Abstract
Unplanned pregnancies in the U.S. disproportionately occur among poor, less educated, and minority women, but it is unclear whether poverty following a birth is itself an outcome of this pregnancy planning status. Using the National Longitudinal Survey of Youth 1997 (n=2,101) and National Survey of Family Growth (n=778), we constructed two-year sequences of contraceptive use before a birth that signal an unplanned versus a planned birth. We regressed poverty in the year of the birth both on this contraceptive-sequence variable and on sociodemographic indicators including previous employment and poverty status in the year before the birth, race/ethnicity, education, partnership status, birth order, and family background. Compared to sequences indicating a planned birth, sequences of inconsistent use and non-use of contraception were associated with a higher likelihood of poverty following a birth, both before and after controlling for sociodemographic variables, and before and after additionally controlling for poverty status before the birth. In pooled-survey estimates with all controls included, having not used contraception consistently is associated with a 42% higher odds of poverty after birth. The positive association of poverty after birth with contraceptive inconsistency or non-use, however, is limited to women with low to medium educational attainment. These findings encourage further exploration into relationships between contraceptive access and behavior and subsequent adverse outcomes for the mother and her children.
Keywords: Fertility, Pregnancy Intentions, Newborn Poverty, Contraceptive Consistency, Education
Introduction
In a recent special journal issue exploring policy proposals to reduce poverty, Wu and Mark (2018) advocate for a policy that would provide all women with free access to the most effective contraceptives, namely long-acting reversible contraceptives (LARC). In quasi-experimental studies, greater access to effective contraception and to abortion has been found to reduce the likelihood of poverty. Bailey (2013) and Browne and LaLumia (2014) found that access to the contraceptive Pill in the 1960’s and 1970’s was associated with lower poverty rates and higher household incomes. Bailey, moreover, found that these effects arose to some extent through reductions in mistimed and unwanted pregnancy. Bailey, Malkova, and McLaren (2018) additionally found that the expansion of Title X programs for low-income women’s family planning in the 1960’s and 1970’s reduced their poverty rates. Finally, having access to abortion has been found to reduce substantially the likelihood of poverty among pregnant, disadvantaged women (Foster et al. 2018).
Underlying these studies and policy prescriptions is the premise that when pregnancies occur unexpectedly, women are likely to be less financially prepared for the birth. We are unaware, however, of any study that has explicitly investigated associations between consistency of contraceptive use and income and poverty following a birth. In the present study, we evaluate links between the planned versus unplanned status of a birth, as inferred from the mother’s pattern of contraceptive use in the two years before the year of the birth, and the poverty status of the family immediately after the birth. We incorporate into our multivariate analyses standard sociodemographic variables plus indicators of whether the woman had stable employment before the birth, her family structure as a teenager, and her poverty status in the year before the birth. Our analyses also include models that allow for the association between contraceptive consistency and poverty after birth to differ by the woman’s educational attainment.
We first review the existing literature on contraceptive access, attitudes, and behavior, pregnancy intentions, and poverty outcomes. We then describe our use of nationally-representative data in two surveys, the National Longitudinal Survey of Youth 1997 (NLSY97) and National Survey of Family Growth (NSFG). We use the two data sources separately and together in pooled-survey estimation. Both the NLSY97 and NSFG questions allow for the constructing of sequences of contraceptive behavior in the two years before the year of the birth in a way that signals whether the birth was unplanned versus planned. Both the NLSY97 and NSFG include measures of poverty-status following a birth, but the NLSY97 alone additionally includes poverty status before the birth. This is a critical variable for our analyses, intended to capture unobserved characteristics that raise the woman’s likelihood of poverty after a birth independently of her contraceptive use or non-use. We use cross-survey multiple imputation (Rendall et al. 2013) to combine these two survey samples to produce pooled-survey estimates of poverty after a birth in a model that accounts for poverty status before the birth in addition to contraceptive-use sequence and sociodemographic characteristics. To anticipate our main results, we find that, compared to women whose patterns of consistent contraceptive use indicate the birth was planned, women with inconsistent use or non-use of contraception have higher poverty rates following a birth, both before and after controlling for sociodemographic characteristics and after additionally controlling for poverty status before the birth. We find that this result holds for women with no more than a high school education or some college, but does not hold for women with a bachelor’s degree. We interpret this as preliminary descriptive evidence consistent with a causal impact of unplanned births on poverty after the birth among women whose socioeconomic characteristics already put them at higher risk of poverty.
Background
Ability to have a planned birth depends on both access to effective contraception and consistent use of contraception. We know of just two studies assessing explicitly the connection between contraceptive access and poverty following a birth, and no studies on consistency of contraceptive use and poverty following a birth. Using pooled U.S. Census data, Browne and LaLumia (2014) found that if a woman lived in a state that had early access to the Pill in the 1970’s, her probability of being in poverty would decrease 0.5%. Bailey, Malkova, and McLaren (2018) analyzed the impact of “Title X” federal family planning programs in the 1960’s and 1970’s. The purpose of these programs was to increase the access to contraceptives, education, and counseling for disadvantaged women. Bailey et al. compared poverty levels of children both in the same counties before and five years after the implementation of the programs. Children born after the introduction of the programs were 7% less likely to live in poor households. The researchers suggest that the leading mechanism explaining the policy impact was the increase in income of disadvantaged parents who avoided unintended pregnancy. These parents would have chances to improve their human capital and socioeconomic conditions before a birth. The selection effect (decreases in births among poor parents) was less significant in explaining the effects of lower poverty. Bailey (2013) used geographic variation in the introduction of the Pill in the 1960’s and 1970’s as a quasi-experiment, finding positive outcomes including increased labor market access, earnings, and household income among parents. Bailey also found that in states where the Pill was permitted, fewer women labeled their births as unwanted and mistimed. We are not aware of any studies specifically of associations between contraceptive access and poverty in periods since the 1970s.
One other relevant study, by Foster et al. (2018), assesses the connection between abortion access and poverty following a birth. They analyzed the time sequences of poverty status of pregnant women presenting at 30 clinics nationwide for an abortion, comparing those women whose gestational duration was within the limit versus just over the limit that allowed her access to abortion at that clinic, with the result in the latter case that she was turned away from the clinic. They found that the poverty rate 6 months later of women who were turned away from the clinic was on average 16 percentage points higher than the poverty rate of women who received their requested abortion at the clinic (61% versus 45%). The difference in poverty rates between these two groups continued to be large at least two years on.
A few studies using recent U.S. data have analyzed more generally the correlates of family poverty immediately following a birth (McKernan and Ratcliffe 2005, Lichter et al. 2018, Thiede, Sanders, and Lichter 2018). McKernan and Ratcliffe (2005) measured poverty before and after a birth and found that having a birth increased the likelihood of poverty by 30%. Foster et al. (2018) found that the increase in household size occurring with a birth was the primary reason for a woman’s being poor after a birth. All of these studies have found significantly higher rates of poverty among women whose births occurred in a context of socioeconomic and family disadvantage. An already socioeconomically vulnerable woman may be especially likely to be economically penalized for an unplanned birth. In-depth interviews of 65 men and women by Kavanaugh et al. (2017) found that respondents felt financially stressed by the addition of a child, the biggest issue stemming from trouble finding childcare.
Pregnancy Planning, Pregnancy Intentions, and Contraceptive Consistency
In reviewing the two concepts of pregnancy intention and pregnancy planning, Klerman (2000) describes the concept of “unplanned” as being manifested in a woman’s contraceptive-use behavior, including not only non-use of contraception but also whether she used a method incorrectly or inconsistently. She finds it “…puzzling that so much more attention seems to be given to…unintended pregnancies compared to…unplanned ones” (p.159). Although the present study focuses on the relationship between contraceptive consistency as a representation of pregnancy planning, we acknowledge that pregnancy planning is difficult to measure. This may reflect complexity and ambivalence underlying women’s pregnancy intentions (Bachrach and Morgan 2013). Thoughts, emotions, and other cognitive processes may not always line up exactly with behavior. Jones (2017) found that a sizable fraction of women both said it was important to avoid pregnancy, but also said they would be happy if they became pregnant; these women were nonetheless more likely to take pregnancy avoidance measures. Having the intention to avoid pregnancy is a consistently strong predictor of contraception use (Bartz et al. 2007, Moreau et al. 2013). Ambivalence about pregnancy, however, may nevertheless present itself through inconsistent use of contraception (Zabin 1999).
Women’s available contraceptive choices and their contraceptive attitudes and behavior are also dependent on broader societal and institutional forces associated with socioeconomic status. This also needs to be carefully accounted for when examining the relationship between contraceptive consistency and subsequent poverty. The theoretical framing of contraceptive behavior as part of pregnancy planning within social class context is developed by England and colleagues (England 2016, England et al. 2016). They describe “efficacy” (planfulness, self-regulation, assertiveness, and beliefs in one’s ability to carry out behaviors that will help realize one’s goals) as being formed through social class-dependent socialization processes. Efficacy is not socialized independently of class; agency and efficacy are driven by confidence in oneself and confidence in institutions (England 2016). This is often instilled from birth. Applying their theoretical framework to contraceptive use, England et al. (2016) find that women who had high efficacy were five to eight times more likely to use contraceptives consistently. Young women’s economic prospects may also serve as a motivator for their contraceptive behavior. Using longitudinal data on 18 to 19 year old women for two and a half years in the Relationship Dynamics and Social Life Study (RDSL), Wu et al. (2016) found that having a GPA of 3.0 or higher and having a job were both associated with being more likely to use contraceptives consistently. Having a high GPA can mean that young women see higher education as a possibility and having a job can imply a chance for a financially stable future.
Current Study
We aim to investigate the role of consistency of contraceptive use, as a proxy for pregnancy planning, in determining the likelihood of poverty after a birth. Using data from two nationally representative surveys covering births in the 2000s through early 2010s, we examine associations of contraceptive consistency in the two years preceding the year in which a woman gives birth with poverty in the year of the birth. We first estimate this association for all women giving birth, controlling for their socioeconomic background and current family status, and alternately before and after controlling for poverty status before the birth. We then allow the association of contraceptive consistency with poverty in the year of the birth to apply differentially for women at low, medium, and high levels of educational attainment. The goal of this study is to begin to shape a story of how persistent disadvantage can be exacerbated and reproduced with an unplanned pregnancy and birth. In a country such as the United States where public assistance for families is low (Brady and Burroway 2012), an unplanned birth may put women and their newborn children in a uniquely vulnerable position. We test the following hypotheses:
H1: A woman whose contraceptive-use indicates the birth was planned will be less likely to be in poverty after a birth than will a woman whose contraceptive use was inconsistent, holding constant her socioeconomic characteristics and poverty status before the birth.
H2: The association between consistency of contraceptive use and poverty after a birth will be strongest among women with the greatest pre-birth socioeconomic disadvantage, as represented by low educational attainment.
