Abstract
Childlessness in the United States nearly doubled between 1980 and 2000. Other dramatic changes in the U.S. population also took place over this period—notably, women’s average educational attainment increased, and the proportion marrying declined—but the impact of these changes on childlessness has not been formally examined. In this article, I use data from the Current Population Survey Fertility Supplements (1995, 1998, 2004, 2008) and logistic regression and regression-based decomposition techniques to assess the contribution of changes in educational attainment, marriage behavior, and racial/ethnic composition on population levels of childlessness in the United States. Results show that increases in the proportion of women unmarried by age 40 contributed most to the increase in childlessness in the late twentieth century, although these increases were offset somewhat by increased childbearing among unmarried women. The rising proportion of women with a college degree also explained a substantial amount of the increase in childless women.
Introduction
Among recent cohorts of women in the United States, nearly 1 in 5 women1 have no children by the end of their childbearing years (Dye 2010). Rates of childlessness have nearly doubled since 1980, matching the historic highs reached by women born in the first decade of the twentieth century, who were in their peak childbearing years during the Great Depression (Casper and Bianchi 2002). The growing number of adults without biological children shifts the structure of family relationships, and childlessness may have consequences for the health and well-being of older adults. Increasing childlessness also implies changes for women’s identity. Although social pressure to have children has declined since the baby boom era, most American women continue to place high importance on having children (McQuillan et al. 2008). More broadly, patterns of childbearing in one cohort alter the normative context for the next cohort and may lead to a self-reinforcing pattern of continuing increases in the proportion of women who do not give birth (Goldstein et al. 2003).
Research on the individual-level determinants of childlessness has found that more-educated women are more likely to be childless (e.g., Abma and Martinez 2006; Bloom and Trussell 1984; Lundquist et al. 2009). This relationship is not a proximate one: it seems unlikely that education has direct effects on female reproductive functioning. Instead, the relationship appears to result from social or ideational incompatibility between motherhood and other pursuits, such as work or education, and is likely to be the product of reciprocal negative relationships between education and fertility (Rindfuss et al. 1980). The association between education and childlessness might explain increases in childlessness since the baby boom because women’s educational attainment has increased steadily since the mid-twentieth century (Buchmann et al. 2008; Jacobs 1996; Snyder and Dillow 2010). However, the contribution of changes in women’s education to changes in childlessness has not been explicitly examined.
In this article, I use data from the Fertility Supplements of four Current Population Surveys (1995, 1998, 2004, 2008) to assess the contribution of compositional changes to trends in childlessness in the late twentieth and early twenty-first centuries. I use logistic regression and decomposition techniques based on logistic regression (Fairlie 1999) to compare the relative impact of changes in educational attainment, marriage behavior, and racial/ethnic composition and in the effects of these factors on childlessness. Results show that increases in women’s educational attainment, especially in college degree receipt, contributed to increases in childlessness, but this contribution was relatively small. Declines in marriage produced a greater proportion of the increase in childlessness, even accounting for the offsetting effects of increasing nonmarital fertility.
Childlessness in the United States
Figure 1 shows trends in the level of permanent childlessness for cohorts of U.S. women born between 1931 and 1968. (The data, sample, and methods for constructing this figure are detailed in the Data and Methods section of this article.) As seen in the figure, childlessness increased steadily for women born between 1931 and 1960; these women were having children from the 1950s through approximately the 1990s and represent the mothers of both the baby boom and the baby bust. Levels of childlessness leveled off and even declined slightly among women born in the 1960s, those who completed childbearing in the first decade of the twenty-first century. These trends in childlessness have been described by others (e.g., Abma and Martinez 2006; Rowland 2007), but the reasons behind them remain poorly understood.
In this article, I examine broad patterns of change in the primary demographic correlates of childlessness in order to determine the main factors producing trends in childlessness across successive birth cohorts of American women. Childlessness is inherently a cohort phenomenon: having no children is the result of decisions and behaviors that cumulate over the reproductive life course. Members of a birth cohort make decisions about childbearing and family formation in response to shared social and economic conditions (Easterlin 1980; Elder 1974/1999; Ryder 1965). Dramatic changes in the social environment faced by cohorts over the late twentieth century, as well as changes in the composition of cohorts of U.S. women, have the potential to explain long-term trends in childlessness in the United States. The goal of this article is not to understand the individual-level determinants of childlessness, but to identify which of the large-scale demographic changes in the second half of the twentieth century in the United States are most closely associated with changing levels of childlessness. I focus on trends in women’s educational attainment and marriage rates as possible factors driving trends in population levels of childlessness. The analyses also account for the changing racial/ethnic composition of the U.S. population.
