Module 11 Kost-Lindberg Article Pregnancy Intentions
Pregnancy Intentions, Maternal Behaviors, and Infant Health
Abstract
The central idea is that a mother's intentions regarding her pregnancy can significantly influence both her behavior and the health outcomes of her infant, a concept with substantial implications for public health policy and research on reproductive behaviors.
Research findings on this topic are not consistently clear, presenting a mix of evidence for and against this premise.
The research uses data from the U.S. National Survey of Family Growth and employs a measure of pregnancy intentions that accounts for the extent of mistiming. This measure is based on a scale developed by Santelli et al. (Studies in Family Planning, 40, 87–100, 2009).
The study analyzes the different characteristics of mothers within various intention status groups.
Propensity score weighting is used to account for the relationship between a mother’s background characteristics and her pregnancy intentions, as well as the outcomes. This method helps to eliminate biases by adjusting for observed differences in demographic and socioeconomic characteristics.
The weighting process reduced the statistical significance of many observed associations between intention status and maternal behaviors or birth outcomes. However, some associations remained significant.
Pregnancies that were mistimed or unwanted were recognized later than intended pregnancies.
Fewer unwanted births received early prenatal care or were breast-fed.
Infants from unwanted births were more likely to have low birth weights than those from intended births.
All births, other than those at the highest level of the desire scale, were less likely to be recognized early in pregnancy and receive early prenatal care.
Keywords
Pregnancy intentions
Unintended childbearing
Unintended pregnancy
Maternal behaviors
Infant health
Introduction
The rate of unintended pregnancies in the United States has remained relatively stable over the past three decades, with 54 per 1,000 women aged 15–44 in both 1981 and 2008 (Henshaw 1998; Finer and Zolna 2014).
Over more than two decades, over one-third of births were reported by mothers as originating from unintended pregnancies (Mosher et al. 2012), a finding supported by reports from fathers (Mosher et al. 2012; Lindberg and Kost 2013).
This persistently high level of unintended pregnancy underscores the importance of understanding the consequences of births resulting from such pregnancies.
The ability to decide whether and when to have a child is a fundamental human right, driving efforts to help women and couples avoid unplanned pregnancies. Efforts to reduce unintended pregnancy are motivated by the potential for improving public health (Institute of Medicine 2011).
Critical reviews of research on the consequences of unintended childbearing indicate that the evidence base is weak and inconsistent due to numerous methodological challenges (Gipson et al. 2008; Logan et al. 2007).
U.S. studies provide mixed evidence of the relationships between unintended pregnancy and maternal behaviors, with weak or no effects of pregnancy intentions on key birth outcomes, such as premature delivery and low birth weight.
This study investigates how mothers’ pregnancy intentions affect prenatal and postnatal health behaviors and infant health at birth in the United States.
The study addresses two key issues that may explain the inconsistent or weak findings in previous research:
The measurement and conceptualization of pregnancy intentions.
The challenge of separating pregnancy intentions from demographic and socioeconomic characteristics that also affect maternal behaviors and infant health outcomes.
The study uses more-nuanced measures of pregnancy intentions than those used in previous studies:
An expanded version of the conventional measure that incorporates the extent of mistiming.
A multidimensional measure of pregnancy desire developed by Santelli et al. (2009).
This research employs propensity scores, a statistical method not previously used in this field, to test and adjust for observed variation in demographic and socioeconomic characteristics of mothers with differing pregnancy intentions.
The goal is to determine whether differences in beneficial maternal behaviors or infant health are due to the characteristics of women in different intention status groups rather than to the direct effect of pregnancy intention.
Background
Conventional Measures of Pregnancy Intentions
The National Survey of Family Growth (NSFG) is the primary data source for measuring pregnancy intentions in the United States.
Since 1973, the NSFG has allowed researchers to classify pregnancies as wanted, unwanted, or mistimed (Campbell and Mosher 2000).
Respondents are asked about their feelings regarding having a baby just before they became pregnant, specifically whether they wanted a baby at any point in the future.
If respondents wanted no children or no more children, the pregnancy is classified as “unwanted.”
