Education Differences in Intended and Unintended Fertility

Abstract

Using a hazards framework and panel data from the National Longitudinal Survey of Youth (1979–2004), we analyze the fertility patterns of a recent cohort of white and black women in the United States. We examine how completed fertility varies by women’s education, differentiating between intended and unintended births. We find that the education gradient on fertility comes largely from unintended childbearing, and it is not explained by child-bearing desires or opportunity costs, the two most common explanations in previous research. Less-educated women want no more children than the more educated, so this factor explains none of their higher completed fertility. Less-educated women have lower wages, but wages have little of the negative effect on fertility predicted by economic theories of opportunity cost. We propose three other potential mechanisms linking low education and unintended childbearing, focusing on access to contraception and abortion, relational and economic uncertainty, and consistency in the behaviors necessary to avoid unintended pregnancies. Our work highlights the need to incorporate these mechanisms into future research.

In recent decades, education differences in the timing and partnership context of fertility have widened. College graduates have postponed childbearing, while less-educated women continue to have their first birth at relatively young ages (Rindfuss, Morgan and Offutt 1996). At the same time, all groups of women have postponed marriage. Combined with greater union instability among the less educated, this has led to growing disparities in single parenthood (Ellwood and Jencks 2004). Recent literature on U.S. fertility has focused so intently on timing and context that the question of social class or education differences in completed fertility has been eclipsed. We return attention to this topic.

U.S. socioeconomic differences in levels of fertility are longstanding, with the poor and less educated generally having more children (Blake 1968; (Freedman, Whelpton and Campbell 1959). In the 1960s and 1970s, scholars and policy makers were concerned about high fertility and the problems of overpopulation. Many saw technological innovation and access to effective contraception as key to reducing unintended pregnancy and driving down fertility, particularly among those with relatively large families (Ryder 1973b; Westoff 1972; Westoff and Bankole 1996). Indeed, with the diffusion of the birth control pill, the incidence of unintended pregnancy in the United States declined in the 1960s and 1970s (Pratt et al. 1984), and fertility dropped to about replacement. But the diffusion did little to level socioeconomic differences in fertility (Sweet and Rindfuss 1983). Further, while all education groups experienced declines in unintended pregnancies up until the 1990s, this decline reversed in the mid-1990s for poor and less-educated women, increasing the education gradient on unintended fertility (Boonstra et al. 2006; Finer and Henshaw 2006). As of the mid-1990s, age-specific fertility rates were such that a woman with a high school degree or less would be expected to have 2.1 children in her life, while college graduates would have 1.6 (Yang and Morgan 2003). The gradient on unintended fertility is steeper, with a much higher proportion of unintended pregnancies to women with low education, and a lower proportion of the unintended pregnancies resolved with abortion. In 2001, 40 percent of births to women with less than a high school education and 10 percent of those to college graduates were unintended (Finer and Henshaw 2006).

The dominant account of education differences in fertility focuses on opportunity costs; women who can earn higher wages are more strongly motivated to limit their fertility because the income they forego for any time taken out of employment for childrearing is greater. A second, cultural view posits that having children is more socially valued in lower classes, in part because alternative sources of meaning are so scarce (Edin and Kefalas 2005). Both of these views appear more relevant in explaining why women have intended births, but they may also bear on unintended fertility. Given large – and growing – education differences in unintended fertility, we set out to better understand differences in the separate components of intended and unintended births. We assess the extent to which education differences can be explained by opportunity costs and values, using longitudinal data that span the reproductive lives of a recent cohort of U.S. women. Women’s education is correlated with their parents’ education and income, with their own occupational status and earnings, and with the education and earnings of their partners. Thus it can be seen, broadly, as an indicator of “class” or SES in a heterodox formulation. Our analysis, however, concerns only effects of education, with select models assessing the role of wage in mediating the education-fertility relationship.

Our analysis is the first to examine how education affects both intended and unintended fertility within a multivariate hazard framework. It is also novel in directly examining the extent to which education differentials in completed (intended and unintended) fertility are explained by differences in women’s wage rates or the number of children they want. To foreshadow, we will find that education differentials in completed fertility operate largely through unintended births, and that these differentials are not explained by either wage rates or number of children wanted. In discussing our results, we draw on diverse literatures to propose three potential explanations for an education gradient in unintended childbearing, focusing on differences in access to contraception and abortion, instability in relationships and economic circumstances, and consistency in the behaviors necessary to avoid unintended pregnancies, or efficacy.

Past Theory and Research

Opportunity Costs

Among economists, the dominant perspective linking education or other measures of earning power to fertility is sometimes called the Columbia-Chicago cost-of-time view (Becker 1981; Hotz, Klerman and Willis 1997; Mincer 1963; Pollak and Watkins 1993). In this view, women decide between alternative uses of their time – in childrearing or market work. The higher their potential wage, the higher the opportunity cost of having a child, assuming they will reduce employment for childrearing. Women with higher potential wages will choose fewer children because the “price” of any hours of employment foregone to engage in childcare is greater for those who can earn more.

Of course, well educated women typically marry better educated husbands (Schwartz and Mare 2005). In the economic view, men’s income increases fertility, just as it increases the purchase of many consumer durables, because it makes having children more affordable (Becker 1960). Indeed, in theory, either men’s or women’s earning power could have a negative opportunity cost effect (also called a “price effect” by economists) or a positive “income effect” on fertility (Macunovich 1996). However, to the extent that women do most of the child-rearing work that involves a reduction of market labor supply, the income effect of men’s and the opportunity cost effect of women’s earnings are expected to dominate.

Given the centrality of the price-of-time perspective, there is surprisingly little direct empirical evidence from the United States to support it. Schultz (1994) used 1980 U.S. Census data and showed that women’s predicted wage (based on education, demographic and state policy variables) had a negative effect on number of children. An earlier time series analysis for 1948–1975 showed that net effects of women’s average wages on fertility were negative, an effect that was stronger when the female employment rate was higher, while men’s average wages had a net positive effect on fertility (Butz and Ward 1979). Using a similar method with data from 1964–1994, Macunovich (1996) found a negative effect of average female wage on fertility early in the period, but a positive effect in more recent years. She concludes that income effects may now dominate the price effects of women’s earnings. Surveying the evidence, Heckman and Walker (1990) conclude that “the main question – do wages and incomes matter? – is still open.”¹

Values – How Many Children are Desired

Demographers have largely been silent on whether education differences in desired family size might explain education differences in fertility. The exception is Judith Blake, a dissenter to the view that contraceptive technology and access drive fertility trends and differentials (Blake 1967; Blake and Das Gupta 1975). She argued for the importance of motivation to have a certain number of children, yet the survey data she assembled from the 1930s through the 1960s suggested that social classes differed little in the number of children they wanted, with the exception that those at the very bottom wanted slightly more (Blake 1967).

