Contraceptive Behavior Dynamics and Unintended Pregnancy: A Latent Transition Analysis

Abstract

The average U.S. woman wants to have two children; to do so, she will spend about three years pregnant, postpartum, or trying to become pregnant, and three decades trying to avoid pregnancy. However, few studies have examined individual patterns of contraceptive use over time. These trajectories are important to understand given the high rate of unintended pregnancy and how little we know about the complex relationship between contraceptive use, pregnancy intention, and patterns of reproductive behavior. We use data from the 2015–2017 National Survey of Family Growth to examine reproductive behavior and pregnancies across three years of calendar data. We identify seven behavior typologies, their prevalence, how women transition between them, and how pregnancies affect transitions. At any given time, half of women are reliably using contraception. A small proportion belong to a high pregnancy risk profile of transient contraceptive users, but some transition to using condoms or other methods consistently. An unintended pregnancy may initiate a transition into stable contraceptive use for some women, although that is primarily condom use. These findings have important implications for the ways contraception fits into women's lives and how that behavior interacts with relationships, sex, and life stage trajectories.

Introduction

Unintended pregnancy¹ is defined as a pregnancy that occurred when a woman wanted to become pregnant in the future but not at the time she became pregnant, or one that occurred when she did not want to become pregnant then or at any time in the future. In 2015, rates of pregnancies that occurred among women who did not want to have a baby ranged from 8.1 to 22.7 per 1,000 women aged 15‒44 (Kost et al. 2023). The U.S. Department of Health and Human Services' initiative to reduce unintended pregnancy is one of their most important national goals (Finer and Zolna 2011; U.S. Department of Health and Human Services 2020). Unintended pregnancy is associated with poor health outcomes for some women, including inadequate or delayed initiation into prenatal care, premature and low birth weight infants, and increased risk for physical and mental health problems for women and their children (Gipson et al. 2008; Herd et al. 2016). Unintended pregnancy rates serve as a proxy measurement to help public health institutions understand the unmet need for contraception services across the United States, and the large numbers of women experiencing such pregnancies is a strong signal that U.S. women are still unable to achieve reproductive autonomy. However, unintended or undesired pregnancies are an inadequate indicator of women's² lived experiences. Examination of this measure alone cannot capture the complexities of women's experiences prior to having an unintended pregnancy, the context in which the pregnancy occurs, or what happens to an individual woman after she experiences a pregnancy she did not want (Guttmacher Institute 2019).

Individual reproductive health trajectories are rarely measured longitudinally, although new studies have identified certain behavior profiles that characterize women's dynamic contraceptive and reproductive health experiences in certain age groups or populations (Brew et al. 2020; Hayford 2009; MacQuarrie et al. 2019). The results from these studies show heterogeneity in fertility preferences and contraceptive trajectories across women of reproductive age throughout the world. However, similar longitudinal contraceptive trajectories have yet to be explored across the larger U.S. population.

The current study uses latent transition analysis (LTA) to estimate longitudinal patterns of contraceptive behavior, sexual activity, and life stage and the probability of transitioning between typical behavior patterns over a three-year period. Instead of directly measuring a single observed variable, such as contraceptive method type, LTA allows us to capture additional nuance of women's reproductive behavior using latent class analysis (LCA) across time. We use a population-based, national sample of U.S. women to explore distinct subgroups of women by reproductive behavior across a three-year period, quantify the change between latent subgroups over time, and explore how pregnancy experiences affect individual change in contraceptive behavior.

Background

Family planning services help women avoid pregnancies they do not want and plan pregnancies they do. Many studies show that programs for affordable and accessible contraception are some of the most cost-effective in existence (Centers for Disease Control and Prevention 2000). In 2014 alone, these services helped U.S. women avoid two million unintended pregnancies (Frost et al. 2016). But the relationship between contraceptive use and pregnancy intention is complex (Borrero et al. 2015; Kost and Lindberg 2015; Trussell et al. 1999). At any given time, a woman's contraceptive use “depends upon the internal balance between positive and negative feelings towards getting pregnant and her positive and negative feelings about a current contraceptive method” (Miller 1986:27). Using a national sample of women aged 18–39, Jones et al. (2015) compared the effects of two measures of pregnancy attitudes—current importance of avoidance of pregnancy (pregnancy avoidance) and unhappiness at the thought of being pregnant (pregnancy feelings)—on consistent contraceptive use over time. Only pregnancy avoidance was associated with consistent contraception use, suggesting that the relationship between pregnancy feelings or intentions and subsequent contraceptive use is not as straightforward as often theorized.

