Previous studies of the fertility decline in Europe are often limited to an earlier stage of the marital fertility decline, when the decline tended to be slower and before the large increase in earnings in the 1920s. Starting in 1860 (before the onset of the decline), this study follows marital fertility trends until 1939, when fertility reached lower levels than ever before. Using data from the Historical Sample of the Netherlands (HSN), this study shows that mortality decline, a rise in real income, and unemployment account for the decline in the Netherlands. This finding suggests that marital fertility decline was an adjustment to social and economic change, leaving little room for attitudinal change that is independent of social and economic change.
Carlsson (1966) published a highly influential paper that classified explanations of the fertility decline into two categories: adjustment and innovation. The view of fertility decline as an adjustment states that fertility control is an adjustment to economic and social change. The view of the decline as a process of innovation, on the other hand, states that the adoption of fertility control represents new behavior attributable to attitudinal change and, more specifically, changes in the acceptability of fertility control on moral grounds. Carlsson interpreted the Swedish evidence as supportive of the adjustment thesis. Most participants in Princeton’s European Fertility Project (EFP), however, have found the adjustment thesis to provide a poor fit for what happened during the transition in Europe. They found correlations between marital fertility and the available social and economic measures to be weak. Their research indicated that the innovation thesis may provide a better guide to what happened (Cleland and Wilson 1987; Knodel and van de Walle 1986). However, others view this conclusion as invalid because of serious reservations about the methods used in the EFP (Brown and Guinnane 2007). The explanatory variables used in most EFP studies, for example, do not support meaningful tests of the role of social and economic change in the fertility transition (Brown and Guinnane 2002). Although their aim was to explain change, most EFP studies used statistical models that do not accurately differentiate cross-sectional from time-series changes. Richards (1977), Galloway et al. (1994), Lee et al. (1994), and Brown and Guinnane (2002) have shown how conclusions can change when more appropriate statistical methods are used.
Some scholars went so far as to propose that innovation theory can substitute for adjustment theory (Cleland and Wilson 1987). Today, however, few would support a “pure” innovation model. The more common stance is that the two sets of explanations—adjustment and innovation—are complementary (Lesthaeghe 1983; Montgomery and Casterline 1996). Cleland (2001b) referred to theories that combine the two explanations as “blended models.” In blended versions of innovation explanations, the fundamental cause of marital fertility decline is reduced demand for children. After the structural conditions are right, marital fertility decline is inevitable, but its timing may be lagged. The diffusion of innovative ideas subsequently conditions the speed and mechanisms of change.
According to Inglehart (1985) and Lesthaeghe and Surkyn (1988:17–23), the mechanism for value change is the demographic dynamic of cohort succession: when an older cohort dies out, it is replaced by a new cohort that holds different values reflecting its unique historical experience. A central sociopsychological postulate is that cohorts tend to be marked for life by the ideas prevalent in their youth (Ryder 1965:851). If value systems are likely to have crystallized by early adulthood and if fertility decline is due to attitudinal change, then there should be cohort influences in fertility decline (Raftery et al. 1995:161).
Previous studies in Europe are often limited to the first stage of the marital fertility decline, when the decline tended to be slower, stopping around 1910 or 1920 (e.g., Brown and Guinnane 2002; Friedlander et al. 1991; Galloway et al. 1994; Reher and Iriso-Napal 1989; van Bavel 2004; Woods 1987). We argue that the limited time window used is likely to have influenced their results, more specifically regarding the role of living standards (see also Boyer and Williamson 1989:109; Crafts 1984). Like Engelen and Hillebrand (1986), this study follows fertility for a longer period. Using data from the Historical Sample of the Netherlands (HSN), Data Set Life Courses Release 2010.01, we show that mortality decline, increased earnings, and unemployment in the 1930s about equally account for the decline before 1940, leaving little room for innovation that is independent of social and economic change. To the extent that attitudinal change is a cohort effect, our empirical results seem to indicate that marital fertility decline in the Netherlands was not the delayed effect of a change in attitudes that occurred earlier in life in response to social and economic change in society at that time.