Using rich, qualitative interviews, Edin and Kefalas (2005) argue that, compared to their middle-class counterparts, low-income women place a higher value on having children. They rely in part on the opportunity cost view, pointing out that the jobs poor women might forego for childbearing pay little anyhow. But they also argue that lower class women place a higher absolute value on children, not merely a higher value relative to their other options (Edin and Kefalas 2005). They recount poignant testimonials from their sample of poor, unmarried mothers who saw children as the greatest source of meaning in their lives. They further cite survey data (Sayer, Wright and Edin 2003) showing that those with low education are dramatically more likely to say that people without children live “meaningless lives.” Lower education is also correlated with more traditional gender role beliefs, which might be expected to encourage fertility (Kane 1995; Kiecolt and Acock 1988).

Education Differences in Unintended Fertility

The contraceptive revolution has brought the United States nowhere near abolishing unintended pregnancy (Boonstra et al. 2006; Finer and Henshaw 2006). To assess whether pregnancies were intended, demographers generally use survey questions that classify pregnancies as unintended if women say that, at the time they conceived, they never wanted to become pregnant again, or that they wanted to become pregnant at some point in the future, but not yet. The first is called “unwanted” and the second “mistimed,” and both are seen as unintended. This classification scheme is imperfect, ignoring the possibility of a continuum with shades of ambivalence in the middle (Bachrach and Newcomer 1999; Edin et al 2007), and ignoring men’s views, but it does allow us to compare educational subgroups.

As noted, recent data show large and increasing differences by education and income in unintended fertility (Boonstra et al. 2006; Finer and Henshaw 2006). At a proximate level, these disparities arise from less consistent and effective contraception (or abortion) by those with less education and income, even when they are not trying to get pregnant (Boonstra et al. 2006; Brown and Eisenberg 1995; Silverman, Torres and Forrest 1987). But studies on this topic are based largely on bivariate associations, so the less proximate mechanisms remain mysterious. At first glance, high levels of unintended fertility seem inconsistent with explanations of education differences based on opportunity cost and values, as these emphasize women’s motivations in getting pregnant. That is, in a rational choice framework, either preferences (child-bearing desires) or pecuniary incentives (based on opportunity cost) should affect whether one wants a pregnancy, and thus affect decisions about whether or not to have unprotected sex, practice contraception or abort. Indeed, economists interpret behavior as “revealed preferences.” Yet, insofar as contraception and abortion have monetary and psychic costs, those whose values or opportunity costs give them only weak motivation to avoid childbirth may not be as vigilant about birth control, leading to some unintended pregnancies as well (Hotz et al. 1997; Miller 1986). Thus we expect values and opportunity costs to have some effect on unintended fertility, but we expect both to have larger effects on intended than unintended fertility.

Research Strategy and Hypotheses

We rely on longitudinal data rich in individual characteristics and changes in characteristics over time, allowing us to directly test two key explanations of education differences in completed fertility: the value women place on children and opportunity costs. If education affects fertility through values, then we would expect to find that a young woman’s desired number of children accounts for some of the education/fertility relationship. If the lower fertility of more educated women is because of their higher earning potential (hence, higher opportunity cost of time spent in childrearing), then we would expect wages to explain the education/fertility relationship. Overall, we expect education differences in child-bearing desires and opportunity costs to operate on intended and unintended fertility, but especially on intended.

We derive estimates of completed cohort fertility by education from discrete-time multinomial hazard models of intended, mistimed and unwanted births. Our models include births of all parities and incorporate interactions between age and education to account for the older age schedule of childbearing among more highly educated women (Martin 2000; Rindfuss, Morgan and Offutt 1996). All analyses are run separately for white and black women because both the timing and number of births vary by race (Morgan 1996; Yang and Morgan 2003).

Data and Methods

National Longitudinal Survey of Youth 1979

We use the NLSY79, a national probability sample of individuals born from 1957 through 1964 and living in the United States in 1979 (U.S. Bureau of Labor Statistics 2006). The sample includes just under 4,000 white and black women ages 14–21 in 1979.² Interviews were conducted every year from 1979 to 1994 and then every other year until 2004 (the survey is ongoing), when sample members were 39–46, or 42 years old on average. Personal interviews were conducted in all survey years except 1987, in which a limited phone interview was done. Attempts were made each year to interview all sample members, regardless of whether they missed prior surveys. Retention rates for the NLSY79 have been relatively high. They exceeded 90 percent through the early 1990s and were 77 percent in 2004. Sample weights are used to adjust for differential attrition over time.

Measures

Intended and Unintended Fertility

Birth histories provide dates of birth for all children born to each respondent. Key to our analysis, the NLSY79 also includes information about the intention status of each pregnancy, based on a series of questions pertaining to contraceptive use and feelings at the time of pregnancy. Women are first asked whether they were contracepting at the time they became pregnant and, if not, whether it was because they wanted to become pregnant. They are then asked: “Just before you became pregnant the (first, second, third, etc.) time, did you want to become pregnant when you did?” Births are intended if a woman reported not using contraception because she wanted to get pregnant or said, irrespective of contraceptive use, that she wanted to get pregnant or felt indifferent about getting pregnant at that time. Births are mistimed if a woman said she did not want a (nother) baby at the time she got pregnant, but did want a (nother) baby at some time in the future,³ and they are unwanted if she said she did not want a (nother) baby at any time in the future. Although the quality of retrospective reports of pregnancy intendedness has been debated (Ryder 1973; Trussell, Vaughan and Stanford 1999; Westoff and Bankole 1996; Williams, Abma and Piccinino 1999), there is ample evidence of its validity. For example, a high proportion of couples who report wanting no more children choose sterilization soon after their last wanted birth (Bumpass 1987). Also, reported birth intendedness is associated with child outcomes later in life (Barber et al. 1999; Baydar 1995; Brown and Eisenberg 1995; Crissey 2005; but see Joyce, Kaestner and Korenman 2000). In an analysis using multiple controls and sibling fixed effects, Barber and East (2009) show that while the effects of a mistimed birth on child outcomes are smaller than those of an unwanted birth, both have negative effects, suggesting that the joint category of “unintended” has some utility.

In the NLSY79, questions about pregnancy intentions are generally asked over a one- or two-year interval.⁴ We were unable to match pregnancy intentions to 8 percent of recorded births (due to either item nonresponse on the intendedness questions or missing reports on pregnancies corresponding to birth dates in the fertility history). Missing information on intention status is fairly evenly distributed across education groups. In descriptive tabulations, we allocate missing information on intention status based on observed race- and education-specific proportions intended, mistimed and unwanted. We exclude these births from our hazard analyses. We model one birth per year, and thus further exclude all but the first birth to occur in any year of age (resulting in a loss of 1 percent of births in our sample). Finally, to allow for lagged predictors in our models, we exclude births prior to 1981, meaning that we exclude all births to women under age 16 (another 1 percent of births in our sample).