Kost et al. (2023) found that rates of pregnancies that occurred among women who did not want to be pregnant fell significantly between 2009 and 2015 among individuals aged 15‒19 and 20‒24. There is evidence that the increase in use of highly effective long-acting reversible contraceptive (LARC) methods in recent years, such as IUDs and implants, has contributed to the decline in the rate of unintended pregnancy (Guttmacher Institute 2014; Mosher et al. 2016). However, several studies have found that the increase in the use of IUDs and implants is not associated with a decrease in nonuse of contraception. Most of the increase in LARC use is attributed to women who were already using some form of contraceptive and who switched to highly effective methods; few women shift from no contraception or from such over-the-counter methods as condoms to LARC methods (Eeckhaut 2022; Nelson et al. 2019). These findings demonstrate that contraceptive trajectories other than transitioning from nonuse to use can lead to reductions in unintended pregnancy by reducing method failures rather than by increasing overall contraceptive use and, hence, underscore that identifying distinct contraceptive trajectories is relevant to our understanding of outcomes like unintended pregnancies.

Expanding the Literature on Longitudinal, Individual Patterns of Reproductive Health

Few studies have examined individual patterns of contraceptive use over time because longitudinal patterns can be difficult to study. For example, an individual's contraceptive use is initiated through a multifaceted decision process. Some methods to prevent pregnancy can be used for minutes, while others last decades. The process is repeated across life stages. Contraceptive use also interacts with fluctuating pregnancy intentions, parity, partnership status, and socioeconomic forces across time (Frost et al. 2007). Contraceptive attitudes also evolve. A woman's contraceptive needs and expectations change as she progresses through the life course (Dehlendorf et al. 2014; Frost et al. 2012). It is important to consider and appreciate the complexity of individual contraceptive use behavior and how it can change over time.

A growing number of recent studies have used longitudinal data to examine contraceptive behavior from a person-centered perspective. The Relationship Dynamic and Social Life study is a panel study that followed 992 Michigan women aged 18–19 weekly for two and a half years. Brew et al. (2020) applied sequence analysis to 581 women in this sample and identified six trajectories of joint relationship, sex, and contraceptive typologies. In a different five-year, retrospective study using Demographic and Health Survey contraceptive history data, MacQuarrie et al. (2019) found several profiles based on contraceptive use and pregnancy experiences in large samples of women aged 15–49 from Burundi and Nepal. Looking at the five-year sequence patterns, the authors used cluster analysis to define contraceptive profiles based on contraception and pregnancy status. They found six distinct subgroups of women's reproductive behavior in these two countries, including “Modern Mothers,” who adopt such short-term modern methods as birth control pills; “Consistently Covered Mothers,” who use long-acting methods; and “Family Builders” and “Quiet Calendar,” both of whom do not use contraception and either become pregnant multiple times (Family Builders) or do not (Quiet Calendar) (MacQuarrie et al. 2021; MacQuarrie et al. 2019; MacQuarrie et al. 2022). These studies used novel methodology to highlight the heterogeneity of contraceptive use behavior and reproductive experiences over time. Brew et al. (2020) focused on young adult experiences over one year in a Michigan county, whereas MacQuarrie et al. (2019) captured a single contraceptive state through five years of calendar data among women from Burundi and Nepal.

Reproductive Life Course

One of the primary reasons for an individual's use of contraception is to avoid pregnancy and control the occurrence of a major life event (i.e., having a child) (Gray and McDonald 2010). While age is a valuable component for understanding contraceptive behavior in the context of the life course (Daniels and Abma 2020; Harris et al. 2021), individual trajectories of reproductive behavior do not follow purely “age-based paths” (Gray and McDonald 2010:44). The life course approach is useful in studying contraceptive behavior because it emphasizes individual experiences and influences over time (Gray and McDonald 2010; Halfon et al. 2014). This study aims to expand on previous research by adding important reproductive life stage contexts to women's contraceptive profiles, including parity and marital status as life stage markers, as well as constructs describing the consistency of contraceptive use behavior. Our goal is to understand how behaviors, exposure, and life stage interact to produce a richer understanding of reproductive behavior typology over time. Additionally, we were motivated to explore the extent to which the experience of an unintended pregnancy is transformative to subsequent contraceptive behavior. Rajan et al. (2017) found that having one unintended birth increased the likelihood of subsequent unintended births and that various characteristics mediate this process (age, race, education, family structure, personality). Our analysis aims to build on that study, specifically by exploring how contraceptive behavior might be altered by the experience of an unintended pregnancy.

Methods

Participants

Data for this analysis come from the National Survey of Family Growth (NSFG) conducted in 2015–2017 (National Center for Health Statistics 2018). The NSFG is conducted in-home via face-to-face interviews of men and women aged 15–49. In total, 5,554 women were interviewed in the 2015–2017 data collection round. We restricted our analytic sample to women respondents with complete contraceptive calendar data. Fifteen respondents (0.3%) were dropped because of missing data, resulting in a final analytic sample of 5,539 women. Four women had completely missing calendars, and 11 had at least one month of contraceptive calendar data missing; none of the covariate data were missing.