Education has consistently been found to be positively associated with childlessness, although associations with the timing of childbearing are stronger than associations with number of children (Abma and Martinez 2006; Bloom and Pebley 1982; Bloom and Trussell 1984; Heaton et al. 1999; Lundquist et al. 2009). Because women’s college enrollment has increased throughout the twentieth century (Snyder and Dillow 2010), educational trends would contribute to increasing childlessness absent other changes. Furthermore, some evidence suggests that the effect of education on fertility increased between the baby boom and baby bust childbearing cohorts (Bloom and Trussell 1984). The impact of a college education on women’s earnings has also increased as women’s educational attainment has increased (DiPrete and Buchmann 2006). If the association between education and childlessness has become stronger over time, this change would exacerbate the impact of rising education on childlessness.
Changes in marriage behavior are also likely to have contributed to trends in childlessness. Childlessness is more common among never-married women than among women who have married (Abma and Martinez 2006; Heaton et al. 1999; Lundquist et al. 2009). After reaching a peak during the baby boom, the proportion of women married has fallen steadily (Goldstein and Kenney 2001); these declines in marriage would be expected to lead to lower birth rates and more childlessness, all other things being equal. However, the association between marriage and childbearing has weakened as nonmarital birth rates have increased since the middle of the twentieth century (Ellwood and Jencks 2004; Smock and Greenland 2010; Ventura and Bachrach 2001; Wu et al. 2001). These changes may counteract the impact of reduced marriage rates on childlessness.
Changes in women’s educational attainment, marriage rates, and fertility patterns in the twentieth century are not independent, and marriage and educational trends likely have interacting impacts on levels of childlessness. Both declines in marriage rates and increases in nonmarital fertility have been concentrated among women with lower educational attainment (Ellwood and Jencks 2004). That is, although changes in the proportion of women who ever marry have been greatest for the least-educated women, the impact of those changes on childlessness is likely to be mitigated by increased nonmarital fertility in this group. Among women with a college degree, the proportion of women marrying has changed less than among less-educated women, but marriage has increasingly been postponed to older ages. Because rates of nonmarital fertility for the most-educated women continue to be low, postponed marriage also implies postponed childbearing. Greater delay in first birth may also result in more childlessness, if women who wait longer to have a first birth are more likely to have difficulty conceiving because of age-related infertility or if delaying childbearing gives them more time to develop life plans that do not involve children.2
Growing immigration from Mexico and Latin America and relatively high birth rates among Hispanic-origin populations in the United States produced a steady growth in the Hispanic population since 1970 (Durand et al. 2006; Martin and Midgley 2003). Overall, Hispanic women have higher birth rates on average than non-Hispanic white women and are less likely to be childless, although there are substantial differences in fertility patterns between foreign-born and U.S.-born Hispanic women and across Hispanic-origin groups (Dye 2010; Landale and Oropesa 2007). Thus, the increasing Hispanic population in the United States is likely to reduce population levels of childlessness. It is not clear how other changes in the racial/ethnic composition of the U.S. population might affect trends in childlessness. In the early twentieth century, rates of childlessness were higher among non-Hispanic white women than among African American women (Bloom and Trussell 1984; Evans 1986). Starting in the 1970s, however, levels converged, and current patterns of childlessness are now similar for white and African American women (Boyd 1989; Dye 2010; Lundquist et al. 2009).