Explanatory Variables

In descriptive tabulations, we rely on years of final completed education, as measured at the last survey. Years of education are categorized into four groups: less than 12, 12, 13–15 and 16 or more. For normative schooling transitions, these correspond to high school drop out, high school graduate, some college and college graduate, respectively, and we will typically refer to them as such. In the early years of the survey, when many women are still in school, observed education may be a poor indicator of educational prospects. We therefore use time-varying predicted years of final completed education in our main analyses, which better capture the effects of a young woman’s educational plans on whether she has a child at a young age. As births may truncate educational experiences, the idea is to create a measure of education that will allow us to examine the effects of education on births without incorporating the effects of births on education. Predicted values are lagged two years from the time a birth is assessed to ensure that they precede conception. They are generated by regressing final completed years of schooling on completed education to date; school enrollment; educational expectations at first interview; the Arms Forces Qualifying Test, a measure of cognitive ability or skill level administered in 1981; and family background, including mother’s and father’s education, parental income in 1979 and family structure at age 14.⁵ Regressions are run separately by five-year age groups and race, and thus are fully interacted along these dimensions. Quantitative predictions are categorized to correspond to less than high school (<12), high school (12–12.99), some college (13–15.99), and college (16+). Despite the richness of the data, the possibility remains that the estimated effects of (predicted) education are subjected to omitted variable bias, reflecting in part other characteristics of women not in our models that cause both their educational attainment and their tendency to have unplanned births.

Our measure of child-bearing desires comes from a question asked in 1979, in the first year of the survey: “How many children do you want to have?” Respondents were in their late adolescence, on average, at the time of the question. For most, this question was asked prior to first birth.⁶ Distinct from child-bearing ideals or expectations, this item comes closest to tapping unconstrained fertility preferences.

The most direct indicator of opportunity cost is the cost of one’s time, or hourly wage rate. Our primary measure of opportunity cost is the natural log of current hourly wage in 2004 dollars. In models including current wage, we also control for whether the respondent is currently in paid employment (wages are ser to the mean for those not employed for pay).⁷ Wages and employment (like education) are time-varying and lagged two years. Although wages may be seen as the most direct measure of opportunity cost, they may be a poor indicator of long-term economic potential (e.g., Xie et al. 2003). We thus examine an alternative measure, which is the natural log of the running average of past wage observations (in constant 2004 dollars). This smoothes out fluctuations in wages and (arguably) provides a better indicator of long-term potential. When we include this indicator in our models, we control for the proportion of all years a woman has spent in paid employment since the start of the survey and flag the cases with no paid work experience.

In addition to these key variables, all models control for school enrollment. Some models include time-varying and lagged indicators of marital status and the natural log of spouse employment income. Specifically, these variables are used to test the sensitivity of our results to the potentially confounding effects of spouse income on the relationship between fertility and own wage. Descriptive statistics of explanatory variables are shown in Appendix Table 1.

Discrete- Time Hazard Models

We examine the education gradient on fertility in a discrete-time hazard framework, which allows us to incorporate influences of multiple explanatory variables that vary over time. We use multinomial logistic regression to examine the relative risks of an intended, mistimed, or unwanted birth of any order, at a given age, relative to having no birth. Suppressing individual subscripts, our full model can be written:

$$\text{log }[{\mathrm{P}}_{\mathrm{j}(\mathrm{t})}/(1-{\mathrm{P}}_{\mathrm{J}(\mathrm{t})})]={\alpha }_{1\mathrm{j}}+{\alpha }_{2\mathrm{j}}\text{age}(\mathrm{t})+{\alpha }_{3\mathrm{j}}{\text{age}}^2(\mathrm{t})+{\mathrm{\beta }}_{1\mathrm{j}}\text{Ed}(\mathrm{t}-2)+{\gamma }_{1\mathrm{j}}\text{Ed}(\mathrm{t}-2)\times \text{age}(\mathrm{t})+{\gamma }_{2\mathrm{j}}\text{Ed}(\mathrm{t}-2)\times {\text{age}}^2(\mathrm{t})+{\mathrm{\beta }}_{2\mathrm{j}}\text{Enrollment}(\mathrm{t}-2)+{\mathrm{\beta }}_{3\mathrm{j}}\text{Desires}+{\mathrm{\beta }}_{4\mathrm{j}}\text{In}(\text{Wages})(\mathrm{t}-2)+{\mathrm{\beta }}_{5\mathrm{j}}\text{Employment}(\mathrm{t}-2)$$ (1)

where the logit or log odds of a birth is an additive function of covariates, j indexes births according to intention status, and t indexes person-years of age from 16–46. The parameters α_1j to α_3j represent the baseline hazard, or the value of the log odds of a birth of intention status j at age t when all other covariates are zero. β_1j is a vector of coefficients for the main effect of predicted education, indicated by high school, some college, and college dummies, all relative to less than a high school degree. The γ_1j and γ_2j are vectors of coefficients for the interactions between education dummies and age and age-squared; these allow education’s effect on fertility to vary by age, accounting for the older ages at which more educated women, and in particular, college graduates, have their children. The β_2j to β_5j represent the estimated effects of school enrollment, child-bearing desires, wages, and employment, respectively. Child-bearing desires are measured only once, at the time of first survey, but all other substantive predictors vary over time and are lagged two years from the assessment of a birth.

A simple transformation of the logit (equation 1) makes it possible to estimate the probability of a birth of intention status j at age t:

$${\mathrm{P}}_{\mathrm{j}(\mathrm{t})}=1/(1+{\mathrm{e}}^{-\text{logit}})$$ (2)

Applying this transformation and filling in parameter estimates from our models, we generate age-specific probabilities of intended, mistimed and unwanted childbearing. We sum these probabilities over ages 16–46 to generate a measure of completed cohort fertility by intention status j,⁸ which can be written:

$${\text{CCF}}_{\mathrm{j}}={\displaystyle \sum _{t=16}^{46}}{P}_{j(t)}$$ (3)

Sample

We include all person-years in which there is information on births, intention status and explanatory variables. For example, if sample members drop out of the survey for a year or more but re-enter, we include all person-years for which there is data. When data are incomplete, we either exclude the affected interval (this is always the case when birth intention status is missing) or use a simple imputation strategy to fill in the gaps. In years in which the survey is not administered (odd years 1994–2004) and in cases of noninterview and item nonresponse, we impute missing values on time-varying explanatory variables by assigning values from the past year.⁹ Because complete, retrospective fertility histories are collected (including on birth intention status), this method allows us to retain most of the off-survey years.

As noted, time-varying explanatory variables are lagged two years from the interval in which a birth is assessed to ensure that their measurement precedes conception. We left-censor births prior to 1981 to allow for this lag, and consequently begin our analysis of births at age 16 (the youngest respondents are 14 in 1979, and thus 16 in 1981). Left censoring does not bias our results, although it may affect the precision of estimates for the youngest age groups, where sample sizes are relatively small.¹⁰

There are 2,543 white and 1,446 black women in our initial sample. We exclude a small number of cases (N = 26) due to missing values on child-bearing desires. We construct a person-year-of-age file for 3,963 individuals with data spanning 1981–2004, yielding 95,112 records. We drop 611 person-years due to missing information on the planning status of births. We censor 8,375 person-years due to attrition from the survey and 9,545 due to missing information on time-varying variables, leaving a final sample of 76,581 person-years (48,930 for whites and 27,651 for blacks). Overall, 3,934 women contribute an average of 19.5 years exposure and a total of 4,939 births.