Analytic Approach: Latent Class and Latent Transition Analysis

LTA describes a type of longitudinal autoregressive model common in social science research in which the outcome variable is a latent categorical variable. The latent construct is captured using a measurement model, which is most commonly an LCA model. The LTA model describes change in outcomes where the outcomes are not directly observed—that is, the outcomes are latent and are indicated by a set of observed variables. LTA is particularly relevant for studying the longitudinal trajectories of contraceptive behavior. First, it is a person-centered, model-based approach that identifies classes of women on the basis of multiple aspects of individual contraceptive behavior, sexual activity, and life stage. Second, LTA addresses change over time and the extent to which other variables are related to that change. With LTA, individuals can move through discrete categories or stages.

In this analysis, LTA is used to identify longitudinal patterns and transitions across classes throughout all three years of the 2015–2017 NSFG contraceptive calendar. Figure 1 shows the specific LTA conceptual model for this analysis. The first step in fitting and evaluating LTA models is to explore the measurement models at each time point of our analysis to inform the number of latent statuses in the longitudinal model. We use LCA models across each of the calendar years. LCA is a measurement model in which individuals can be classified into mutually exclusive and exhaustive types using their patterns of answers on a set of indicator variables. The starting point for conducting an LCA on empirical data is a contingency table formed by cross-tabulating all the observed indicator variables. A latent class model is then made up of the estimated latent class prevalence rates and item-response probabilities, which can be used to obtain expected cell proportions for this contingency table.

Latent transition analysis then allows us to characterize change between latent classes across time. We use “status” to describe the unobserved groups in this LTA model. For example, if an individual is in a particular latent status at study year 1, what is the probability that the individual will be in that latent status at study year 2, and what is the probability that this individual will be in a different status? The probability of membership in latent status s at time t is a function of the probabilities of membership in each latent status at time t – 1 and the conditional probability of transiting from each latent status into latent status s between times t – 1 and t. To interpret this change over time, measurement invariance—or constraining the item-response probabilities in LTA to be equal across times—is an important assumption. The assumption of full measurement invariance facilitates straightforward discussions about transitions among the statuses because the statuses are always the same across the three-year period. Formal measurement invariance testing in the form of likelihood ratio tests is used to statistically assess the plausibility of measurement invariance. We expect measurement invariance in this model. Measurement may vary by an individual woman's age; however, there is less reason to suspect that measurement would vary from year to year in any meaningful way given that the women in our sample are heterogeneous with respect to age within any given year.

LCA and LTA Model Specification and Analysis

Models were based on Ryoo et al. (2018) and Nylund et al. (2007). Stata 16.0 was used for data management and analytic sample creation and regression, while Mplus 8.4 (Muthén and Muthén 2019) was used for the LTA model.

We used the eight indicators of contraceptive behavior, sexual activity, and life stage described below for the LCA. This was done for each study year. Covariates were not considered in this part of the analysis. After latent statuses were defined, we formally tested the transition probability invariance from year 1 to year 2 to year 3.

If model entropy is high, the classification of most likely status in the LTA is close to the true classification. Under these conditions, we can use multinomial logistic regression models to explore determinants of transitions between the latent statuses. We ran two multinomial models in which the dependent variable was the status in year t and the main independent variables of interest were status in year t ‒ 1 and pregnancy experience in year t ‒ 1, where t  =  2, 3. We also controlled for the following socioeconomic variables: age group at start of calendar year t (13–17, 18–22, 23–27, 28–32, 33–37, 38–42, and 43–47 at t  =  1), education level (less than high school, high school graduate, some college, and college degree or more), race and ethnicity (White non-Hispanic, Black non-Hispanic, Hispanic, and other), federal poverty level (≤ 138%, 139–199%, 200–299%, 300–399%, and ≥ 400% of federal poverty level), and religion (no religion, Catholic, Protestant, and other). The sample for each model was women who experienced a pregnancy in year t ‒ 1. We present predicted probabilities for transitions between the latent statuses from year t ‒ 1 to year t (t  =  2, 3) derived from the two multinomial logistic models for different subgroups of women classified by their experience of unintended and intended pregnancy in year t ‒ 1.

Measures Used to Define Latent Statuses

The NSFG does not follow the same women over time. However, the survey has a retrospective longitudinal component: the contraceptive calendar. This calendar captures the contraceptive method(s) female respondents used each month for the time period from the January three years prior to the interview date through the interview date. For example, for interviews done in October 2017, the contraceptive calendar collects monthly information from January 2014 through October 2017. The first three years of contraceptive data were used in this analysis, excluding the interview year, to standardize the amount of time in the survey for each woman. For a participant who was interviewed in October 2017, this study would capture her contraceptive behavior from January 2014 through December 2016. We summarized these data by calendar year and used the calendar year as the unit of time in the final model. Year profiles were done to align with the NSFG's reported pregnancies measure (as described below) and with common clinical practice cycles (e.g., annual gynecology exams).