Data and Methods
Data
In order to track changes in childlessness in the late twentieth and early twenty-first centuries, I use data from the June Fertility Supplements of four Current Population Surveys (1995, 1998, 2004, 2008). The Current Population Surveys are nationally representative monthly household surveys designed primarily to track employment and economic indicators. In June of some years, a supplemental module administered to adult women asks questions about past childbearing. These questions were asked of all women up to age 65 in 1995, but only women of childbearing age (45 and younger) in later surveys.3
Although the structure of the fertility supplement varies across survey years, respondents were asked in all surveys how many children they had given birth to. The dependent variable in this analysis is a dichotomous measure of childlessness, with the value 1 if a woman had no children and 0 if a woman had ever given birth. Analysis is limited to women aged 40 or older at the time of the survey. Birth rates in the United States to women aged 40 and older are low, and first-birth rates are lower—9.2 births per thousand women and 2.0 births per thousand women, respectively, in 2005, and lower still in previous years (Martin et al. 2007). The majority of women aged 40 and older can thus be assumed to have completed childbearing, and childlessness among these women can be analyzed as permanent childlessness.4
The analytic sample for this analysis is therefore women aged 40–44 in the 1998, 2004, and 2008 surveys and women aged 40–64 in the 1995 survey. This analysis focuses on childbearing behavior in the United States, so women who immigrated to the United States as adults are not included in the analysis. However, women who were born abroad but came to the United States before age 15—that is, who spent all of their reproductive years in the United States—are included in the analysis. The final analytic sample included 35,150 women (1995: 19,956; 1998: 5,021; 2004: 5,489; 2008: 4,684). These women were grouped into seven birth cohorts: 1931–1939, 1940–1944, 1945–1949, 1950–1954, 1955–1959, 1960–1964, and 1965–1968. (Because of small sample sizes in the early cohorts, women born before 1940 were combined into a single group.) These cohorts span several distinct eras of childbearing patterns: the baby boom (1931–1939 and 1940–1944 birth cohorts, who were having children in the 1950s and 1960s), the baby bust of the 1970s (1950–1954, 1955–1959), and the more recent cohorts of women who were having children during the 1980s and 1990s, when birth rates were relatively stable.
The goal of this article is to assess the impact of large-scale population changes on childlessness in the United States. Independent variables are therefore limited to the primary stratifying factors in U.S. fertility: educational attainment, marital status, and race/ethnicity (Morgan 1996; Smock and Greenland 2010). Educational attainment is measured in the CPS with a question about the respondent’s highest level of school completed or degree received. This analysis divides respondents into four categories: no high school diploma, high school diploma or GED, college attendance but no bachelor’s degree (including associate’s degree and vocational degrees), and bachelor’s degree or higher. Using the CPS marriage history variables, women are classified as ever-married or never-married. Four racial/ethnic groups are identified in the analysis: non-Hispanic whites, non-Hispanic blacks, Hispanic women, and women of other racial groups. Although the CPS sample sizes are large, the numbers of members of some racial/ethnic groups (Asian American, Native American) in the samples are small. Women reporting a race/ethnicity other than white, African American, or Hispanic are therefore classified as “other race” for this analysis. After the 2000 census, which for the first time allowed respondents to select more than one race, the CPS also permitted the reporting of multiple races. The number of women in the analytic sample reporting more than one race is relatively small (N = 197 in 2004 and 2008 combined). These women were assigned to a single racial/ethnic group using an algorithm developed by Liebler and Halpern-Manners (2008) to identify the racial group that would have been most likely to be picked by the respondent under the single-race guidelines used before 2000. Only one woman could not be assigned to a single racial/ethnic category using the Liebler-Halpern-Manners algorithm. This woman, who reported more than three races, was assigned to the “other race” category.
Methods
I approach questions about trends in childlessness from two perspectives. The first uses individual-level analyses to describe cohort changes in childlessness. These analyses use logistic regression to predict childlessness among women who have completed childbearing in the seven aforementioned birth cohorts. The goal of these analyses is not to show sociodemographic variation in childlessness, which has been analyzed in previous studies. Instead, the outcome of interest is change in the coefficients for the cohort variables. Change in these coefficients is used to assess the degree to which change over time is accounted for by change in the explanatory variables. Results are presented as a series of nested models. The first model includes only dummy variables for birth cohort; education, race/ethnicity, and marital status are added in subsequent models. In exploratory models, I also tested interactions between explanatory variables. Including these interactions did not change coefficients for the cohort variables, so I do not include them in the final models. In order to observe change over time in the associations between independent variables and childlessness, I then run the full model separately for the first birth cohort, the last birth cohort, and a cohort in the middle of the period of study. To more fully understand change in the predictors of childlessness, I include interactions between education and marital status in these cohort-specific models. Because of small sample sizes for some racial/ethnic groups, I do not include interactions with race in these models.
The second approach examines childlessness at the population level and assesses the contribution of compositional changes to trends in childlessness more formally, using regression-based decomposition (Fairlie 1999). Decomposition techniques break down change over time or differences across groups into components attributable to population composition or group-specific rates. These components can be interpreted as the amount of change that would have taken place had only the specified characteristic changed—for instance, only the educational composition of the population, or only the relationship between education and childlessness.