Theories of Fertility Decline
There has been a proliferation of theoretical statements on the fertility decline (Burch 1996). Hence, in any review, an organizing principle is essential (e.g., Alter 1992; Bulatao 2001; Freedman 1979; Friedlander et al. 1999; Hirschman 1994; Kirk 1996). As an organizing principle, we use the intellectual roots of theories in the paradigms or major perspectives of modern sociological theory. Although the intellectual roots of adjustment theory—functionalism and theories of rational choice—are well known, those of innovation theory—conflict theory—may be less obvious. We argue that the intellectual roots of fertility theories matter. Popper (1963:33–39) asserted that theories should be rejected when they are inconsistent with the data. In practice, however, theories may also be rejected because the scientific paradigm in which they are embedded has been abandoned (Kuhn 1962). Thus, even without the EFP (and political pressure), we argue that demographic transition theory would have come under increasing pressure simply because of its intellectual roots in functionalism.1
Functionalism: Demographic Transition Theory
In the same year, Frank Notestein (1945) and Kingsley Davis (1945) published what are generally considered to be classic formulations of demographic transition theory (Szreter 1993:661). The intellectual roots of demographic transition theory are well known. Davis (1959) explicitly identified sociology with functionalism. In analyzing demographic systems, functionalists emphasize the interdependence of the system’s parts, the existence of a state of equilibrium, and how all parts of the system reorganize to bring things back to normal (Wallace and Wolf 1995:18). Accordingly, in demographic transition theory, mortality decline is a key factor triggering fertility decline (Cleland 2001a; Davis 1945, 1963; Mason 1997:446; Reher and Sanz-Gimeno 2007:704).
Conflict Theory: Innovation
Conflict theory is the major alternative to functionalism as an approach to study the structure of societies. Where functionalists see interdependence and unity in society, conflict theorists see groups fighting for power. Conflict theorists emphasize coercion rather than consensus as the cause of social order. Conflict theory is increasingly popular in modern sociology at the expense of functionalism (McFalls et al. 1999; Wallace and Wolf 1995:76).
The intellectual roots of innovation theory become apparent in a search for the origin of innovative ideas. Imitation is central to innovation theory (Knodel and van de Walle 1986:412). However, without a class structure, there would not be any group to imitate and, hence, no diffusion of innovative ideas. For Livi-Bacci (1986), attitudinal change is initiated by elites and gradually spreads downward to the lower classes. Following Tarde (1890) and Sorokin (1947), Lesthaeghe and Surkyn (1988:17) identified the educated elite as the source of cultural innovations that eventually diffuse to lower classes (Treas 2002:270). In Caldwell’s Theory of Intergenerational Wealth Flows (1980:228), schools impose middle-class values on the working class.
Lesthaeghe links ideational change with the process of secularization (Lesthaeghe and Surkyn 1988; Lesthaeghe and Wilson 1986). Empirical studies seem to indicate that religious commitment is largely determined in adolescence, remaining more or less constant for the rest of the life cycle. Thus, the decline in religiosity appears to be mostly a cohort effect (Argue et al. 1999; Chaves 1989; te Grotenhuis and Scheepers 2001; Tilley 2003; Voas and Crockett 2005).
Cultural factors, such as language, may create barriers to the diffusion of attitudinal change. Hence, innovation theory predicts that the diffusion of attitudinal change will be conditioned by regional factors that cannot be completely explained away by socioeconomic characteristics at the individual level (Anderson 1986).
Theories of Rational Choice
Theories of rational choice are guided by the assumption that people base their actions on what they perceive to be the most effective means to their goals. Instead of imitation, fertility decline is largely seen as a rational process based on individual calculations that lower fertility makes sense for economic, social, and psychological reasons (Bulatao 2001:11). Theories of this type are most closely associated with economics (Wallace and Wolf 1995:279). There are two major economic approaches to the study of fertility change. The first is the “new home economics” associated with Gary S. Becker (1981) and T. W. Schultz (1974). In this approach, reduced demand for children as determined by income, prices, and tastes is the basic driving force in the transition. The second approach is the supply and demand framework of Richard Easterlin (1975). Supply factors are the elements that constrain natural fertility and child survival (Hirschman 1994:214–215; Kirk 1996:369–370). Like demographic transition theory, some theories of rational choice have room for culture and attitudes. These determine the “cost of fertility regulation” in the supply and demand framework of Easterlin. Thus, cultural factors may influence the timing of the onset and pace of decline by increasing the cost of fertility regulation.
In sociology, the best-known examples of rational choice approaches are those associated with “exchange theory” (Wallace and Wolf 1995:280). Social exchange theories of fertility emphasize the ability of children to create access to critical material resources through ties of kinship and other personal relationships made possible by children (Schoen et al. 1997). Parents may place a high value on children as security assets when there are no real alternatives to children as sources of security (Cain 1983).
Whereas Easterlin’s supply and demand framework is not specific about the role of income, according to his relative cohort size model, decisions about childbearing are made on the basis of income potential relative to an expected standard of living. The expected standard of living is determined through childhood economic socialization. The standard of living experienced in childhood and adolescence produces aspirations that are then compared with income potential in young adulthood, which is a function of relative cohort size. A decline in income potential relative to the expected standard of living is predicted to lower fertility (Pampel 1993:496). Macunovich (2000) tested the Easterlin hypothesis and demonstrated how changes in relative cohort size appear to have triggered fertility declines.
