Approximately three-quarters of all ever-married women spend at least 15 years at the end of their reproductive lives exposed to the risk of unwanted fertility (childbearing that is not wanted then or at any point in the future).1 The population of all those at risk of unintended childbearing includes not only those who desire no more children, but also those wishing to delay childbearing until a more ideal time in the future. Young women, who constitute part of this population, spend more time now than ever before both sexually active and unmarried.2 The years during which women and couples are at risk of an unintended birth (either unwanted or mistimed) thus begin with adolescence and continue throughout the reproductive years.

Most research examining correlates of the intention status of women's pregnancies has been based on data collected retrospectively. However, for years, retrospective reports of pregnancy wantedness have been recognized to be subject to recall error or rationalization. As long as three decades ago, it was noted that respondents might not admit to what could be perceived as an inadequacy, or that they might change their attitude based on positive experiences with a child resulting from a pregnancy that, at conception, would have been considered unintended.3 The presumed result of these deficiencies in retrospective studies would be the underreporting of unintended pregnancies or the overreporting of intended ones.

Prospective studies of unintended childbearing are not beset by issues of rationalization or biased recall, but are subject to at least one important limitation in their predictive ability: An intervening situational change might affect a respondent's desire to have a child (i.e., an attitude and corresponding behavior might change over the course of an interval).

In this article, we use the 1988 round of the National Survey of Family Growth (NSFG) and a 1990 telephone reinterview of 1988 respondents to identify which of the women who reported in 1988 that they wanted to avoid childbearing for good or to postpone a birth for at least three years were most successful at avoiding a birth in the interval between the two surveys. The data permit a prospective look at the control of unintended fertility during the late 1980s, and enable us to compare our findings with analyses that have relied on retrospective reports.

Background

Our study highlights variables that have been found to be important predictors of unintended childbearing in retrospective analyses: education, income, race, parity and marital status.4 We also examine labor force status and, to the extent that is possible, change in marital status (arguably one of the most important situational factors affecting childbearing decisions). In addition, we control for contraceptive use and assess the importance of attitude certainty among women wishing to have no more children. Both of these factors are thought to be important in determining the connection between fertility attitudes and behavior. Moreover, measures of certainty of intention to avoid a pregnancy are often unavailable for analysis in studies of this sort.

Socioeconomic Characteristics

Education has long been known to be a critical factor influencing fertility5—by its impact, for example, on exposure to the risk of pregnancy and on desired family size. In addition, it has been argued that women's access to and control over financial resources (usually their own earnings or their family's income) affect their fertility intentions. Greater access to both education and financial resources may reduce unintended pregnancy, at least partly by increasing a woman's decision-making power, both directly and indirectly.6 For example, raising a woman's educational and financial status may enhance her self-esteem and self-confidence, and may thereby increase her ability to influence important decisions, including those concerning contraception and fertility.

Race and income are highly correlated;7 it has been shown that black and Hispanic women have markedly lower levels of income and education and less access to health care and health insurance than do white women.8 In the 1980s, black women were more likely than either white or Hispanic women to have experienced an unwanted birth,9 and more recent retrospective data show the same pattern. The unintended pregnancy rate of black women has been estimated to be almost three times that of white women,10 despite substantial declines in the percentage of black women who were sexually active but not using any method of contraception between 1988 and 1995.11 The differences in fertility intendedness between black and white women, however, may be the result of the strong association between race and socioeconomic factors rather than due to race alone.12

Additionally, unmarried women, often teenagers, are more likely than married women to experience an unplanned birth. Since marital unions are generally considered to be more stable than consensual unions, which in turn tend to be more stable than those in which a residence is not shared,13 the extent to which births are planned and considered wanted by both partners is likely to be greatest in the former and to be least in the latter. Previous research has demonstrated the importance of marital status as a predictor of intention status.14

The respondent's behavioral status (whether he or she is starting, continuing or intensifying a particular behavior)15 also may predict behavioral outcomes; for example, a first birth is likely to change a couple's life more than a subsequent birth. Parity, which has been found to be associated with unintended childbearing, is thus included in this analysis.16 Finally, as childbearing and employment are at least partially incompatible,17 a woman's labor force status also can be expected to affect her fertility intentions and behavior, at least in the short term, and so is examined in this research.

Changing Circumstances

The impact of changing circumstances on attitudes and behavior has been discussed in detail in the social psychology literature. Attitudes or intentions may not be consistent with subsequent behavior, as "there are almost always situational forces working for and against behavioral realization of attitudes."18 Controlling for changes in these situational factors, therefore, should be important in prospective assessments of unintended childbearing.