Regression-based decomposition for continuous variables was developed by Blinder (1973) and Oaxaca (1973). These techniques measure the contribution of changes in regression coefficients and changes in the mean of independent variables to change over time or differences between groups. Essentially, changes in the mean of the outcome variable can be expressed as the sum of changes in the means of the independent variables (compositional changes) multiplied by the coefficients of those variables at time 0 and changes in the coefficients (changes in effects) multiplied by the means of the independent variables at time 1. Alternatively, compositional changes can be evaluated using coefficients at time 1 and changes in effects evaluated using variables at time 0. These two ways of decomposing change provide different results; both expressions are valid, but they “weight” composition or rate effects differently depending on the reference point.
The Oaxaca-Blinder decomposition is based on the linear relationship between independent and dependent variables; that is, it is based on the fact that the mean of the predicted dependent variable is equal to the regression equation evaluated at the mean of each independent variable. For a dichotomous variable using logistic regression, the average predicted probability is not equal to the regression equation evaluated at the mean of the independent variables, so the technique cannot be directly applied to dichotomous outcomes. Instead, I use an extension of the Oaxaca-Blinder technique, developed by Fairlie (1999), that has been used to study a range of social phenomena (e.g., Stearns et al. 2007; Van Hook et al. 2004). Rather than relying on the means of the independent variables, Fairlie’s extension uses a random assignment process to match cases at time 0 with cases at time 1. Values of the independent variables from the matched cases are then used to evaluate the effects of compositional changes (see Fairlie (1999) or Van Hook et al. (2004) for details of the decomposition procedure). For each time period, a random subset of cases is chosen (here, a subset of 5,000 cases). The logistic regression equation is estimated for these subsets, and each individual is assigned the predicted probability from the regression. The subsets are then sorted by predicted probability and matched in order, such that the women with the highest predicted probability from each subsample are matched, and values from the matched cases are used to evaluate compositional change. Of course, the results of the decomposition are then sensitive to the random subset. The procedure is therefore repeated multiple times (here, 1,000 times), and the average across all repetitions is reported in the final results. The standard errors of these estimates were quite small (less than one one-hundredth of a percentage point), indicating that the estimates were closely clustered.
Results
Descriptive Results
Table 1 shows changes in the composition of women of reproductive age and in the association between sociodemographic characteristics and childlessness between the birth cohorts of 1931–1939 and 1965–1968. The population trends shown in the top panel have been studied extensively elsewhere (e.g., Buchmann et al. 2008; Durand et al. 2006; Ellwood and Jencks 2004; Goldstein and Kenney 2001; Martin and Midgley 2003), so I describe them only briefly. There are three key observations here. First, the proportion of the population that is white declined from 82 % in the earliest cohort to 69 % in the cohort born 1965–1968. The size of all other racial/ethnic groups increased. Second, the percentage of women in the cohort who had been married declined from 96 % to 86 % over the period under study. Further tabulations (not shown) indicate that this decline was sharpest among women with a high school diploma or less. Third, women’s average educational attainment increased dramatically. The proportion of women without a high school diploma fell by more than half, from 23 % to 9 %, while the proportion who had obtained a bachelor’s degree more than doubled (from 15 % to 32 %). The proportion of women with some college education also increased slightly.
The second panel presents overall and group-specific changes over the seven cohorts in the proportion of women aged 40–44 who are childless. Population levels of childlessness increased steadily across the pre-1940 to 1955–1959 cohorts, rising from just over 10 % to just under 20 % of women. After the 1955–1959 cohort (who finished childbearing in the late 1990s), however, this increase leveled off and showed signs of reversing by the 1965–1968 cohort (who were 40–44 in 2008). Trends for African American and non-Hispanic white women followed the population pattern, although the plateau appears to have been reached later for African American women than for white women. Hispanic women had levels of childlessness that were equivalent or slightly higher than non-Hispanic women in the early cohorts and lower in the later cohorts; levels of childlessness for Hispanic women did not increase substantially until the 1955–1959 birth cohort (late 1990s). Never-married women were more likely to be childless than ever-married women at every time period, but rates of childlessness among never-married women declined consistently over the seven cohorts. Trends in childlessness were similar for all education groups, and the most-educated women (women with a bachelor’s degree or higher) had the highest rates of childlessness across the period. There is little difference in rates of childlessness between the two lowest education groups (women with no high school diploma and high school graduates). In the early period, levels of childlessness among women with some college education were similar to those among less-educated women. However, childlessness increased more rapidly among women with some college education than among women who did not attend college.