Although mortality decline is a key factor in demographic transition theory, rational choice theories are more likely to stress the role of rising real wages. There are at least three ways in which rising real wages may induce fertility decline. First, higher wages for women may raise the cost of children relatively more (substitution effect) than they raise household income (income effect), thus leading to a decline in the demand for children (Galor and Weil 1996). Second, if rising real wages change the importance of education, parents may invest more in education (“child quality”) than in child quantity (Becker et al. 1990; Lesthaeghe and Neels 2002:345). Third, as real income increases, the demand for children may decline because parents become less dependent on children as secondary workers. According to the “economic independence hypothesis,” adolescents and young adults living at home were able to renegotiate their contribution to the parental household as alternatives to living with parents became more affordable (Schellekens 1993).
Data and Variables
The data used in the analysis come from the Historical Sample of the Population of the Netherlands (HSN), Data Set Life Courses Release 2010.01. The HSN is a national database with information on the complete life history of a 0.5% random sample (78,105 birth records) of men and women, or research persons (RPs), born in the Netherlands between 1812 (the introduction of the vital registration system) and 1922. In all Dutch provinces, a random sample of births was drawn, which was stratified by period of birth and level of urbanization of the municipality (Mandemakers 2000, 2001). We make use here of RPs born only after 1850. The selected individuals could be followed in the municipal population registers from birth to death and from one municipality to the next. The population registers record the complete fertility history of RPs who were born after 1850 (Mandemakers 2000, 2001).2 The absence of a fertility history does not necessarily mean that the RP did not marry; some of the RPs may have been lost to follow-up as a result of poorly recorded migration. About one-half of the RPs are male. Our analysis is based on the fertility history of female RPs and the wives of male RPs. Because the sample fraction is small, the chance of a couple consisting of two RPs is very small. We included only first marriages of RPs in our analysis. Thus, our fertility estimates may be biased upward. The HSN did not record the fertility history of RPs born before 1850. In order to add cohorts that were born before 1850, we also included the fertility history of the mothers of RPs, regardless of whether the RP ever married. Although the fertility histories of the male and female RPs constitute a random sample, this cannot be said of the mothers of RPs. Clearly, mothers with a large number of offspring are more likely to have been selected. We partially control for this selection effect by adding birth cohort dummy variables to the analysis.3 We used information only on the marriage of the mother who produced the RP because the fertility history of other marriages of the mother is not known if it occurred in a household that did not include the RP at some stage. Mothers of RPs whose date of marriage is unknown were not included in the multivariate analysis.
Continuous population registers in the sense of bound documents with nonremovable pages were prescribed in the Netherlands by Royal Decree of December 22, 1849. The registers had to record the population residing within the municipality. The returns from the census of 1849 were copied into the population register, and afterward, all changes occurring in the population during the following decades were recorded in the register. For each individual, date and place of birth, relation to the head of the household, sex, marital status, occupation, and religion were recorded.
For the town of Tilburg, Janssens (1994) checked the quality of the population registers in recording demographic events by checking the registration of births in the population register against the birth registers. She found that at most, 0.2% of births were not entered in the population register, all such cases being children dying soon after birth. The omission of less than 1% of births should not affect the quality of our results.
Figure 1 presents Coale’s marital fertility index (Ig)—the ratio of the number of births occurring to married women to the number that would occur if married women were subject to maximum fertility—in census years as estimated by Engelen and Hillebrand (1986) for the whole country and the 11 provinces. Figure 1 shows that in 1930, marital fertility was lowest in the province of North Holland (which includes Amsterdam) and highest in the Catholic provinces of North Brabant and Limburg. Innovation theory predicts that the diffusion of attitudinal change will be conditioned by regional factors (Anderson 1986). Hence, we added 10 dummy variables to indicate province at marriage for RPs and province of birth of the RP for his/her parents, the province of South Holland being the reference category.
Innovation theory predicts that attitudinal change is initiated by elites and gradually spreads downward to the lower classes. Hence, we included a social classification scheme based on occupation. The occupation attributed to each household is the first occupation of the husband listed in the population registers. We used a new international social classification scheme for occupational data—the social power scheme (SOCPO)—developed by van de Putte and Miles (2005). The fundamental organizational principle of their scheme is the potential to influence one’s destiny through control of resources. We split their fourth category into farmers and others. Hence, the categories used in the analyses are
1. Unskilled workers
2. Semi-skilled workers and the small-scale self-employed
3. Manual skilled workers, and supervisors of semi- and unskilled workers
4a. Manual super-skilled and nonmanual skilled workers, supervisors of skilled workers, and local businessmen
5. Nonmanual super-skilled workers, supra-local businessmen, executives and those with general policy tasks, and nobility
6. Without occupation, and unknown
Urban populations are thought to be forerunners (Livi-Bacci 1986). We used the percentage employed in agriculture in 1930 in the municipality at marriage for RPs and municipality of birth of the RP for his/her parents as a measure of urbanization. The percentage agrarian in the 1930 census was obtained from the Historical Ecological Database (Department of Social Geography, University of Amsterdam; see Knippenberg and de Vos 2008:106).