Results

Table 1 shows the average number or intended, mistimed and unwanted births to white and black women in 2004, at the last year of the survey. Recall that respondents are 39–46 years old at last interview, i.e., at or very near the end of their child-bearing years. These are thus reasonable estimates of completed child-bearing for this cohort. The first column of Table 1 shows average fertility levels pooled across education groups. As is well documented (Yang and Morgan 2003), whites have lower overall fertility than blacks, although neither group is far from replacement level, with whites averaging 1.85 children per woman and blacks, 2.18. The share of all births that are unintended differs by race, with 23 percent (i.e., .42/1.85) mistimed and 4 percent unwanted among whites, compared to 38 percent and 18 percent, respectively, among blacks.

Table 1.

Observed Completed Fertility by Education

	All	< High School	High School	Some College	College	Ratio of < High School to College
Whites
Intended births	1.36	1.38	1.36	1.35	1.34	1.03
Mistimed births	.42	.56	.55	.43	.23	2.39
Unwanted births	.08	.13	.10	.07	.04	3.30
Total births	1.85	2.06	2.01	1.86	1.61	1.28
N	2,029	123	819	490	597
Blacks
Intended births	.94	1.27	.96	.86	.85	1.49
Mistimed births	.83	1.15	.93	.78	.48	2.39
Unwanted births	.40	.84	.41	.39	.14	6.24
Total births	2.18	3.26	2.30	2.04	1.47	2.22
N	1,197	116	504	391	186

Notes: Means are weighted and Ns are unweighted. Sample includes women at last interview, in 2004, ages 39–46. Missing data on birth intention status is imputed based on the observed proportion of intended, mistimed and unwanted by education and race.

Columns 2–5 (Table 1) show average completed fertility by years of completed education in 2004; the final column shows the ratio of total fertility for the lowest to highest education group, providing a measure of the education gradient on fertility. Summing over all births, while high school dropouts have 1.28 times as many children as their college-educated peers, and black high school dropouts have 2.22 times as many. This overall fertility gradient, however, masks substantial variation by the intention status births. For both race groups, the education gradient on fertility increases substantially for mistimed and unwanted births, relative to intended births. Indeed, white high school dropouts have virtually the same number of intended births as college graduates, but they have 2.39 times as many mistimed and 3.30 times as many unwanted births. For blacks, the ratios for high school dropouts compared to college graduates are 1.49, 2.39 and 6.24 for intended, mistimed and unwanted births, respectively. Clearly, education differences in fertility are driven largely by differences in unintended childbearing.

Next we examine education gradients in a multivariate framework. All subsequent analyses rely on our measure of predicted ultimate education. Table 2 reports estimates of total intended, mistimed and unwanted births by education derived from our multinomial hazard models. These are obtained by generating age-specific predicted birth probabilities from model coefficients for each of our four education groups, holding all other variables at their (weighted) mean values, and then summing birth probabilities from age 16 to 46.¹¹ Model 1 is our most basic model, which includes a quadratic function of age, predicted education, predicted education by age and age-squared, and school enrollment status. Model 2 adds the respondent’s early report of her desired fertility, and Model 3 adds our most direct indicator of opportunity cost, hourly wage. Results for whites and blacks are shown separately, and the complete models are reported in Appendix Table 2.

Table 2.

Predicted Completed Fertility by Education

	< High School	High School	Some College	College	Ratio of < High School to College
Whites
Model 1
Intended births	1.07	1.24	1.15	1.25	.86
Mistimed births	.62	.46	.39	.21	3.02
Unwanted births	.19	.08	.07	.03	6.68
Total births	1.88	1.78	1.60	1.48	1.27
Model 2: Model 1 + Desires
Intended births	1.08	1.24	1.14	1.24	.87
Mistimed births	.63	.46	.39	.20	3.07
Unwanted births	.18	.08	.07	.03	6.52
Total births	1.89	1.78	1.59	1.47	1.29
Model 3: Model 2 + Wage
Intended births	1.12	1.28	1.14	1.17	.96
Mistimed births	.55	.43	.38	.21	2.55
Unwanted births	.16	.08	.07	.03	5.43
Total births	1.83	1.78	1.58	1.41	1.29
Blacks
Model 1
Intended births	1.03	.76	.73	.76	1.36
Mistimed births	.80	.85	.69	.47	1.69
Unwanted births	.56	.29	.34	.08	7.33
Total births	2.39	1.90	1.75	1.31	1.83
Model 2: Model 1 + Desires
Intended births	1.05	.77	.72	.74	1.42
Mistimed births	.82	.86	.68	.47	1.75
Unwanted births	.55	.29	.34	.08	7.15
Total births	2.42	1.92	1.74	1.28	1.88
Model 3: Model 2 + Wage
Intended births	1.10	.80	.73	.71	1.56
Mistimed births	.74	.80	.66	.48	1.54
Unwanted births	.48	.28	.34	.08	5.77
Total births	2.32	1.88	1.74	1.27	1.83

Notes: Predictions are derived from (weighted) multinomial logistic regression models. Total fertility is estimated as the sum of age-specific predicted birth probabilities varying education while holding all other predictors at their weighted mean values. Model 1 includes age, age-squared, age by predicted education interactions and school enrollment; Model 2 adds child-bearing desires; Model 3 adds In of current wage and employment status.

Model 1 serves as a baseline estimate of the association between education and fertility. The education gradient derived from our model (as measured by the ratio of total fertility for the least to highest educated) is less than one on intended births to white women. That is, white college graduates are predicted to have somewhat higher completed intended fertility than their less-educated peers. White high school dropouts are predicted to have 3.02 times as many mistimed births and 6.68 times as many unwanted births as white college graduates. Results are similar for blacks, although the education gradient is always greater than one, i.e., high school dropouts have higher fertility than college graduates, irrespective of intention status: 1.36 times as many intended, 1.69 times as many mistimed, and 7.33 times as many unwanted.

Model 2 additionally controls for fertility desires, measured in the first year of the survey. If education differences in fertility desires were an important source of the education/fertility gradient, we should see a lessened education gradient after desires are controlled. But Model 2 (Table 2) shows that the differentials are virtually identical with and without controlling for fertility desires. The reason that early child-hearing desires do not explain any of the education gradient on fertility is that, despite the fact that desires are modestly related to fertility (Appendix Table 2), they are hardly correlated with education. For all education groups and both races, the median and modal desire is two children. What education differences there are in desires show college graduates wanting somewhat more children (Appendix Table la). Those white women who go on to be college graduates initially desired an average of 2.61 children, compared to 2.40 among those who drop out of high school; the analogous numbers for blacks are 2.68 for those who become college graduates and 2.02 for those who drop out of high school.