If they wanted a baby at some point in the future, the pregnancy is classified as “wanted.”
Respondents with wanted pregnancies are asked whether the pregnancy occurred sooner than they would have liked, at about the right time, or later than they would have liked.
Pregnancies that occurred sooner than preferred are classified as "mistimed."
This retrospective “conventional” measure assesses a woman’s control over her reproductive life by determining whether she wanted a child just before becoming pregnant and whether she became pregnant at the right time for her.
Other surveys, such as the Pregnancy Risk Assessment Monitoring System (PRAMS) and the Demographic and Health Surveys (DHS), phrase the questions slightly differently but generally adopt the same categories of wanted, mistimed, or unwanted.
Most often, pregnancies are classified as either “intended” or “unintended.”
Intended pregnancies are those wanted at, or sooner than, the time they occurred.
Unintended pregnancies include unwanted and mistimed pregnancies.
Pregnancies where a woman reports that she “didn’t care” whether she had a baby are typically labeled as intended.
The terms “intended” and “unintended” may imply planning or intentional behavior that may or may not have occurred.
Women are asked whether they wanted a baby, not whether they intended a pregnancy.
Most studies on the consequences of unintended pregnancy distinguish between two (intended or unintended) or three (intended, mistimed, or unwanted) categories of intention status.
Because unintended pregnancies combine wanted (mistimed) with unwanted pregnancies, this dichotomous measure may be too broad for investigating the impact of childbearing intentions on maternal behaviors and infant health.
The three-category measure of pregnancy intentions (intended, mistimed, or unwanted) may also be too inclusive, combining pregnancies that were only moderately mistimed with those that were greatly mistimed.
A four-category intention status measure can be constructed using responses to an additional question asked of women reporting mistimed pregnancies: how long she had wanted to wait (measured in weeks or months).
Research has found meaningful distinctions between pregnancies that were “slightly” mistimed (by less than two years) and those that were greatly mistimed (by two or more years) (Lindberg et al. 2008; Mosher et al. 2012; Pulley et al. 2002).
The four categories of this measure are:
Wanted at that time or sooner (intended).
Mistimed by less than two years.
Mistimed by two or more years.
Unwanted at any time.
The question used is, “Right before you became pregnant with your (Nth) pregnancy (which ended in [date]), did you yourself want to have another baby at any time in the future?”
The four answer choices are “Yes,” “No,” “Not sure/Don’t know,” and “Didn’t care.”
In the NSFG, very few respondents chose “Not sure/Don’t know” ( for our analysis of births; coded as missing).
Those responding “Didn’t care” can be classified as intended, unwanted, or missing, depending on the focus of the analysis ( for our analysis; coded as intended).
This construct expands the conventional measure of pregnancy intentions by allowing for variation in wantedness and timing.
The conventional measure of intention status from retrospective surveys has been used to inform policies and programs (Campbell and Mosher 2000; Finer and Kost 2011; Klerman 2000). However, there are growing concerns about its use for individual-level analyses, and numerous critiques have been published (see Bachrach and Newcomer 1999; Klerman 2000; Luker 1999; Miller and Jones 2009; Peterson and Mosher 1999; Santelli et al. 2003).
These critiques have formed their own body of research (see also Bachrach and Morgan 2013; Barrett and Wellings 2002; Fischer et al. 1999; Gerber et al. 2002; Higgins et al. 2012; Kaufman et al. 1997; Kavanaugh and Schwarz 2009; Kendall et al. 2005; Lifflander et al. 2007; McCormick et al. 1987; Moos et al. 1997; Petersen and Moos 1997; Poole et al. 2000; Santelli et al. 2006; Santelli et al. 2009; Stanford et al. 2000; Trussell et al. 1999; Westoff and Ryder 1977).
Alternative Measures of Pregnancy Intentions
Conventional measures of pregnancy intention may not adequately capture gradations in attitudinal dimensions of childbearing.