In the case of fertility decisions, a change in marital status is likely to be one of the most important situational changes that may influence childbearing intentions. Women still appear to view out-of-wedlock childbearing as less desirable than childbearing within marriage, as marriage is still viewed by many as a "normative license" for parenthood.19 A change in marital status would be expected either to widen or to narrow the range of options a woman sees as available to her, although the extent to which that effect is observed may vary by race and other factors.20

Attitude Certainty

The extent to which attitudinal reports, including those pertaining to childbearing, will be consistent with subsequent behavior is thought to be influenced by the individual's assessment of the importance of the issue, the intensity of the respondent's opinion of the topic, and the individual's knowledge of or level of involvement with the issue.21 For example, a 1990 study of diaphragm use found that more strongly held attitudes were significantly correlated with more consistent behavior.22

Attitude intensity may not be the best predictor of behavior over the long term, however, even though it is closely linked to the notion of attitude strength. Attitude intensity may correspond to level of emotionality, yet given constraints, people who feel more intensely about everything may be no more able than individuals who feel less intensely to behave in certain ways.23 Attitude certainty has been argued to be a better predictor of behavior; desires that are reported to be very certain are thought to be more stable over time than those that are less certain, and less likely to fluctuate with routine changes in daily circumstances. Therefore, we expect that stronger certainty of intention to avoid childbearing would reduce the likelihood of a birth in the interval between interviews.

Contraception

In fertility research, attitudes toward childbearing (e.g., number or timing) are considered to be among a bundle of psychological traits that are translated first into fertility desires and then into fertility intentions, such as those measured in the NSFG. Intentions, in turn, are thought to determine "instrumental behaviors [such as contraceptive use], which themselves determine whether a fertility outcome occurs."24 Because contraception is an important intervening variable between the intention to avoid a pregnancy and whether a birth occurs, we also take contraceptive use into account in our analyses.

Methodology

Analytic Approach

Our analysis compares women's fertility intentions reported in 1988 with their actual fertility during the short period between interviews. We focus on those who claimed in 1988 that they wanted to avoid childbearing for good, as well as those who wanted to postpone a birth for at least three years. We ascertain to what extent these women succeeded in avoiding a birth, and determine which women were most successful in meeting this objective. To allow for the possibility that the processes affecting the "postponers" and those forgoing all future childbearing might differ significantly, we examine the two groups separately in all analyses.

We refer to births occurring in the interval, despite intentions to the contrary, as unpredicted. This terminology aids in maintaining a distinction between prospective analyses of fertility behavior and traditional retrospective studies of intendedness. In a prospective analysis, an unpredicted birth could actually be intended if, for example, a woman's pregnancy intentions had changed during the interval between interviews.

We expect that the intention to postpone versus to forgo future births would be related both to life-course markers and to socioeconomic activities. Some characteristics probably influence postponing behavior more than stopping behavior, such as being young, unmarried, poor or of low parity. Being older, formerly married or of higher parity are likely to be more closely related to an intention to avoid all future childbearing. If intentions lead to related instrumental behaviors (such as contraceptive use), as is argued in the literature,25 both postponers and those forgoing all future childbearing should be likely to practice contraception to lower their risk of an unpredicted birth.

We expect that situational characteristics, such as a change in marital status, should influence the fertility of women postponing births more than the fertility of those stopping or forgoing childbearing. Women who would like a child or another child but who judge the present not to be the ideal time may envision more ideal conditions in the future; such circumstances may manifest themselves sooner than expected. On the other hand, postponing women may also begin to perceive even less ideal conditions to be important or imminent (e.g., age-related infertility). As a result of either scenario, original timing intentions may be amended.

Data

Our analysis is based on data from Cycle 4 of the NSFG and a subsequent computer-assisted telephone reinterview, both of which were conducted by the National Center for Health Statistics. The Cycle 4 data were collected between January and August 1988, using a multistage area probability sample of noninstitutionalized U.S. women aged 15-44. Interviews were conducted with 8,450 women of all marital statuses; 2,771 were black, 5,354 were white and 325 identified themselves with some other racial or ethnic group.