Model 3 adds indicators of opportunity cost. Our most straightforward measure of opportunity cost is the natural log of the respondent’s hourly wage two survey years before the hazard of a birth is assessed. Because women who are not employed have no wage, we set their wage to the mean and control for paid employment status. We examine change in the education gradient that results from adding these labor force variables to assess whether education differences in fertility arise from the way education affects the opportunity cost to women (and their families) of any time they take out of the labor market for childrearing. If wages deter fertility, the education gradient on predicted fertility derived from Model 3 should be lower relative to that from Model 2 (Table 2), i.e., we would expect to see a reduction in the ratio of fertility of the bottom to the top education group. In fact, we find no change in the education gradient for whites (the ratio of completed fertility for white high school dropouts to college graduates remains constant at 1.29 across models 2 and 3), and we find a very small decline for blacks (the ratio decreases from 1.88 to 1.83).

Why do wages fail to explain the education gradient on fertility? Of course wages vary by education, but they are, surprisingly, only weakly associated with fertility. Model 3 estimates in Appendix Table 2 reveal that wage coefficients are not always statistically significant, nor are they consistently negative, as predicted by economic theories of opportunity cost. Indeed, under a rational choice model, the negative effect of women’s wages should be largest on intended fertility, but here we find positive associations. We explore the relationship between fertility and wages in greater detail in Table 3. This table shows the predicted value of fertility, varying wages under alternative model specifications. To generate predictions, we hold education constant at the level of high school graduation and set school enrollment, child-bearing desires and employment status to their weighted mean values. The first set of results is based on Model 3 (which uses current wage to measure opportunity costs). Among whites, increasing wages from the 25^th to the 75^th percentile increases intended fertility by an estimated .14 births (from 1.22 to 1.36). Wage effects on intended fertility among blacks are in the same direction and are of a similar magnitude, but miss statistical significance. Higher wages are associated with reduced unintended fertility, although changes are statistically significant only in the case of black mistimed fertility, with an increase from the 25^th to the 75^th percentile of wage reducing mistimed fertility by .09 births. In all cases, the ratio of fertility at the 25^th percentile of wage to fertility at the 75^th percentile of wage is close to one.

Table 3.

Predicted Completed Fertility by Wages

	Intended	Mistimed	Unwanted
Whites
Model 3: Model 2 + Current Wage
25th percentile wage ($7.85)	1.22	.44	.08
75th percentile wage ($14.30)	1.36	.41	.08
Ratio of 25th to 75th percentile	.90	1.07	1.05
Model 4: Model 3 + Spouse Income
25th percentile wage ($7.85)	1.26	.46	.08
75th percentile wage ($14.30)	1.43	.44	.08
Ratio of 25th to 75th percentile	.88	1.06	1.05
Model 5: Model 2 + Average Past Wage
25th percentile wage ($6.11)	1.22	.44	.08
75th percentile wage ($10.70)	1.35	.44	.07
Ratio of 25th to 75th percentile	.91	1.01	1.09
Model 6: Model 5 + Spouse Income
25th percentile wage ($6.11)	1.27	.47	.08
75th percentile wage ($10.70)	1.40	.47	.07
Ratio of 25th to 75th percentile	.90	1.00	1.10
Blacks
Model 3: Model 2 + Current Wage
25th percentile wage ($6.88)	.77	.85	.29
75th percentile wage ($11.36)	.84	.76	.27
Ratio of 25th to 75th percentile	.91	1.12	1.08
Model 4: Model 3 + Spouse Income
25th percentile wage ($6.88)	.79	.85	.27
75th percentile wage ($11.36)	.86	.75	.26
Ratio of 25th to 75th percentile	.92	1.12	1.07
Model 5: Model 2 + Average Past Wage
25th percentile wage ($5.90)	.77	.88	.27
75th percentile wage ($8.94)	.80	.84	.27
Ratio of 25th to 75th percentile	.97	1.04	1.00
Model 6: Model 5 + Spouse Income
25th percentile wage ($5.90)	.79	.87	.26
75th percentile wage ($8.94)	.81	.84	.26
Ratio of 25th to 75th percentile	.98	1.05	.99

Notes: Predictions are derived from (weighted) multinomial logistic regression models. Total fertility is estimated as the sum of age-specific predicted birth probabilities varying wages while holding education at the level of high school graduation and all other predictors at their weighted mean values. Model 2 includes age, age-squared, age by predicted education interactions, school enrollment and child-bearing desires; Model 3 adds In of current wage and employment status; Model 4 adds marital status and In of spouse income. Starting again with Model 2, Model 5 adds In of average past wage, proportion years employed, and whether never employed; Model 6 adds marital status and In of spouse income.

Next we performed sensitivity tests to see if alternative procedures would support predictions about opportunity costs. In Model 4 (Table 3), we include the same labor force variables as Model 3, i.e., current wage and employment status, but we also control for marital status and spouse employment income (all lagged by two years). Women with high earnings tend to marry men with high earnings and, in economic theory, men’s wages are presumed to have an income effect, whereas women’s are thought to have a price effect. Thus, without controlling for spouse income, we might mask any true effect of women’s wage on fertility. Results based on Model 4 suggest that this is not the case. Women’s own wages have little effect, whether or not we control spouse income.¹² Controlling spouse income, we find the same small, positive coefficients of own wage on intended fertility (statistically significant for whites) and negative coefficients on unintended fertility (statistically significant for black mistimed births).

Of course, wages fluctuate, and women’s current wages may be imperfect predictors of their future wages, particularly if more education gives women access to jobs with much on-the-job training that offer lower starting wages but steeper wage trajectories. We thus try another sensitivity test, based on the idea that an average wage across past years in which the woman worked may be a better indicator of her potential wage. Instead of current wage, Model 5 (Table 3) includes the natural log of average past wage (in constant dollars) for all years. And instead of current employment, we control for the proportion of all past survey years that the respondent reports a wage and flag cases with no paid work experience (as in previous models, labor force indicators are lagged two years). Here again, we see little change in predicted fertility varying average past wage from the 25^th to the 75^th percentile and holding all other variables constant. Average past wage has the same estimated positive effect on white intended fertility as current wage; none of the other changes (for whites or blacks) are statistically significant.¹³ Adding controls for marital status and spouse income in Model 6 (Table 3) does not alter these findings. In sum, our results suggest that wages do little to deter fertility; indeed, they are very modestly, positively associated with intended fertility. As far as we can discern by our measures, opportunity cost does not explain the association between education and fertility.¹⁴

We briefly comment on racial differences here, although our focus is on education differences within racial groups. Overall, black women average higher fertility than white women (2.18 compared to 1.85 children). However, highlighting the importance of separate analyses by race, Table 1 shows that race and education interact, such that among the least educated women, blacks have substantially higher fertility than whites (3.26 vs. 2.06), whereas among college graduates the difference reverses, with black women averaging fewer births (1.47 vs. 1.61). The strong education gradient on unintended births that we emphasize here holds for both races, but even more strongly for blacks. The other striking race difference is that, within each education level, a higher proportion of black than white births are unintended (mistimed or unwanted, but especially the latter). These differences may reflect the large racial differences in union status across the education distribution; black women are much less likely to be married, and married women are much more likely to intend their births.

