Fertility Following an Unintended First Birth

Abstract

Research on unintended fertility tends to focus on births as isolated events. This article expands previous research by examining the relationship between early unintended childbearing and subsequent fertility dynamics in the United States. Data from the 2002 National Survey of Family Growth show that 27.5% of mothers report an unintended first birth. We use event history methods to show that these women are significantly more likely than women with an intended first birth to have an unintended second birth than to have either no second birth or an intended second birth, net of sociodemographic characteristics. An unintended first birth also increases the risk of having an unintended third birth relative to no birth or an intended birth, independent of the intendedness of the second birth. We conclude that early unintended fertility is a strong signal of high risk for subsequent unintended fertility.

Introduction

Unintended fertility in the United States is high relative to levels in other Western industrialized countries, with about 50% of recent pregnancies and 38% of live births unintended (Barber and Emens 2006; Finer and Henshaw 2006; Henshaw 1998). Because unintended childbearing is negatively associated with well-being among women and children, the U.S. Department of Health and Human Services has declared the reduction of unintended pregnancies—and thus of unintended childbearing—a national health goal (U.S. Department of Health and Human Services 2000).

Unintended fertility is a well-studied topic, and research has generally fallen along two lines. The first type of research describes aggregate-level trends in unintended fertility rates or group-level correlates of unintended fertility (e.g., Finer and Henshaw 2006; Kissin et al. 2008). The second type of research is micro-level research, either focusing on the individual-level predictors of unintended fertility and studying births as independent events (e.g., Hayford and Guzzo 2006; Musick 2002; Speizer et al. 2004) or on the link between an unintended birth and subsequent well-being (e.g., Barber et al. 1999; Crissey 2005). However, recent evidence suggests a growing concentration of unintended fertility in the United States (Wildsmith et al. 2010). More than 40% of women who have one unintended birth have another, and this proportion appears to have increased in recent cohorts. To understand the cumulative process of repeat unintended fertility, we take a life course perspective on unintended childbearing, examining the relationship between unintended births and subsequent childbearing.

The life course perspective centers on the notion that early events in an individual’s life shape subsequent outcomes (Elder 1998). This perspective implies that the conditions in which women begin family formation are likely to influence later family behaviors. Here, we focus on how unintended births at low parities are related to fertility trajectories, using data from the 2002 National Survey of Family Growth to examine parity-specific patterns of intended and unintended childbearing. We expect that compared with women who have an intended first birth, women with an unintended first birth have a high risk of experiencing a subsequent unintended birth. This association is likely driven in part by causal processes: having an unintended birth alters women’s relationship, educational, and career pathways in ways that may increase the risk of later unintended births. The association may also reflect processes of selection into unintended fertility if structural or psychological factors that increase the risk of unintended births persist over the life course. Although we do not explicitly model causal relationships, we use variation by parity in the patterns of fertility after an unintended birth to assess which relationships are likely to be causal, distinguishing between short-term (direct) associations and more distal relationships. We find that having an early unintended birth is a strong predictor of subsequent unintended fertility, even net of sociodemographic controls, and that having a first unintended birth has persistent associations with later intendedness.

Measuring Unintended Fertility

Unintended childbearing has traditionally been divided into two categories: unwanted births and mistimed births. Unwanted births are those for which women reported that right before they became pregnant, they did not want to have any births at any point in the future (a number failure), while mistimed births are those identified as occurring any time earlier than desired (a timing failure). Unintended births are then the sum of all births identified as unwanted or mistimed. A distinction is usually made between unwanted and mistimed births because in theory, they reflect different concerns over the life course and by parity, and unwanted births tend to be more strongly correlated with negative outcomes than mistimed births (Barber et al 1999; Santelli et al. 2003). However, these definitions have long been recognized as problematic for studying unintended fertility in the late twentieth and early twenty-first centuries (e.g., Klerman 2000; Santelli et al. 2003). The categories were originally established in a context where most unintended fertility came at the end of the childbearing years and therefore focused on unwanted births as the most problematic. In the current context, a substantial amount of unintended fertility takes place at younger ages and low parities, as a timing issue. The mistimed category combines teenage births, which are likely to be strongly disruptive, with births to older women that occur only a few months earlier than planned. Furthermore, cognitive interviews suggest that survey respondents do not interpret the unwanted category in the way that survey designers intended (Klerman and Pulley 1999). Surveys show consistently high reported unwanted fertility among young women—on the order of one in five births to women age 15–19 in the period 1997–2001 (Chandra et al. 2005)—even though most young women report wanting to have children someday.