Six contraceptive behavior indicators were created for years 1, 2, and 3 of the NSFG calendar data to describe contraceptive behavior (see Table 1). In the calendar, women were able to report up to four methods of contraception used during each month and sexual activity with any person of the opposite sex in that month. Two life course variables—marital status and parity—were also included in our analysis by pulling data from the larger NSFG survey and constructed to reflect the status of these variables during each calendar year of the three-year period. The NSFG does not distinguish months in which the participants were pregnant in the calendar data. Therefore, women who experience any months of pregnancy throughout the three-year period were characterized using other indicators (e.g., not using any contraception, months of unprotected sexual activity).

Most Frequent Type of Method

Participants were able to report up to four methods of contraception used in each calendar month.³ Using the first method reported in each month,⁴ methods were grouped into four categories of effectiveness and then aggregated over each year to represent the most frequent type of method used that year (Trussell et al. 2018):

1. Most effective: sterilization and vasectomy, IUDs, coil, loop, and hormonal implants.

2. Moderately effective: birth control pills, Depo-Provera, contraceptive patch, and vaginal contraceptive ring.

3. Least effective: barrier methods (female and male condoms), withdrawal, foam, jelly, suppositories, diaphragms and caps, and natural family planning (calendar/rhythm, standard days, cycle beads, and safe periods).

4. No method.

Frequency of Emergency Contraception Use

Use of emergency contraception (EC) was tallied as the number of months in which the method was used during each year. This variable considers all four methods that a woman could report over the course of one month. For example, if a respondent said she used withdrawal as her first method and then said she used emergency contraception as the second, emergency contraception use would be reported within that study month for that participant. This variable is categorized as 0 for no EC use during the calendar year and 1+ for any EC use during each year. The inclusion of EC use captures a distinct, postcoital contraceptive behavior; EC helps women prevent pregnancy in cases of forced sex, contraceptive failure, lack of use of contraceptives, or incorrect use (Larson et al. 2020).

Frequency of Contraceptive Method Switching

This measure captures the frequency of switching method types, based on the four categories of contraception described above. Given that there are 12 months in a year, a participant may switch methods up to 11 times in one study year. We collapsed the variable to 0, 1, or 2 or more method changes during each study year.

Parity

Parity was measured by the total number of pregnancies that ended in a live birth. Parity during each year was constructed by subtracting the number of live births since the start of each calendar year from the parity at the time of the interview. This variable is categorized as 0, 1, 2, or 3 or more live births.

Marital Status

Participants' marital status was assessed during each calendar year. A woman could be categorized as “single in the study year, never been married,” “single in the study year, had a previous marriage,” “married throughout the year,” and to indicate a relationship transition period within the study year, “got married in the study year” and “had a marriage dissolution in each study year.”⁵ A participant might have multiple marriage events in a study year (e.g., got married and had a marriage dissolution in the study year). The participant was categorized by the last event that occurred in that year. For example, if she entered the study year married, divorced within the year, and remarried in the year, then she would be categorized as “got married in the year.”

Condom Use Frequency

This variable represents the number of months the participant used condoms during each year and covers all four methods that a woman could report over the course of one month. For example, if a respondent said she used the pill as her first method and then said she used condoms as the second method, condom use would be reported within that study month for that participant. This variable was recorded as 0–12 months of condom use during each year.

Condom use was included as a distinct variable because consistent and correct use of condoms is an effective contraceptive and also reduces the risk of STIs (Centers for Disease Control and Prevention 2021). Condoms are relatively cheap and easy to find over the counter and are used at the time of intercourse. These characteristics may imply that condom users exhibit unique sexual and reproductive health behaviors that are relevant to their contraceptive trajectories. Condom use may imply less calculated, long-term prevention of pregnancy along with intentional prevention of STIs in the short term.

Unprotected Sexual Activity

Unprotected sexual activity accounts for the number of months a participant was sexually active and not using a method within each study year. This variable includes months in which women were pregnant or trying to become pregnant, as well as months in which women were sexually active, not using a method, and not trying to get pregnant. This variable is categorized as 0–12 months of inconsistent contraceptive use during each year.

Sexual Activity Frequency

The frequency of sexual activity represents the number of months the participant was sexually active with a partner of the opposite sex during each year. The NSFG asks about any occurrence of sexual intercourse during that month and does not distinguish vaginal intercourse from other forms of sexual intercourse. This variable is categorized as 0–12 months of sexual activity during each year.

Predictor of Transitions: Pregnancy Experience

The NSFG collects data on the year each pregnancy ended, the outcome of each pregnancy within the calendar (e.g., live birth, stillbirth, miscarriage, induced abortion), and the timing/intention/desired status of each pregnancy within the calendar. This information was used to classify pregnancies within study years 1 and 2 as either unintended or not. Unintended pregnancies include those that women reported as having occurred too soon or as being mistimed or unwanted, as well as those about which they reported feeling indifferent or that they “didn't care,” “don't know,” or were “not sure.” The not unintended—or intended—category includes pregnancies that women reported as having occurred at the right time, later than wanted, or as overdue. Two variables were then created indicating whether each woman experienced no pregnancies, only intended pregnancies, or any unintended pregnancies in study years 1 and 2 (see Table 2). These variables were used as covariates to predict latent status transitions in the subsequent year.