Regression Results
Table 2 shows results from a set of five logistic regression models predicting childlessness among women aged 40–44. Model 1 includes only dummy variables for the seven cohorts; Model 2 adds race/ethnicity; Model 3 adds marital status; Model 4 adds education; and Model 5 includes all independent variables. Coefficients for sociodemographic characteristics are consistent with previous research and with the bivariate statistics; they are not discussed here. The primary results of interest are the coefficients for the dummy variables for cohort and their changes across models as additional controls are added.
Model 1, the unconditional model, shows a time trend consistent with that shown in Fig. 1: there is almost no difference in levels of childlessness between the pre-1940 cohorts and the 1940–1944 cohort, but subsequent cohorts show steadily increasing levels of childlessness until the 1965–1968 cohort, who experienced levels of childlessness that were slightly lower than those of the cohort before them. The cohort coefficients are essentially unchanged in Model 2, which controls for race/ethnicity; the changing racial/ethnic composition of the U.S. population does not explain trends in childlessness after 1970. In Model 3, which controls for marital status, the coefficients for the time trend are reduced substantially, especially in the later cohorts, demonstrating that increasing proportions of never-married women do explain some of the increase in childlessness in the late twentieth and early twenty-first century. Controlling for educational attainment (Model 4) also reduces the cohort coefficients, though not as much as controlling for marital status does. Finally, Model 5 includes all explanatory variables. In this model, cohort differences are reduced below the level of any of the previous models. The further reduction in coefficients suggests that the roles of education and marital status trends in explaining time trends in childlessness are to some extent independent. As noted previously, interactions between explanatory variables were tested in exploratory analyses but did not further explain change over time and so were not included in the final models.
These models constrain the effects of the explanatory variables to be equal over time, but Table 1 suggests that these associations vary. Table 3 contains results from the full model estimated separately by cohort and depicts variation over time. For simplicity, I show results for only three of the seven cohorts, the earliest and latest cohort and a cohort in the midpoint of the period that approximates the point at which trends began to change. For these models, the goal is to explain associations with childlessness within a given cohort rather than changes across cohorts. I therefore include interactions between education and marital status. The associations between sociodemographic characteristics and childlessness are not constant but shift over this period. (I tested for statistical significance of coefficient changes across cohorts by running pooled, fully interactive models; these results are not shown but are discussed where relevant.) For example, racial/ethnic differences are largest in the middle cohorts. For the 1931–1939 birth cohort, there are no statistically significant differences across racial/ethnic groups. By the 1955–1959 cohort (women having children in the 1970s and early 1980s), both Hispanic women (b = –0.39, OR = exp(–0.39) = .68) and African American women (b = –0.90, OR = .41) are substantially less likely to be childless than non-Hispanic white women. These changes are statistically significant (p < .01) for black women but not for Hispanic women, probably because of relatively small sample sizes resulting in large standard errors in the first cohort. Racial/ethnic differences decline in later cohorts, and these declines are statistically different from zero for both African American and Hispanic women; in the 1965–1968 cohort, differences between Hispanic and non-Hispanic white women are no longer statistically significant.
Coefficients for education and marital status and interactions between the two suggest complex patterns of association between these characteristics and childlessness. Although coefficients change slightly across cohorts, these changes are statistically significant only for the main effect for marital status (p < .01, 1931–1939 cohort to 1955–1959 cohort; p < .001, 1931–1939 cohort to 1965–1968 cohort). The main effect for marital status—the association between nonmarriage and childlessness for women with a high school diploma—is large, positive, and statistically significant in all three cohorts, but it declines steadily across the period. That is, over this period, when nonmarital birth rates were increasing, differences in childlessness between ever-married and never-married women in the middle education category declined. (Note that this is attributable to increases in childlessness among married women as well as decreases among unmarried women (Table 1); the de-linking of marriage and childbearing works both ways.) The main effect coefficients for the education variables represent differences in childlessness among ever-married women with different levels of educational attainment. Ever-married women with at least a bachelor’s degree were significantly more likely to be childless than ever-married women with a high school diploma in all cohorts. Differences among ever-married women with a high school diploma, some college, and less than a high school diploma are small in magnitude (and declining over time) and generally not statistically significant. Education–marital status interactions are negative for women with less than a high school diploma, although this interaction is statistically significant only in the earliest cohort, and positive for women with a bachelor’s degree or more. (Interactions for women with some college are positive but small in magnitude and not statistically different from zero.) That is, the difference in levels of childlessness between ever-married and never-married women is smaller for the least-educated women and larger for the most-educated women, consistent with education gradients in nonmarital fertility rates. These interactions also mean that education differences in childlessness are larger among never-married than among ever-married women.
