born in 1892-1896 was 26.9% for all women and 18.8% for ever-married women.
39Analysing the fertility of ever-married women only might appear to be a more rigorous approach (because the data are better and the indicator more robust since ever-married women are, in principle, all exposed to the risk of pregnancy). For the most recent cohorts, however, it is important to measure overall fertility, since there is no longer a distinction between marital and non-marital births.
Parity distribution of all women, by cohort (1850-1966)*
Parity distribution of all women, by cohort (1850-1966)** The mean age at childbirth was 28 years for the oldest cohorts, and 27 years for the more recent ones (INSEE, population estimates and civil records). For ease of interpretation, an average age of 28 was added to the birth cohorts to give an idea of the reproductive period.
2 – Parity progression ratios
40In addition to family size, which indicates how many women have had exactly one child, two children, and so forth, during their lifetimes, parity progression ratios measure the percentage of women who have increased their family size from one child to two, from two to three and so on (Figure 6). For all childless women, the ratio a0 expresses the risk of having at least one birth (its complement to 1 (1 – a0) represents the risk of remaining childless). For all women who have had at least one child, the probability a1 expresses the risk of having at least two children, and so forth. This indicator is useful in that it shows how many women are mothers of at least one child, at least two children, etc., without taking subsequent births into account. It is surprising to observe that while the probability of having a second child is the same in the cohorts born around 1960 as in those born a century earlier, there are twice as many families of two children in the most recent cohorts (39%) as in the cohorts born around 1860 (18%).
41We can see immediately that parity progression ratios decreased across all cohorts of women born between 1850 and 1891. The pattern changed for the cohorts between 1892 and 1911 who had their children during the interwar period, among whom the probability of having a first child increased. The probability of having a second child began to increase with the 1902-1906 cohort and that of having a third child with the following cohort. Starting with the cohorts born in 1927-1931, the probability of having a first child and then a second was higher than in the cohorts born in the mid-nineteenth century. Conversely, the probabilities of higher parities fell, indicating a shift in fertility behaviour which subsequently became increasingly dependent on the number of children already born, as is characteristically the case under regimes of more controlled fertility (Festy, 1979). Lastly, starting with the women born in the 1950s, who began having children in the early 1980s, behaviours have tended to stabilize: 87% of these women have at least one child, and they had the highest probability of having a second child (78%). By contrast, very few have more than two children: the probabilities of having a third child (a2 = 44%) or more than three (a3 = 33%) has dropped sharply. The stabilization of the parity progression ratios a2 and a3 in the cohorts born after the Second World War may indicate a polarization effect, with some women still having large families. Large families are less common than in the previous cohorts but probably reflect a choice not to limit family size. This may be a consequence of international immigration and the permanent settlement in France of communities with higher fertility than the French average (Masson, 2013).
Parity progression ratios of all women born between 1850 and 1966*
Parity progression ratios of all women born between 1850 and 1966** The mean age at childbirth was 28 years for the oldest cohorts, and 27 years for the more recent ones (INSEE, population estimates and civil records). For ease of interpretation, an average age of 28 was added to the birth cohorts to give an idea of the reproductive period.
Legend: In the 1892-1896 birth cohorts, the probability of having a first child was 73% (a0), that of having a second child 68% (a1) and so forth.
Note: The parity progression ratios for the cohorts born after 1962 are for reference only.
42Although the preceding comments are also valid for ever-married women (Appendix Figure A.6), the parity progression ratios of never-married women follow quite a different pattern (Figure 7). Firstly, the probability of having a (first) child increased from 17% to 60% between the never-married women born in the early 1920s and those born in the late 1960s. Even more importantly, the probability of having a second child increased steadily between those cohorts, to the extent that never-married women born in the 1960s and mothers of one child have almost all had a second child. However, only 35% have a third child and, of that group, only the same percentage have a fourth child. While never-married women are still much less likely to have a first and a second child than ever-married women, their behaviours are nevertheless moving closer to those of ever-married women, in particular for those who already have two children. Never-married women who already have three children are just as likely to have a fourth child as ever-married mothers of three children.
Parity progression ratios of never-married women born between 1850 and 1966*
Parity progression ratios of never-married women born between 1850 and 1966** The mean age at childbirth was 28 years for the oldest cohorts, and 27 years for the more recent ones (INSEE, population estimates and civil records). For ease of interpretation, an average age of 28 was added to the birth cohorts to give an idea of the reproductive period.