In an attempt to more appropriately classify unintended births in terms of the severity of their impact, recent research has proposed a new categorization system that combines unwanted births with some mistimed births in an effort to capture the most disruptive unintended births while moving beyond the timing/number distinction. Births (and pregnancies) mistimed by two or more years (labeled as “seriously mistimed”) tend to resemble unwanted births (and pregnancies) in terms of outcomes, whereas those that are mistimed by less than two years are closer to intended births (and pregnancies) in terms of outcomes (Abma et al. 2008; Chandra et al. 2005; Lindberg et al. 2008; Pulley et al. 2002). For example, Lindberg et al. (2008) found that the proportion of pregnancies mistimed by less than two years that were carried to term (63%) was closer to the proportion of intended pregnancies carried to term (78%) than the proportion of pregnancies mistimed by more than two years that were carried to term (39%), which was close to the proportion of unwanted pregnancies carried to term (45%). Moreover, the difference in the proportion carried to term between the two mistimed groups was statistically significant, but the differences between the slightly mistimed and the intended pregnancies and between the seriously mistimed and the unwanted pregnancies were not. Similarly, women were less likely to breast-feed if they categorized their pregnancy as seriously mistimed or unwanted than if they reported their pregnancy to be wanted or slightly mistimed, but the likelihood of breast-feeding did not differ between unwanted and seriously mistimed or between slightly mistimed or wanted (Pulley et al. 2002). Following this recent research, we define unintended births as those characterized as unwanted or seriously mistimed, and intended births are those that are wanted or slightly mistimed.

There has been considerable debate about the validity of retrospective reports on birth intendedness. There is a tendency in retrospective accounts to rationalize births and a reluctance to identify a child as unwanted (Musick 2002; Trussell et al. 1999; Williams et al. 1999). Still, the face validity of these measures of unintendedness has generally been shown to be high (Bachrach and Newcomer 1999; Joyce et al. 2000). Measuring unintended pregnancy is more difficult. Many unintended pregnancies end in abortion, and abortion is known to be underreported in U.S. surveys (Jones and Kost 2006). As a result, most research on unintendedness, including this analysis, focuses on births instead of pregnancies. The underreporting of abortion makes it impossible to pinpoint the effects of abortion access or attitudes on unintended fertility, and analyses conflate the factors leading to unintended conception and those associated with the decision to carry an unintended pregnancy to term. For these reasons, we focus on births rather than pregnancies.

Predictors of Unintended Fertility

The high levels of unintended fertility in the United States mean that unintended births take place to women across the spectrum of age, relationship status, and socioeconomic characteristics (Barber and Emens 2006).¹ Still, a large body of research has consistently found that certain factors are associated with higher risks of unintended fertility. On average, births to young women are more likely to be unintended than births to older women (e.g., Logan et al. 2007). Unmarried women report more of their births as unintended than married women, with cohabiting women falling somewhere between (e.g., Finer and Henshaw 2006). Finer and Henshaw (2006) demonstrated that black and Hispanic women have higher unintended birth rates than non-Hispanic white women and that women with family incomes below the poverty level and women without a high school diploma are more likely to have an unintended birth than women with higher incomes and more education.

Analyses of sociodemographic variation in unintended fertility have been largely descriptive, with little multivariate analysis, and the causal processes behind these patterns of unintended fertility are not well understood. Unintended pregnancies may result either from a woman’s failure to plan childbearing or from a failure to carry out plans. These mechanisms are conceptually distinct but difficult to distinguish empirically with survey data; the following discussion addresses both processes. Some sociodemographic differences in unintended fertility are likely due to differences in the acceptability of childbearing in different contexts. Although an increasing proportion of births take place outside marriage, American women continue to report that marriage is their preferred setting for childbearing (Thornton and Young-DeMarco 2001). Thus, married women may be more likely to plan births. Behavioral differences may also contribute; for example, older women and women in more stable relationships are more likely to use highly effective coitus-independent methods of contraception, such as hormonal methods and intrauterine devices (IUDs) (Mosher et al. 2004). Differential access to contraception has been proposed as an explanation for the higher unintended birth rates among low-income women (e.g., Frost et al. 2007). However, qualitative and quantitative studies of women reporting unintended births show that lack of access to contraception is not a primary cause of unintended conception (Edin and Kefalas 2005; Sable et al. 2000). Instead, women attribute their non-use or inconsistent use of contraception to low motivation to avoid pregnancy and decision-making factors around contraceptive use. If women see little cost to unintended fertility—for example, if they have low chances for securing high-earning jobs and thus low opportunity costs to childbearing—they may devote little effort to contracepting effectively.