Data and Methods
Data
Data for this study come from U.S. women who reported giving birth between panel waves through to the final annually-spaced wave in 2011 in the National Longitudinal Survey of Youth 1997 (NLSY97) and from women who reported giving birth in the year before survey in the National Survey of Family Growth (NSFG), 2002 and 2006-2010 cycles. These two surveys have the major strengths of collecting both contraceptive-use histories and poverty status, and covering approximately the same time period. The NSFG collects more detailed contraceptive histories than does the NLSY97. Only the NSLY97, however, collects poverty status across multiple time points (annually). Combining the two data sources has three advantages. First, it allows for replication of our main results of differences in poverty in the year of the birth by contraceptive consistency before the birth across two data sources. The NSFG, moreover, allows for an alternative, more detailed specification of our contraceptive-consistency explanatory variable. Second, it allows for a regression specification that includes a crucial control variable of poverty status in the year preceding the birth. Third, with the application of cross-survey multiple imputation (described below), our combined-survey approach allows for estimation of the fuller regression specification using a sample size of the two surveys combined, thereby providing more precise estimates of the relationship between contraceptive consistency and poverty after giving birth.
Starting in 1997, the NLSY97 (Bureau of Labor Statistics 2017) interviewed a nationally representative sample of 8,984 individuals who were between ages 12 and 16. 92.1% of eligible respondents completed the first round, 1997, interview. Black and/or Hispanic and Latino populations were oversampled. Respondents were interviewed annually until 2011 and biennially since. Approximately 83% of the 1997 sample was interviewed in 2011. The sample includes all women who gave birth having been sexually active in one or more of the years preceding the 1999 through 2011 survey waves and who had non-missing values on the outcome and predictor variables. Because we build the contraceptive-consistency and birth sequence from three consecutive years of data, we are not able to use data after the biennial interview schedule began after 2011. We need two years of data before a birth to observe contraceptive use, therefore, the first year we can use to observe a birth is the 1999 wave (unless birth happened in 1998 wave and respondent reported having sex for the first time in 1997 wave). The women in our analytic sample are interviewed at ages immediately after the birth that range from age 17 to 31. A woman is represented as a separate observation at each birth that meets these criteria. We exclude births that were the result of contraceptive failure during consistent contraceptive use, which is defined as using contraceptives 100% of the time in the year before birth among women who gave birth in any of the first nine months of the year of birth exposure (discussed in more detail below).
The NSFG is a cross-sectional representative survey of the household population of women between 15 and 44 years old, and is conducted by the National Center for Health Statistics (Center for Health Statistics 2003, 2011). The 2002 cycle used a periodic strategy (each sample was interviewed in the course of one year), whereas since 2006, the NSFG has followed a continuous interviewing design, with the 2006-2010 period constituting the first cycle under this new, continuous design. Women in the NSFG are interviewed on topics including individual history of family life, marriage and divorce, pregnancy, infertility, use of contraception, and general and reproductive health. The questions are answered by in-person interview, but more sensitive questions are answered privately by self-administration. Hispanic, Black, and teen women were oversampled. The overall response rate was 79% in the 2002 cycle and 78% in the 2006-2010 cycle. We select women who had a birth in the year before the interview, since poverty is assessed in only the year before interview in the NSFG. To conduct our analysis on a sample of women approximately comparable to those in the NLSY97, we use data from the female respondents between ages 17 and 31 years old.
Measures
Contraceptive Consistency
As noted above, our measures of contraceptive consistency preceding a birth require three consecutive years of observation. The third year of observation, denoted (t-1,t), is the year in which the birth occurred. The two preceding years, (t-3,t-2) and (t-2,t-1), are those in which a woman’s contraceptive use is observed. In addition, if a woman gives birth in one of the last two months of (t-1,t), we also use observed contraception in the first months of year (t-1,t). In general, we assume that year (t-2,t-1) was the year of exposure to either a planned or unplanned conception. The two surveys capture contraceptive use differently. The NSFG questions are monthly based, whereas the NSLY97 questions are yearly and sexual-episode based. Also, whereas the NSFG’s respondents need to recall their contraceptive use in the last three years, the NSLY97’s respondents report in each annual wave on their contraceptive use in the previous year. We tested the comparability of both measures for a larger sample (women 17-35 years sexually active, years 2001-2015), not restricted to women with recent births, and found no statistically significant differences between the surveys in terms of annual contraceptive use/non-use (results not shown).
Contracepting consistently is defined as using contraception at every sexual encounter in the case of the NLSY97, or at every sexually-active month in the case of the NSFG. The first step in defining our multi-year categories of contraceptive consistency prior to the birth is to divide women into three categories of contraceptive use in a given year: 0% contraceptive use is non-use; 1-99% is inconsistent use; and 100% is consistent use. This strict definition of “consistent use” matches that used by Manlove et al. (2007) and Sipsma and Ickovics (2015). We use up to two years before the year in which the birth occurred to observe contraceptive consistency, conditional on the women having had sex for the first time. Using the NSFG, Glei (1999) showed that long-term contraceptive non-users were at most risk for an unintended birth, followed by inconsistent users, and then least at risk were effective users. Following this framework, we classified women into never-consistent and ever-consistent contraceptors in the two years prior to the birth year. Additionally, we define as non-use, married those women who do not use contraceptives at all in both (t-3, t-2) and (t-2, t-1) but are married in (t-1, t). These women do not fit in perfectly with the ‘planners’; they also however do not fit well with the non-planners, as married women have some of the lowest rates of unintended pregnancies (Finer and Zolna 2016).
The ever-consistent women are those who in the year that is two years before the birth year use contraceptives 100% of the time. In the year before birth year, their use of contraceptives will generally be less than 100% and possibly not at all, indicating they stopped using contraception in that year with the intention of becoming pregnant. These women represent the highest level of planning in our grouping, similarly to Glei’s sample of effective users who most often avoided an unintended birth. Never-consistent women are those with any combination of non-use and inconsistent use in the two years before the birth year. In the NLSY97, using contraceptives 100% of the time means the woman uses contraceptives at every sexual encounter, whereas in the NSFG this means that a woman used contraceptives in 100% of the months that she was sexually active. This difference in definitions would be expected to make the ever-consistent category more difficult to achieve in the NLSY97 measure than in the NSFG measure. In the weighted distributions presented in Table 1, we indeed find this to be the case: 46.3% are ever-consistent in the NSFG versus 39.8% in the NLSY97. These are nevertheless relatively small differences. To further test for the sensitivity of our analyses to the stricter NLSY97 criterion, we relaxed the NLSY97’s definition of consistent use to 90-100% of sexual episodes and found this did not substantially affect our results (results not shown).
Table 1:
Characteristics of women giving birth between ages 17 and 31, 1999 to 2011, proportions
| NSFG | NLSY97 | NSFG+NLSY97 | |||
|---|---|---|---|---|---|
| Model 1&2 | Model 2a | Model 1, 2, & 3 | χ^2 | Model 1&2 | |
| Education | ** | ||||
| High School or Less | 0.572 | 0.553 | 0.558 | ||
| Some College | 0.184 | 0.253 | 0.234 | ||
| Bachelors Degree + | 0.244 | 0.194 | 0.208 | ||
| Race/Ethnicity | ** | ||||
| White | 0.536 | 0.713 | 0.664 | ||
| Black | 0.162 | 0.153 | 0.155 | ||
| Hispanic | 0.229 | 0.121 | 0.151 | ||
| Other | 0.073 | 0.013 | 0.030 | ||
| Age | 25.3 | 24.5 | 24.7 | ||
| Partnership Status | |||||
| No Resident Partner | 0.192 | 0.207 | 0.203 | ||
| Cohabiting | 0.206 | 0.216 | 0.213 | ||
| Married | 0.602 | 0.577 | 0.584 | ||
| Birth Order | ** | ||||
| One | 0.423 | 0.469 | 0.456 | ||
| Two | 0.318 | 0.338 | 0.332 | ||
| Three Plus | 0.259 | 0.193 | 0.212 | ||
| Lived with biological parents at 18 years old | 0.437 | 0.417 | 0.423 | ||
| Mother's Education | * | ||||
| High School or Less | 0.609 | 0.602 | 0.604 | ||
| Some College | 0.223 | 0.277 | 0.262 | ||
| Bachelors Degree + | 0.168 | 0.121 | 0.134 | ||
| Had 6+ months of full-time employment | 0.679 | 0.821 | ** | 0.782 | |
| Contraceptive Consistency | * | ||||
| Ever Consistent | 0.463 | 0.398 | 0.416 | ||
| Sometimes Consistent | 0.263 | - | - | ||
| Always Consistent | 0.233 | - | - | ||
| Never Consistent | 0.436 | 0.403 | 0.508 | 0.488 | |
| Non-Use Married | 0.101 | 0.101 | 0.094 | 0.096 | |
| Poor at Birth | 0.332 | 0.228 | ** | 0.257 | |
| Poor in year immediately prior to birth | - | 0.174 | - | ||
| Unweighted N | 778 | 2,101 | 2,879 | ||
Notes: All proportions are weighted
Group differences from chi-squared (NLSY97 vs. NSFG)
p<0.01
p<0.05
p<0.1
Sources: National Longitudinal Survey of Youth 1997 (NLSY97) and the National Survey of Family Growth 2002, 2006-2010
In the NLSY97, our annual contraceptive-use measure is built using several questions in the Self-Administered Questionnaire. First, the respondent was asked how many times in the last year have they had sexual intercourse. Next, the respondent was asked how many times out of all the times that they have had sexual intercourse in the last year did the respondent use any method of birth control. In 2002 to 2011, before being asked about general birth control, the respondent was first asked how many times they had used condoms. If the respondent stated using condoms 100% of the time, they were not asked further about other birth control method use. To obtain a percentage, we divided the number of times a woman used birth control by the times she has had sex in the last year. To minimize missing data, we also took advantage of questions used as follow-ups if the respondent did not know how often they used birth control (1998 and onward). The follow-up questions allowed for a respondent to give a percentage, 0 to 100%, as an estimate of how many times they used birth control in the last year.
Based only on contraceptive-use information in the year before the year of the birth, about 13% of the births to women in the NLSY97 would appear to be attributable to contraceptive failure. That is, the woman reported 100% contraceptive use before the year of birth (results not shown). This is comparable to the national average of 10-12% (Sundaram et al. 2017). However, we want to minimize the possibility that we are over-identifying contraceptive failures due to the year-to-year observations: births occurring in the last three months of year (t-1,t) that may have been conceived in one of the first three months of year (t-1,t), following 100% contraceptive use in year (t-2,t-1). To address this issue, we code as ever-consistent, women who are consistent contraceptive users (100%) in (t-2, t-1) but have their birth in the last two months of (t-1, t). This will include in our analytic sample women who may have had at least two to three months of intentional non-use before getting pregnant early in the year and who went on to give birth in the last two months of the (t-1, t). Similarly, we include women who had sex for the first time in the year before the birth, reported consistent contraceptive use in the year before the birth, and had a birth in the last two months of (t-1,t). Distributions of the NLSY97 contraceptive sequences making up each multi-year prior contraceptive-use category are shown in Table A1 of the Appendix.