Summary, Conclusions and Policy Implications

Using panel data from the NLSY79, we assessed the effect of women’s education on whether she had no birth relative to the competing hazards of intended, mistimed and unwanted births. If the mother reported that at the time she got pregnant she wanted another child someday in the future but not right then, the birth was classified mistimed; if she said she never wanted another child, it was considered unwanted. These categories have in common that they are unintended at the time they occur. We wanted a measure of the effects of education on births not contaminated by effects of births on education. To try to get at this, we lagged our explanatory variables two years behind the wave in which the birth was reported and relied on predicted education (based on socioeconomic background, test scores, educational aspirations and education to dare). Predicted education also captures the effects of a young woman’s realistic knowledge that she is likely to go to college on whether she has a baby in her early years.

We find that (predicted) education is negatively associated with completed fertility among both black and white women. Our basic model predicts that white women who do not complete high school will have 1.88 children, on average, and those who complete college will have 1.48; for blacks, the analogous decrease across education categories is from 2.39 to 1.31 (Table 2, Model 1). These figures include intended and unintended births, and obscure the fact that what education mainly deters is unintended births. Indeed, our key finding is that the education gradient on unintended fertility, and especially on unwanted fertility, is much steeper than on intended fertility. The least educated white women are predicted to have .86 times as many intended, 3.02 times as many mistimed, and 6.68 times as many unwanted births as their counterparts who have graduated from college (Table 2, Model 1). The least educated black women are predicted to have 1.36 times as many intended, 1.69 times as many mistimed, and 7.33 times as many unwanted births as their counterparts who have graduated from college (Table 2, Model 1). In sum, the completed fertility gradient comes largely (and in the case of whites, entirely) from the tendency of less-educated women to have more unintended births.

Do Child bearing Desires Explain Education Differences in Fertility?

We found that controlling for how many children women said they wanted at the first wave of the survey (when they were 14–21) does nothing to reduce the educational differentials. This is because less-educated women do not aspire to more children. At the median, all eight race-by-education groups want exactly two. Qualitative accounts depict children as the main source of meaning for less-educated women (Edin and Kefalas 2005), whereas they are one among a number of meaning-making projects for the better educated. Desires for children relative to other pursuits may explain education differences in fertility; education differences in early reports of how many children women want, however, do not.

Do Opportunity Costs Explain Education Differences in Fertility?

Economists argue that having a child has higher opportunity costs for well educated women because they have access to jobs with higher earnings, so they have more to lose from any scaling back of employment – whether dropping out entirely for a time, or reducing their hours of paid work. We noted that even the past literature, with its mixed results, gives us reason to doubt this explanation. Some authors have cautioned that the theory does not clearly predict that high wages of women will deter fertility because the income effect (such that earning more makes raising children more affordable) might override the opportunity cost effect (Macunovich 1996), especially as purchasable substitutes for mothers’ time are more available (Joshi 2002). Moreover, the theory seems to apply most clearly to intended fertility, whereas the educational gradient comes largely from unintended births.

Our study casts further doubt on the opportunity cost explanation of fertility differentials. Education does, of course, raise women’s wages. However, when we enter wage rate as a covariate in our hazard models, it explains little to none of the educational gradient on fertility (Table 2, moving from Model 2 to 3). The reason that wage does not explain education effects is that estimated wage effects on fertility are small, even if we control for husband’s employment income (and marital status) (Table 3). Further, these effects are positive on intended fertility, opposite the opportunity cost prediction. One might criticize using current wages to test opportunity costs; wages fluctuate a good deal for young people, while the opportunity cost calculations envisioned by the economic model are based on long-term earning power. Thus, as a sensitivity test, we created an alternative measure of earning power – the average of all past years’ wage rates (in constant dollars).This variable did no better than current wage at accounting for the education gradient on fertility. Thus, we doubt that the higher opportunity cost faced by more educated women is why they are having fewer children.

What Explains the Education Differences in Unintended Fertility?

We have shown that neither wanting more children (they do not) nor lower wages (they are only weakly associated with fertility) explains the higher fertility of less-educated women, so what does? Because the higher fertility of the less educated comes mainly from unintended fertility, the question is really why the less educated have more unintended births of both the mistimed and unwanted variety. We propose three possibilities: socioeconomic differences in access to contraception or abortion, instability in relationships and economic circumstances, and efficacy. We call for further research to shed light on unintended fertility – one of the “persistent empirical puzzles” in the social demography of fertility (Bachrach et al. 2007:2).

Access to Contraception

The simplest hypothesis is that unintended fertility is higher among the less educated because they do not know about or cannot afford or access contraception or abortion. Those with low education and income often have no private health insurance,¹⁵ and while most clinics accept Medicaid and many offer sliding scales, many private physicians do not accept Medicaid patients (Silverman et al. 1987). One survey of low income, fertile, sexually active women found that about a quarter of those who did not want to get pregnant were nonetheless not contracepting. But only about 9 percent of nonusers mentioned a reason that could be construed as about cost or availability of services (Silverman et al. 1987). A recent qualitative study corroborates this; unmarried low-income parents were asked if they had ever been in a situation where they wanted to use birth control but did not because they could not afford it. None of those with unintended pregnancies said that money had ever been a barrier to contraceptive use (Edin et al. 2007).

There is stronger evidence that access limits use of abortion. The political climate has become increasingly unfavorable to abortion since about 1980, just the period in which NLSY79 data were collected, and many women have to travel a considerable distance to obtain one (Boonstra et al. 2006). Medicaid has not paid for abortions for decades except in states that contribute their own funds for what the federal government will not cover as a matter of policy (Boonstra et al. 2006). One quasi-experimental study gained analytical leverage from the fact that North Carolina provided funding for abortions for poor women, but did so each year only until funds ran out. The study assessed whether the dates of funding cutoff, which varied by year, corresponded with trends in abortions and births among poor women, concluding that not funding abortions for poor women decreases abortions and increases birth (Morgan and Parnell 2002).

Relational and Economic Uncertainty

Most demographic thinking into the 1970s assumed that childbearing largely took place within long-term, stable marriages, so that the two important questions were how children were spaced, and at what number the couple would stop. Today, the relationship context of childbearing is more complicated; an unintended birth, for example, may mean that a woman wanted a child but not with this partner (Zabin et al. 2000). Fertility and marriage have decoupled, with 4 in 10 births now to unmarried women (Hamilton, Martin, and Ventura 2009), and rates much higher for less-educated women. While non-marital births are often to cohabiting couples, these couples have high rates of separation (England and Edin 2007; Wu and Musick 2008). Women with low education are increasingly less likely to marry (Goldstein and Kenney 2001) and more likely to divorce (Martin 2006) than those with higher education, and thus are now much less likely to be making child-bearing decisions within a continuous relationship. Increased inequality and the decline of men’s earnings at the bottom (Morris and Western 1999) have also increased the economic instability in which partnerships exist.