Results

Table 1 presents sample characteristics by calendar year. The mean age was approximately 30, with ages ranging from 13 to 49 over the study years. About a quarter of women used the most effective method types during the majority of each year, while about 18% and 15% used moderately effective and least effective methods, respectively. A large proportion of the sample—40‒45%—used no method during the majority of each year. More than three quarters of the sample never used condoms, while, on average, 7% used condoms every month. Most women in the sample never used emergency contraception. A large proportion of women (47‒48%) had sex every month of each study year, however, another large group (23‒30%) reported no sexual activity in each year. The majority of women were either not having sex or were using contraceptives every month that they did have sex, while about 7% had sex every month of the year and never used contraception. Regarding method switches, on average, about 75% of women never switched contraceptive method during each study year. A majority of the sample (more than 50%) were single and never married during the study, about 30% were married, and about 10% were single but previously married. Almost half of the sample had no children, and about equal proportions had one child, two children, or three or more. Table 2 shows that 91% of women did not report a pregnancy in years 1 or 2. The remaining women reported pregnancies that were intended (5%) or unintended (3.5%).

The seven status-solution models were favored with respect to the fit indices for all three time periods (see Tables A1 and A2, online appendix). Table 3 shows the results of the seven solution models' latent statuses with item probabilities and means for each latent status. On the basis of the item probabilities, we classified the seven different statuses of women (ordered by their predicted prevalence in the population; see Table 4): Single Abstinent Women (LS1); Consistently Covered Mothers (LS2) (MacQuarrie et al. 2019); Stable Users (LS3); Sexually Active Nonusers (LS4); Condom Users (LS5); Intermittent Users (LS6); and Transient Users (LS7). The first latent status, Single Abstinent Users, was characterized by abstinence. They had the lowest mean (0.0) of sexual activity of all seven statuses, and most were single and never married (85%), had no children (80%), and used no contraception (81%). The next two statuses, Consistently Covered Mothers and Stable Users, looked similar in terms of high mean months of sexual activity and low levels of unprotected sex. However, all of Consistently Covered Mothers had a least one child and 82% were using the most effective methods of contraception; this latent status consisted mostly of married women (64%) or previously married women (17%), but a small number (15%) were single and never married. Stable Users were identified by their high level of consistent sexual activity but low probability of having unprotected sex throughout the study period. They used a range of contraceptive methods: 51% used moderately effective methods, 21% used the most effective methods, and 16% used the least effective methods.

The fourth latent status, Sexually Active Nonusers, comprised women with a high probability of becoming pregnant in any given study year. They consistently had unprotected sex, more than any other latent status group. More than 75% of women in this status used no contraception and did not switch methods (71%). This status could include women who were pregnant, postpartum, or trying to become pregnant, as well as those who were having unprotected sex and did not want to become pregnant. Fifty percent were married and 36% were single and never married. Condom Users comprised the fifth latent status, and they had the highest mean number of months of condom use, almost equal to their mean months of sexual activity; they reported low levels of unprotected sexual activity. Half of the women in this status had no children and 62% were single and never married.

The sixth latent status, Intermittent Users, was composed of 75% single, never married women and was characterized by their low mean number of months of sexual activity over each year and low levels of unprotected sex. Even though most of the women in this status were using no method for most of the year (68%), they were not having frequent sex. Five percent of Intermittent Users used EC in the past year. Finally, the last latent status is Transient Users: most of these women switched contraceptive methods at least once during the year. They had sex frequently, and about half the time were unprotected. The majority used the least effective methods (48%) or no method (39%) throughout the year and had the highest probability of using EC (7%) of any latent status. About 60% of these women had at least one child, 58% were single and never married, and 30% were married.

Table 4 summarizes the latent status prevalence and the transitions over the three time points. The prevalence rates (δ estimates) indicate that the probability of being in the Single Abstinent Women status was the highest across the time periods but decreased across time (.29 at year 1, .26 at year 2, and .23 at year 3), whereas the probability of being in the Transient Users was the lowest but increased across time (.03 at year 1, .05 at year 2, and .05 at year 3). Other statuses remained relatively consistent in prevalence: the probability for Stable Users was .15 at years 1 and 2 and .16 at year 3; for Consistently Covered Mothers it was .18 across all years; for Sexually Active Nonusers it was .14 at year 1, .15 at year 2, and .17 at year 3; for Condom Users it was .13 at year 1 and .11 at years 2 and 3; and for Intermittent Users it was .08 at year 1 and .10 at years 2 and 3.