The NSFG collects information of monthly sexual intercourse and contraceptive method used for the time period from the January three years prior to the interview date (National Center for Health Statistics 2003, 2011). For example, for interviews conducted in August 2010, information covers from January 2007 through August 2010. The number of months reported for each respondent depends on the date (month/year) of her interview, but ensures a window of at least three years of contraceptive questions for each respondent. Restricted to women who declare at least one experience of sexual intercourse with a man, the survey collects monthly information for this three-year period on whether they had sexual intercourse and on the contraceptive methods used. The woman’s annual measure of contraceptive use is then whether she used contraceptives 0% of the months, 1%-99% of the months, or 100% of the months. We combined information on annual contraceptive use in the two years before the year of the birth into the three categories of never-consistent, ever consistent, and non-use, married in the same way as for the NLSY97. As with the NLSY97, we also use information on contraceptive use in the first months of year (t-1,t) to code as ever-consistent, women who gave birth in the last two months of (t-1, t) and reported consistent use in year (t-2,t-1). Distributions of the NSFG annual sequences making up each multi-year prior contraceptive-use category are found in Appendix Table A2.
Although with the NLSY97 contraceptive questions referring to an annual period, we choose to define contraceptive consistency in the NSFG also on an annual basis for our main analyses, we also took advantage of the NSFG’s monthly data to conduct a check on the validity of our analyses that are based on this annual coding. For this check, we refined our ever-consistent versus never-consistent categorization into a three-way categorization of always-consistent, sometimes-consistent, and never-consistent (see Appendix A3). The monthly-based always-consistent measure allows us to develop are more refined identification of women who are planning a pregnancy than the annually-based ever-consistent measure. There may be a possibility that women who are always-consistent users are stronger “planners” than those who seem to oscillate between consistent and inconsistent use before getting pregnant.
In the monthly-based measure, we again analyze sequences of contraceptive use in the months when the interviewees reported having had sexual intercourse, beginning with the first month of the (t-3,t-2) year, and including the sexually-active months of the (t-2,t-1) and (t-1,t) year up to the month of conception (which may be either in the (t-2,t-1) or (t-1,t) year). We define women as always-consistent contraceptive users when they first have a sequence of contraceptive use for at least 6 consecutive months in which they were sexually active, followed by a sequence of non-use for every month up to and including the month of conception. The initial sequence of contraceptive use begins either in the first month of (t-3, t-2) or in the first subsequent month in which they were sexually active. A second category, “sometimes-consistent” users, are defined as women who use contraceptives for at least the first 6 months, but whose subsequent sequence of months, beginning with the first month of nonuse and ending in conceptive, included at least one month of contraceptive use. That is, they transitioned from a pattern of consistent contraceptive use to a pattern of inconsistent use. The third monthly-based category, which we again refer to as never-consistent, includes at least one month of non-use in the first 6 months of (t-3, t-2).
Note that our contraceptive-consistency variable does not explicitly consider the contraceptive method(s) used. To check that we did not overlook the relationship between socioeconomic status and contraceptive method, and therefore potentially an additional important dimension of contraceptive consistency, in analyses not reported here we controlled for the use of moderately-effective or LARC methods versus less-effective methods of contraception. We did not find a significant difference between ever-consistent users of moderately-effective or LARC methods and ever-consistent users of less-effective methods, when contrasted with never-consistent contraceptive use.
We also note that, because of data limitations (Lindberg et al. 2020), we investigate consistency of contraceptive use only for pregnancies carried to term. We note in this context the findings of Foster et al. (2018) that disadvantaged women who tried unsuccessfully to get an abortion were more likely to face economic hardship after a birth. This has relevance to our Hypothesis 2, to the extent that higher education women may be more likely to get an abortion because they often have more access to those services (Reeves and Venator 2015), and thereby avoid adverse economic consequences of getting pregnant because of inconsistent contraceptive use.
Poverty
Our measure of poverty is similar to that used in the official poverty rate, in which a family resources numerator is matched to a poverty threshold denominator, but improves upon it in two important ways. First, in both our NLSY97 and NSFG measures, cohabitors are included as members of the family unit. Doing so follows the recommendation of the National Academy of Sciences review (National Research Council 1995), subsequently used in the Census Bureau’s Supplementary Poverty Measure (Short 2011; Fox et al. 2015), and reduces substantially the poverty rates estimated for cohabiting-couple families compared to those estimated using the official poverty measure (Iceland 2000). The family income reported in the NSFG in both the 2002 and 2006–2010 cycles includes all relatives as well as a resident nonmarital partner. Rounds 1–7 (1997–2003) of the NLSY97 collected income information on all members of the household. After 2003, due to change of question wording, income information was only collected for family members, including resident nonmarital partners. The poverty status measure therefore reflects a household level from 1998–2003 and a family level from 2004–2011.
Second, in the NLSY97, though not in the NSFG, we include the cash value of in-kind benefits from government programs, specifically the Food Stamps/SNAP and Women, Infants, and Children (WIC) nutritional assistance programs. In the NLSY97, respondents are asked for the approximate cash value of resources from SNAP/Food Stamps, WIC, and Other Welfare, and these are incorporated into the “gross household income” amount used in the calculation of the income-to-poverty ratio in the NLSY97, together with income from wages, child support, interest, stocks, and other assets (Bureau of Labor Statistics, nd). The NSFG instead prompts the respondent to consider income components “including income from all the sources you just went through, such as wages, salaries, Social Security or retirement benefits, help from relatives, and so forth. Please enter the amount before taxes” (National Center for Health Statistics, nd). Receipt of welfare or public assistance is asked of NSFG respondents only following the “total family income” question, and without the respondent being asked for dollar values. In part as a result of the fuller inclusion of resources from government programs in the NLSY97, the poverty rates estimated in our study for women who just gave birth are lower than those estimated for women who just gave birth in the NSFG. We handle this discrepancy in poverty levels in our regression analyses by incorporating an “NSFG Survey” indicator (see Analyses subsection below). We show that, as expected, this coefficient is positive (higher likelihood of poverty in the year of birth when this is measured in the NSFG than in the NLSY), and is statistically significant. When we predict poverty rates, we predict them at levels for the reference-category, NLSY97 women.
In neither the NLSY97 nor the NSFG data are there variables for the value of family-based tax rebates such as the Earned Income Tax Credit. Wimer et al. (2016) found for the period of our study (the decade of the 2000s) that addition of estimated cash value of tax rebates and in-kind benefits and subtraction of expenses for work, childcare, and medical care, together with defining cohabitors as part of the family unit, impacted substantially the trend, but not the overall levels of poverty rates of the households with children under 5 years old, as compared to those of the official poverty measure. A consideration additional to the definition of what belongs in a poverty measure, however, is underreporting. Household surveys have been shown to underreport information on income (Hurst et al. 2014) and in particular, income from welfare programs (Meyer et al. 2015). There are no obvious solutions to this problem, but it should be noted as a factor inducing upward bias in poverty measurement in our study.
We take advantage of the panel character of the NLSY97 to incorporate poverty status both in the year of the birth and in the year before, with poverty in the year before serving as a lagged dependent variable. Because of item non-response on some or all of the income-component questions, 12.5% of our NLSY97 sample has missing poverty status in the year of the birth, and an additional 14.9% of the sample has missing values for poverty status in the year before1. We exclude from our analyses the 12.5% without a valid poverty status on the outcome variable (poverty status in the year of the birth), but include those cases in which poverty status is missing only in the year before the birth year. We do this by using multiple imputation (Little and Rubin 2002). The imputation equation is estimated from the “complete” NLSY97 cases that include poverty status both in the year of the birth and the year before. Poverty status in the year before is then multiply-imputed to the 14.9% of sample women who constitute the “incomplete” cases in which poverty status is observed in the year of the birth but not in the year before. This procedure mirrors that used on the NSFG, for which all cases are “incomplete” with respect to poverty status in the year before the birth year, using a logistic model appropriate to imputation for monotone missingness (see Analyses subsection below). Alternative procedures to impute missing data on other covariates or on poverty in the year of the birth would involve imputation for arbitrary missingness patterns, thus engaging the additional assumptions needed for a multivariate normal approximation of the joint distribution imputation (Schafer and Graham 2002). Our conservative approach of limited imputation for item non-response, moreover, is guided by findings that when data are missing not at random (MNAR), listwise deletion may result in lower bias than multiple imputation (Pepinsky 2018).
Social, economic, and demographic characteristics
In order to best isolate the effects of contraceptive use on poverty after birth, we control for other possible predictors of poverty and contraceptive-use influencers. Own education, mother’s education, birth order, age, and partnership status in the model are measured at the end of the year of the birth. A mother’s low education can indicate a higher chance of a respondent growing up in an economically disadvantaged household, and therefore having a higher chance of being vulnerable to poverty herself (Hoynes, Page, and Stevens 2006; Kaushal 2014). The NSFG and the NLSY97 do not provide sufficient data to code fathers’ education. Several studies have shown that going through family-structure disruptions is associated with adverse life outcomes (McLanahan and Percheski 2008; Rosenfeld 2015; McLanahan and Jacobsen 2015). Therefore, we include a measure of family structure during youth. We used the best-matched variable between the NLSY and NSFG, which was whether the respondent lived with both biological parents at 18 years old.
We include a variable for presence of having ever had a spell of at least 6 months of continuous full-time employment before the birth, coded from a single question in the NSFG and from detailed employment histories in the NLSY97. Alon, Donahoe, and Tienda (2001) and Lu et al. (2017) have shown that when women have an established work record, they tend to stay on that trajectory. Rendall and Shattuck (2019) show more specifically that women's achieving stable employment before becoming a mother is associated with a higher employment probability in years following a birth. Active participation in the labor market, not surprisingly, decreases the chances of a child being in poverty (Chen and Corak 2008).