Expanded questions on childbearing intentions were added to more recent rounds of the NSFG to provide more-refined measures (Klerman and Pulley 1999; Peterson and Mosher 1999; Mosher et al. 2012), including two questions measured on a Likert-type scale to assess how much women had wanted to avoid or have a pregnancy, and how much they had been trying to avoid or become pregnant.
While the wanting scale is an expanded measure of the same dimension included in the conventional measure, the trying dimension is new.
This question asks women to think back to before they were pregnant and assess how much effort they had put into reaching their childbearing goals.
Although these two scales have been included in the NSFG since 2002, few studies have used these measures instead of, or in addition to, the conventional measure of intention (Miller and Jones 2009; Mosher et al. 2012; Santelli et al. 2009).
Staff at the National Center for Health Statistics now use this four-category measure in their publications of NSFG analyses and advise all users of the NSFG to do so (personal communication, William D. Mosher, October 17, 2012).
The Pregnancy Risk Assessment Monitoring System (PRAMS) initiated the additional question on timing in 2012.
There is some implied gradation of wantedness in the conventional measures, with overdue pregnancies presumed to have the highest level of desire and mistimed pregnancies presumed to reflect somewhat less desire than on-time pregnancies. However, the validity of this assumed gradation is unknown.
These questions were based on the psychosocial theories of Warren Miller, which posit multiple dimensions of pregnancy attitudes that affect motivations and behavior before, during, and after pregnancy (Miller 1992, 1994, 1998; Miller and Pasta 2002; Miller et al. 2004).
The specific wording for the wanting scale is as follows: “0 means you wanted to avoid a pregnancy and a 10 means you wanted to get pregnant. If you had to rate how much you wanted or didn’t want a pregnancy right before you got pregnant (that time), how would you rate yourself?”
The wording for the trying scale is, “0 means trying hard not to get pregnant and a 10 means trying hard to get pregnant. If you had to rate how much you were trying to get pregnant or avoid pregnancy right before you got pregnant (that time), how would you rate yourself?”
Another scaled measure, included beginning with the 1995 NSFG, gauges women’s happiness when they discovered they were pregnant.
Many women who experience unintended pregnancies nonetheless report high levels of happiness (Hartnet 2012; Lindberg et al. 2008; Trussell et al. 1999).
This measure of happiness can be a stronger predictor of women’s behaviors during pregnancy than their reported intentions (Blake et al. 2007; Sable and Libbus 2000; Santelli et al. 2009).
From 2002 onward, the NSFG survey also asked the woman about her male partner at the time of the pregnancy and whether she had wanted to have a baby with him, building from influential research on a clinical sample by Zabin et al. (2000) finding that women expressed not wanting to get pregnant “with this partner.”
How the woman felt about becoming a parent with a current partner may be another dimension contributing to how much she had wanted to have or avoid a pregnancy (Kroelinger and Oths 2000).
Santelli et al. (2009) developed an improved multidimensional measure of unintended childbearing using the additional NSFG questions.
They devised a “desire scale” by combining all the aforementioned measures:
The wantedness component from the conventional measure (wanted/unwanted).
The three Likert-scale questions on wanting, trying, and happiness when pregnancy was discovered.
The question on whether the woman had wanted to have a baby with that partner.
The desire scale parsed into seven ordinal categories had a strong relationship with pregnancy outcomes, such that women who had low levels on the desire scale were more likely to obtain an abortion than women with higher levels (Santelli et al. 2009).
While timing is not part of the desire scale, Santelli et al. (2009) identified the extent of mistiming as a unique dimension predictive of the decision to abort or continue the pregnancy.
Desire Scale Formula
Respondents who had wanted the pregnancy at the time it occurred receive 1 point (“on time” in the preceding equation), and those for whom the pregnancy was unwanted lose 1 point (“unwanted” in the equation).
Testing Two Expanded Measures of Pregnancy Intentions
This analysis examines whether maternal behaviors and birth outcomes differ by pregnancy intentions, comparing findings using two measures of pregnancy intentions:
The conventional measure expanded to four categories (intended, mistimed by less than two years, mistimed by two or more years, or unwanted).
A version of the multivariable desire scale proposed by Santelli et al. (2009).
These measures share some traits but not others.