In demographic transition theory, mortality decline is a key factor triggering fertility decline. Childhood mortality may influence fertility in three ways (Doepke 2005; Palloni and Rafalimanana 1999). First, the death of an infant to a breastfeeding mother will shorten the postpartum infecundable period. To control for this physiological effect, we included a dummy variable indicating whether an infant death occurred in year t – 1. Second, the death of a child may also increase fertility through the replacement of the child that has died by the birth of an additional child. To capture this effect, we also included a dummy variable indicating whether a child aged 1–4 had died in year t – 1. The strength of the replacement effect may be a function of family size and the number of children a family has lost, the likelihood of replacement declining as family size increases and with each additional child death. Hence, we included a variable counting the number of child (0–4) deaths before year t.
Third, infant and early childhood mortality at the regional level may also influence reproductive behavior. In societies with high mortality, there may be an “insurance” or a hoarding effect, with couples attempting to bear as many children as possible in anticipation that not all will survive. Hence, we included measures of infant and early childhood mortality at the national level measured in year t – 1. We used a time series of the infant mortality rate (IMR) published by the Central Bureau of Statistics (2001). We computed a time series of early childhood mortality (1 m4) from data published by Tabeau et al. (1994). Figure 2 shows that early childhood mortality had already started to decline in the 1870s, whereas IMR started to decline in the 1880s.
The “new home economics” approach predicts that an increase in income will trigger fertility decline. Measures of income are rarely included in the analysis of the European fertility decline (e.g., Galloway et al. 1994:146). We used the gross domestic product (GDP) per capita in thousands of 1913 guilders as a proxy for average income (Central Bureau of Statistics 2001). GDP per capita sharply increased between 1918 and 1928. The rise in labor market activity of married women postdates the onset of marital fertility decline by several decades (Clark 2005:514; Goldin 1990; van Poppel et al. 2009). Hence, the growth in GDP per capita until 1940 probably reflects growth in male wages to a considerable extent. GDP per capita measures only one aspect of living standards. Hence, we also used a time series of the percentage unemployed estimated by den Bakker and de Gijt (1994) and Smits et al. (2000). Both macro-level variables are measured in year t – 1 to model births in year t. Unfortunately, we were unable to obtain regional measures of GDP and unemployment. Some regional variation, however, may be picked up by the regional dummy variables.
Religion may influence the cost of fertility regulation. During the transition, Roman Catholics in the Netherlands consistently had a higher level of marital fertility than other religious denominations (van Poppel 1985). Schellekens and van Poppel (2006:34) showed that childhood mortality and the occupation of the father do not attenuate the effect of being Catholic at all, suggesting that the effect of being Catholic was at least partly ideological. Jews are thought to have led the rest of the population in achieving lower levels of reproduction (Livi-Bacci 1986). Hence, the religion of the head of the household was used to compute two dummy variables indicating whether the household was Roman Catholic or Jewish.
According to Inglehart (1985) and Lesthaeghe and Surkyn (1988:17–23), the mechanism for attitudinal change is the demographic dynamic of cohort succession. Hence, we added cohort dummy variables to the analysis. Of course, the absence of cohort effects does not imply that there was no attitudinal change because innovation may also be a period effect. To the extent that economic and demographic variables explain the decline, however, there will be limited room for attitudinal change that is independent of social and economic change.
We controlled for the following demographic variables in the analysis: age of the woman, marital duration, number of children ever born in year t – 1, and a dummy variable indicating first year of marriage. After contraceptive behavior is controlled for, the influence of the age of the woman should mostly reflect physiological factors, such as sterility and intrauterine mortality. In most cases, coital frequency declines with marital duration (Brewis and Meyer 2005; van Bavel 2003). The inclusion of crude parity or the number of children ever born is essential in order to control for fecundability and secondary sterility (van Bavel 2003). The probability of a birth in the first calendar year of marriage is relatively low when there is no exposure to intercourse in the months preceding marriage. We added a dummy variable indicating the first year of marriage to take this into account.