Analyses
We estimate three regression models, for the NSFG and NLSY97 samples separately and then in pooled-survey analyses. In the descriptive statistics and regression models, we adjust for the complex survey design, including the use of each survey’s sampling weights that are first normalized to have a mean of 1 in each of the NLSY97 and the NSFG 2002 and NSFG 2006-10 cycles (following Rendall et al. 2008). We follow guidance from the NLSY97 and NSFG documentation to incorporate stratum and sampling cluster (PSU) in all our variance estimation of the descriptive statistics and analysis equations. Incorporating clustering by PSU, however, means that we are unable to additionally include NLSY97 within-woman clustering in the case of multiple observed births to the same woman between 1999 and 2011. In prioritizing adjustment for clustering within PSU, we follow the guidance of Cameron and Miller (2015, p.333) who recommend incorporating clustering at the highest level of aggregation when more than one level of clustering is present in the data.2
The first model (Model 1) includes only the contraceptive-consistency variable among the regressors. Model 2 includes contraceptive consistency plus the sociodemographic variables, but not pre-birth poverty status. Model 2a for the NSFG-only also includes the separation of the contraceptive consistency variable into never-consistent, sometimes-consistent, always-consistent, and non-use married, with always-consistent as the reference category. Model 3 in the NLSY97 adds pre-birth poverty status. This is our preferred specification, as pre-birth poverty status is expected to capture unobserved characteristics that raise the woman’s likelihood of poverty after the birth independently of her contraceptive use. Pre-birth poverty status, however, is available only in the NLSY97. We used cross-survey multiple imputation (MI) to impute values of this variable from the NLSY97 to every observation in the NSFG. The statistical theory of cross-survey MI is developed in Gelman et al. (1998) and Rendall et al. (2013). Applications in social demography include Baker et al. (2015) and Capps et al. (2018). Like within-survey MI for item non-response, cross-survey MI accounts for increases in variance of the estimates induced by including imputed values. It has a major advantage over within-survey MI, however, of more easily satisfying the missing at random (MAR) assumption needed for unbiased MI, since the reason the value on the variable is missing is that the respondent was randomly sampled into the survey that does not include the question.
We also perform within-survey multiple imputation of pre-birth poverty status for women in the NLSY97 who do not provide that information, thus further increasing the sample size. Because the variable imputed serves as a control variable in our analysis, conducting within-survey and cross-survey MI substantially reduces sampling variability about our main estimate, of the relationship of contraceptive consistency on poverty after birth, after appropriately accounting for increases in variance induced by imputing poverty before the birth for about two fifths of the observations. Because we are also then able to use our preferred regression model specification for a sample that includes both NSFG and NLSY97 observations, cross-survey multiple imputation method overcomes what would otherwise be potentially substantial omitted-variable bias if the less preferred specification of only variables in common between the NLSY97 and NSFG (that is, the Model 2 specification) were used in our pooled-survey estimation.
Because our estimation combines observations from two nationally representative surveys of approximately the same ages and years, we begin by assuming that they sample from a common social process except for a potential difference in levels of the outcome variable (postbirth poverty status). We test the validity of this assumption by conducting diagnostics under a model-fitting framework, following Rendall et al. (2013), and used also in Baker et al. (2015) and Capps et al. (2018). As we show below in the Results section, we find model-fit improvement only when adding an intercept shift variable for overall post-birth poverty rate differences between the surveys, and not when adding a full set of covariate interactions with “survey.” Model fit improvement in the latter case would be evidence calling into question the appropriateness of a pooled-survey method.
One of the strengths of the multiple imputation method is its separation of the imputation step from the estimation of the analysis model. This allows for greater flexibility in the implementation of the analysis model. For example, the complex sample designs of both surveys are handled easily using the SAS SURVEYLOGISTIC procedure in the analysis model. The standard errors are adjusted using the primary sampling unit and stratum in both the NLSY97 and the NSFG. We are able to implement both the imputation and analysis steps using statistical package software from SAS Version 9.4, respectively using the PROC MI (with the MONOTONE LOGISTIC option) and PROC MIANALYZE commands. Estimates are weighted in the analysis model (but not in the imputation model) using survey weights provided by the respective (NSFG and NLSY97) data producers to account for oversampling, differential non-response and attrition. PROC MIANALYZE allows us to estimate 95% confidence intervals, standard errors, and significance tests that are adjusted for the additional uncertainty introduced by the cross-survey imputation process. We follow Ratitch, Lipkovich, and O’Kelly (2013) in calculating the imputation-adjusted 95% confidence intervals around the odds ratios after first using PROC MIANALYZE to calculate the imputation-adjusted standard errors.
For our second hypothesis, we use the pooled data to estimate the interaction effects between own education and contraceptive consistency. Our purpose is to allow the difference in poverty between women with never-consistent versus ever-consistent contraceptive use to vary by their educational attainment. For this, we use the Model 2 specification that does not require that poverty before the birth is first multiply imputed from the NSLY97 to the NSFG. Using the margins command in Stata, we obtain the predicted probabilities and 95% confidence intervals of being in poverty after a birth for women in each combination of education and contraceptive use. To test for statistical significance between never-consistent and ever-consistent for each education group, we use the Stata lincom command.
Results
In the weighted univariate distributions presented in Table 1, we show that the NSFG and the NLSY97 compare reasonably well on most of the sociodemographic predictors. The NSFG education distribution has slightly more women having a bachelor’s degree or more, at 24.4% compared to 19.4% in the NLSY97. The NLSY97 race/ethnic distributions have notably more (non-Hispanic) White women and fewer Hispanic women than the NSFG. This reflects changes in the U.S. population composition between the 1997 year of the NLSY’s sampling frame and the 2002 and 2006-2010 years of the NSFG’s sampling frames. The NLSY97 and the NSFG are similar in their partnership status distribution and mean age. The birth-order distributions are substantively similar between the NSFG and NLSY97, with first births respectively accounting for 42.3% and 46.9% of births. Among family background variables, differences in distributions of mother’s education parallel the differences of distributions seen in the mother of the newborn’s own education, with more high school educated women in the NSFG, and more with some college in the NLSY97. The proportion of women who had already experienced a continuous period of 6 or months full-time employment is substantially higher when estimated from the NLSY97 data (82.1%) than when estimated from the NSFG data (67.9%). This will be partly due to the higher proportion Hispanic in the NSFG, a race/ethnic group previously shown to experience lower proportions achieving stable employment before a first birth (Shattuck and Rendall 2017).
The percent of women in poverty after a birth in the NSFG is 33.2%, which is substantially and statistically significantly higher than the NLSY97 (22.8%). This difference is consistent in direction with differences between the surveys on some sociodemographic variables, notably the higher percentages of disadvantaged minorities (Hispanic and Black) in the NSFG than in the NLSY97, and the higher percentage of women with 6 or more months of full time employment before the birth in the NLSY97. However, in the multivariate models that control for these variables as described below, we find that poverty is still substantially more likely to be experienced in the year of the birth in the NSFG than in the NLSY97. Survey measurement differences therefore also are likely to play a significant role. These may include the NLSY97’s using household poverty (which included non-family members as well) for the first seven rounds and its asking in more detail than in the NSFG the different potential sources of household or family income. Also, the NLSY97 offers the respondent to include a more comprehensive list of income sources (cash and non-cash) than the NSFG, which may contribute to the lower poverty levels for women in the NLSY.
Our measures of contraceptive consistency are overall similar between the NLSY97 and NSFG, though more women are estimated to be never-consistent in their contraceptive use using the NLSY97 (50.8%) than using the NSFG (43.6%). Offsetting this, 46.3% of women are ever-consistent contraceptive users before the birth when estimated using the NSFG versus 39.8% who are ever-consistent when estimated using the NLSY97. The higher ever-consistent proportion in the NSFG may be due to contraceptive consistency being measured as “every month” in the NSFG versus every episode of sexual intercourse in the NLSY97. For regression Model 2a in the NSFG, we use adjusted definitions of our contraceptive sequences: sometimes-consistent, always-consistent, never-consistent, and non-use married. We see that 26.3% of women are sometimes-consistent, 23.3% are always-consistent, and 40.3% are never-consistent users. The third category, non-use, married, has accounts for 10.1% in the NSFG and a similar fraction in the NLSY97. More detailed distributions of these contraceptive use sequences are shown in Tables A1 and A2 of the Appendix. In particular, in both the NSFG and NLSY97 the modal two-year sequence in the never-consistent category is women who are inconsistent contraceptive users in both (t-3,t-2) and (t-2,t-1), and the modal two-year sequence in the ever-consistent category is women who are consistent contraceptive users in (t-3,t-2) and inconsistent contraceptive users in (t-2,t-1). The inconsistent-use category will include women who were consistent contraceptive users for a first portion of the year (t-2,t-1) and consistent non-users for the remaining portion of that year, leading to the birth in year (t-1,t).
Table 2 shows the logistic regression models for the binary outcome of poor/non-poor after the birth. Results are presented as odds ratios (OR) with 95% confidence intervals (95% CI). The baseline model (Model 1) shows that women who have a pre-birth contraceptive-use sequence in which they never display 100% contraceptive use (never-consistent) are substantially more likely to be poor after the birth than are women who use contraceptives 100% of the time in at least one of the two years before the birth (ever-consistent). These results are seen in both the NSFG (OR 2.39; 95% CI 1.66, 3.45) and NLSY97 (OR 1.76; 95% CI 1.41, 2.18). Women with a contraceptive sequence of non-use while married do not differ significantly in their likelihood of poverty after birth compared with women whose contraceptive use is ever-consistent. When adding in the sociodemographic predictors (Model 2), the odds ratio for never-consistent contraceptive sequence decreases to 2.02 in the NSFG, meaning women who have a contraceptive sequence that is never-consistent are about 2 times more likely as ever-consistent women to be in poverty after birth, after controlling for their sociodemographic characteristics. In the NLSY97, the never-consistent Odds Ratio also decreases with inclusion of socio-demographic characteristics, with never-consistent users being 36 percent more likely to be in poverty than ever-consistent users (OR 1.36; 95% CI 1.04, 1.77). Neither the NSFG and the NLSY97 shows statistically significant contrasts for the non-use, married group relative to the ever-consistent group.