Only the conventional measure includes the timing dimension, but it is more limited than the desire scale in its ability to capture variation in the strength of pregnancy wantedness.
The desire scale includes information gleaned from several measures, including happiness about being pregnant and the woman’s attitude toward having a baby with the father.
Previous Findings
Unintended pregnancy rates in the United States differ sharply by demographic and socioeconomic subgroup (Finer and Zolna 2014).
Pregnancy intentions are strongly related to women’s demographic characteristics (age, marital status, race, ethnicity, and parity) and socioeconomic characteristics (educational attainment, income, and poverty status) (D’Angelo et al. 2004; Hayford and Guzzo 2010; Joyce et al. 2000b; Kost and Forrest 1995; Pulley et al. 2002; Williams 1991; Williams et al. 1999).
These characteristics can also predict:
Late recognition of pregnancy (Ayoola et al. 2009).
Later initiation or lower levels of prenatal care (Ayoola et al. 2010; Centers for Disease Control and Prevention 2002; Taylor et al. 2005).
Initiation, continuation, and exclusive use of breast-feeding after delivery (Ahluwalia et al. 2003; Centers for Disease Control and Prevention 2002; DiGirolamo et al. 2005; Jones et al. 2011; Li et al. 2005; McDowell et al. 2008; Merewood et al. 2006; Singh et al. 2007; Thulier and Mercer 2009).
Poor birth outcomes, such as small for gestational age (McCowan and Horgan 2009), low birth weight (Ashdown-Lambert 2005; Blumenshine et al. 2011; Keeton and Hayward 2007), or preterm births (Afable-Munsuz and Braveman 2008; Blumenshine et al. 2011; El-Sayed et al. 2012; Keeton and Hayward 2007).
The effects of pregnancy intentions on these outcomes are likely confounded with the effects of the mother’s background characteristics.
Researchers have used multivariate regression to control for mothers’ characteristics when testing for a relationship between pregnancy intentions and maternal behaviors or infant health (Altfeld et al. 1997; Baydar 1995; Joyce et al. 2000a; Korenman et al. 2002; Kost et al. 1998a, 1998b; Marsiglio and Mott 1988; Mohllajee et al. 2007; Pulley et al. 2002; Weller et al. 1987).
Prior Studies Relating Pregnancy Intentions to Maternal Behaviors and Infant Health
Research on the consequences of unintended childbearing dates back to Forssman and Thuwe’s (1966) 21-year follow-up study of 120 births in 1939–1942 to Swedish women denied abortions and matched control births born at the same hospital on the same day.
The two groups differed significantly in both their background characteristics and a range of child outcomes, including health, educational and occupational attainment, public assistance, and criminal, military, and social services records.
The researchers concluded that an unwanted child is “born into a worse situation” than other children and “runs a risk of having to surmount greater social and mental handicaps than its peers” (p. 87).
However, all statistical comparisons were tests of differences in outcomes for the two groups with no attempt to control for differences in background characteristics.
Henry David and colleagues (1988) studied 220 births in 1961–1963 to women twice denied an abortion in Prague, Czechoslovakia.
Births to these mothers were matched to control births whose mothers did not try to terminate the pregnancy.
Births were matched in pairs, using age, birth order, number of siblings, and school class, and were followed into childhood.
This matching strategy was an early attempt to control for the differing background characteristics underlying the intention status groups.
The researchers found that unwanted births faced significantly more disadvantages on health and school performance measures.
Following these studies, other researchers have investigated the impact of pregnancy intentions on maternal behaviors and infant health.
In 1995, the Committee on Unintended Pregnancy, convened by the Institute of Medicine at the National Academies of Science, reviewed research to date and concluded that:
The consequences of unintended pregnancy are serious, imposing appreciable burdens on children, women, men, and families.
A woman with an unintended pregnancy is less likely to seek early prenatal care and is more likely to expose the fetus to harmful substances (such as tobacco or alcohol).
The child of an unwanted conception especially (as distinct from a mistimed one) is at greater risk of being born at low birthweight, of dying in its first year of life, of being abused, and of not receiving sufficient resources for healthy development.