We formally tested the transition probability invariance (likelihood ratio statistic between the transition invariant model and the free transition model, p value = .25; see Table A2, online appendix). The results indicated that the Bayesian information criterion and the likelihood ratio difference test favored the transition invariant model, whereas the Akaike information criterion favored the free transition probability model. We had no reason to believe that contraceptive behaviors would change in a different manner from one typical calendar year to another, and the Akaike information criterion is known to favor overly complex models (Vrieze 2012), so we chose the model that restricted changing latent statuses across time (i.e., transition probabilities are the same for year 1 to year 2 and for year 2 to year 3).

The second panel of Table 4 summarizes the transition matrix. Women in several statuses (Single Abstinent Women, Consistently Covered Mothers, Stable Users, and Sexually Active Nonusers) remained relatively stable within their statuses from years 1 to 2 to 3 (transition probability for remaining in the same class was >.70). We focus our discussion here on transitions where the transition probability is at least .10, representing the more common types of transitions that occur. About 13% of Single Abstinent Women moved into Intermittent Users status. A small percentage of women (11–13%) moved from Stable Users into Sexually Active Nonusers and vice versa. Around 12% of Condom Users transitioned to Transient Users. The last two statuses (Intermittent Users and Transient Users) transitioned more frequently to other classes across the three study years. Intermittent Users tended to move to Single Abstinent Women (16%) or Condom Users (13%), although 46% of Intermittent Users remained in that status from year to year. For Transient Users, most of these women tended to move to Stable Users (12%), Sexually Active Nonusers (28%), or Condom Users (16%). Only a third (32%) of Transient Users remained in that status from year to year.

Table 5 presents the predicted transition probabilities calculated from the multinomial logistic regression analysis for women who experienced intended and unintended pregnancies in study years 1 (panel A) and 2 (panel B). Very few women who have pregnancies remain in the same status from year to year, as indicated by each matrix's diagonal numbers. The samples for the models are small and we do not have sufficient power to draw conclusions about the statistical significance of differences between women who experienced intended versus unintended pregnancies. Therefore, this analysis should be interpreted as exploratory. Women who had intended pregnancies have higher transition probabilities to the Sexually Active Nonusers (LS4) status than women who had unintended pregnancies, especially for those who were Condom Users (LS5), Intermittent Users (LS6), or Stable Users (LS3) in the previous year. For Single Abstinent Women (LS1) who had unintended pregnancies, 42% and 36% remained Single Abstinent in years 2 and 3, respectively, 32% and 26% became Stable Users (LS3), and 18% and 15% became Consistently Covered Mothers (LS2). Women who had an unintended pregnancy in year 1 were more likely to transition to Stable Users (LS3) in year 2 than women who had an intended pregnancy, especially for those who were Consistently Covered Mothers (LS2), Sexually Active Nonusers (LS4), and Condom Users (LS5) in year 1. Transient Users (LS7) who had an unintended pregnancy in year 1 were also more likely to transition to Intermittent Users (LS6) in year 2 than women who had an intended pregnancy, although these results are not robust compared with year 3. Additionally, Single Abstinent Women (LS1), Stable Users (LS3), and Sexually Active Nonusers (LS4) who had an unintended pregnancy in year 2 were more likely to transition to Consistently Covered Mothers (LS2) in year 3 than women who had an intended pregnancy in year 2.

Discussion

Few studies have explored changes in individual contraceptive behavior over time. We found seven distinct typologies of contraceptive and reproductive behavior over three years in a nationally representative sample of U.S. women. This formative analysis extends previous research on contraceptive typologies by examining how U.S. women's contraceptive and reproductive behavior changes from one typical calendar year to the next.

Prevalence and Transition Rates

Our study found that approximately half of U.S. women are reliably and efficiently using contraception. The most prevalent statuses—Consistently Covered Mothers, Single Abstinent Women, and Stable Users—were all using effective contraception or not having sex at all. These statuses account for more than 50% of the entire study sample each year. The statuses were also relatively stable. More than 90% of Consistently Covered Mothers remained in that status throughout the three-year study period, making that status the most constant. These women used long-term or permanent methods of contraception and had at least one child, and a large proportion (47%) had three or more children. Some of these women might consider themselves to be at the end of their reproductive life course, while others might be using effective, long-term methods for childspacing. The prevalence of Single Abstinent Women declined from the first year to the third (from 29% to 23%), suggesting that as time went on, more transitioned out of that status. This makes sense for the oversampled 15–19 age group in the 2015–2017 NSFG, who, as they aged, transitioned into having sexual relationships. For older women in this status, their categorization here might represent a short-term period of abstinence. About 13% of Single Abstinent Women switched to Intermittent Users every year, implying that, at the start of their reproductive life course, women are more likely to intermittently use contraception or use methods that are easily reversible as they transition in and out of sexual relationships. For Stable Users, 76% remained in that status from year to year. These are women who are effectively using contraception year to year when they have sex but tend to have fewer children than Consistently Covered Mothers and use moderately effective or short-term hormonal methods. Interestingly, over the three-year period, 13% of Stable Users transitioned into Sexually Active Nonuser status, possibly describing users who deliberately started having unprotected sex to get pregnant.