Table 2:
Logistic Regression of Poverty at Birth on Contraceptive Consistency and Socio-demographic variables, ages 17 to 31, 1999 to 2011, Odds Ratios
| NSFG | NLSY97 | NSFG+NLSY97 | |||||||
|---|---|---|---|---|---|---|---|---|---|
| Model 1 | Model 2 | Model 2a | Model 1 | Model 2 | Model 3 | Model 1 | Model 2 | Model 3 | |
| Contraception sequence before birth (Reference: Ever Consistent) a,b | |||||||||
| Never Consistent | 2.39 ** [1.66, 3.45] | 2.02 ** [1.33, 3.06] | 2.66 ** [1.67, 4.25] | 1.76 ** [1.41, 2.18] | 1.36 * [1.04, 1.77] | 1.25 [0.95, 1.64] | 1.95 ** [1.61, 2.35] | 1.50 ** [1.20, 1.88] | 1.42 ** [1.10, 1.84] |
| Non Use Married | 1.29 [0.61, 2.72] | 1.43 [0.53, 3.84] | 1.58 [0.59, 4.22] | 0.72 [0.45, 1.16] | 1.44 [0.86, 2.43] | 1.24 [0.71, 2.19] | 0.90 [0.60, 1.35] | 1.38 [0.85, 2.23] | 1.27 [0.74, 2.19] |
| Sometimes Consistent | 1.59 [0.90, 2.81] | ||||||||
| Other Predictors | |||||||||
| Age | 0.98 [0.90, 1.06] | 1.00 [0.92, 1.09] | 0.86 ** [0.81, 0.91] | 0.89 ** [0.83, 0.94] | 0.90 ** [0.86, 0.94] | 0.93 ** [0.88, 0.98] | |||
| Race/ethnicity (Reference: White) | |||||||||
| Black | 1.37 [0.76, 2.48] | 1.49 [0.83, 2.66] | 1.69 ** [1.21, 2.35] | 1.37 + [0.99, 1.90] | 1.67 ** [1.25, 2.24] | 1.37 * [1.00, 1.87] | |||
| Hispanic | 2.00 * [1.10, 3.66] | 2 24 ** [1.24, 4.06] | 0.83 [0.59, 1.15] | 0.67 * [0.49, 0.91] | 1.17 [0.85, 1.62] | 0.96 [0.67, 1.37] | |||
| Other | 2.32 + [0.94, 5.72] | 2.48 * [1.04, 5.91] | 0.76 [0.31, 1.86] | 0.53 + [0.26, 1.10] | 1.32 [0.71, 2.44] | 1.00 [0.48, 2.07] | |||
| Education (Reference: High school or less) | |||||||||
| Some College | 0.45 ** [0.25, 0.81] | 0.41 ** [0.22, 0.74] | 0.63 ** [0.44, 0.88] | 0.73 + [0.52, 1.04] | 0.55 ** [0.41, 0.74] | 0.65 ** [0.48, 0.89] | |||
| Bachelor degree + | 0.09 ** [0.04, 0.20] | 0.08 ** [0.04, 0.19] | 0.22 ** [0.09, 0.54] | 0.26 ** [0.11, 0.62] | 0.15 ** [0.08, 0.27] | 0.17 ** [0.09, 0.32] | |||
| Partnership Status (Reference: Married) | |||||||||
| No Resident Partner | 2.76 ** [1.51, 5.05] | 2.69 ** [1.47, 4.92] | 3.01 ** [2.18, 4.17] | 2 41 ** [1.69, 3.43] | 2.86 ** [2.14, 3.83] | 2.35 ** [1.70, 3.25] | |||
| Cohabiting | 1.18 [0.60, 2.31] | 1.12 [0.59, 2.15] | 2.21 ** [1.62, 3.02] | 1.88 ** [1.32, 2.67] | 1.80 ** [1.34, 2.41] | 1.56 ** [1.12, 2.17] | |||
| Birth Order (Reference: One) | |||||||||
| Two | 1.51 [0.88, 2.58] | 1.38 [0.84, 2.29] | 2.27 ** [1.68, 3.06] | 2.22 ** [1.57, 3.13] | 1.97 ** [1.50, 2.59] | 1.92 ** [1.40, 2.64] | |||
| Three Plus | 2.74 ** [1.43, 5.26] | 2.35 * [1.22, 4.49] | 4.99 ** [3.49, 7.15] | 3.78 ** [2.58, 5.56] | 4.01 ** [2.89, 5.57] | 3.02 ** [2.08, 4.39] | |||
| Had 6+ months of employment before birth | 0.56 * [0.35, 0.91] | 0.51 ** [0.31, 0.84] | 0.54 ** [0.36, 0.79] | 0.64 * [0.44, 0.94] | 0.54 ** [0.40, 0.73] | 0.63 ** [0.45, 0.88] | |||
| Lived with biological parents at 18 years old | 1.12 [0.67, 1.86] | 1.14 [0.69, 1.89] | 0.64 ** [0.48, 0.86] | 0.69 * [0.50, 0.93] | 0.72 ** [0.56, 0.92] | 0.77 + [0.58, 1.02] | |||
| Mother's Education (Reference: High school or less) | |||||||||
| Some College | 1.51 [0.83, 2.77] | 1.52 [0.82, 2.81] | 0.65 * [0.46, 0.93] | 0.68 * [0.48, 0.97] | 0.81 [0.61, 1.07] | 0.85 [0.63, 1.14] | |||
| Bachelor degree plus | 0.64 [0.32, 1.28] | 0.62 [0.31, 1.22] | 0.59 [0.29, 1.20] | 0.61 [0.29, 1.30] | 0.61 + [0.36, 1.04] | 0.65 [0.37, 1.15] | |||
| Poor before birth | 6.99 ** [4.96, 9.85] | 6.64 ** [4.76, 9.28] | |||||||
| NSFG indicator | 1.74 ** [1.32, 2.29] | 1.91 ** [1.39, 2.64] | 1.94 ** [1.36, 2.77] | ||||||
| Sample N | 778 | 778 | 778 | 2,101 | 2,101 | 2,101 | 2,879 | 2,879 | 2,879 |
Notes: Estimates are weighted and confidence intervals are in brackets
p<0.01
p<0.05
p<0.10
Contraceptive consistency sequence is defined by the longer of two years or time since first sex.
Source: NLSY97 and NSFG (2002, 2006-2010)
Reference Category for contraceptive consistency NSFG Model 2a is "always consistent".
Other strong predictors of poverty after birth that are consistent across the two surveys are having no more than a high school education and not having had a spell of at least 6 months consistent full-time employment before the birth, not having a resident partner, and having three or more children. The relationship of own education to poverty is both strong and monotonic. Compared to women with no more than a high school education, women with some college education are estimated to be 55% less likely to be poor in the NSFG (OR 0.45; 95% CI 0.25, 0.81) and 37% less likely to be poor in the NLSY97 (OR 0.63; 95% CI 0.44, 0.88), and women with a bachelor's degree or more are 91% less likely to be poor in the NSFG (OR 0.09; 95% CI 0.04, 0.20) and 78% less likely to be poor in the NLSY97 (OR 0.22; 95% CI 0.09, 0.54). In both the NLSY97 and the NSFG, women with no resident partner are nearly three times as likely to be poor than those who are married.
Some predictors are consistent in direction between the NSFG and NLSY97 but are statistically significant in their association with poverty in only one of the two surveys. Having a cohabiting partner (compared to being married) increases the likelihood of poverty greatly in the NLSY97 (OR 2.21; 95% CI 1.62, 3.02), but at OR=1.18 in the NSFG this contrast with married women’s poverty is not statistically significant. Similarly, living with both biological parents at 18 years old reduces the likelihood of poverty by approximately a third in the NLSY97 (OR 0.64; 95% CI 0.48, 0.86), but is not statistically significant in the NSFG. Finally, relative to a first birth, the likelihood of poverty after a second birth is greater in the NLSY97 (OR 2.27; 95% CI 1.68, 3.06), but at OR=1.51 in the NSFG this contrast is not statistically significant. Hispanic women are significantly more likely to be poor in the multivariate Model 2 for NSFG women, but are not statistically different from White women in the NLSY97. By contrast, in the NLSY97, Black women are significantly more likely to be poor while in the NSFG this value is not significant.
Model 2a for the NSFG uses a monthly-defined contraceptive consistency measure, with always-consistent as the reference category rather than ever-consistent, and with an in-between ‘sometimes-consistent’ category as an additional regressor. In a specification that is otherwise the same as for Model 2, we find that never-consistent users have a higher increase in likelihood of poverty after birth (OR 2.66; 95% CI: 1.67, 4.25). That is, the sharper distinction between always-consistent use and never-consistent use allowed by adding the ‘sometimes-consistent’ category in the NSFG results in a greater estimated never-consistent contraceptive use association with poverty. We also see that being ‘sometimes-consistent’ relative to ‘always-consistent’ is associated with an elevated risk of poverty (OR 1.59), but this difference falls just outside statistical significance (p=.108).
As we showed in Table 1, the overall poverty levels of women who just gave birth differ between our surveys, about 10 percentage points higher in the NSFG sample. We evaluate whether this survey difference is maintained in a multivariate model predicting poverty after birth by pooling the NSFG and NLSY97 observations and including an NSFG indicator variable among the regressors (see the ‘NSFG+NLSY97’ columns in Table 2). Comparing the magnitude of the odds ratio for the NSFG indicator between Model 1 and Model 2, we see that the difference does not go away, and that only a small portion of the difference is attributable to the differences in sociodemographic variables between the two surveys. In Model 1, the odds of poverty are 74% higher (OR 1.74; 95% CI 1.32, 2.29) when using the NSFG data before including sociodemographic variables among the regressors. After including these sociodemographic variables as regressors in the model (Model 2), the poverty likelihood is increased to being 91% higher among women sampled into and measured in the NSFG relative to women sampled into and measured in the NLSY97 (OR 1.91; 95% CI 1.39, 2.64).
As shown in Appendix Table A4, the model fit is always improved by including an “NSFG survey” intercept, but always worsens when including interactions between the regressors and “survey.” This implies that pooling observations across the two surveys is statistically appropriate when estimating Models 1, 2, and 3 (Rendall et al. 2013, pp.497-498).
The final model, Model 3, includes as a regressor, poverty in the year before the year of the birth. This is our preferred model specification. We present estimates of this model both for the NLSY97 observations alone and for the pooled NSFG+NLSY97 observations. As expected, being in poverty in the year immediately prior to the birth increases greatly the likelihood of being in poverty after the birth (NLSY97-only OR 6.99; 95% CI 4.96, 9.85). The similarity of this odds ratio in the multiply-imputed data (NSFG+NLSY97 OR 6.64; 95% CI 4.76, 9.28) is evidence of statistically unbiased imputation of values of this variable to the NSFG observations. Moreover, as expected, the magnitude of contrast in the odds of poverty after birth between never-consistent and ever-consistent contraceptive use is smaller than before inclusion of poverty in the year before the birth, implying omitted-variable bias in the models that omit this regressor. In the NLSY97-only estimation of Model 3, the odds ratio falls from 1.36 to 1.25 (95% CI 0.95, 1.64). At mean or modal categories of the other characteristics,3 this Model 3 odds ratio estimated on the NLSY97 data corresponds to a difference in poverty probabilities of 14.2% for never-consistent versus 12.0% for ever-consistent women. In the pooled NSFG+NLSY97 estimation, the odds ratio falls from 1.50 to 1.42 (95% CI 1.10, 1.84) in Model 3. Again at modal categories of the other characteristics, and for the reference NLSY97 survey observation of the overall poverty rate, this corresponds to a difference in poverty probabilities of 16.5% for never-consistent versus 12.0% for ever-consistent women. That is, never using contraception consistently is associated with 42% higher odds of poverty after birth relative to ever having used contraception consistently in the two years before the birth, corresponding to a 4.5 percentage point, or one third, higher poverty risk. The Model 3 contrast between married nonusers of contraception and ever-consistent contraceptive users does not attain statistical significance in either of the single-survey estimates nor in the pooled-survey estimates.4
Among the variables substantially and statistically-significantly associated with poverty after the birth in the full model (Model 3) with pooled-survey estimation include being Black (37% higher odds of poverty than White), having more education (35% and 83% lower odds of poverty respectively for some college and bachelor’s degree than the chance of poverty among women with no more than a high-school education), and being single or cohabiting (respectively 135% and 56% higher likelihoods of being poor than a married woman. Finally, the background factors of whether the woman had stable employment before the birth and whether she lived with both biological parents at age 18 remain protective against poverty after the birth (37% and 23% less likely to be poor, respectively).