43Despite the difficulties stemming from the diversity of available sources and the heterogeneity of results, it is nonetheless possible to assess trends in the lifetime parity distribution of women born between 1850 and 1966. This historical analysis of parities and parity progress ratios refines the conclusions drawn from averages (completed fertility) and shows that parity distributions should always be taken into account when analysing fertility (Anderson, 1998). It provides a clear picture of the considerable decline in large families (four or more children) among the cohorts born between 1850 and 1900, the high prevalence of childlessness and one-child families in the cohorts who had their children in the interwar period, and the sharp increase in families of two children starting with the cohorts born in the 1920s. Consideration of family size is therefore important for fertility analysis because it shows that for the same completed fertility, the parity distribution may vary substantially, revealing different conceptions of family. It has particular value for the historical analysis of the family and of the place of children and women in society. Longitudinal research is also essential for understanding recent fertility trends and putting them into perspective. Indeed, while current fertility levels are low, they are not exceptional on the scale of a century of cohort fertility, and the fertility of the most recent cohorts is characterized less by low fertility than by the high prevalence of two-children families.
44Lastly, this research highlights the importance of distinguishing between the fertility of all women, of ever-married women and of never-married women, particularly for the analysis of childlessness and low fertility (one child). This differentiation confirms the overwhelming impact of the fertility of ever-married women until the cohorts born in the 1920s and the subsequent impact on fertility of new union formation preferences.
45The renewal of interest in fertility in the twentieth century  reflects the continuing quest to understand the end of the demographic transition (Brée et al., 2016a, 2016c; Breschi et al., 2016), the baby boom (Duvoisin et al., 2016; Gauvreau and Laplante, 2016) and the baby bust (Nomès and Van Bavel, 2016). Retrospective analysis of family sizes by educational level or religion, or in relation to marriage patterns, offers new elements for understanding changes in fertility. Likewise, analysis at the sub-national level makes it possible to explore the heterogeneity of behaviours in more detail, as recommended by Van Bavel and Reher (2013). The contribution of qualitative approaches (Bonvalet et al., 2011; Rebreyend, 2003; Rusterholz, 2015, 2017; Sohn, 1996) is also increasing. Clearly, a combination of these methods is needed in order to understand fertility in the twentieth century and up to the present day.
Parity distribution (%) and completed fertility
Parity distribution (%) and completed fertility
Distribution of the number of women aged 45 at most at the time of the survey, by five-year cohort
Distribution of the number of women aged 45 at most at the time of the survey, by five-year cohort
Parity distribution of never-married women (1912-1916 to 1962-1966 cohorts)
Parity distribution of never-married women (1912-1916 to 1962-1966 cohorts)
Parity distribution of all women
Parity distribution of all womenNote: The results of the 1931 and 1946 censuses are provided for reference only because they were calculated on the assumption that all never-married women were childless (which explains the much bigger differences for 0 and 1 child than for the other parities).
Completed fertility of the cohorts according to different estimates
Completed fertility of the cohorts according to different estimatesInterpretation: “If never-married women all had at least one child”: 90% with one child, 5% with two children, 3% with three children and 2% with four or more children.
Simulation of completed fertility and parities for ever-married women without the increase in large families between the 1897-1901 and 1942-1946 cohorts*
Simulation of completed fertility and parities for ever-married women without the increase in large families between the 1897-1901 and 1942-1946 cohorts** The mean age at childbirth was 28 years for the oldest cohorts, and 27 years for the more recent ones (INSEE, population estimates and civil records). For ease of interpretation, an average age of 28 was added to the birth cohorts to give an idea of the reproductive period.
Note: This estimate produces a linear decrease in the percentage of large families from 13.25% for the 1937-1941 cohorts to 11.89% for the 1947-1951 cohorts, which completely wipes out the increase observed between those two groups of cohorts. The remaining share is distributed across the other parities according to the actually observed distribution.
Parity progression ratios for ever-married women born between 1850 and 1966*
Parity progression ratios for ever-married women born between 1850 and 1966** The mean age at childbirth was 28 years for the oldest cohorts, and 27 years for the more recent ones (INSEE, population estimates and civil records). For ease of interpretation, an average age of 28 was added to the birth cohorts to give an idea of the reproductive period.