Individual psychological characteristics such as self-efficacy and risk-taking tendencies have also been proposed as causes of unintended fertility (Brown and Eisenberg 1995). Self-efficacy is an individual’s belief that he or she has the ability to act in order to influence life events and outcomes. The perception of individual control over desired outcomes is related to motivation in general and has been shown to be associated with a variety of health-related behaviors, including contraceptive use among adolescents (Grembowski et al. 1993; Longmore et al. 2003; Schwarzer and Fuchs 1996). Women who are more risk-tolerant may also be more likely to engage in unprotected sex even if they do not want to become pregnant. Measures of risk-taking tendencies have been shown to be positively associated with sexual activity in adolescence, although not with contraceptive use at last intercourse (Kowaleski-Jones and Mott 1998; Raffaelli and Crockett 2003).

Unintended Fertility and Subsequent Fertility

In this article, we hypothesize that an unintended birth early in a woman’s childbearing career is associated with subsequent fertility and intentionality. As Morgan and Rindfuss (1999) argued, because the occurrence, timing, and sequencing of fertility represent a “nonreversible event” that affects other behaviors (such as schooling and employment), early family-formation behaviors are very likely to affect later ones. A large literature supports this idea, with much of it focusing on fertility following a teenage birth (see, e.g., Hofferth 1987; Kalmuss and Namerow 1994; Ribar 1996) or a nonmarital birth (e.g., Driscoll et al. 1999; Guzzo and Furstenberg 2007).

The same underlying characteristics may drive both low-parity births and subsequent fertility behavior. For example, women with low educational attainment are likely to have low economic prospects, which may reduce their motivation to contracept at both low and high parities. Some psychological characteristics may also be long-term risk factors: although risk-taking behavior and self-efficacy can evolve over the life course (Mirowsky and Ross 2007), they can also be persistent aspects of personality. Having an unintended birth might thus be considered a “signal” for characteristics that are difficult to measure in large surveys but result in higher chances of having more births, particularly unintended births. It is also possible that some women are consistently more willing to report a birth as unintended; consistent reporting differences would produce associations between birth intendedness across parity.

Unintended births at the start of women’s childbearing may also change women’s behaviors or characteristics in ways that make later unintended births more likely. The most direct effect of mistimed fertility is to shift childbearing to earlier ages relative to ideal ages. By definition, women with mistimed births have children earlier than planned. These women may reach their desired family size at an earlier age than women with intended births (as they shift all of their childbearing earlier than they would have otherwise) and will therefore spend more time “at risk” of an unwanted birth at the end of the childbearing years. Other effects of unintended fertility may be either positive or negative, and these effects are likely to vary across individuals. An unintended birth may derail women’s educational or employment trajectories. Reducing women’s attachment to school and work may reduce the perceived costs of additional childbearing and thus increase subsequent fertility, both intended and unintended. Alternatively, the disruption caused by an unintended birth may increase women’s motivation to avoid subsequent unintended births: that is, some women may go to great pains to avoid another “mistake.” In either case, the direct effect of an unintended birth is likely to take place in the short term. Unintended births may have longer-term implications as well, but these effects are likely to be mediated by measurable factors such as relationship status, subsequent educational attainment, and the timing of intermediate births.

We predict that women who have one unintended birth will be more likely to have another unintended birth, both because of selection processes (individual characteristics that persist from birth to birth) and because of causal processes (changes in life circumstances brought about by the first unintended birth). We do not attempt to distinguish between these mechanisms analytically. However, we note that stable individual characteristics that persist from the first to the second birth are also likely to be present at the third birth. Selection processes will therefore be present at all parities. Causal processes, in contrast, will dissipate over subsequent births.

Data and Methods

We use the 2002 cycle of the National Survey of Family Growth (NSFG), a nationally representative survey of U.S. women of age 15–44 designed to measure levels and trends in fertility. The NSFG includes detailed birth and relationship histories as well as measures of sociodemographic characteristics and family background. The 2002 cycle interviewed 7,639 women. We limit our analyses to those who, if they had a birth, had valid information on the intendedness of that birth (excluding 125 women) and further restrict our analyses to Hispanic, non-Hispanic white, and non-Hispanic black women because of small sample sizes and the diversity of women in the “other” racial category (excluding 380 women). This leaves a sample of 7,134 women, of whom 4,067 were mothers.

The NSFG is the primary national source of information on birth intendedness, having included questions regarding the intendedness of births since its inception in 1973 (London et al. 1995; Ventura et al. 2008). The NSFG does not directly inquire whether a birth was intended or wanted. Instead, wantedness and intendedness are constructs based on responses to a series of questions asked for every birth. Wantedness is derived from the question, “Right before you became pregnant, did you yourself want to have a(nother) baby at any time in the future?” A negative answer would be characterized as an unwanted birth. If a woman responds affirmatively, she is asked about the timing of the pregnancy: “So would you say you became pregnant too soon, at about the right time, or later than you wanted?” Births that are identified as too late or at about the right time are considered wanted and intended. Births that are identified as occurring too soon are asked a follow-up question regarding the extent to which the births were too soon: “How much sooner than you wanted did you become pregnant?” We consider births occurring two or more years too soon as seriously mistimed and thus unintended (according to the operational definition used here), while those occurring less than two years too soon are considered slightly mistimed and thus intended. Our categorization decision was confirmed with exploratory analyses using a more detailed classification system (later than wanted, wanted or on time, slightly mistimed, seriously mistimed, unwanted). In preliminary analyses predicting intendedness of first births, we found very few differences in the predictors between births characterized as wanted or on time, later than wanted, or two or fewer years too early, so we grouped these types of births. There were some slight differences in the predictors of higher-order seriously mistimed and unwanted births, but for the sake of ease of presentation and interpretation of data, we decided to group these two categories as well. We return to these differences briefly later in the article.