The mother may be at greater risk of depression and of physical abuse herself, and her relationship with her partner is at greater risk of dissolution.
Both mother and father may suffer economic hardship and may fail to achieve their educational and career goals (Brown and Eisenberg 1995:250–251).
Most importantly, the committee also concluded that “Unintended pregnancy is not just a problem of teenagers or of unmarried women or of poor women or minorities; it affects all segments of society” (Brown and Eisenberg 1995:250).
This report concluded that the negative effects of unintended childbearing were not simply due to underlying maternal characteristics.
Each of the maternal behaviors and birth outcomes addressed in this article—early pregnancy recognition, early initiation of prenatal care, breast-feeding, low birth weight, and preterm delivery—has a substantial body of literature linking it with infant and child health (American Academy of Pediatrics 2012; Ayoola et al. 2009; Ayoola et al. 2010; Callaghan et al. 2006; Kramer and Kakuma 2004; MacDorman et al. 2013; McDowell et al. 2008; U.S. Department of Health and Human Services 2011).
The relationship between pregnancy intentions and these outcomes remains less clear.
In 2008, Gipson and colleagues provided an updated review of the research literature, focusing specifically on methodologically rigorous studies that attempted to control for sociodemographic background characteristics, but they found mixed evidence for the effects of pregnancy intentions on early pregnancy recognition, early prenatal care initiation, and measures of infant health at birth (Gipson et al. 2008).
Numerous U.S.-based studies showed an association between pregnancy intentions and delayed initiation of prenatal care, but several studies also found that the relationship was diminished in multivariate analyses that included measures of the mother’s demographic and socioeconomic characteristics.
Cheng et al. (2009) examined PRAMS data limited to births in Maryland and found that both mistimed and unwanted births were significantly less likely to initiate prenatal care in the first trimester of pregnancy, after controlling for maternal age, race/ethnicity, education, marital status, Medicaid status, and parity.
After examination of both U.S. and European studies, Gipson et al. (2008) concluded that there was consistent evidence of a relationship between pregnancy intentions and breast-feeding, but one of the five U.S.-based studies reviewed did not find a significant relationship overall (Marsiglio and Mott 1988), and one did not find significantly lower probabilities of breast-feeding among mistimed births (Kost et al. 1998b).
Although many studies have found bivariate associations between pregnancy intentions and breast-feeding initiation and duration, the effects of pregnancy intentions on breast-feeding can be greatly diminished by covariates included in the analyses (Cheng et al. 2009; Joyce et al. 2000a).
A meta-analysis of the effects of pregnancy intention on low birth weight and preterm birth found consistent evidence of a bivariate relationship across studies: mistimed and unwanted births were more likely to be low birth weight and preterm than were intended births (Shah et al. 2011).
Among studies that included background characteristics of the mothers, only two were based on nationally representative data from the United States and examined births occurring in the 1980s and early 1990s (Kost et al. 1998b; Joyce et al. 2000a).
Joyce et al. (2000a) found no association of pregnancy intentions with an infant’s risk of low birth weight; the analyses did not examine preterm births.
Kost et al. (1998b) found no significant relationship of pregnancy intention with a combined measure of preterm birth, low birth weight, or small for gestational age after maternal behaviors during pregnancy and sociodemographic characteristics were included in the model.
A recent population-based study of births in Ireland also found no significant relationship of pregnancy intentions with low birth weight or preterm birth after adjusting for mothers’ background characteristics, although births from mistimed and unwanted pregnancies were combined in one unintended category (McCrory and McNally 2013).
In the only nationally representative study to investigate the impact of pregnancy intentions on the timing of the mother’s pregnancy recognition, Kost et al. (1998a) used data from the 1988 National Maternal and Infant Health Survey and found that mothers of mistimed and unwanted births were significantly less likely to recognize that they were pregnant within the first six weeks of pregnancy than mothers of intended births, even after the researchers controlled for numerous background characteristics.
Data and Methods
Data for this study come from pregnancy histories of women surveyed in the 2002 and 2006–2010 NSFG conducted by the National Center for Health Statistics (NCHS).