To evaluate our second hypothesis of greater association of less consistent contraceptive use with higher poverty among women with the lowest education levels, we first estimate differences in contraceptive use and its relationship to poverty after a birth by the three levels of own education. Figure 1 shows the bivariate distribution of contraceptive use sequences before a birth by own educational attainment, estimated separately using the NSFG and NLSY97. As expected, never-consistent contraceptive users are most common in the lowest education group of women with no more than a high school graduate education. Never-consistent contraceptive users represent about half of all women in this lowest education category in both surveys. Ever-consistent contraceptive users are increasingly more prevalent as educational attainment increases, constituting about one half of college graduate women. Nevertheless, never-consistent contraceptive users are common too in this highest education group, at between a third and two fifths of all women.
FIGURE 1: Distribution of Contraceptive Use Sequence by Education, Women aged 17-31 in 1999-2011 giving birth.
Notes: Results are weighted
Source: NLSY97 (n=2101) and NSFG 2002, 2006-2010 (n=778)
We next use the combined NSFG and NLSY97 data in a Model 2 regression specification (that is, excluding poverty status in the year before the birth), but with the addition of interactions of contraceptive consistency indicators with own education. Results are presented in the form of average predicted probabilities of poverty for women in each educational group, controlling for birth order, age, marital status, mother’s education, employment, family structure, and race. This corresponds to the “average discrete change” method described by Long and Mustillo (2018), averaging over the observed values on other variables among women with that education category. Among women in both the low (no more than high school graduate) and medium (some college) educational attainment groups, the poverty rate is in both cases substantially higher among never-consistent contraceptive users. For women with no more than high school graduate educational attainment, the poverty rates of never-consistent and ever-consistent contraceptive users are respectively 33.2% and 26.0%. For women with some college education, the poverty rates of never-consistent and ever-consistent contraceptive users are respectively 24.0% and 17.6%. For both education groups these differences, of between six and seven percentage points greater poverty risk for never-consistent than ever-consistent contraceptive users, are statistically significant despite the partial overlap of their 95% confidence intervals (see Schenker and Gentleman 2001). For women with a bachelors degree or more, however, we find not only that poverty rates after a birth are much lower, but also that this poverty rate is statistically no different by pattern of contraceptive use before the birth (6.5% and 9.0% respectively for never-consistent and ever-consistent contraceptive users).
Discussion and Conclusions
In this study we used nationally-representative retrospective (NSFG) and panel (NLSY97) data sources of longitudinal contraceptive-use information to investigate unplanned births and their potential consequences for newborn child poverty. Our major finding, consistent with our first hypothesis, was that women whose behavior does not include consistent contraceptive use in the two years preceding a birth (never-consistent users) have a higher chance of being in poverty after giving birth than women whose contraceptive use in those two preceding years indicates that the birth was more likely to have been planned (ever-consistent users). We did not find an elevated chance of poverty for women who do not use contraceptives in the years before the birth but were married at the time of birth. The latter group, accounting for only 1 in 10 women in this population (Table 1), we conjecture to have had a greater likelihood of not using contraceptives because they were trying for, or at least were not trying to avoid having, a child. Additionally, married-couple poverty rates are in general much lower than those of either cohabiting couples or single mothers (Short 2011). Because this is a smaller group, however, our statistical power to distinguish their poverty outcome from that of ever-consistent women is also reduced.
In our multivariate analyses, the additions of sociodemographic factors, which partially explain contraceptive use in previous studies (Glei 1999), reduced but did not remove the association of inconsistent contraceptive use with poverty after a birth. Nor did inclusion of poverty in the year before the year of the birth remove this positive association of inconsistent contraceptive use with poverty after the birth. Other factors that we found to be protective against poverty after a birth are being White, having fewer children, living with both biological parents at age 18, having had a spell of full-time employment for at least 6 months, being highly educated, and being married. Although none of these findings on other factors associated with poverty after giving birth are surprising, they are reassuring with respect to our data sources and analytical methods that control for them.
A woman’s educational attainment is expected to have particular significance for our analyses because of its hypothesized relationship to both a woman’s labor market prospects and to knowledge and socialization about contraception and about planned behavior in general (England et al. 2016). When we viewed the distribution of contraceptive use behavior prior to a birth by education, we found that as education increased, a smaller percentage of women fell into the never-consistent category. Women, regardless of socioeconomic status, may be sexually active, but unintended births are more common among disadvantaged women (e.g, Reeves and Venator 2015; Finer and Zolna 2016; but see also Wise et al. 2016). When predicting the risk of poverty after a birth allowing for an interaction of education level and contraceptive-use pattern, we found that, consistent with our second hypothesis, the association of consistency of contraceptive use with poverty after birth is stronger among less educated women. Women with a bachelor’s degree or more having significantly different outcomes than all other women is a phenomenon not unique to poverty after a birth. Women with a bachelor’s degree are more likely to get married and stay married (Lundberg and Pollak 2015, Cherlin 2004, Kennedy and Ruggles 2014) and have fewer children and postpone childbearing (Musick et al. 2009). In the present study, we found that not only do women with a bachelor’s degree have significantly lower poverty levels after a birth than a woman with some college or no more than a high school education, but also that contraceptive consistency in the two years preceding a birth was not predictive of this group’s poverty status after giving birth. Inconsistent contraceptive use, meanwhile, was associated with a poverty rate between 7 to 9 percentage points higher than consistent contraceptive use for women with a high-school-only or some-college level of educational attainment. We interpret these results to indicate that highly-educated women have access to resources that can buffer the possible effects of using contraception inconsistently in a way that less-educated women cannot. This is an added privilege to an already more privileged group. We note here also that Wise et al. (2016) found that, after controlling for other factors, women who are highly educated were actually more likely to have a mistimed birth than those with low education levels. Consistent with this, our study suggests that a major advantage experienced by higher-educated women is having the resources to avoid at least the most severe negative economic outcomes of a birth after inconsistent contraceptive use.
Using quasi-experimental methods, previous research has addressed how contraceptive access effects poverty using data from the 1960’s and 1970’s, taking advantage of the family planning policy shifts (Browne and LaLumia 2014, Bailey, Malkova, and McLaren 2018). To our knowledge, our study is the first to explore the relationship of contraceptive access or contraceptive use with poverty after a birth using recent data. The finding of a link between inconsistent contraceptive use and newborn child poverty thereby offers indirect support to studies calling for improved access to IUDs and implants (LARC, Wu and Mark 2018, Trussell et al. 2013), as these methods are not only highly effective, but also obtain their increased effectiveness in part because they do not rely on user consistency for pregnancy prevention once inserted (Grimes 2009). As Wu and Mark and Trussell et al. argue, reducing unplanned births may reduce not only newborn child poverty but also public expenditures associated with poverty and adverse child health (see also Joyce et al. 2000). If efforts to support women’s contraceptive access and consistency were more widely funded and supported, women may also have enhanced agency with respect to achieving their fertility goals, including those related to their fertility timing. In addition, we also point to contraceptive use operating together with education, where highly educated women are able to avoid the most adverse economic outcomes of a birth after inconsistent contraceptive use. Encouraging consistent contraceptive use is therefore only a partial poverty-prevention policy strategy, with policies that increase pre-birth economic resources a needed additional instrument.
FIGURE 2: Average predicted probability of women in poverty after a birth for Never and Ever consistent contraceptive users by education, ages 17-31 1999-2011.
Notes: Estimates are weighted and confidence intervals represented by bars.
*= The difference between Never and Ever consistent is significant at p<0.05
Predicted probabilities calculated within education level and controlling for: marital status, race, age, mothers education, and birth order
Source: NLSY 1997 (n=2101) and NSFG 2002, 2006-2010 (n=778)
Acknowledgments
Declarations: We are grateful for comments received from the discussant and participants at the 2019 Population Association of America Annual Meeting, and for support from the National Science Foundation BIGDATA: Applications program, grant NSF IIS-1546259, and from the Eunice Kennedy Shriver National Institute of Child Health and Human Development, grants R03-HD084974 and P2C-HD041041.
Appendix
Table A1:
Contraceptive-Consistency Category by Contraceptive Sequence, NLSY97
| N | Proportion Group |
Proportion All |
Contraceptive use | Birth(t-1,t) | ||
|---|---|---|---|---|---|---|
| (t-3,t-2) | (t-2,t-1) | |||||
| Ever Consistent | ||||||
| 198 | 0.2319 | 0.0942 | Consistent | Nonuse | ||
| 575 | 0.6733 | 0.2737 | Consistent | Inconsistent | ||
| 75 | 0.0878 | 0.0357 | Consistent | Consistent | * | |
| 6 | 0.0070 | 0.0029 |
|
Consistent | * | |
| Total | 854 | 0.4065 | ||||
| Non-use Married | ||||||
| 178 | 0.9780 | 0.0847 | Nonuse | Nonuse | ** | |
| 4 | 0.0220 | 0.0019 |
|
Nonuse | ** | |
| Total | 182 | 0.0866 | ||||
| Never Consistent | ||||||
| 111 | 0.1042 | 0.0528 | Nonuse | Nonuse | *** | |
| 8 | 0.0075 | 0.0038 |
|
Nonuse | *** | |
| 499 | 0.4685 | 0.2375 | Inconsistent | Inconsistent | ||
| 38 | 0.0357 | 0.0181 |
|
Inconsistent | ||
| 122 | 0.1146 | 0.0581 | Nonuse | Inconsistent | ||
| 287 | 0.2695 | 0.1366 | Inconsistent | Nonuse | ||
| Total | 1065 | 0.5069 | ||||
| Total All | 2101 | |||||
Notes: Block represents that respondent was not sexually active during (t-3, t-2)
= Gave births [t-2months, t]
=Married at t
=Unmarried at t
Table A2:
Contraceptive-Consistency Category by Contraceptive Sequence, NSFG
| N | Proportion Group |
Proportion All |
Contraceptive use | Birth(t-1,t) | ||
|---|---|---|---|---|---|---|
| (t-3,t-2) | (t-2,t-1) | |||||
| Ever Consistent | ||||||
| 18 | 0.0508 | 0.0231 | Consistent | Nonuse | ||
| 262 | 0.7401 | 0.3368 | Consistent | Inconsistent | ||
| 68 | 0.1921 | 0.0874 | Consistent | Consistent | * | |
| 6 | 0.0169 | 0.0077 | Consistent | * | ||
| Total | 354 | 0.3599 | ||||
| Non-use Married | ||||||
| 64 | 0.3516 | 0.0823 | Nonuse | Nonuse | ** | |
| 7 | 0.0385 | 0.0090 |
|
Nonuse | ** | |
| Total | 71 | 0.0913 | ||||
| Never Consistent | ||||||
| 51 | 0.0479 | 0.0656 | Nonuse | Nonuse | *** | |
| 9 | 0.0085 | 0.0116 |
|
Nonuse | *** | |
| 156 | 0.1465 | 0.2005 | Inconsistent | Inconsistent | ||
| 17 | 0.0160 | 0.0219 |
|
Inconsistent | ||
| 38 | 0.0357 | 0.0488 | Nonuse | Inconsistent | ||
| 82 | 0.0770 | 0.1054 | Inconsistent | Nonuse | ||
| Total | 353 | 0.4537 | ||||
| Total All | 778 | |||||
Notes: Block represents that respondent was not sexually active during (t-3, t-2)
= Gave births [t-2months, t]
=Married at t
=Unmarried at t
Appendix A3: Contraceptive-Consistency for Model 2a, NSFG
Based on the information of contraceptive use per month for years (t-3,t-2, and t-2,t-1), we break down the Never- and Ever-consistent categories into three categories: Always, Sometimes and Never consistent. Then we analyze the sequences of contraceptive use per month between (t-3,t-2) and the month of conception. The month of conception is available on the Female Pregnancy data for both rounds (2002 and 2006-2010). The number of months with information is conditional on the number of months with sexual intercourse. That is, when selecting the months for input to sequence coding, we essentially skip over months in which there is no sexual intercourse reported. For example, in identifying the use of contraception in the first 6 months beginning in the first month of year (t-3,t-2), we use as input to the sequence coding the first 6 months in which she reports sexual intercourse. If in these first six months of reported sexual intercourse, she used contraception in all six months, her sequence will either be coded as always-consistent or sometimes-consistent. If in any of these first six months of reported sexual intercourse, she reported not using contraception, her sequence will be coded as never-consistent.