Discrete-time event history models were used to assess the effects of the independent variables on the probability of giving birth. Discrete-time models are estimated using logistic regressions of woman-year data. This approach allows considerable flexibility in handling time-varying covariates, such as age and marital duration, and censored observations (Allison 2010:236–240). In particular, it is a convenient way to model age, period, and cohort effects (Raftery et al. 1995).
Event history models were introduced by Cox (1972) as a synthesis of regression models and life tables, initially to analyze nonrepeatable events such as death. By contrast, the nature of fertility is that births are repeatable events. Because we are not interested in any specific birth interval, birth intervals were pooled. Pooling birth intervals is also statistically more efficient. Unfortunately, pooling introduces a problem: dependence among birth intervals. Hence, random effects were added to control for unobserved heterogeneity between women. The GLIMMIX procedure in SAS 9.2 was used to estimate regression coefficients (Allison 2010:260–280).
The models assume that the hazard for a birth is constant within annual intervals but is otherwise unconstrained. A woman may have twins or two births in a single year. However, we counted only 2,313 of 64,040 births (3.6%) that occurred in the same calendar year. Hence, we modeled ties as single events.
The dependent variable is the log odds of a woman giving birth in a specific calendar year. The unit of analysis is the woman-year; that is, each woman contributes as many units to the analysis as the number for which she is observed. Woman-years below age 20 were omitted from the analysis. After left-truncation at January 1, 1860, 17,444 women contributed 288,159 years to the analysis. Female RPs and the wives of male RPs contributed 205,799 years, and the mothers of RPs contributed 82,360 years.
Our review of the literature identified two theories of cohort replacement: Easterlin’s relative cohort-size model and Lesthaeghe’s secularization hypothesis. Age-period-cohort models are particularly useful to detect the distinct impacts of age, period, and cohort on some outcome of interest. Disentangling the distinct effects of age, period, and cohort, however, involves a methodological problem because the three are perfectly correlated. There are at least three conventional strategies for identification and estimation: (1) constraining two or more of the age, period, or cohort coefficients to be equal; (2) transforming at least one of the age, period, or cohort variables so that its relationship is nonlinear; and (3) assuming that the cohort or period effects are proportional to certain measured variables (Yang and Land 2006).
Mason et al. (1973) pointed out that the identification problem can be solved by imposing equality constraints on categories of age, period, and/or cohort. One criticism of this method is that estimates of model effect coefficients are sensitive to the arbitrary choice of the identifying constraint. Mason et al. (1973) designed their strategy for aggregate data, such as mortality rates. The identification problem for aggregate population data, however, does not necessarily transfer directly to individual-level data. One can use different temporal groupings for the age, period, and cohort variables—for example, single years of age, and five-year intervals for time periods and birth cohorts—to break the linear dependency (Yang 2008:210).
A second strategy is to transform at least one of the age, period, or cohort variables so that its relationship to others is nonlinear (Mason et al. 1973; Raftery et al. 1995; Yang 2008). Although the use of a nonlinear function solves the problem of the arbitrary choice of the identifying constraint, this approach still is not very informative about the mechanisms by which period-related changes and cohort-related processes act on the dependent variable of interest.
Period dummy variables are a poor proxy for some set of contemporaneous influences, and cohort dummy variables are an equally poor proxy for influences in the past. When these influences can themselves be directly measured, there is no reason to probe for period or cohort effects (Hobcraft et al. 1982). Hence, a third strategy is to constrain the effects of period and/or cohort to be proportional to some other substantive variable. Heckman and Robb (1985) termed this the “proxy variable” approach because period and cohort are represented by some other variable. Adopting the third strategy, we use four proxies for the period effect: real GDP per capita, the unemployment rate, IMR, and the rate of early childhood mortality (1 m4). The proxy variable approach, however, also has its drawbacks. Although replacing an accounting dimension with measured variables solves an identification problem, it makes room for specification errors (Smith et al. 1982). Replacing the period dummy variables by proxies may lessen the rigorousness of the control for the period effects on cohort differences (O’Brien 2000:125). In other words, we may overestimate cohort effects.
If we do not want to generalize our results to cohorts that were not included in the analysis, then conventional statistical methodology guidelines suggest that it might be more appropriate to model them with a fixed-effects specification. Hence, we model cohort influences as fixed effects. Yang and Land (2006, 2008), however, argued that when sample sizes within each cohort are unbalanced, mixed (fixed- and random-effects) models use the available information in the data more efficiently than fixed-effects models. They warn that the standard errors of estimated coefficients of conventional fixed-effects regression models may be underestimated, leading to inflated t ratios and actual α levels that are larger than nominal levels of significance. Thus, we are likely to see cohort influences where there are none.
Stopping or Spacing?