The NSFG is a national probability survey of the noninstitutionalized population aged 15–44 in the United States (Groves et al. 2009; Lepkowski et al. 2006, 2013).
Following recommended protocols, observations from the 2002 () and 2006–2010 () surveys were pooled (National Center for Health Statistics 2011).
The study focused on women’s behaviors during pregnancy and immediately after the birth, as well as infant health outcomes.
The analysis is limited to pregnancies ending in a birth.
The conventional measure of pregnancy intention is applied to every reported pregnancy, while the trying, wanting, and happiness scales are limited to pregnancies that occurred within three years of the interview.
To compare findings for the two measures of intentions, the unit of analysis is nonmultiple live births in the three years prior to the survey interview. This time frame reduces the risk of retrospective reporting bias.
Women can contribute more than one birth to the analysis; 27% of births have at least one sibling in the analytical sample.
All births in the three-year window were included because the pregnancy histories of women interviewed in the NSFG are expected to be representative of all births in the United States near the time of the survey (Joyner et al. 2012; National Center for Health Statistics 2011).
Potential autocorrelation among births with the same mother was accounted for by including the mother’s unique identification code as a cluster indicator in complex survey design commands, in addition to the other design variables specified for use with the NSFG.
Analyses were limited to births of mothers age 20 or older at conception. Teen pregnancies were excluded because with 73% of births to teen mothers unintended, it would be difficult to distinguish the role of intention status from the role of age.
The number of observations for analysis was 4,297, representing all singleton live births to women aged 20–44 in the periods 1999–2002 and 2004–2010.
In analyses, the four-category conventional measure of pregnancy intentions and the desire scale, ranging from 0 to 6, were used.
This continuous measure was parsed into a five-category desire scale based on quintiles of its frequency distribution.
Outcome measures were constructed with binary responses (no/yes) following guidelines in Healthy People 2020 (U.S. Department of Health and Human Services 2011).
Two measures of maternal health behaviors during pregnancy were used:
Whether the mother recognized she was pregnant within the first six weeks of the pregnancy.
Whether prenatal care was initiated in the first trimester.
One measure of a maternal health behavior following pregnancy: breast-feeding.
Whether the baby was ever breast-fed for any length of time; among those who breast-fed, whether the infant was exclusively breast-fed for at least six months (limited to births with age greater than six months at interview), and whether breast-fed for at least one year (limited to births with age greater than 12 months at interview) were examined.
Finally, two measures of infant health at birth were examined: preterm delivery and low birth weight.
A further 36 births with missing data were omitted in analyses of the conventional measure, and 41 births were missing a value for the desire scale measure (only three births were missing values for both measures).
Quintiles produced upper-level cutoff values of 2.69, 4.35, 5.09, 5.91, and 6.0. The substantive meaning of these values is unknown, as is that for the integer values used by Santelli et al. (2009). However, quintiles may capture underlying groupings of similar desire categories more accurately than arbitrary levels of integers.
Preterm is defined as a delivery occurring at or before 36 weeks of pregnancy; low birth weight is defined as ≤88 ounces or ≤2,500 grams (<5 lbs., 8 oz.).
Analytic Strategy
First, it was examined whether dependent variables—maternal health behaviors and birth outcomes—vary by demographic and socioeconomic characteristics of the mothers because these characteristics are also associated with pregnancy intentions and would be potential confounding variables.
Variation in the distribution of the mothers’ demographic and socioeconomic characteristics across the four intention status groups of the conventional measure and the five categories of the desire measure was identified.
These distributional differences motivated efforts to separate the effects of pregnancy intentions on the outcomes from those attributable to the background characteristics.
Inverse propensity (probability) weights, an adaptation of propensity score analysis, were employed.
Propensity score methods are used for adjusting the distribution of characteristics of two groups (a treatment and a control group) so that they are matched (“balanced”) with respect to observed characteristics that are relevant to group assignment but also affect the outcome of interest (Austin 2011; Rosenbaum and Rubin 1983; Stuart 2010).