Decomposition Results
Results from these logistic regressions show that compositional changes account for some of the increase in childlessness in the late twentieth century, but also that the effects of sociodemographic characteristics have changed over the period under study. Decomposition analysis can calculate the amount of change attributable to compositional changes and coefficient changes for each variable. Table 4 shows estimates of the relative contribution of these factors from a decomposition analysis based on the regression models presented in Table 3. For the decomposition analysis, change over the seven cohorts was broken into two sections: the 1931–1939 cohort to the 1955–1959 cohort, a period over which childlessness increased substantially; and the 1955–1959 cohort to the 1965–1968 cohort, which saw a slight decline in childlessness. The first panel of Table 4 decomposes change over the first period; the second panel decomposes change over the second period.
The right-most column (column 5) in Table 4 shows the proportion of change associated with each category. A negative proportion means a decline in childlessness, and a positive proportion means an increase in childlessness. Increases and decreases together add up to 1 (in the case of net increase) or –1 (net decrease). In both periods, the largest proportion of change is attributable to the intercept. This type of change—essentially, change in the reference category for all independent variables—can be thought of as secular change experienced across the population. That is, most of the increase in childlessness comes from general changes and not behavior specific to a particular segment of the population. Still, demographic changes explain a substantial proportion of change in the early period: 25 % (1 – .75 = .25) of the change can be attributed either to compositional change or to change in the coefficients of the measures included in the analysis.
For the first period, changes related to race/ethnicity would have produced an overall decline in childlessness absent any other changes. These changes are largely rate effects attributable to increasing racial/ethnic disparities in childlessness (Tables 1 and 3) rather than effects resulting from the changing racial/ethnic composition of the population. The total impact of marital status on trends in childlessness includes both the main effect and the education interactions.5 The main effect is substantially larger than the interaction components, indicating that most of the impact of marital status was common across all education groups. This component accounts for about 28 % of the overall increase. Looking at the first four columns, most of this increase stems from increases in the proportion of women who never married (Table 4, columns 1 and 2). Reductions in rates of childlessness among never-married women would have produced a slight decline in childlessness if those had been the only changes (Table 4, columns 3 and 4), but the decline was not enough to balance the contribution of compositional changes.
The contribution of education to trends in childlessness is also split across main effect and interaction terms. As with marital status, the main effect terms are larger than the interaction terms, although the marital status–bachelor’s degree interaction is relatively larger than other education interactions. That is, changes associated with education were, for the most part, shared across ever-married and never-married women. For women with a bachelor’s degree or more, a relatively larger proportion of change—about one-third (.07 / (.16 + .07))—was specific to never-married women. Between the 1931–1939 birth cohort and the 1955–1959 birth cohort, changes in educational attainment and fertility behavior among women with a high school diploma, women without a high school diploma, and women who had attended some college would have produced a slight decline in childlessness, net of other changes. In contrast, increases in the proportion of women with a college degree and in levels of childlessness among these women (Table 1) account for about 23 % (.16 + .07 = .23) of the total increase in childlessness. The relative contribution of composition and rate changes to changes associated with college degree receipt depends on which cohort is taken as the base. The impact of compositional change is greater when the coefficient from the 1955–1959 cohort (column 2) is used, since the educational gradient in childlessness is greater for these women (see Tables 1 and 3), and almost all the change specific to never-married women is due to compositional change.
Between the 1955–1959 and 1965–1968 birth cohorts, there was little overall change in the proportion of women who were childless, and the portions of change associated with individual characteristics had opposing effects. Convergence in rates of childlessness across racial/ethnic groups would have led to an increase in childlessness if no other factors had changed, although this convergence was to some extent offset by increases in the proportion of the population that was not white. Although the positive association between never marrying and childlessness was reduced in magnitude over this period, the proportion of women who never married increased; these changes would also have led to increased population levels of childlessness between the 1955–1959 and 1965–1968 cohorts if no other components had changed. This pattern was especially marked among women with some college attendance but no bachelor’s degree. These shifts were counterbalanced by a reduction in the positive association between college attendance and childlessness, as well as a secular reduction in childlessness in the population as a whole.