Summary measures of marital fertility, such as Ig, and the total marital fertility rate (TMFR), do not indicate whether marital fertility declined because of stopping behavior or the spacing of births. Coale and Trussell (1974) described a two-parameter model of marital fertility that differentiates between stopping and spacing behavior. The first parameter, M (uppercase M), can be considered as an index of the level of fertility in the population in relation to the level of fertility in natural fertility populations. M is influenced by those factors that affect the level of fertility, such as birth spacing, but that are not motivated by parity dependent efforts to stop childbearing. Birth spacing may be achieved either through deliberate interventions (such as contraception) or through less-obviously volitional practices (such as breastfeeding) (Wilson et al. 1988). The other parameter, m (lowercase m), is interpreted as a measure of the extent of stopping behavior, although it may be an insensitive indicator of stopping behavior because small increases in m may come about as a result of significant increases in the proportion stopping (Okun 1994:221). Figure 3 shows that M and m both increased until the last decade of the nineteenth century. Afterward, m continued to increase, whereas M declined. The decline in M suggests that spacing played an important role in the marital fertility decline in the Netherlands (compare with van Bavel and Kok 2004). We can only speculate about the reasons for the rising M values. Perhaps these are the result of a decline in breastfeeding.
Adjustment or Innovation?
Innovation theory predicts that attitudinal change is initiated by elites and gradually spreads downward to the lower classes. According to Woods (1987:290), the decline in marital fertility in England affected all occupational groups at the same time. Haines (1989:321), however, concluded that the decline in England was more rapid and possibly occurred earlier in the middle and upper classes. Figure 4 presents estimates of the total marital fertility rate at age 20 by 10-year period and social class in the Netherlands. Using period measures instead of the cohort measures of marital fertility used by Woods (1987) and Haines (1989), the HSN data indicate that fertility declined earlier and more rapidly among the upper class (SOCPO level 5), while the decline among the middle class (SOCPO level 4a) preceded the decline among farmers (SOCPO level 4b) and working classes (SOCPO levels 1, 2, and 3). Thus, a diffusionist view is consistent with Fig. 4. It is still possible, however, to argue against innovation and in favor of adjustment to the differential impact of changing social and economic circumstances.
Table 1 presents descriptive statistics of the variables used in the multivariate analysis. Table 2 presents two logistic regression models of the decline in marital fertility. Coefficients are presented as odds ratios or exponents of the raw logistic coefficients. The odds ratios are multiplicative effects on the odds of giving birth in any one-year interval. A coefficient of 1.00 represents no effect, a coefficient greater than 1.00 represents a positive effect, and a coefficient less than 1.00 represents a negative effect on the odds.
The first model includes individual-level variables, household characteristics, and cohort dummy variables, but no proxies for period effects. To control for period effects, the second model adds four proxies for period effects. Except for the coefficients of age and cohort dummy variables, coefficients in both models are very similar.
In demographic transition theory, mortality decline is a key factor triggering fertility decline. The death of an infant in the previous year has the expected positive effect. The extent to which this is a physiological as well as a replacement effect is uncertain. If there is a replacement effect, the death of older children is also expected to have a positive effect. Our results indicate that the death of a child aged 1–4 in the previous year also has a significant effect. Controlling for the number of children ever born, however, the total number of deceased children aged 0–4 has a negative effect. Thus, to the extent that a child death had a replacement effect, this effect seems to have declined as a function of the number of previous deaths.
Infant and early childhood mortality at the regional level may also influence reproductive behavior. After we control for infant mortality at the household level, IMR also has a positive effect. Hence, the hypothesis that infant mortality affects marital fertility through an insurance effect is consistent with the data. Although significant, the effect of the early childhood mortality rate is not in the expected direction. We suspect multicollinearity and will consider herein the combined effect of the two early childhood mortality measures.
The new home economics approach predicts that an increase in income will induce a fertility decline. As predicted, GDP per capita has a negative effect on marital fertility. The unemployment rate also has a negative effect on marital fertility.
Innovation theory predicts that attitudinal change is initiated by elites and gradually spreads downward to the lower classes. Hence, we included a social classification scheme based on occupation (SOCPO). The multivariate analysis shows that the differences shown in Fig. 4 between unskilled laborers and other socioeconomic groups (except for farmers) are significant. When the middle class replaces unskilled laborers as the omitted category, most differences remain significant (result not shown).