Imbens (2000) extended the logic of propensity score methods for applications involving more than two groups by estimating group-specific propensity scores.
The inverse of these probabilities can be applied to weight observations and create balanced comparison groups (McCaffrey et al. 2013). This method was used for the analysis of multiple intention status groups.
Intended births were weighted by the inverse of the propensity of having been intended; births mistimed by less than two years were weighted by the inverse of the propensity of being in that mistimed group, and so on.
Inverse propensity weighting gives greater weight to observations in each intention status group that have a low probability of being in that group so that they represent a larger proportion of the births in their group in weighted analyses.
The inverse weighting creates distributions of characteristics for each group that resemble the full sample, thus equating the groups.
The propensity scores used for weighting were estimated from a multinomial logistic regression, with intention status as the dependent variable (intended births were the reference category).
The independent variables included in the model were all available demographic and socioeconomic measures that prior research has found to be related to intention status:
Age of the mother at conception (20–24, 25–29, 30–44, as well as a continuous measure of single year of age).
Maternal union status at conception (married, cohabiting, not in union).
Maternal race/ethnicity (non-Hispanic white, foreign-born Hispanic, native-born Hispanic, non-Hispanic black, non-Hispanic other races).
Whether the mother is foreign-born (no, yes).
Whether the mother was a high school graduate at interview (no, yes).
Because education measured after the birth cannot predict the pregnancy intention of the birth, it was reasoned that for most mothers in our sample (age 20 or older at conception), very few who had not completed high school before age 20 would complete it after that age.
The respondent’s mother’s education (less than high school, high school graduate or GED, some college or more).
The order of the birth (first birth, second birth, third or higher-order birth).
Whether the delivery was paid for by Medicaid, as a rough proxy for economic status near the time of the birth because income—and poverty status—are measured only at the time of interview and could be affected by the birth or subsequent ones.
Variables to account for survey implementation and other potential biases: three periods covered by the NSFG surveys (women interviewed in 2002, 2006 to the first half of 2008, and the last half of 2008 to 2010) and the natural log of the length of recall (measured in months, from birth to mother’s date of interview).
Unlike multivariate regression used for explanatory purposes, the regression model used for estimating propensity scores did not seek parsimony. All measures potentially predictive of intention status were included, regardless of statistical significance.
The propensity score estimation process was assessed by calculating a “standardized bias,” defined as the absolute value of the difference in means of each of the paired intention status groups divided by the standard deviation of the mean for all births (with each of the three unintended birth groups compared separately with the intended birth group).
This measure is recommended in the statistical literature, and unlike standard statistical tests (e.g., t-tests), the standardized bias is not affected by sample size (Stuart 2008, 2010).
The propensity estimation model was finalized, and the adjusted distributions of characteristics across intention status groups were considered balanced after all estimates of standardized biases fell below .25 (see Tables 6 and 7 in the appendix).
This process was iterative: various interaction terms as well as the optimum array and form of variables were tested.
The propensities of being intended, mistimed by less than two years, mistimed by two or more years, and unwanted were estimated individually.
The distributions of characteristics across intention status groups were compared before and after weighting to ensure that the weighting procedure balanced the distributions.
Results
This section presents results from bivariate analyses of maternal behaviors and birth outcomes by pregnancy intention status, using both the conventional measure and the desire scale.
It also shows results from multivariate analyses, using propensity score weighting to adjust for observed differences in demographic and socioeconomic characteristics across intention status groups.
Descriptive Statistics
Table 1 presents the distribution of births by intention status for the four-category conventional measure and the five-category desire scale.
Under the conventional measure, 60% of births were intended, 18% were mistimed by less than two years, 11% were mistimed by two or more years, and 11% were unwanted.
The distribution of births across the five categories of the desire scale was fairly even, ranging from 16% to 24%.
Table 2 presents descriptive statistics for the full sample of births.
About 16% of new mothers did not recognize their pregnancy until after six weeks, and about 21% did not begin prenatal care in the first trimester.
Sixty-nine percent of women breast-fed their infants, but among those who breast-fed, only 19% breast-fed exclusively for at least