Analytic Plan

We first describe the distribution of unintended fertility among women overall and then use discrete-time event history models to examine how the intendedness of births is related to subsequent fertility. We run separate models by parity, looking at the association between first and second births and between first, second, and third births. We also predict having a first birth by intendedness as a baseline model of sorts to examine which individuals are selected into unintended fertility. By establishing a baseline, it is possible to determine whether the same characteristics that select women into starting their fertility careers with an unintended birth continue to be associated with the risk of subsequent unintended fertility. The dependent variable for the analysis of first births has three categories: no birth, an intended birth, or an unintended birth. In the analyses predicting higher-parity births, we run two sets of models: one predicting any birth, which serves to relate intendedness to overall fertility, and one using the three-category dependent variable accounting for intendedness. We use logistic regression in predicting any birth and multinomial logistic regression for the intendedness of the birth.

All analyses use person-months as the unit of analysis. In the model predicting the first birth, women enter the analysis when they turn 12 and exit the month of their first birth or at the time of survey if they have not had a birth. For models predicting higher-parity births, women enter the month of the preceding birth (i.e., women enter the month of their first birth for models predicting a second birth, and the month of their second birth for models predicting a third birth) and leave when they have a birth or at the time of the survey if they have not had a birth.

Our key independent variables are indicators of whether prior births were unintended. For models predicting the second birth, we control for whether the first birth was unintended or intended. For third-birth models, we use a set of four dummy variables distinguishing women with no unintended births, women with two unintended births, women with an unintended first birth and an intended second birth, and women with an intended first birth and an unintended second birth. Our overarching hypothesis is that unintended fertility at low parities is associated with unintended fertility at higher parities. We believe that an association between fertility and the intendedness of the most recent birth is an indicator of causal forces, while an association between fertility and intendedness of earlier births captures unobserved heterogeneity.

Independent Variables

We include a range of socioeconomic and demographic control variables. In the model predicting first birth and intendedness, we include age as a time-varying categorical variable (less than 18, 18–19, 20–24, 25–29, and 30 or older). In the models predicting higher-parity births, we include a control for the age at last birth, and duration since last birth is specified as a piecewise, time-varying linear spline (less than 24 months, 24–48 months, and more than 48 months) because of the discontinuities between duration since last birth and fertility. We include variables for non-Hispanic white women, native-born non-Hispanic black women and foreign-born non-Hispanic black women, native-born Hispanic women, and foreign-born Hispanic women. We distinguish between native- and foreign-born Hispanic women following research that indicates variation in family behavior according to nativity status (Landale and Oropesa 2007). Exploratory analysis indicated differences by nativity status for black women (but not white women) as well (about 10% of blacks are foreign-born); research has noted that foreign-born blacks often display dissimilar behaviors relative to their native-born counterparts in terms of educational attainment (Bennett and Lutz 2009; Massey et al. 2007) and labor force participation (Gore 2005), which may reflect underlying differences in motivation that could affect fertility behaviors.

Because the 2002 cycle of the NSFG did not include a detailed education or employment history as in other cycles, we have limited time-varying measures of socioeconomic status. We use data on the month when a high school diploma was received to construct a time-varying measure of education (high school diploma or GED/no diploma). In addition, we account for family background, which may also influence women’s fertility behaviors and the acceptability of childbearing in different circumstances (Musick 2002), through measures of family structure at age 14 (intact, stepfamily, or other), respondent’s mother’s education, and whether the respondent’s mother had a birth prior to age 18. Because women in relationships are more likely to have a child, all models include a time-varying indicator of whether the woman was cohabiting or married during the month. In analyses not shown here, we modeled relationships in greater detail to examine the impact of transitions in and out of relationships and changes in partners. We found that it mattered more whether women simply were in a relationship, and the type of relationship, than whether they were moving in and out of different types of relationships with the same or different partners. As such, we are using a simple indicator of relationship type rather than the more detailed relationship indicators because the main results of interest (the association between early and later unintended fertility) were not affected by relationship status, the interpretation of relationship status effects on fertility was less straightforward, and the model fit was not significantly improved.