To code these variables, we use the sequence commands in STATA (SQ-Ados). We classify women following these rules:
1. Always Consistent:
Beginning in the first month of year (t-3,t-2) a sequence of use every month for at least 6 months AND beginning with the first month of non-use, a sequence of non-use for every month up to and including the month of conception.
Example (1=use; 0=non-use):
| 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 |
If they were classified as never-consistent on the original annual variable, but they start with at least 6 consecutive months of use on (t-3,t-2), and between (t-2, t-1) and the month of conception, all months in which she had sex they were non-users.
Example:
| 1 | 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 | 0 | 0 |
OR
| 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 0 |
2. Sometimes Consistent:
Beginning in the first month of the year (t-3,t-2) a sequence of use every month for at least 6 months; AND in the sequence of months beginning with the first month of non-use and ending in the month of conception, at least 1 month of contraceptive use.
Example 1:
| 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 0 | 1 | 0 | 0 |
In the above example, she is a contraceptive user since first month with information in year (t-3,t-2), and she has at least one ‘0’ (non-use) followed by a ‘1’ (use) up to and including the month of conception. The woman is an ever-consistent contraceptive user under the annual definition, and is a sometimes-consistent user under the monthly definition.
Example 2:
| 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 0 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 0 | 1 | 0 | 0 |
In the above example, she is a contraceptive user in all of the first 6 months with information in year (t-3,t-2), but has at least one month without contraceptive use in year (t-3,t-2), and she has at least one ‘0’ (non-use) followed by a ‘1’ (use) up to and including the month of conception. She is a sometimes-consistent on a monthly basis, whereas she is never-consistent on an annual basis.
3. Never Consistent:
The never-consistent monthly-basis cases start off with at least one 0 in the first 6 months of (t-3,t-2).
Example:
| 0 | 1 | 0 | 0 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
OR
| 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 |
Table A4.
Model Fit Statistics for Pooled Logistic Regressions of Poverty on Contraceptive Consistency
| Model 1 | Model 2 | ||||||
|---|---|---|---|---|---|---|---|
| Pooled NLSY97 and NSFG Model Fit statistics |
no NSFG intercept or regressor interaction |
NSFG intercept, no regressor interaction |
NSFG intercept and regressor interaction |
no NSFG intercept or regressor interaction |
NSFG intercept, no regressor interaction |
NSFG intercept and regressor interaction |
NSFG intercept and main variable interaction |
| AIC | 3,246.1 | 3,213.6 * | 3,213.8 | 3,246.1 | 2,553.5 * | 2,661.4 | 2,554.9 |
| BIC | 3,264.0 | 3,237.5 * | 3,249.6 | 3,264.0 | 2,660.9 * | 2,828.5 | 2,674.2 |
best fitting model (lower = better fit)
Sources: National Longitudinal Survey of Youth 1997 (NLSY97) and the National Survey of Family Growth 2002, 2006-2010
Notes: "NSFG intercept" indicates that the pooled-survey (NLSY97 and NSFG) regression model specification includes a dummy variable for the observation's coming from the NSFG sample. "NSFG intercept and regressor interaction" indicates that the pooled-survey (NLSY97 and NSFG) regression model specification includes a dummy variable for the observation's coming from the NSFG sample plus an interaction variable for NSFG*<regressor> for each of the regressors in the Model 1 or Model 2, respectively. "NSFG intercept and main variable interaction" indicated an interaction between the NSFG intercept and contraceptive consistency only. See Table 3 for Model specifications.
Footnotes
If the respondent is not independent from their parents, but their parents did not fill out a “Parent Interview” we do not know the income and poverty status of the respondent, and therefore have missing poverty information. 27.4% of women had missing data on either the year of the birth or the year before the birth.
We also compared standard error inflation between incorporating PSU versus individual woman clusters and found similar magnitudes.
White, married, first birth, 24.6 years old, have had a full-time job, did live with biological parents at 18 years old, high school or less education, mom completed high school or less education.
In an alternate, expanded classification, we also included separately a “non-use, unmarried” category of women. They are included among the never-consistent group in the results presented in Table 2. The never-consistent odds of poverty in Model 3 were little changed by this recategorization (1.40 in place of 1.42 in the NSFG+NSLY97 estimate, and 1.20 in place of 1.25 in the NLSY97-only estimate). The alternate NSFG-only Model 2a, moreover, shows that the never-consistent category that retains the “non-use, unmarried” in it, while separating out sometimes-consistent contraceptive users, exhibits a stronger contrast between never-consistent and ever-consistent women’s post-birth poverty risk.
References
- Alon S, Donahoe DA, & Tienda M (2001). The Effects of Early Work Experience on Young Women’s Labor Force Attachment. Social Forces 79(3):1005–1034. [Google Scholar]
- Bachrach CA & Newcomer S (1999). Intended Pregnancies and Unintended Pregnancies: Distinct Categories or Opposite Ends of a Continuum?” Family Planning Perspectives 31(5): 251–252. [PubMed] [Google Scholar]
- Bachrach CA & Morgan SP (2011). Is the Theory of Planned Behaviour an appropriate model for human fertility? Vienna Yearbook of Population Research 9: 11–18. [Google Scholar]
- Bachrach CA & Morgan SP (2013). A Cognitive-Social Model of Fertility Intentions. Population and Development Review 39(3):459–485. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Bailey MJ (2013) Fifty Years of Family Planning: New Evidence on the Long-Run Effects of Increasing Access to Contraception. Brookings Papers on Economic Activity 341–410. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Bailey MJ, Malkova O, & Mclaren ZM (2018) Does Access to Family Planning Increase Children’s Opportunities? Evidence from the War on Poverty and the Early Years of Title X. Journal of Human Resources 55(1). [DOI] [PMC free article] [PubMed] [Google Scholar]
- Baker EH Rendall MS, & Weden MM (2015). Epidemiological paradox or immigrant vulnerability? Obesity among young children of immigrants” Demography 52(4):1295–1320. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Bartz D, Shew M, Ofner S, & Fortenberry JD (2007) Pregnancy Intentions and Contraceptive Behaviors Among Adolescent Women: A Coital Event Level Analysis Journal of Adolescent Health 41(3): 271–276. [DOI] [PubMed] [Google Scholar]
- Brady D and Burroway R. (2012) Targeting, Universalism, and Single-Mother Poverty: A Multilevel Analysis Across 18 Affluent Democracies. Demography 49:719–746. [DOI] [PubMed] [Google Scholar]
- Brown SS & Eisenberg L (1995). The Best Intentions: Unintended Pregnancy and the Well Being of Children and Families. Washington D.C.: National Academy Press. [PubMed] [Google Scholar]
- Browne SP & LaLumia S (2014). The Effects of Contraception on Female Poverty. Journal of Policy Analysis and Management 00(0):1–21. [DOI] [PubMed] [Google Scholar]
- Bureau of Labor Statistics (2015). National Longitudinal Survey of Youth 1997 cohort, 1997-2013 (rounds 1-16). Produced by the National Opinion Research Center, the University of Chicago and distributed by the Center for Human Resource Research, The Ohio State University. Columbus, OH. . [Google Scholar]
- Bureau of Labor Statistics (nd) National Longitudinal Survey of Youth 1997 cohort, Codebook Supplement, Household Income – Appendix 5. Retrieved from: https://www.nlsinfo.org/content/cohorts/nlsy97/other-documentation/codebook-supplement/appendix-5-income-and-assets-variab-3
- Cameron AC & Miller D (2015). A Practitioner’s Guide to Cluster-Robust Inference. Journal of Human Resources 50(2):317–372. [Google Scholar]
- Capps R, Bachmeier JD, and Van Hook J (2018). Estimating the Characteristics of Unauthorized Immigrants Using U.S. Census Data: Combined Sample Multiple Imputation. The ANNALS of the American Academy of Political and Social Science 677(1):165–179. [Google Scholar]
- Chen W & Corak M (2008). Child Poverty and Changes in Child Poverty. Demography 45(3):537–553. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Cherlin AJ (2004). The Deinstitutionalization of American Marriage. Journal of Marriage and Family 66(4):848–61. [Google Scholar]
- Edin K & Kefalas MJ (2005) Promises I can Keep: Why Poor Women Put Motherhood before Marriage. Berkeley, CA: University of California Press. [Google Scholar]
- England P (2016). Sometimes the Social Becomes Personal: Gender, Class and Sexualities. American Sociological Review 81(1):4–28. [Google Scholar]
- England P, Caudillo ML, Littlejohn K, Conroy Bass B, & Reed J (2016). Why Do Young, Unmarried Women Do Not Want to Get Pregnant Contracept Inconsistently? Mixed-method Evidence for the Role of Efficacy. Socius 2:1–15. [Google Scholar]
- Finer LB & Zolna MR (2011). Unintended Pregnancy in the United States: Incidence and Disparities, 2006. Contraception 84:478–485. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Finer LB & Zolna MR (2016). Declines in Unintended Pregnancy in the United States, 2008-2011. New England Journal of Medicine 374(9):843–852. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Foster DG, Biggs MA, Ralph L, Gerdts C, Roberts S, & Glymour MM (2018). Socioeconomic Outcomes of Women who Receive and Women Who are Denied Wanted Abortions in the United States. American Journal of Public Health 108(3):407–413. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Fox L, Wimer C, Kaushal N, & Waldfogel J (2015) Waging War on Poverty: Poverty Trends Using a Historical Supplemental Poverty Measure. Journal of Policy Analysis and Management 34(3):567–592. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Gelman A, King G, and Liu C (1998). Not Asked and Not Answered: Multiple Imputation for Multiple Surveys. Journal of the American Statistical Association 94:846–57. [Google Scholar]