Sexually Active Nonusers made up about 15% percent of the sample, and this percentage increased with time. This status includes women who were already pregnant or trying to become pregnant, as well as women who were ambivalent about pregnancy or contraceptive use or did not believe they could get pregnant. It also includes women who did not want to become pregnant. Some 73% of women remained in the status over the three calendar years, which may reflect the time it takes to get pregnant, pregnancies running over multiple calendar years, and the postpartum period until the eventual return to fertility. This status may also include women who are unable to get pregnant, including older women. Eleven percent of the Sexually Active Nonusers switched to Stable Users every year. These women may have been postpartum and using effective contraception to prevent a short interpregnancy interval.

Condom Users were also a consistent group over the three-year study period, making up 11–13% of the sample in each year. Twelve percent of Condom Users transitioned to Transient Users each year. The Transient Users were the smallest status in this sample (3–5%), and they transitioned in and out of other statuses frequently. They may have included women who could not find a method that worked for them, as over 85% of these women switched methods in a year. They were having sex frequently and having unprotected sex about half of the time. A large proportion of this latent status, 28%, moved into Sexually Active Nonuser status, suggesting that Transient Users may have become pregnant after a period of unstable contraceptive use. Another 16% of these women transitioned to Condom Users, while a small proportion, 12%, became Stable Users. This is a small group of women who might be interested in a tailored, clinical intervention to help them address their dynamic contraceptive needs. Indeed, every woman should be empowered “to choose a method of birth control that she can use correctly and consistently over time” (Dehlendorf et al. 2014:10). When a woman presents signs of being a Transient User, that may indicate that she is at high risk of unintended pregnancy or is dissatisfied with her family planning options. Clinicians may initiate more in-depth contraceptive counseling that includes developing a personal and trusting relationship with clients, optimizing decision making, communication of side effects and risks, and personalized discussion regarding future fertility preferences (Dehlendorf et al. 2014).

Finally, Intermittent Users accounted for 10% of our sample, but only 46% remained in this status in any given year. These women—unlike women in the sample who had sex most months or not at all—were having sex infrequently throughout the year. However, 16% of Intermittent Users became Single Abstinent Women in each year. One hypothesis is that these women had a romantic or sexual partner for a short time and then returned to abstinence when that relationship ended. Also, 13% of Intermittent Users became Condom Users, suggesting that some of these individuals go on to have increased sexual activity and are consistently using condoms with these partners.

By design, measures of life stage, parity, and marital status were an integral part of the formation of the reproductive states. The results show that there was a relatively high degree of separation by life stage variables. Single women with no children tended to be Single Abstinent Women, Intermittent Users, or Stable Users, while most married women with children used a long-acting or permanent contraceptive method as Consistently Covered Mothers. Condom Users and Intermittent Users used short-term, reversible methods, which is consistent with other reproductive life stage research. In a large national survey from Australia, women with fewer than two children were more likely to use oral contraception, while women with two or more children were more likely to use longer term or permanent methods (Gray and McDonald 2010). Additionally, in Harris et al.'s LTA model using data on 8,197 women from Australia, they found that contraceptive behaviors tend to stabilize as women age, with women often switching to more effective methods in their late 30s and 40s (Harris et al. 2021). Our findings corroborate these studies, demonstrating that life course factors play an important role in understanding contraceptive preferences and reproductive behavior dynamics (Gray and McDonald 2010; Harris et al. 2021).

Changes in Contraceptive Behavior After a Pregnancy

This study showed how women change contraceptive and reproductive trajectories in response to unintended and desired pregnancies. Because of the small sample sizes of women who had unintended pregnancies in study years 1 and 2, these results are suggestive of the influence of women's pregnancy experiences but not definitive. It should also be noted that the results do not allow us to disentangle directionality of what state an unintended pregnancy originated from because the NSFG public use data file contains only the year in which a pregnancy ended.

In our study, most women transitioned into different statuses in the year after an intended pregnancy. For some women who were intermittent or transient contraceptive users, an unintended pregnancy led to stable contraceptive use, although our study suggests that is often condom use. Women who were already consistently using contraception but experienced an unintended pregnancy most often remained stable users of contraception. This finding is consistent with a life course perspective in which unintended fertility has a causal impact on subsequent trajectories (Rajan et al. 2017). Although Rajan et al. examined how an unintended birth has a powerful influence on subsequent unintended births, our results suggest that for a subset of women, an unintended pregnancy may influence subsequent decisions about contraceptive behaviors.