Summary
Increases in women’s educational attainment contributed to increases in childlessness over the second half of the twentieth century (from the 1931–1939 birth cohort to the 1955–1959 cohort). Results from a logistic regression show that controlling for education accounts for about 15 % of the coefficients for time trends. However, reductions in marriage rates appear to have contributed more to changes in childlessness; including controls for marriage in a logistic regression model reduces coefficients by close to 30 %. Results from regression-based decomposition are consistent with the conclusion that changes in marriage rates played a more substantial role in trends in childlessness than changes in women’s educational attainment. Interactions show that education and marriage trends both contributed independently to changes in childlessness, but also that the impact of changes in education are unevenly distributed across ever-married and never-married women and, similarly, that the impact of marriage changes is not equal across education groups.
The rising number of women with a bachelor’s degree, along with the increasing association of college education with childlessness, explained about 23 % of the overall increase in childlessness between the 1930–1939 and 1955–1959 cohorts in the decomposition model. (Note that this is an absolute increase of about 19 women per thousand in the level of childlessness.) This change was largely reversed in the period between the 1955–1959 and 1965–1968 birth cohorts; over this period, decomposition results show that changes in the receipt of bachelor’s degrees would have produced a decrease of about 7 childless women per thousand, largely because of reduced levels of childlessness among college-educated women over this period. Although the proportion of women receiving a bachelor’s degree continued to increase, the proportion of these women who were childless declined.
Across both of these periods, continued declines in the proportion of women marrying would have produced substantial increases in childlessness, net of other changes. In fact, changes in marriage rates mediate time trends in regression results and explain the largest proportion of the increase in childlessness over the second half of the twentieth century in a decomposition analysis. Increases in nonmarital fertility rates offset this increase somewhat and are likely to offset it more in the future, but they have not been able to absorb the rapid increase in the proportion of women who are unmarried by age 40.
Discussion and Conclusions
Overall, increases in childlessness after the baby boom are primarily explained by secular change, secondarily by reductions in the marriage rate, and third by increases in women’s education. Since the baby bust, the positive association between college degree receipt and childlessness has declined, as have rates of childlessness among women with a bachelor’s degree. Although nonmarital fertility has increased, marriage continues to be strongly associated with childlessness, and trends in marriage are an important predictor of trends in childlessness.
The changing racial/ethnic composition of the U.S. population has had a relatively small impact on trends in childlessness. This analysis was limited to women who were born in the United States or arrived in the United States as children; thus, it does not account for the short-term impact of immigration on fertility. Over time, continued declines in the proportion of the non-Hispanic white population may contribute to decreased childlessness. However, if levels of childlessness continue to converge across racial/ethnic groups, as they have over the past few decades, the relevance of the racial/ethnic composition of the population to fertility levels may recede.
This article, like other large-scale quantitative studies of childlessness (e.g., Abma and Martinez 2006; Lundquist et al. 2009), focuses on childlessness among women and excludes childless men from analysis. Some recent research suggests that well-educated men are more likely to reach their desired number of children than well-educated women (Morgan and Rackin 2003; Quesnel-Vallée and Morgan 2003). Rates of childlessness for college-educated men may, therefore, be lower than for college-educated women. On the other hand, educational homogamy in the United States is strong and getting stronger (Schwartz and Mare 2005). College-educated men are therefore likely to marry college-educated women, and trends in childlessness may be similar for men and women. Historical data on men’s fertility are unlikely to appear, but expanding current data collection on men’s childbearing may make it possible to study gender differences in childlessness. Differences between men and women in levels and trends in childlessness would point to gender-specific factors causing incompatibility between, for example, professional employment and parenthood, whereas similarity would suggest that more general economic conditions explain changes in childlessness. However, continued challenges in collecting data on men’s fertility (Joyner et al. 2012) and the relatively greater importance of social parenthood (e.g., stepparenting) for men than for women might make such comparisons difficult.
