Contraceptive Selection and Practice: Associations with Self-Identified Race and Socioeconomic Disadvantage

Abstract

Many researchers and policymakers have linked contraceptive programs to improvements in women’s and children’s socioeconomic outcomes. However, these studies have overlooked how socioeconomic status may be an initial driver of contraceptive choice and behavior. Here, I examine the relationship between a comprehensive measure of socioeconomic disadvantage, self-identified race, and contraceptive method selection at enrollment in a unique longitudinal study of contraceptive clients who received a new type of method at no cost. I then examine whether socioeconomic disadvantage has an association with contraceptive switching or discontinuation. I demonstrate that socioeconomic disadvantage decreases the chance of selecting any IUD, while Black racial membership increases the chance of selecting the 3-month injectable and Multiracial membership increases the chance of selecting the Vaginal Ring. I then demonstrate that socioeconomic disadvantage and self-identified race have intersectional and variable associations with switching, and, to a lesser extent, discontinuing methods. These findings offer an important insight for implementation in contraceptive programs: eliminating financial barriers to access contraceptive services does not eliminate the socioeconomic contexts that influence method selection and use that occur as part of everyday lived experiences. Taken cumulatively, these results suggest that contraceptive services should be offered to women in ways that ensure access to reproductive justice without obscuring the need for social changes in the institutions that create disadvantage and shape contraceptive use itself.

INTRODUCTION

For decades, population researchers and policymakers have suggested that contraceptive use can lead to reductions in inequality and poverty in the United States (Nathanson 1991; Furstenberg, 2003; Furstenberg, 2007; Parks and Peipart, 2016). A significant portion of contraceptive literature echoes this “contraception-as-poverty-reduction” argument but rarely accounts for how socioeconomic disadvantage prior to contraceptive use structures actual contraceptive dynamics (Bailey, 2006; Dehlendorf, 2010; Parks and Peipart, 2016). Failing to address these structures participates in an ongoing history of reproductive injustice that systematically sets up people who can give birth as the perpetrators of their own poverty (Roberts, 1999). Framing the provision of contraceptive services as a type of deliverance from disadvantage can have the dangerous effect of making contraceptive users seem responsible for their own continued hardships.

I address this critical issue by asking whether disadvantage is associated with the type of new contraceptive method selected, free of charge, among a large cohort of women from diverse socioeconomic backgrounds in a longitudinal study in Salt Lake City, Utah—the HER Salt Lake Contraceptive Initiative (Sanders et al., 2018). This study is distinct from the existing literature because, while most studies use race and socioeconomic status to predict current method use, i.e. contraceptive use prevalence, I use the structure of this longitudinal data to examine how a priori socioeconomic disadvantage shapes the choice that users make about which new type of contraceptive method they select. This is an important distinction—while existing studies accurately represent cross-sectional contraceptive prevalence at the population level, they do not represent the change that is inherently part of choosing a new method, as well as continuing that method or not over time. How respondents progress through these changes may be more important to their long-term reproductive outcomes than their contraceptive use at any given moment. I then ask whether pre-existing socioeconomic disadvantage interacts with self-identified race, survey wave, and method, to shape whether respondents switch or discontinue their initial method over the first year of use. Using these analyses, I show that pre-existing disadvantage has a unique relationship with the selection of a new type of contraceptive method and with contraceptive practices over time, net of the ability to “afford” contraception. By providing this evidence, I contribute to work on stratified reproduction that situates racialized and classed discourses about contraceptive behaviors within structures of inequality, and encourage providers, researchers, and policymakers to offer full, free, and informed access to contraception that is not contingent on its poverty-reduction potential (Ginsburg and Rapp, 1995).

BACKGROUND AND PRIOR RESEARCH

The relationship between social disadvantage and health outcomes is persistent throughout literatures on health, suggesting that socioeconomic status is a fundamental cause of population health disparities (Adler and Rehkopf 2008; Phelan, Link and Tehranifar 2010). This literature increasingly shows that the link between socioeconomic status and health is dependent on other indicators of social location, such as race (Colen et al. 2018). It suggests that health outcomes are unequally patterned across sub-groups in ways that exacerbate inequalities and interact with discrimination and psychosocial stressors (Colen et al. 2018). This inequality occurs through a variety of mechanisms, including residential segregation, differential access to institutions and public goods, and lived experiences of discrimination (Chae et al. 2018).

While there is little to suggest that the relationship between disadvantage and contraceptive use would be dissimilar to the relationship with health, there are factors unique to contraception that could shape how disadvantage and contraceptive use interact. These factors include the highly politicized and regulated nature of reproduction and that women of color may prefer user-directed contraceptive methods (versus provider-directed methods such as IUDs, which require insertion), as U.S. institutions have histories of regulating the fertility of women of color, particularly Black women, via welfare regulations and unethical experimentation (Roberts, 1999). Another distinction may be that reproduction’s presumed outcome—pregnancy—is not an inherently “negative” outcome in the way that disease diagnoses are.