Results

Descriptive Statistics

Table 1 displays weighted statistics describing sociodemographic characteristics and the distribution of unintended fertility across the life course for all mothers in the 2002 NSFG. More than one-half (57%) of mothers began childbearing in their 20s, with about one-third having children in their teens, and only 13% beginning childbearing at age 30 or later. More than one-third of the mothers had their first child outside of a coresidential union, while 13% were cohabiting, and one-half were married. The average number of children a mother had in 2002 was 2.19, with an average of 0.62 unintended births.² Slightly more than one-quarter of all mothers reported that their first birth was unintended. (This proportion is lower than the traditional definition of unintended; using the definition that groups slightly mistimed with seriously mistimed and unwanted, about 41% of births are unintended (not shown).) Slightly more than 40% of women with children reported having any unintended births, suggesting that the majority of women who experienced unintended fertility experienced it with their first child.

Of women who had two (or more) births, more than one-half (59%) reported that both their first and second births were intended. About one-third (33.5%) of women reported an unintended first birth; 19.4% had an unintended first birth followed by an intended second birth, and 14.1% had two unintended births. Relatively few mothers reported that their second birth was unintended after an intended first birth. Looking at how an unintended first birth relates to subsequent fertility overall, among women with an intended first birth, only 5.5% of all their births were unintended, compared to three-fourths of all births to women with an unintended first birth. Women who begin their fertility careers with an unintended birth have more children, on average, than women whose first child was intended; this difference is statistically significant, although the magnitude is modest. It appears that women with an initial unintended birth often go on to have subsequent unintended births, while a higher-parity unintended birth after an intended first birth seems to be somewhat rare, and these differences suggest that the first birth serves as a strong signal of subsequent unintended fertility. However, it is not clear to what extent the intendedness of a first birth predicts later unintended fertility net of risk factors for the first unintended birth. We consider this question in the following multivariate analyses.

Multivariate Results

Table 2 shows relative risk ratios from multinomial logistic regression predicting the risk of having a first birth by intention status using discrete-time event history models. Because our focus is on unintended fertility, this discussion will focus mostly on the last column, which compares the risk of having an unintended first birth relative to an intended first birth. Consistent with previous research, the risk of having an unintended first birth relative to an intended first birth declines with age, largely because women are less likely to have mistimed births as they age (not shown). There are significant race/ethnicity/nativity differences as well. Native-born non-Hispanic black women are 1.7 times as likely to have an unintended first birth rather than an intended birth compared with non-Hispanic white women. (As the first two columns indicate, native-born non-Hispanic black women are also more likely overall to have a birth than white women.) Foreign-born non-Hispanic black women and Hispanic women have lower odds—by about 40% and 60% lower, respectively—of having an unintended versus intended birth relative to non-Hispanic white women. In contrast, the odds of having an unintended first birth rather than an intended first birth for native-born Hispanic women are not significantly different from those of non-Hispanic white women, although native-born Hispanic women are more likely to have a birth overall than non-Hispanic white women. There is no statistically significant relationship between family structure at age 14 or educational attainment (at least as measured by high school graduation) and the intendedness of a first birth. However, women whose own mother had a birth prior to age 18 are more likely to have a birth than no birth, and they are 1.2 times as likely to report this birth as unintended. Both cohabiting and married women are more likely to have a birth overall compared with their noncohabiting, nonmarried counterparts, referred to here as “single” for brevity. Cohabiting women are no more likely than single women but about three times as likely as married women (not shown) to have an unintended first birth than an intended first birth, while married women carry a much lower risk of an unintended first birth than single women (relative risk ratio (RRR) = 0.296).

These models can be considered as baseline models for understanding the process of selection into an unintended first birth. They demonstrate strong associations between the intendedness of a first birth and age, race, and union status, even controlling for other factors. Turning now to predicting higher-parity births based on the intendedness of the first birth, Table 3 shows four sets of models. Model 1 is a standard event history model predicting any birth but including no covariates, and Model 2 adds covariates. Model 3 is a multinomial model predicting the intendedness of a birth with no covariates, and Model 4 adds covariates to the multinomial model. In Models 2 and 4, age is modeled as two components: categorized age at first birth (fixed) and interval since first birth (time-varying).

In Model 1, women whose first birth is unintended are significantly less likely to have a second birth (OR = 0.87) than women with an intended first birth. Additional models (not shown) demonstrate that this association largely works through relationship status because women who have an unintended first birth are far less likely to be in a coresidential relationship. When sociodemographic characteristics are controlled (Model 2), the intendedness of the first birth does not have a statistically significant relationship with overall second-birth hazards. In the full model, Model 2, several socioeconomic and relationship covariates are statistically significant and in the expected direction (i.e., age at last birth, relationship status), consistent with the results for first birth.