This article also neglects the biological causes of involuntary childlessness. It is not clear how trends in sterility may have contributed to change over time in rates of childlessness. Some scholars have argued that high levels of childlessness in the early twentieth century were due to poor health conditions, ranging from malnutrition to untreated sexually transmitted diseases (e.g., Cutright and Shorter 1979; McFalls and McFalls 1984). Contemporary rates of sterility from these sources are low, but growing postponement of childbearing means that more women may be subject to age-related difficulty in conceiving and carrying a child to term (American Society for Reproductive Medicine 2008; Anderson et al. 2000; McFalls 1990; Menken et al. 1986). However, analysis of trend data suggests that infertility among married couples declined between 1982 and 2002 (Stephen and Chandra 2006). In addition, the development of assisted reproductive technologies may reduce the impact of impaired fecundity on population levels of childlessness. To the extent that biomedical factors related to childlessness are shaped by social structures, changes in these factors will be captured by measures of changing sociodemographic characteristics of the U.S. population. A direct analysis of trends in involuntary childlessness is not possible with existing data, although information from the National Surveys of Family Growth could be used to approximate these trends starting in the 1970s.
Levels of childlessness fluctuated during the twentieth century. During each moment of increase, social scientists and social observers expressed concern about the social consequences of reduced childbearing and framed their concerns as worries about women’s choices. In the late nineteenth and early twentieth centuries, scholars interpreted childlessness as the result of selfishness (Popenoe 1936), or of biological incompatibility between higher education for women and childbearing (Clarke 1873). These concerns were assuaged by rebounding birth rates during the baby boom. More recently, similar worries have been expressed as fears about the emotional consequences of forgone childbearing among highly educated professional women (e.g., Grigoriadis 2002; Hewlett 2002). The concept of “opting out” of paid employment (Belkin 2003) expresses the flip side of these concerns: if highly educated women opt out of employment because motherhood and career are incompatible, then highly educated women who continue paid work must be “opting out” of motherhood. But just as empirical research has found little evidence of opting out of the workforce (Percheski 2008; Stone 2007), this article finds little evidence that trends in childlessness are solely or even largely driven by the behavior of highly educated women.
In contrast, this article shows a substantial contribution of trends in marriage to changes in childlessness. Furthermore, because analyses do not account for changes in the timing of marriage, results are likely to underestimate the total impact of changes in the marriage regime to trends in childlessness. Over the past 40 years, much of the public conversation about the relationship between marriage and fertility has focused on the “social problem” of rising nonmarital fertility. Consistent with the dramatic increase in childbearing outside of marriage, rates of childlessness among never-married women have declined. Yet childlessness among the never-married continues to dwarf childlessness among ever-married women. Despite the rising prevalence and acceptability of childbearing outside of marriage, the powerful symbolic link between marriage and childbearing persists.
Acknowledgments
An earlier version of this article was presented at the 2010 annual meeting of the Population Association of America. I am grateful to Sam Hyun Yoo for superb research assistance, to Jenny Trinitapoli and the Social Dynamics Writing Group for helpful comments, to Jennifer Glick for brilliant creative contributions, and to Jennifer Van Hook for sharing SAS programs. Any remaining imperfections are, of course, my own.
Notes
As the number of childless women grows, the number of childless men has probably increased as well. (An increase in the number of childless women but no increase in the number of childless men is theoretically possible, if some women have children with multiple men and other women have no children.) Fatherhood is important in men’s lives (e.g., Eggebeen and Knoester 2001; Marsiglio and Pleck 2005; Nock 1998; Townsend 2002), and the consequences of possible increased childlessness for men may be significant. However, reliable data for studying trends in permanent childlessness among men do not exist. Therefore, this article discusses childlessness among women only.
Postponement is, of course, mechanically linked to childlessness: women who have an early birth cannot then be childless, and childless women are those who have avoided having children first at young ages and then at successively older ages. Research also suggests that intentions to be childless are rare at young ages, and most permanently childless women reach terminal childlessness by repeatedly postponing the first birth (Hagewen and Morgan 2005; Hayford 2009).
Some bias may be introduced by the fact that women are observed for a longer period in the 1995 survey than in the later surveys. The proportion of women who had a first birth between ages 45 and 65—that is, the proportion of women for whom a longer period of observation would lead to different conclusions—is 0.43 % of the sample in the 1995 CPS. Analyses treating these women as childless lead to nearly identical results.
The Fertility Supplements of the Current Population Survey do not include questions about adoption, stepchildren, or foster children. As a result, analyses are limited to biological childlessness and do not address the distribution of social parenthood.
In exploratory analyses, I conducted a decomposition analysis based on models without marital status–education interactions. In that analysis, the effects of education and of marital status were almost identical to the net effects (calculated by adding main effects and interaction terms) in the model with interactions. The similarity across models with and without interactions and the relatively small size of the interaction terms suggest that the impact of education and marital status trends on childlessness are mostly independent rather than joint.