Note: The parity progression ratios for the cohorts born after 1962 are for reference only (source: Famille et logements, 2011).
46Enquête Famille (Family survey), 1954, INSEE, ADISP-CMH. Enquête Famille (Family survey), 1975, INSEE, ADISP-CMH. Enquête Famille (Family survey), 1982, INSEE, ADISP-CMH. Enquête Famille (Family survey), 1990, INSEE, ADISP-CMH.
47Étude de l’histoire familiale (Family history survey), 1999, INSEE, ADISP-CMH.
48Enquête Famille et logements (Family and housing survey), 2011, INSEE, ADISP-CMH.
49French population census, 1931.
50French population census, 1946, INSEE, Statistique générale de la France.
Centre for Demographic Research, Université catholique de Louvain, Belgium.
Correspondence: Sandra Brée, Centre de recherche en démographie, Université catholique de Louvain, 1 place Montesquieu, B1348 Louvain-la-Neuve, Belgium, email: firstname.lastname@example.org
Patrick Festy’s vast study (1979) of fertility in Western countries between 1870 and 1970 was based on a diagonal reading of five-year age-specific fertility rates collected by Pierre Depoid (1941) to estimate the fertility of French cohorts, a method that Fabienne Daguet (2002) also applied to the data from INED’s report on the demographic situation of France in 1974 (INED, 1976).
Reproductive life is generally measured between the ages of 15 and 50. It was limited here to age 45 (also commonly used) so that another five-year cohort could be included in the analysis with almost no impact on the results (because very few children are born to women aged 45-50).
The question “How many children do you have?” is asked in the censuses starting in 1886 but only provides information about surviving children, not all live-born children, which can create a considerable bias (for Paris, see Brée, 2011, 2017). For a history of family statistics, see the first chapter on family statistics from the 1946 French population census (INSEE, 1946, pp. VIII-XII).
The sample size per cohort for the different surveys is presented in Appendix Figure A.1.
The 1962 survey, considered to be biased, is no longer distributed by INSEE.
It was Louis Henry who recommended remaining “at the national level” [with no more than] a broad classification into major occupational or […] socioeconomic groups” (Henry, 1953, p. 490).
The Family survey was compulsory until 1990. In 1999, given the largely retrospective nature of the survey, its designers dropped this requirement (Héran, 2005). This change probably explains the significant difference between the non-response rates in the 1982 and 1990 surveys and the more recent 1999 and 2011 surveys.
For the 1999 and 2011 surveys, the weightings proposed by the survey designers were used (poidsm5 for the 1999 survey and poids_ind for the 2011 survey).
The average of 1902-1906 and 1912-1916 was used for the cohort of 1907-1911.
Illegitimate births were slightly more frequent in the cohorts born between 1860 and 1890 (roughly 9% according to Nizard and Maksud, 1977) than in the later cohorts (7% for the cohorts born around 1900; 8% for those born in the 1920s; source: INSEE, civil registration).
We chose to use the levels of permanent celibacy from the population censuses (INSEE) rather than from the surveys (which, according to Toulemon, 1996, are underestimated). For the cohorts from 1850 to 1922-1926, the level used is that published in Chasteland and Pressat (1962).
The cohort fertility of all women cannot be estimated from the 1931 and 1946 censuses because only married, widowed or divorced women were surveyed, which explains why this method was chosen.
The method consists in summing the age-specific rates for the same cohort (e.g. for the cohort born in 1930, the fertility rate at age 20 in 1950 is added to the fertility rate at age 21 in 1951, and so forth).
The issue of how stillborn children are counted is irrelevant here because the question is only asked in relation to live-born children.
In both cohorts, one-third of women were mothers of three or more children.
On the links between fertility and crises, see Eggerickx et al. (2016).
Acts of 1920 prohibiting the advertisement and sale of contraceptive methods, and the Act of 1923 that downgraded abortion from a crime to a misdemeanour, with a lighter penalty.
Mother’s Day and family allowance in 1932 (the latter was made universal in 1939).
On the contribution of illegitimate fertility to overall fertility in the nineteenth century, see Brée (2014).
See, for example, issue 2016-2 of Les Annales de démographie historique.