- Glei DA (1999). Measuring Contraceptive Use Patterns Among Teenage and Adult Women. Family Planning Perspectives 31(2):73–80. [PubMed] [Google Scholar]
- Grimes DA (2009) Forgettable contraception. Contraception 80:497–499. [DOI] [PubMed] [Google Scholar]
- Hoynes HW, Page ME, & Stevens AH (2006). Poverty in America: Trends and Explanations. Journal of Economic Perspectives 20(1): 47–68. [Google Scholar]
- Hurst E, Li G, and Pugsley B (2014) Are household surveys like tax forms? Evidence from income underreporting of the self-employed. The Review of Economics and Statistics 96(1): 19–33. [Google Scholar]
- Iceland J (2000) The ‘family/couple/household’ unit of measurement in poverty estimation Journal of Economic and Social Measurement 26:253–265. [Google Scholar]
- Jones RK (2017). Change and consistency in US women’s pregnancy attitudes and associations with contraceptive use. Contraception 95:485–49. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Joyce TJ, Kaestner R, & Korenman S (2000). The Effect of Pregnancy Intention on Child Development. Demography 37(1):88–94. [PubMed] [Google Scholar]
- Kaushal N (2014). Intergenerational Payoffs of Education. The Future of Children 24(1). [DOI] [PubMed] [Google Scholar]
- Kavanaugh ML, Kost K, Frohwirth L, Maddow-Zimet I, & Gor V (2017). Parents’ experience of unintended childbearing: A qualitative study of factors that mitigate or exacerbate effects. Social Science & Medicine 174(2017): 133–141 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Kennedy S & Ruggles S (2014). Breaking Up Is Hard to Count: The Rise of Divorce in the United States, 1980-2010. Demography 51(2):587–98. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Klerman L (2000). The Intendedness of Pregnancy: a Concept in Transition. Maternal and Child Health Journal 4(3): 155–162. [DOI] [PubMed] [Google Scholar]
- Lichter DT, Sanders SR, & Johnson KM (2018). Hispanics at the Starting Line: Poverty among Newborn Infants in Established Gateways and New Destinations. Social Forces 94(1): 209–235. [Google Scholar]
- Lindberg L, Kost K, Maddow-Zimet I, Desai S, & Zolna M (2020). Abortion Reporting in the United States: An Assessment of Three National Fertility Surveys. Demography 57:899–925. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Little RJA and Rubin DB. (2002). Statistical Analysis with Missing Data (2nd ed.) Hoboken, NJ: John Wiley. [Google Scholar]
- Lu Y, Wang JS, & Han W (2017). Women’s Short-Term Employment Trajectories Following Birth: Patterns, Determinants, and Variations by Race/Ethnicity and Nativity. Demography 54:93–118. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Long JS, & Mustillo SA (2018). Using Predictions and Marginal Effects to Compare Groups in Regression Models for Binary Outcomes. Sociological Methods and Research. online access [Google Scholar]
- Lundberg S & Pollak RA (2015). “The Evolving Role of Marriage : 1950 – 2010.” The Future of Children 25(2):29–50. [Google Scholar]
- Manlove J, Logan C, Moor KA, & Ikramullah E (2008). Pathways from Family Religiosity to Adolescent Sexual Activity and Contraceptive Use. Perspectives on Sexual and Reproductive Health 40(2):105–117. [DOI] [PMC free article] [PubMed] [Google Scholar]
- McKernan S, & Ratcliffe C (2005). Events that trigger poverty entries and exits. Social Science Quarterly 86: S1146–69. [Google Scholar]
- McLanahan S, & Percheski C (2008). Family Structure and the Reproduction of Inequalities. Annual Review of Sociology 34:257–276. [Google Scholar]
- McLanahan S, & Jacobsen W (2015). Diverging Destinies Revisited. Families in an Era of Increasing Inequality Amato PR et al. (eds). National Symposium on Family Issues; 5: 201–212. [Google Scholar]
- Meyer Bruce D., Mok WKC, & Sullivan JX(2015). Household Surveys in Crisis. Journal of Economic Perspectives. 29(4):199–226. [Google Scholar]
- Moreau C, Hall K, Trussel J, & Barber J (2013). Effect of prospectively measured pregnancy intentions on the consistency of contraceptive use among young women in Michigan. Human Reproduction 36(2): 211–251. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Musick K, England P, Edington S, & Kangas N (2009). Education Differences in Intended and Unintended Fertility. Social Forces 88(2):543–72. [DOI] [PMC free article] [PubMed] [Google Scholar]
- National Center for Health Statistics. (2011). 2006–2010 National Survey of Family Growth Public Use Data and Documentation. Hyattsville, MD: CDC National Center for Health Statistics. Retrieved from http://www.cdc.gov/nchs/nsfg/nsfg_2006_2010_puf.htm [Google Scholar]
- National Center for Health Statistics. (2003). 2002 National Survey of Family Growth Public Use Data and Documentation. Hyattsville, MD: CDC National Center for Health Statistics. Retrieved from https://www.cdc.gov/nchs/nsfg/nsfg_cycle6 [Google Scholar]
- National Center for Health Statistics (nd) 2006-2010 NSFG Questionnaires, Female Questionnaire, Female J Retrieved from https://www.cdc.gov/nchs/data/nsfg/NSFG_2006-2010_Y1_FemaleJ_CRQ.pdf
- National Research Council (1995) Measuring Poverty: A New Approach Washington, DC: The National Academies Press. [Google Scholar]
- Pepinsky TB (2018) A note on listwise deletion versus multiple imputation Political Analysis 26:480–488. [Google Scholar]
- Ratitch B, Lipkovich I, & O’Kelly M (2013). Combining Analysis Results from Multiply Imputed Categorical Data. PharmaSUG 2013 - Paper SP03. [Google Scholar]
- Reeves RV & Venator J (2015). Sex, contraception, or abortion? Explaining class gaps in unintended childbearing. Washington, D.C: Brooks Institution. [Google Scholar]
- Rendall MS, Admiraal R, DeRose A, DiGiulio P, Handcock MS, and Racioppi F. (2008). Population constraints on pooled surveys in demographic hazard modeling. Statistical Methods and Applications 17(4):519–539. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Rendall MS, Ghosh-Dastidar B, Weden MM, Baker EH, & Nazarov Z (2013). Multiple imputation for combined-survey estimation with incomplete regressors in one but not both surveys. Sociological Methods and Research 42(4):483–530. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Rendall MS, & Shattuck RM (2019) First birth before first stable employment and subsequent single-mother ‘disconnection’ before and after the Welfare Reform and Great Recession. Journal of Poverty 23(2):83–104. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Rosenfeld M (2015). Revisiting the Data from the New Family Structure Study: Taking Family Instability into Account. Sociological Science 2:478–501. [Google Scholar]
- Schafer JL and Graham JW. (2002). Missing Data: Our View of the State of the Art. Psychological Methods 7:147–77. [PubMed] [Google Scholar]
- Schenker N and Gentleman JF (2001). On Judging the Significance of Differences by Examining the Overlap Between Confidence Intervals. American Statistician 55(3):182–186. [Google Scholar]
- Shattuck RM, and Rendall MS (2017) Retrospective reporting of first employment in the life courses of U.S. women. Sociological Methodology 47:307–344. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Short K (2011). The Research Supplemental Poverty Measure: 2010 Current Population Reports P60-241. Washington, DC: U.S. Government Printing Office. [Google Scholar]
- Sipsma HL & Ickovics JR (2015). The Impact of Future Expectations on Adolescent Sexual Risk Behavior. Journal of Youth Adolescence 44(1): 170–183. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Sundaram A, Vaughan B, Kost K, Bankole A, Finer L, Singh S, & Trussell J (2017). Contraceptive failure in the United States: Estimates from the 2006–2010 National Survey of Family Growth. Perspectives on Sexual and Reproductive Health 49(1):7–16. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Thiede BC, Sanders SR, & Lichter DT (2018). Born Poor? Racial Diversity, Inequality, and the American Pipeline. Sociology of Race and Ethnicity 4(2):206–228. [Google Scholar]
- Trussell J, Henry N, Hassan F, Prezioso A, Law A, & Filoneko A (2013). Burden of unintended pregnancy in the United States: potential savings with increased use of long-acting reversible contraception. Contraception 87:154–161. [DOI] [PMC free article] [PubMed] [Google Scholar]
- U.S. Census Bureau (2017). Children Characteristics: 2011-2015 American Community Survey 5-year Estimates. https://factfinder.census.gov/faces/tableservices/jsf/pages/productview.xhtml?src=CF Accessed 10/10/17.
- Williams L, & Abma J (2000). Birth Wantedness Reports: A Look Forward and a Look Back. Social Biology 47(3-4):147–165. [DOI] [PubMed] [Google Scholar]
- Wimer C, Nam J, Waldfogel J, and Fox L (2016) Trends in child poverty using an improved measure of poverty Academic Pediatrics 16:S60–S66. [DOI] [PubMed] [Google Scholar]
- Wise A, Gerinimus AT, & Smock PJ (2016). The Best of Intentions: A Structural Analysis of the Association between Socioeconomic Disadvantage and Unintended Pregnancy in a Sample of Mothers from the National Longitudinal Survey of Youth (1979). Women’s Health Issues 27(1):5–13. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Wu J, Meldrum S, Dozier A, Stanwood N, & Fiscella K (2008). Contraceptive nonuse among US women at risk for unplanned pregnancy. Contraception, 78(4), 284–289. [DOI] [PubMed] [Google Scholar]
- Wu JP, Kusunoki Y, Ela EJ, & Barber JS (2016). Patterns of contraceptive consistency among young adult women in southeastern Michigan: Longitudinal findings based upon journal data. Women’s Health Issues 26(3):305–312. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Wu LL & Mark NDE (2018). Could we level the playing field? Long-Acting Reversible Contraceptives, Nonmarital Fertility, and Poverty in the United States. RSF: The Russell Sage Foundation Journal of the Social Sciences 4(3): 144–166. [Google Scholar]
- Zabin LS (1999). Ambivalent Feelings about Parenthood May Lead to Inconsistent Contraceptive Use-and Pregnancy. Family Planning Perspectives 31(5): 250–251. [PubMed] [Google Scholar]