Models 3 and 4, which compare having no birth, having an intended birth, and having an unintended birth, show that an unintended first birth has opposite associations with the risk of intended and unintended second births. The null relationship shown in Model 2 is the result of these opposing relationships balancing each other out in the presence of socioeconomic and demographic controls. Compared with a woman whose first birth was intended, a woman with an unintended first birth is less likely to have an intended second birth (RRR = 0.67) and more likely to have an unintended second birth (RRR = 2.05) relative to no birth, net of a range of sociodemographic characteristics. Women with an unintended first birth are nearly three times as likely (RRR = 3.05) to have an unintended second birth than an intended second birth relative to their counterparts who began childbearing with an intended birth. Note that having an unintended first birth is a stronger predictor of having an unintended second birth—both relative to intended births and relative to no birth—than any other variable in the model, including race, age at first birth, relationship status, and education, all of which are powerful determinants of fertility trajectories. This association could be seen as evidence of causal effects of an early unintended birth on later births. However, it is also consistent with the notion that women whose first birth is unintended have some unobserved characteristic that also makes them more likely to have another unintended birth.

Many factors associated with the intendedness of a second birth are similar to those associated with the intendedness of a first birth. There are some differences, however. Although there is no statistically significant difference in the intendedness of first births to single and cohabiting women, second births to cohabiting women more closely resemble those of married women. Cohabiting women are less likely than single women to have an unintended birth relative to an intended birth, and the difference between cohabiting and married women is not statistically significant (not shown). In predicting first births, blacks and Hispanics (regardless of nativity) are more likely to have an intended first birth relative to no birth in comparison with non-Hispanic whites, but there are no race/ethnicity/nativity differences in predicting a second intended birth relative to no birth. However, compared with non-Hispanic whites, all other race/ethnicity/nativity groups are more likely to experience an unintended birth relative to no birth and relative to an intended birth.

Table 4 shows results for third births. Again, results from both dichotomous (any birth vs. no birth) and multinomial (no birth vs. intended birth vs. unintended birth) models are included. In the unconditional model, Model 1, women who have an unintended first and second birth are roughly 42% more likely to have a third birth relative to women whose first two births were intended. However, in the presence of socioeconomic and relationship controls, the intention status of the first two births is not predictive of the likelihood of a third birth (Model 2). The positive association seen in Model 1 is largely driven by the earlier age of childbearing among women whose births were not intended (not shown). That is, early unintended births seem to increase third birth rates by increasing the amount of time that women are at “risk” of subsequent fertility.

In multinomial models, intendedness of prior births works in different and countervailing directions for the likelihood of and intendedness of a third birth (Model 4). As with second births, women with earlier unintended births are more likely to have unintended third births and less likely to have intended third births; similarly, the intention status of early births are the strongest predictors in the model of the intention status of the current birth. Compared with women with no unintended births, women with an unintended first birth or an unintended second birth have a higher risk of an unintended third birth relative to an intended third birth. The strongest association with subsequent fertility occurs among women whose first two births were both unintended; compared with women with no unintended births, these women are more than five times more likely to have an unintended versus intended third birth.

The continued relationship between the intention status of the first birth and higher-parity fertility suggests an important role of unobserved heterogeneity in explaining these relationships. If having an unintended first birth had a purely causal effect on subsequent birth timing or intendedness, this causal effect would be mediated in third-birth models controlling for the age and intention status of second birth. Instead, the intention status of first birth appears to capture some characteristic of women’s reproductive behavior or attitudes that has continuing associations with later births. However, the relative magnitude of the coefficients for first and second births in the third-birth models also points to some causal relationship. In fact, the magnitude of the association between an unintended second birth and the risk of an unintended third birth relative to an intended third birth (RRR = 3.29) is close in size to the magnitude of the association between an unintended first birth and the risk of an unintended second birth relative to an intended second birth (RRR = 3.05, Table 3, Model 4), suggesting a fairly straightforward and consistent connection between the immediate prior birth and the subsequent birth. If having an unintended birth were only an indicator of some other characteristic (including willingness to report an unintended birth), women with an intended first birth and an unintended second birth should resemble women with an unintended first birth and an intended second birth. The stronger relationship between the more recent birth and third-birth intendedness (RRR = 3.29, compared with RRR = 2.15) implies some distinct relationship between recent fertility and third births. However, those women with two unintended births have the highest risk of a subsequent unintended birth, suggesting that a small subset of women have great difficulty in managing reproductive behavior throughout their childbearing years.

Coefficients describing the relationship between sociodemographic characteristics and the intention status of third births are generally the same sign as coefficients predicting second births. However, the magnitudes of some coefficients are smaller, and some associations that are statistically significant in the second-birth models are not statistically different from zero in the third-birth model. In particular, race/ethnicity/nativity differences in unintended fertility are smaller for third births than for first and second births. This attenuation may be due to the smaller sample size in the third-birth model. It is also possible that the selection of women who have had a first and a second birth into the model reduces variation in the third-birth models relative to the second-birth models.