During the transition, Roman Catholics in the Netherlands consistently had a higher level of marital fertility than other religious denominations. Catholics also have above-average marital fertility in the HSN data. Our results seem to indicate that Jews were not forerunners in the marital fertility transition (see also Schellekens and van Poppel 2006). Urban populations are also thought to be forerunners (Livi-Bacci 1986). Our results indicate that marital fertility was higher in more agricultural communities. Innovation theory predicts that the diffusion of attitudinal change will be conditioned by regional factors that cannot be completely explained away by socioeconomic characteristics at the individual level (Anderson 1986). After we control for individual and household characteristics and the percentage employed in agriculture, significant differences remain between provinces. Marital fertility in the Catholic provinces of North Brabant and Limburg, however, is not significantly higher. The remaining differences between provinces may reflect differences in living standards. Unfortunately, however, no real GDP per capita series are available at the provincial level.
If cohorts tend to be marked for life by the ideas prevalent in their youth and if fertility decline were due to attitudinal change, cohort influences should be evident. At least one rational choice theory (Easterlin’s relative cohort size theory) also predicts that fertility decline is a cohort effect. The cohort dummy variables in the first model have a negative effect. Cohort differences in the first model, however, may reflect period effects. To control for period effects, the second model adds four proxies for period effects. After the inclusion of the period proxies, there are no more negative cohort effects. Thus, we reject the hypothesis that marital fertility decline is a cohort effect, even though we may have overestimated cohort effects and the α level of the cohort coefficients may be too small, as explained in the methodological section. To the extent that we did not overestimate cohort effects, there are positive cohort influences during the decline. Perhaps these cohort influences are the result of a decline in breastfeeding.
Although the fertility decline is not a cohort effect, we cannot reject the innovation hypothesis because attitudinal change might be a period effect. Herein, we will discuss the extent to which attitudinal change was a period effect that is independent of social and economic change.
Figure 5 presents observed and predicted values of TMFR by five-year period. A comparison of the observed (thick line) with the predicted (thin line) series indicates the extent to which the second model is able to predict the timing of the onset and pace of the marital fertility decline. The second model predicts a slightly earlier decline than the one observed. Over time, the fit of the model improves.
An easy way to interpret the regression model results is to simulate what marital fertility levels would have been under constant social and economic circumstances. Hence, we computed predicted series of TMFR that factor out change in several independent variables from the predicted series of TMFR, using the coefficients in the second model. We call the result a “counterfactual-predicted series” (see Fig. 5). For each year t, TMFRs were simulated by computing for each woman the predicted probability that she will give birth in year t. These probabilities were summarized to obtain the predicted number of births for each age group in year t. To obtain predicted age-specific marital fertility rates, we divided the number of births in each group by the observed number of women in the appropriate age group. Figure 5 presents average values of TMFR—the sum of the age-specific fertility rates—for five-year periods.
The first counterfactual-predicted series assumes no decline in either infant or early childhood mortality rates after 1880–1884 (thick dotted line). The second counterfactual-predicted series adds the assumption of no rise in GDP per capita after 1880–1884 (thick dash-dot line), and the third counterfactual-predicted series adds no rise in unemployment after 1880–1884 (thick dashed line).
Although the death of a child has a significant effect on marital fertility, the decline in infant and early childhood mortality at the household level provides only a marginal explanation for the decline in marital fertility (result not shown). Hence, there are no assumptions concerning infant and early childhood mortality at the household level in the first counterfactual-predicted series. When infant and early childhood mortality rates are held constant at the level of 1880–1885, TMFR would have declined much less before 1940 (thick dotted line). When in addition to the infant and early childhood mortality rates, GDP per capita is also held constant at the level of 1880–1885, the second model would not predict any decline in TMFR before 1930 (thick dash-dot line). Thus, mortality decline and rising real wages seem to account for the decline before 1930, leaving little room for any omitted variable to explain part of the decline before that date. Finally, if in addition to childhood mortality and GDP per capita, unemployment levels were also to remain constant at the level of 1880–1885, the second model would not predict a decline in TMFR before 1940 (thick dashed line). Combined, mortality decline, increased earnings, and unemployment during the economic crisis of the 1930s account for the decline before 1940.
Conclusion and Discussion
Two major findings emerge from our analysis. First, social and economic change accounts for the marital fertility decline that occurred before 1940. Second, ideational change that is independent of social and economic change was not a major determinant of the decline. Thus, using micro-level data, our analysis replicates the findings of macro-level studies, such as Richards (1977), Engelen and Hillebrand (1986), Galloway et al. (1994), and Brown and Guinnane (2002), showing that the EFP downplayed the role of social and economic development. Like them, we reject pure versions of the innovation model.
Blended models often assume a delay between social and economic change and marital fertility decline. When the structural conditions are right, marital fertility decline is inevitable, but its timing may be lagged because of the pace of attitudinal change. Our results indicate, however, that social and economic change in the Netherlands accounts for the decline before 1940, leaving little room for attitudinal change that was independent of social and economic change. In Brazil, ideational change also was not independent of economic and social change (Potter et al. 2002).