The telephone reinterview took place between July and mid-November 1990.* The final sample, which consisted of a supplement of teenagers to account for the aging of the original sample, consisted of 5,686 women; 1,594 were black, 3,902 were white and 190 identified themselves as other. The final response rate for the telephone reinterview was 67.5%; this figure includes the response rate from the original NSFG interview (82.1%).26

Variables

We use the term "unpredicted birth" to describe any birth that occurred in the interval between the 1988 interview and the 1990 reinterview to women who had said in 1988 that they intended to postpone childbearing at least for the next three years, as well as to those who said they wanted no more children. We focused on births, rather than on all pregnancies, because of the widespread underreporting of abortion in the NSFG and other national survey data.

The strength of the respondents' preference to have no more children was measured from a question asked after women reported their intentions to forgo or delay future childbearing.§ The question allowed us to create a measure of certainty of intention for those who intended to have no more children only. Unfortunately, we lacked an analogous measure of certainty for women who wanted more children, but not within the next three years. This would have required a survey item ascertaining how certain they were that they would delay having the next child for the specified amount of time. The measure for our subsample of forgoers is coded as: very sure they want no more children, or not very sure.**

Age was classified by the age of the respondent as of March 15, 1988. Respondents were classified as black, white or other, depending on self-reports in the 1988 interview. Because the number of respondents designated as "other" was very small, percentages are generally presented for black women and white women only; totals include the "other" racial or ethnic groups, however. Parity was determined by the number of live births the woman reported at the time of the 1988 interview.

Women were classified by marital status in 1988 as married, widowed, divorced, separated or never-married. Those who reported that they were not married but were living with their sexual partner were considered to be cohabiting. We also included a measure of change in formal marital status between the two interviews, which allowed us to disaggregate changes that occurred after a conception leading to a birth and those that occurred before a conception. The variable consisted of three categories: no formal change in marital status occurred, or the change occurred after the conception leading to the birth in question; the respondent got married or resumed living with her husband before she conceived, or she got married or resumed living with her husband and no birth occurred; or the respondent's marriage dissolved (she was either widowed, divorced or separated) before she conceived, or the marriage dissolved and no birth occurred.††

Income in 1988 was measured in accordance with the census definition of poverty status; it was calculated by dividing the total family income by the weighted average threshold income of nonfarm residents in households headed by persons under age 65.27 The three income categories were: income below poverty level; income at 100-199% of poverty; and income at 200% or more of poverty. Education was measured in years of school completed as of the 1988 survey, and was coded as: less than a high school education; high school graduate; or one or more years beyond high school.

Working status was based on items detailing the respondent's labor force participation in the week before each survey. Labor force status in 1988 was classified as: working full-time (35 hours or more); working part-time (1-34 hours); or not working (unemployed, laid off, looking for work, keeping house, in school or other). We omitted from the analysis those on temporary leave, for whom active labor force participation status was ambiguous.

A woman's contraceptive status also was defined as of the 1988 interview, and was coded into one of four groups. Women using highly effective methods were those relying on methods with failure rates of less than 5%, even under conditions of "typical (i.e., imperfect) use"28—male or female contraceptive sterilization, the pill and the IUD. Users of less effective methods included women relying on any of the 12 methods that have greater than 10% failure rates, given typical use. Women "not using contraceptives and not at risk of unpredicted birth" were those who were not practicing contraception and who were either postpartum at the time of the interview, seeking pregnancy, sterile for noncontraceptive reasons, had not had intercourse since first menstruation, had had intercourse only once or had not had intercourse in the last three months.‡‡ Those classified as "not using contraceptives, at risk of unpredicted birth" included women not reporting any contraceptive use in 1988, those who did not fall into any of the "not at risk" categories and those who had had intercourse in the three months prior to the interview. These four categories correspond closely to those used in an earlier analysis of contraceptive use and exposure to the risk of unintended pregnancy.29

Our sample excluded 16 respondents who reported contradictory information during the 1988 interview about whether they were seeking pregnancy at the time. We also excluded 1,306 women who were contraceptively sterile or whose husbands were contraceptively sterile (26% of the sample) and 279 who were noncontraceptively sterile or sterile for unknown reasons (5.6% of the sample). These women were not asked about future birth intentions and were not at risk of an unpredicted pregnancy in the interval.

Eight of the respondents who claimed to want to avoid giving birth for at least three years (four women who wanted to avoid future childbearing and four who wanted to postpone) were coded as "not avoiding pregnancy." These seemingly contradictory cases remain in the sample because respondents were classified as "not avoiding pregnancy" if either they or their husband or partner wanted a pregnancy as soon as possible. These respondents wanted to postpone or forgo childbearing, while their husband or
partner did not.