To assess the relative importance of selection and unobserved heterogeneity, we ran multilevel models (not shown) pooling person-months of exposure to the risk of unintended births, with women as the level-two unit of analysis. Person-months were nested within women, and each woman had a unique random intercept (representing individual-level variation in overall propensity to have an unintended birth). Although the individual-level random effect accounted for about 10% of the variance in likelihood of a higher-parity unintended birth, incorporating these individual unobserved effects in the model did not fully account for the observed relationship between having an unintended first birth and later unintended fertility. Because unobserved heterogeneity did not seem to completely account for the relationship between unintended early fertility and subsequent unintended fertility, we chose to use the current models because they allowed us to model third-parity births separately and consider the possible effects of both first- and second-birth intendedness on third-birth intendedness.

In models not shown here, we also disaggregated seriously mistimed and unwanted births in both the independent and dependent variables in the higher-parity models. Although both seriously mistimed and unwanted births predicted the risk of subsequent seriously mistimed and unwanted births, each specific type of unintended birth more strongly predicted the same type of unintended birth. That is, although a seriously mistimed prior birth significantly increased the risk of an unwanted birth (perhaps women who have a birth far earlier than they intended reach or reduce their desired family size earlier and thus face more years of risk for an unwanted birth), it more strongly and significantly increased the risk of a seriously mistimed birth, and the same is true for unwanted births. This lends further credence to the notion that the conditions of prior births are strongly likely to be repeated for subsequent births.

Discussion

After we account for socioeconomic and demographic characteristics, we find that the intention status of first and second births is not related to the overall risk of a subsequent birth, but having an unintended birth at any parity is associated with a higher risk of a subsequent unintended birth relative to no births or an intended birth. These results suggest both causal relationships and unobserved heterogeneity at work. Having an unintended first birth is associated with unintended fertility for both second and third births, even net of mediating factors, such as age, relationship status, and subsequent birth intendedness. This association is likely due to persistent unobserved factors. At the same time, intendedness of the second birth is a stronger predictor of third-birth intendedness than is intendedness of first birth, suggesting a direct causal relationship between one birth and the next as well.

Possible persistent factors associated with unintended fertility over the life course might include psychological traits, such as self-efficacy and planfulness as well as attitudes toward contraception and abortion. Although these factors may change over time as individuals age and mature (Mirowsky and Ross 2007), there is likely some stability in these factors because they often are innate personality characteristics. Most of these factors are not easily measurable in a survey setting. Similarly, abortion data is difficult to obtain in surveys, but more reliable data on abortion would allow researchers to assess whether the association between early and subsequent unintended fertility is driven by pregnancies (suggesting contraceptive use as a pathway) or by pregnancies carried to term (suggesting abortion as a pathway).

Comparing full models with the unconditional models shows that age and relationship status are important mechanisms for the relationship between early unintended fertility and later unintended fertility. In particular, being in a cohabiting or marital relationship strongly predicts women’s risk of having a birth and their characterization of their births. Having a birth outside a committed union is less than ideal, and our results show that it was unlikely that women who had such births intended to become pregnant at the time. Having a birth outside marriage decreases the likelihood of marriage in the future (Upchurch et al. 2001), and women who have one birth outside of marriage are increasingly likely to have any and all subsequent births outside of marriage (Hoffman and Foster 1997; Wu et al. 2001). As such, an early unintended birth, especially one that occurs outside a coresidential relationship, can increase the risk of a subsequent unintended birth to the extent that it limits women’s future prospects for more serious unions without necessarily impacting their exposure to subsequent fertility.

Another potential pathway for an effect of early unintended birth on later fertility may arise from how such births are received by the mother, family, friends, and the larger community. In a study of socially disadvantaged single mothers, Edin and Kefalas (2005) found that while many unintended births are not initially welcomed, mothers often find within themselves reserves of strength they did not know they had. Expectant mothers “rise to the occasion” and meet their responsibilities, and this brings them an enormous sense of empowerment. Moreover, even among families that may initially respond negatively, family and friends usually rally around an expectant mother to provide her (and her child) support. The outpouring of support may send a message that unintended childbearing is not a big deal—that it can be managed and perhaps may even become an unexpected source of joy. These potential positive outcomes, combined with ambivalence about contraceptive use from partners, concerns (or misconceptions) about side effects (Kendall et al. 2005), and widespread examples of individuals with unintended (usually nonmarital) childbearing both within the community and in society overall, may simply lessen low-income women’s motivation to avoid pregnancy. It is not clear whether this possible pathway is relevant across other socioeconomic strata as well as for disadvantaged women.