Lesthaeghe and Surkyn (1988) proposed cohort replacement as a determinant of the pace of attitudinal change. At least one rational choice theory—Easterlin’s relative cohort size theory—also predicts that fertility decline is a cohort effect. Based on individual-level data from the Netherlands, the empirical results of this study suggest that marital fertility decline was not a cohort effect. One of the few previous studies to present an age-period-cohort model of marital fertility decline reported a similar finding for Iran (Raftery et al. 1995). Of course, the absence of negative cohort influences does not necessarily imply that marital fertility decline was simply an adjustment to social and economic change and not attributable to attitudinal change. Attitudinal change may have been a period effect. Our results indicate, however, that in the Netherlands, social and economic change accounts for the decline before 1940, leaving limited room for attitudinal change that is independent of social and economic change.
Contemporary demographers implied secularization as a causal factor in the low fertility of the 1930s (van Bavel 2010:7–8). However, te Grotenhuis and Scheepers (2001) showed that the rate of secularization in the Netherlands before 1955–1960 was very low. Our results seem to indicate that unemployment was a major contributing factor to relatively low fertility in the 1930s.
Mortality decline is a key factor in demographic transition theory. Childhood mortality may influence fertility in three ways: through the physiological, the replacement, or the insurance effect. We found indirect evidence of a replacement effect of child deaths. Although early childhood deaths had a positive effect on fertility, childhood mortality decline as measured at the household level provides only a marginal explanation for the decline in marital fertility. Thus, mortality decline does not seem to explain marital fertility decline through physiological or replacement effects. In Latin America, Palloni and Rafalimanana (1999) also found little evidence for these pathways between early childhood mortality decline and marital fertility decline. In the Netherlands, changes in infant and early childhood mortality as measured at the national level, on the other hand, explain roughly one-half of the decline until 1930. Hence, provided that mortality at the national level is not a proxy for unobserved variables affecting both mortality and fertility, mortality decline mostly contributed to marital fertility decline in the Netherlands through a hoarding effect.
Although it is almost standard procedure to include mortality in models of fertility decline, measures of real wages and income are much less often included in studies of the European fertility decline. Galloway et al. (1994) included income but subsequently found that income did not make an important contribution to the decline in Prussia. Their analysis, however, stopped in 1910. In the Netherlands, increased earnings seem to have played a more important role. Possibly, this finding occurred because of covering a longer period. In a macro-level time-series analysis, Eckstein et al. (1999) followed the Swedish fertility decline for a much longer period and reported that wage increases and reductions in child mortality accounted for a large part of the fertility decline. Although GDP per capita may be a proxy for unobserved variables, the hypothesis that increased earnings played a major role in the decline is the most parsimonious, or least complex, explanation for the correlation between the increase in GDP per capita and marital fertility decline.
Although we found socioeconomic class differences in marital fertility decline, innovation theory does not seem to be consistent with our empirical results because there is no evidence of attitudinal change that is independent of social and economic change. Demographic transition and rational choice theory, on the other hand, are both consistent with our empirical finding that mortality decline is strongly correlated with fertility decline. Although demographic transition and rational choice theory are both consistent with our empirical finding that the increase in income is strongly correlated with fertility decline, demographic transition theory is much less informative about the possible mechanisms through which rising real wages might induce fertility decline.
The rise in labor market activity of married women postdates the onset of marital fertility decline by several decades (Clark 2005:514; Goldin 1990; van Poppel et al. 2009). Thus, the decline is unlikely to be the result of the substitution effect of women’s wages. When discussing the correlation between rising real wages and fertility decline, the investment in education—child quality—is perhaps the mechanism most often mentioned. However, just how many households made this type of calculation is difficult to determine. “The pressures to substitute child quality for child quantity are presumably strong only where pressures for equity among children within a given family are strong, that is, where investments in all children are expected to be more or less equal. Where gender stratification is strong, parents may feel little pressure to educate their daughters and may therefore pursue a strategy of relatively high fertility combined with educating sons only” (Mason 2001:167). In Europe, parents may have felt little pressure to educate their daughters, given that most female careers were terminated by marriage (Vincent 2000:75). Although the economic independence hypothesis may be consistent with the qualitative evidence, such as presented by de Regt (2004), a statistically rigorous test using the HSN data is not possible.
We are grateful to two anonymous referees for providing us with very valuable comments.
More information on the HSN is available online (http://www.iisg.nl/~hsn).
We did not include a variable indicating whether observations were based on the fertility history of the RPs or that of their mothers because such a dummy variable would be correlated with the period proxies and cohort dummy variables.