Limitations

The current study used eight items to capture contraceptive behavior, exposure, and life stage development. It also used data on pregnancy desire. To measure contraceptive behavior and the extent to which pregnancies were desired, the NSFG asks women to retrospectively recall their behavior and intentions. These retrospective measures have the potential for recall bias, especially for pregnancy intention, given that women's perceptions of a past conception can change over time (Hall et al. 2019; Joyce et al. 2002; Ralph et al. 2020; Rocca et al. 2019). A second limitation is the retrospective nature of the calendar data, which are subject to respondents' quality of recall. Several studies have assessed the quality of contraceptive use data collected as calendar data from low- and middle-income countries (Callahan and Becker 2012; Strickler et al. 1997; Tumlinson and Curtis 2021). Accuracy in reporting autobiographical events is dependent on several factors and has been shown to be positively associated with education level (Angel et al. 2010). Contraceptive calendar reliability may look different in a U.S. context, where there are higher levels of female literacy. Additionally, there have been a growing number of innovations in the life history calendar applications in the NSFG (Axinn et al. 1999). For one, the calendar provides a matrix of visual cues that respondents can use to help them recall the timing of life events. This data collection tool increases the utility of the life history calendar to document the timing and sequences of contraceptive behavior and sexual activity in the U.S. population.

Another limitation is that the 2015–2017 NSFG dataset reports only the year in which each pregnancy ended for respondents' reported pregnancies. We do not have the details of pregnancy timing, so we cannot describe contraceptive profiles of individual women immediately before they became pregnant. With a more specific time frame, future studies may be able to differentiate among Sexually Active Nonusers (e.g., those who were deliberate in moving into a higher probability of pregnancy to get pregnant).

An important limitation of this analysis involves the underreporting of induced and spontaneous abortions within the dataset. Although the NSFG is the most comprehensive source of information on pregnancy and contraceptive use among reproductive age women in the United States, abortion underreporting is a major issue (Lindberg et al. 2020). Underreporting introduces measurement error in some of the derived variables for the analysis, including use and types of contraceptive methods during each month. This type of measurement error should have a small effect on the LTA because the percentage of months of data that would be affected by abortion misreporting is small relative to the total number of months of use in the sample. In addition, abortion and other underreported unintended pregnancies may bias our results on how women transition in their contraceptive status when they have an unintended pregnancy. Again, our results are suggestive and not definitive. Transition patterns of other women who have unintended pregnancies but omit these from their history limit the extent to which we can describe the experiences of unintended pregnancy. Therefore, these results should be interpreted with some caution.

There are also limitations with the constructed indicators and the latent transition model. First, LCA models were used to capture the underlying latent construct of contraceptive behavior each year, over the course of the three years of calendar data. We chose to model contraceptive behavior in calendar years because it is a natural unit of time, there is a seasonality to childbirth, and most health insurance plans cover an annual well-woman exam (Dahlberg and Andersson 2018; Mathers and Harris 1983). However, that unit of time is ultimately arbitrary and another time frame could have been modeled. Second, the LTA models categorical, stage-sequential development among our eight indicator variables. In using LTA, we inevitably simplify and collapse the ways in which women of reproductive age in the United States have sex, use contraception, get married, and have children. In future studies, other important indicators should be included in the analysis, such as contraception and pregnancy desires and pregnancy intentions over time. Future studies could also explore such distal outcomes as STIs or negative birth outcomes that might illuminate the conditions under which certain patterns of contraceptive behavior are tied to other reproductive health outcomes.

Implications

Contraceptive use is nearly universal in the United States, yet little is known about the dynamics of contraceptive use over time at the individual level. This study moves away from examining macro-level, cross-sectional average contraceptive behaviors to describing dynamic patterns of contraceptive use at the individual level using an advanced statistical technique, LTA. The findings provide insights into the ways in which contraceptive behavior fits into the lives of women and how that behavior interacts with relationships, sexual activity, and life stage trajectories. Our results provide evidence of the heterogeneity of lived experiences that underscores the value of women's having access to contraceptive care that is focused on quality of care, rights, health, and autonomy.

Identification of the existence, relative size, and trajectories of interpretable subgroups helps in the development of new hypotheses about women's contraception behavior and experiences. There is evidence that unintended pregnancy in the United States has become increasingly concentrated among subsets of women (Sonfield et al. 2014; Wildsmith et al. 2010). Our findings show that a relatively small proportion of women belong to the high pregnancy risk profile of transient contraceptive users each year but that this group is dynamic, with women moving in and out of it over time. More attention needs to be paid to addressing the context of individual contraceptive trajectories, including sexual beliefs and behaviors, risk-taking behaviors, and access to and use of contraceptives and other reproductive health services. Clinicians can aim to empower women to reflect on their own trajectories and uncertainties, including their feelings and desires toward both contraceptive use and childbearing throughout their lives.

Overall, these trajectories show that contraceptive and reproductive health experiences—including intended and unintended pregnancies—are heterogeneous in the United States. Our analysis provides a new perspective on the dynamics of individual contraceptive behavior and underscores the importance of capturing the complexities of women's fertility-controlling experiences and enhancing longitudinal integration of contraceptive behaviors within family planning research and clinical practice.

Acknowledgments

The authors thank Cathy Zimmer and Chris Wiesen of the Odum Institute at the University of North Carolina at Chapel Hill for their expert statistical and programming guidance.

Notes

References