Recent qualitative studies suggest that this climate of economic and relational instability may create more ambivalence about pregnancies, leading to more pregnancies that are reported as unintended. Edin and Kefalas’ (2005) study, mentioned earlier, found that poor women see the ideal context for childbearing as an economically stable married couple with a house. But when neither economic stability nor the house seem within reach anytime in the foreseeable future, valuing children greatly, they sometimes “roll the dice” and stop contraception, leading to conceptions that are not explicitly planned, but often welcomed. In another qualitative study using in-depth interviews of a subsample of unmarried couples from the Fragile Families survey, Edin et al. (2007) classified parents’ non-marital births by intendedness, finding that the more serious and stable the relationship was, the further the pregnancy was along the continuum from unplanned to ambivalent. (Few were explicitly planned.) Thus a subset of the non-marital births they analyzed seem to fit the mold suggested by Edin and Kefalas. Anthropologist Johnson-Hanks (2005) argues that people facing great uncertainty do not set fertility targets in the way presumed by family planning programs. The women in Cameroon that she studied could not tell her how many children they wanted because the answer depended on particular – and unpredictable – configurations of relationships, income, and kin support. Yet they did not act randomly; she characterizes their decisions as “judicious opportunism,” seizing good-enough opportunities for family building when they arose.

These qualitative studies suggest that the relational and economic instability now faced by less-educated women often creates ambivalence about the appropriateness of childbearing, leading at times to couples opening themselves to pregnancy in less than ideal circumstances. Some of these pregnancies will be classified as “unintended” by the usual survey methods, and, indeed, they are not explicitly intended, even though they may be at least ambivalently wanted at the time of conception. Of course, if such ambivalent pregnancies were the main explanation of the education gradient in births classified in surveys as unintended, it would call into question the measurement of intendedness that we (and other demographers) have used. We believe that this conclusion is premature, but call for more qualitative studies that probe the intentionality around conception.

Efficacy

Another possible explanation for education disparities in unintended fertility is that those with less education are less apt to believe that their own actions can have effects, and also have less of the self-regulation needed to engage in the sometimes onerous behaviors that further one’s long-term goals. Applied to contraception, these two factors lead to less consistency in contraception and more unintended pregnancies. Mirowsky and Ross (2007) show that those with more education perceive themselves to have more control over their lives. They theorize that a greater subjective sense of control is built up from experiences with challenges that one can master, more autonomous jobs, the control that having money brings and better health. The disadvantages of the childhood and adult circumstances of women with little education may mitigate against a sense of control. Perceived lack of control may in turn be a self-fulfilling prophesy, discouraging one to undertake actions that would pay off. For example, Ross and Mirowsky (2003) show that those with less education are less likely to engage in health-promoting behaviors (e.g., exercising or refraining from smoking), and that these create some of the education differentials in health. In the qualitative interview study of low-income, unmarried parents, Edin et al. (2007) concluded that about a quarter of the pregnancies they analyzed were cases in which there was not even an ambivalent desire to have a child at the time of conception, yet the couple was not contracepting; the authors saw these cases as indicative of lack of efficacy because behavior was not aligned with goals.

Something akin to this explanation of education or class differentials in fertility can be found in decades-old qualitative research by Rainwater (1960), which found many working class and poor couples with more children than they wanted. Operating in an era before the birth control pill, and before sterilization became common, the main means of birth control were condoms and the diaphragm. Most people in all classes knew about available methods, and money to buy them was not mentioned as the limitation. Nor did opportunity costs seem a factor as few women thought of employment as an option unless income was desperately needed. But education related to consistency in the use of these contraceptives. Some of the education differences in contraceptive use leading to unintended pregnancies may be cases in which individuals do nor believe they can control their fertility and/or are lacking the self-regulation to consistently engage in the hassle of contraception, even when they do not want a pregnancy.¹⁶

Implications for Public Policy

The issue of high unintended fertility concentrated among disadvantaged populations is more pronounced in the United States than other affluent nations (Brown and Eisenberg 1995). No doubt related to its high level of unintended childbearing, the United States has a relatively high rate of early and nonunion fertility (Kiernan 2002; Heuveline, Timberlake and Furstenberg 2003), also concentrated among the disadvantaged (Ellwood and Jencks 2004). Despite the dramatic reduction in U.S. teen fertility in the 1990s and early 2000s (Hamilton, Martin and Ventura 2009), far from converging with other nations, there are signs that education and income inequalities in U.S. family patterns are growing. For example, education differences in mother’s age at birth, divorce and single parenthood have widened (Martin 2006; McLanahan 2004; Raley and Bumpass 2003), as have differences by education and income in the proportion of births that are unintended (Boonstra et al. 2006; Finer and Henshaw 2006).

Several decades ago, demographers saw the high rates of unintended fertility in the United States and elsewhere in the context of fears about world overpopulation. They envisioned that contraceptive availability could eliminate most unintended births and usher in zero population growth. Instead, the United States got to about zero population growth even while the proportion of unintended births remained high, especially for those with low education (Finer and Henshaw 2006). Some women are having larger families than they intended. But, overall, U.S. women have .33 fewer births at the end of their child-bearing years than their stated expectations early in life (Quesnel-Vallee and Morgan 2003). Our Table 1 shows that of all the eight race-by-education groups, only black women in the lowest education group have fertility appreciably above the replacement level of 2:1. U.S. fertility among this most recent cohort to reach the end of their child-bearing years would fall well below replacement without unintended childbearing.

Well-educated women are relatively successful at avoiding unintended fertility, but they often have fewer children than they wanted early in life. Our analysis suggests that it is not their high wages leading to their low fertility, nor does it seem to be a lack of stable, economically secure marriages. College graduates have adopted a cultural schema that demands intensive mothering (Hays 1996; Lareau 2003), thus many wait until they are established in careers and marriages before having children. We suspect that it is their late start at marriage and fertility, as well as the difficulty they experience when trying to combine paid work and parenting, that leads many of them to end up with fewer children than desired. The lack of male participation in parenting and limited parenting accommodations by employers may discourage the transition to motherhood and the decision to have a second child.

Women with less education are much less likely to put off childbearing and are more likely to have unintended pregnancies. Their fertility, however, averages only slightly above replacement; most women in this group are not having more children than they wanted, more typically, they are having them at times and in situations that they view as less than ideal. Our analysis casts doubt on the usual explanations of the fertility-education relationship. Less-educated women do not start out wanting more children than those who get more education, nor do their lower wage levels explain much of their higher unintended fertility. Since our analysis was unable to explain educational differentials in unintended fertility, we could only speculate on their causes. We have suggested three mechanisms that deserve further study – that lack of economic and geographical access to abortion is a barrier, that the difficult economic circumstances common among less-educated women lead them to open themselves to pregnancies that they ambivalently want when the right circumstances seem unlikely, and that these same difficult circumstances erode the sense of control and self-regulation needed for contraceptive efficacy when women are clear that they do not want a conception. Assessing the merits of these hypothesized explanations will help to inform policies aimed at reducing tin intended fertility.