Results

Table 1 shows the weighted percentages of respondents who intended to avoid a birth for at least three years following the 1988 interview.§§ Roughly two-thirds of women in the final sample intended to avoid a birth, at least for the next few years: Forty-one percent were postponing and 26% wanted no (or no more) children. Respondents most likely to want to stop childbearing were older women, those with at least a high school education, higher parity women and those who were no longer married. Respondents most interested in postponing childbearing for three years or more were younger women (especially those aged 15-19), those with lower household incomes, women with less than a high school degree, those who had not yet started having children or who had had only one child, those who had never been married (and, to a lesser extent, those who were cohabiting) and women not working or working part-time. A number of these findings are clearly age-related.

In general, women who were within the prime childbearing ages, those at lower parities, those who were married or cohabiting and those who had higher incomes were least apt to be avoiding pregnancy at the time of the 1988 interview. Younger, unmarried women and those with lower household incomes were more inclined to report that they intended to put their reproduction on hold, at least temporarily.

Table 2 shows that the large majority of women wishing to avoid childbearing for at least three years were successful within that relatively brief window of time: Only about 10% of postponers and 8% of those claiming to want no (or no more) children had births that they did not intend to have at the time of the 1988 interview. There were variations, however, among subgroups. For both groups of women avoiding childbearing, those most likely to experience an unpredicted birth in the interval between the two interviews were women with household incomes below the poverty level, higher parity women and those who were not using any contraceptive method in 1988, despite being at risk of pregnancy.

Among those interested in postponing a birth for at least three years, white women and those who had more than a high school education were more successful than other women in avoiding an unpredicted birth in the interval between interviews. Among women who intended to forgo or discontinue childbearing entirely, those working full-time were less likely than those who were unemployed to have given birth in the time between interviews. Women aged 35 and older also were more successful than were women in their 20s at avoiding an unpredicted birth between interviews.

We conducted separate logistic regression models of the effects of social and demographic characteristics on the likelihood of an unpredicted birth among those who wanted to postpone a birth (Table 3) and among those who wanted to forgo or discontinue childbearing (Table 4). In both tables, the first model includes only women's characteristics as of the 1988 interview and is the most comparable to research that has been done using retrospective data. The second model adds change in marital status, and the third adds contraceptive status and (for the analysis of women wanting to forgo future childbearing) certainty of intentions.

All three models in Table 3 show that among those interested in postponing a birth, women with incomes below the poverty level in 1988 were more than twice as likely as those with incomes between 100-199% of poverty to experience an unpredicted birth. For those wanting to forgo future births (Table 4), the size of the difference approaches threefold. While these results are striking, the results for education are mixed. Among respondents intending to postpone childbearing, those with a college education were clearly more successful at avoiding a birth than were those with less education (Table 3); however, education shows no effect among those women who intended to forgo or discontinue childbearing (Table 4). Similarly, race is a statistically significant factor only among those women desiring to postpone a birth (and then only in Model 1). Race lost significance when change in marital status and contraceptive status were added to the statistical models.

Women aged 35 and older who wanted to avoid future childbearing were about 60% less likely than women in their 20s to experience an unpredicted birth. Among those who wanted to postpone future births, women in their early 30s were about 70% less likely than women in their 20s to experience an unpredicted birth.

Parity was a stronger predictor of births to women who were hoping simply to postpone further births (Models 1 and 2) than it was in any of the models for women intending to discontinue or forgo childbearing (Table 4). Women wishing to postpone future childbearing who had had no live births were 50-60% less likely than comparable women who had had one child to experience an unpredicted birth in the interval. However, the differences between women with one previous birth and those at higher parities were not significant in any model in Table 3. This is consistent with the idea that the transition from no children to the first birth is the most important.

Never-married women who were postponing childbearing were more than 60% less likely to have a birth in the interval than were married women, even when contraceptive status and at-risk status were controlled for. Among women who wanted to forgo future childbearing entirely, marital status had no statistically significant impact on the likelihood of experiencing an unpredicted birth; similarly, a change in marital status had no significant impact on the likelihood of experiencing a birth.

Not surprisingly, contraceptive status had a pronounced effect on the likelihood of having an unpredicted birth. Among women wishing to postpone and those wishing to have no future births, those who were not using a contraceptive method while at risk of pregnancy were 2-3 times as likely as users of effective methods to report an unpredicted birth.

After we controlled for all other characteristics, employment status in 1988 had no statistically significant effect on the likelihood of having a birth in the interval; it is therefore shown only for Model 1 in both tables. Certainty of intentions also failed to predict success or failure in avoiding a birth after the effects of other relevant variables were controlled for.