Socioeconomic status may contribute to both selective and causal mechanisms linking early and later unintended fertility. Family income during childhood and adolescence is related to later economic and educational attainment: women who face financial barriers to accessing reproductive health care in adolescence and early adulthood are disproportionately likely to face similar barriers later in life. At the same time, unintended fertility has negative effects on educational attainment and may also slow professional advancement. Thus, having an unintended birth may reduce women’s access to economic resources that facilitate effective use of contraception in the future, either by preventing contraceptive use (for instance, if a woman cannot afford to fill contraceptive prescriptions every month and thus cycles on and off hormonal contraception) or forcing women to choose less-expensive and less-efficacious methods (such as condoms) over methods with greater reliability and less potential user-error (such as IUDs). Because of data limitations, we are unable to empirically examine these pathways.

The cross-sectional nature of the NSFG makes it impossible to identify socioeconomic conditions prior to or at the time of births, and the lack of an employment or educational history precludes the inclusion of time-varying socioeconomic factors. In general, we are limited by the lack of socioeconomic control variables (including insurance and other potential access factors) because it has been well established that women at risk of an unintended birth are disproportionately of low income. Better measurement of socioeconomic status might have allowed us to more accurately identify selection mechanisms.

The cross-sectional design also means that we do not know women’s actual fertility intentions prior to having children, and as with any work on fertility intentions, there are always concerns about retrospective accuracy. As noted earlier, women often are reluctant to label births as unintended, and there is also some evidence that reports of unintendness may shift over time as recall error, rationalization, and other factors change. As such, our measures may underestimate unintended fertility, especially early unintended fertility, and that may affect our estimation of fertility trajectories and our coefficients. It is also worth noting that our definition of unintended as including only those births identified as unwanted or seriously mistimed may make it difficult to compare our work with earlier studies that also include slightly mistimed births; however, as more researchers adopt this convention, this may be less of a problem. Finally, we should reiterate that we are focusing on births rather than pregnancies because of the underreporting of abortions and pregnancies not carried to term. Had we explored pregnancies instead of births, we would generally expect to see the same pattern and perhaps even a stronger relationship, given the evidence on repeat abortions (Jones et al. 2006). However, it is also possible that attitudes toward or access to abortion is an unobserved characteristic driving these results. If some women are consistently unable or unwilling to terminate unintended pregnancies, these women will be more likely to have unintended births than women who experience unintended pregnancies and end those pregnancies by abortion.

Method choice may also impact the association between early and subsequent unintended fertility, particularly if a woman becomes motivated to change her contraceptive usage in response to an unintended pregnancy (either to prevent future births more efficiently or to adjust her desired fertility preferences regarding timing and family size). Unfortunately, the 2002 NSFG collects detailed contraceptive use data only from January 1999 to the time of survey, so we cannot include time-varying measures of method use as the births in our analytical sample extend back to 1969. Although there are some data on whether women were using contraception when they became pregnant, there is no information on efficacy and consistency of use. Moreover, many women reported they were not using contraception at all; for women who were trying to get pregnant, the lack of contraceptive use is surely endogenous, whereas the reasoning behind not using contraception among those who did not want to get pregnant is far from clear. Thus, we were unable to test an underlying hypothesis that an unintended birth may affect a women’s usage of contraception.

Conclusion

These findings have implications for policy makers interested in preventing unintended births. Although it would be preferable to prevent women from having an unintended first birth, the lack of large and sustainable decreases in unintended fertility in the past suggests that this a formidable goal. Given relatively limited funds, it might be best to focus on a smaller and more readily identifiable group of women. As such, because having an unintended birth is a strong predictor of later unintended fertility, targeting resources toward women who report an earlier unintended pregnancy may be an effective prevention technique. Regardless of whether causal or selective (or both) mechanisms are behind the association, our results suggest that interventions to prevent unintended pregnancies might be made more efficient by focusing on women who have already had one unintended birth.

Our results are also useful in understanding the dynamics of fertility and family formation in the United States. Early research on unintended fertility focused on unwanted births late in the childbearing career. As delayed marriage and childbearing have become more common, an increasing proportion of unintended births occur at low parities. The need to reassess the measurement and definition of unintended fertility in light of this shift has long been recognized (see, e.g., Klerman 2000; Santelli et al. 2003). The strong correlation between the intention status of early and later births suggests that we may need to further reorient thinking toward a conceptualization of unintended births not simply as events that occur at the extreme ends of the childbearing career, or as isolated events, but rather as markers that characterize the entirety of the childbearing career.

Acknowledgments

A previous version of this article was presented at the 2009 annual meeting of the Population Association of America, April 30–May 1, Detroit, MI. We are grateful for helpful comments from Lawrence Finer and from participants in the session, as well as input from Elizabeth Wildsmith.

Notes

References