Appendix

a. Time-Invariant Sample Characteristics
	Whites	Blacks
Childbearing Desires in 1979 – All Women	2.57	2.33
Childbearing Desires in 1979 – by Final Education
< High school	2.40	2.02
High school	2.53	2.26
Some college	2.64	2.34
College	2.61	2.68
Variables in Predicted Education Regressions
Educational expectations in 1979	13.98	13.94
Standardized scores on AFQT	.55	−.54
Mother's education	11.96	10.81
Father's education	12.18	10.40
Parents' income in 1979	24,159	12,203
Lived with both parents until age 14	.80	.49
Total Sample Size Pooled Over Person-Years	48,930	27,651

b. Time-Varying Sample Characteristics at Three Observation Points and Over All Years
	Whites				Blacks
	1981	1991	2001	All Years	1981	1991	2001	All Years
Completed Education to Date
< High school	.59	.08	.06	.13	.68	.15	.11	.20
High school	.28	.44	.40	.42	.22	.43	.42	.41
Some college	.12	.23	.24	.24	.10	.29	.32	.29
College	.01	.25	.30	.21	.00	.12	.15	.10
Predicted Completed Education
< High school	.09	.07	.06	.07	.16	.15	.11	.14
High school	.19	.33	.39	.32	.24	.38	.42	.34
Some college	.53	.33	.25	.34	.54	.33	.32	.38
College	.19	.27	.30	.27	.07	.14	.15	.14
Age	19.29	29.52	39.55	30.24	19.14	29.53	39.46	30.32
Currently enrolled in school	.65	.08	.07	.17	.68	.06	.05	.15
Current wage (if employed)	8.46	13.82	15.91	12.98	9.16	11.20	12.30	11.19
Ln of current wage (if employed)	1.97	2.42	2.56	2.36	2.06	2.28	2.37	2.25
Currently employed	.62	.88	.86	.85	.35	.80	.85	.74
Average past wage (when employed)	8.46	9.66	9.14	9.07	9.16	8.43	7.43	8.03
Ln of average past wage (when employed)	1.97	2.18	2.12	2.10	2.06	2.03	1.89	1.96
Proportion years employed from survey start	.62	.85	.86	.83	.35	.70	.73	.67
Never employed	.38	.01	.00	.05	.65	.05	.02	.13
Spouse wage income (if married)	27,242	49,543	57,091	46,592	22,847	33,114	39,427	32,697
Ln of spouse wage income (if married)	9.79	10.29	10.55	10.31	9.52	9.39	9.97	9.57
No spouse income (if married)	.03	.11	.08	.10	.06	.25	.12	.20
Married/ spouse present	.13	.62	.72	.55	.05	.30	.31	.25
N	1,909	2,243	1,855	48,930	1,040	1,277	1,052	27,651

Notes: Means are weighted and Ns are unweighted. Characteristics are shown for our analysis sample, pooled over all person-years.

Notes: Means are weighted and N's are unweighted. Characteristics are shown for our analysis sample, including all valid person-years in the years indicated. All wages/income are in constant 2004 dollars.

a. Multinomial Logistic Regression Models 1–3 of Intended, Mistimed and Unwanted Births – Whites
	M1			M2			M3
	Intended	Mistimed	Unwanted	Intended	Mistimed	Unwanted	Intended	Mistimed	Unwanted
Age	.19	.04	.33	.18	.04	.34	.18	.06	.38
Age-squared	−.01*	.00	−.01	−.01*	.00	−.01	−.01*	.00	−.01
Predicted Education (< high school omitted)
High school	−6.86*	−5.68*	−4.70	−7.08*	−5.87*	−4.52	−6.95*	−5.78*	−4.25
Some college	−11.38*	−9.00*	−7.03	−11.54*	−9.16*	−6.86	−11.28*	−9.26*	−6.65
College	−26.10*	−19.89*	−10.47	−26.29*	−20.07*	−10.21	−25.71*	−20.03*	−9.94
High school X age	.47*	.41	.23	.49*	.42	.22	.48*	.42	.21
High school X age-squared	−.01*	−.01	.00	−.01*	−.01	.00	−.01*	−.01	.00
Some college X age	.75*	.62*	.31	.76*	.63*	.30	.74*	.64*	.30
Some college X age-squared	−.01*	−.01*	.00	−.01*	−.01*	.00	−.01*	−.01*	.00
College X age	1.62*	1.22*	.46	1.64*	1.23*	.45	1.60*	1.24*	.44
College X age-squared	−.02*	−.02*	−.01	−.02*	−.02*	−.01	−.02*	−.02*	−.01
Currently enrolled in school	−.58*	−.08	.35	−.58*	−.08	.35	−.57*	−.13	.30
Childbearing desires				.09*	.07*	−.11	.09*	.07*	−.11
Ln of current wage							.19*	−.10	−.07
Currently employed							−.09	−.29*	−.44*
Imputation flag	.14	.12	−.01	.13	.11	−.01	.11	.06	.15
Intercept	−3.54	−2.31	−8.08*	−3.61	−2.38	−7.97*	−3.94*	−2.31	−8.13*

b. Multinomial Logistic Regression Models 1–3 of Intended, Mistimed and Unwanted Births – Blacks
	M1			M2			M3
	Intended	Mistimed	Unwanted	Intended	Mistimed	Unwanted	Intended	Mistimed	Unwanted
Age	.06	.25	.01	.06	.25	.01	.06	.25	.03
Age-squared	.00	−.01*	.00	.00	−.01*	.00	.00*	−.01*	.00
Predicted Education (< high school omitted)
High school	−8.24*	1.65	−8.54*	−8.30*	1.57	−8.51*	−8.31	1.39	−8.80*
Some college	−9.03*	−1.45	−4.43	−9.11*	−1.55	−4.38	−9.06	−1.87	−4.67
College	−16.97*	−8.10	−5.66	−17.01*	−8.21	−5.62	−16.60	−8.72	−5.83
High school X age	.60*	−.18	.58*	.60*	−.18	.57*	.60	−.16	.60*
High school X age-squared	−.01*	.00	−.01*	−.01*	.00	−.01*	−.01*	.00	−.01*
Some college X age	.60*	.01	.23	.60*	.02	.23	.60	.04	.26
Some college X age-squared	−.01*	.00	.00	−.01*	.00	.00	−.01*	.00	.00
College X age	1.06*	.52	.20	1.06*	.53	.20	1.03	.57	.23
College X age-squared	−.02*	−.01	.00	−.02*	−.01	.00	−.02*	−.01	.00
Currently enrolled in school	−.26*	−.27	−.01	−.26*	−.26	−.01	−.25	−.29*	−.06
Childbearing desires				.06*	.05*	−.03	.06*	.05*	−.03
Ln of current wage							.19	−.25*	−.16
Currently employed							−.04	−.10	−.26*
Imputation flag	.18	−.08	−.11	.18	−.07	−.11	.32	−.02	.04
Intercept	−1.95	−3.98	−2.19	−2.01	−4.01	−2.16	−2.47	−3.52	−2.04

Notes: An imputation flag is included for person-years in which any time-varying variable is imputed using data from the previous year.

^*p < .05