Discussion

In this prospective analysis of unintended fertility, we were specifically interested in comparing women's intentions to avoid a birth with their fertility during the interval between surveys, so that comparisons could be drawn between the predictors of unintended childbearing that are identified prospectively and those that have been identified retrospectively.* We looked at women who wanted to postpone childbearing separately from those who wanted to forgo future childbearing: Respondents who were unsuccessful at postponing a birth could be considered comparable to women who had experienced mistimed births in retrospective studies; respondents who were unable to avoid future births altogether could be considered comparable to those who had experienced unwanted births.†§

Many of the effects of the social and demographic variables remained significant in our analysis, even when changes in marital status, certainty of intention and contraceptive use were controlled. While the results demonstrate the importance of contraceptive use in mediating intention and behavior, they also indicate that some women are less likely to have an unpredicted birth due to factors in addition to contraceptive status alone.

Many of our results are consistent with those of previous retrospective studies. For example, we too found effects of education and age on unpredicted fertility among women seeking to postpone a future pregnancy, and the effects were in the expected directions.

Some of our findings are not consistent with those of earlier retrospective analyses, however. Interestingly, our findings suggest that older women who wanted no (or no more) children were significantly less likely than women in their 20s to experience the equivalent of an unwanted birth in the interval, even when at-risk status was controlled for, and the effect of parity was not significant. Race was significant only among women interested in postponing a birth and only in Model 1, which most closely approximates studies using retrospective data. The lack of effect of race among those forgoing childbearing is surprising, as race is often one of the strongest predictors of unwanted childbearing when it is assessed retrospectively, net of other factors. Further, those want-
ing to postpone a birth who were never- married were less likely to experience an unpredicted birth in the interval than were married women, even when at-risk status was controlled. These findings are less consistent with earlier analyses that have examined this issue retrospectively.

Where differences exist between the two types of data, we can assume one of several things. First, some women may be more apt to fail to report births as unintended when asked about their attitudes retrospectively. Second, others may be more likely to experience situational changes that had not been anticipated and that made childbearing more desirable in the time period between interviews when the analysis is done prospectively. If only the first condition were true, we might conclude that white women, women in their 20s and more highly educated women are particularly likely to underreport unwanted births. In addition, married women might be relatively more likely than never-married women to underreport births as mistimed. If only the second condition were true, we could conclude that those same groups of women were particularly likely to experience a situational change that made childbearing more desirable in the interval than it had been at the time of the first interview (and that they were able to become pregnant within the short time frame between the two surveys).

The literature we reviewed suggests that a comparison between stated intentions and subsequent fertility behavior should focus on the social and demographic characteristics of the respondents, as well as on changing circumstances in their lives and the certainty of their intentions to avoid a birth. Past research implies that consistency between intention and behavior would be greatest when the intention was one of some certainty and when major changes had not occurred in the person's life to bring about a change in attitude.

We were able to analyze data that begin to assess both of these factors, although our measures clearly do not capture all of the necessary information about the context in which fertility decisions are made (or not made). The failure of attitude certainty to contribute more to our analysis may suggest that it is more subject to change upon reflection or under evolving circumstances than has been thought. An assessment of fertility at progressively longer durations, in addition to the two-year interval, might also improve the test of the importance of certainty of intentions.

As we have noted, we also had hoped to examine a number of changes in the respondents' life circumstances that might have altered their attitudes toward near-term childbearing (such as household income, marital status, labor force participation and schooling). If respondents experienced a change in circumstances that made childbearing in the time frame desirable or at least acceptable, we expected that some women's attitudes toward pregnancy avoidance might genuinely have changed during the interval. Unfortunately, with the exception of marital status, the NSFG data lack the level of detail that would establish temporal ordering of births and such status changes in the interval. We were therefore limited to looking at changes in marital status, obviously a potentially important determinant for fertility phenomena. Surprisingly, we found no effect of certainty of intentions or change in marital status on the odds of experiencing an unpredicted birth.

Despite the limitations of the data, this analysis updates our knowledge of the linkages between intentions to avoid childbearing and subsequent fertility behavior in the near term in the United States. In so doing, it should contribute to ongoing attempts at improving data quality. Future prospective studies that allow for analysis of larger samples of unpredicted births with more detailed information about sequences of events, and perhaps with data allowing multi-item constructs on attitudes,30 will help answer some of the questions raised here.