There is extensive literature demonstrating that people’s feelings about unintended pregnancies are often ambivalent, even among those intending to prevent pregnancies (as an example see Higgins, Popkins, and Santelli, 2012). The spectrum of desirability for the outcome of pregnancy may make social patterning of contraceptive use distinct in unidentified ways.

Predicting Contraceptive Method Use Using Socioeconomic Status and Self-Identified Race

Contraceptive literature often examines self-identified race and individual measures of socioeconomic status, such as educational achievement, employment status, or receipt of public assistance. Some distinct method-based racial patterns emerge among White, Black, and Hispanic women for long-acting reversible methods (LARCs) and the pill, and there are distinct educational gradients in contraceptive use, despite few consistent findings for other measures of socioeconomic status. Only one known study examines intersectional experiences between self-identified race and socioeconomic status. This study, Kramer et al. (2018), found no significant relationship between race interacted with poverty level and education on current LARC use.

Among nationally representative studies that examine LARC use, current LARC use was found to be 1) similar across Black, Hispanic, and White women (Daniels and Abma, 2018), increased among Black women (Frost and Darroch, 2008), and decreased among Black and Hispanic women (Rocca and Harper, 2012; also see Sangi-Hanghpeykar et al, 2006 for similar non-representative findings). These studies also suggest that having any college experience or having completed college significantly increases current use of LARCs relative to those who have no college or did not complete it (Daniels and Abma, 2018; Frost and Darroch, 2008).

Findings by race and socioeconomic status for current pill use are the most consistent in the literature using nationally representative surveys: Black and Hispanic women are significantly less likely to ever use or currently use the pill relative to White women (Daniels and Abma, 2018; Frost and Darroch, 2008; Littlejohn, 2012; Rocca and Harper, 2012). Increased education is positively and significantly associated with pill use (Daniels and Abma, 2018; Frost and Darroch, 2008; Rocca and Harper, 2012), with one exception (Littlejohn, 2012). However, findings for other hormonal methods, such as the injectable, the vaginal ring, or emergency contraception, do not have consistent relationships with socioeconomic status or race (Littlejohn, 2012; Rocca and Harper, 2012). Many of these studies do not examine other hormonal methods besides “LARCs” (which groups several methods), the pill, and injectables.

Predicting Contraceptive Method Practice Using Socioeconomic Status and Self-Identified Race

Measuring contraceptive continuation, discontinuation, or switching—what is often called contraceptive dynamics or practice—is difficult to accomplish, as it requires access to expensive, prospective, panel studies, and is limited by how accurately researchers can measure the timing of these dynamics. If we think about switching and discontinuing not as negative contraceptive outcomes in and of themselves, but as indicators that patients’ needs (related to menstruation, side effects, method characteristics, access, etc.) either were not met initially or were not met as they sought to continue a method, then this mismatch could result from structural inequalities in accessing contraceptive care, provider bias, or socio-cultural preferences or influences (Higgins, Kramer, and Ryder, 2016; Littlejohn, 2013; Jackson et al., 2016; Polis, Hussain, and Berry, 2018). Further, we can imagine switching as a process whereby a patient is initially mismatched with a method but is then able to re-establish their needs and preferences via access to their healthcare providers, while discontinuation may be an indicator that physical or institutional access was a major barrier to continuing contraceptive use.

Studies have found mixed relationships between lower socioeconomic status and discontinuation: some found that low-income or lower educated women are more likely to discontinue, while others found no relationship (Dehlendorf et al., 2011; Littlejohn, 2012; Ananat, Gassman-Pines, and Gibson-Davis, 2013; Kusunoki et al., 2016; Kramer et al., 2018). Only one study using nationally representative data addresses discontinuation (Littlejohn, 2012). Littlejohn (2012) finds no significant racial results for method discontinuation but does find that women who are less educated are more likely than college-educated women to discontinue a hormonal contraceptive method because of dissatisfaction with that method, net of race (Littlejohn, 2012). Kusunoki et al’s (2016) long-term panel data among 18-19-year-olds also finds no difference across racial groups for discontinuation, but does find a decreased likelihood of switching among Black respondents. These studies suggest that more evidence is needed to predict the magnitude and direction of associations between sociodemographic characteristics and contraceptive practices.

Limitations within the Literature

There are several limitations within studies that examine socioeconomic correlates with current contraceptive use and contraceptive practice. The first is that these studies often focus on single or grouped methods. Most studies examine correlates with pill use, others that examine more than one method frequently lump together types of methods (i.e. LARCs or “hormonal”). Additionally, many studies focus on a small set of racial groups, predominantly young White, Black, or Hispanic women. Most studies do not explicitly model the intersectional experience of how socioeconomic disadvantage may interact with race and method choice to produce differential contraceptive practices. In the present study, I include the six most-commonly selected methods by respondents in the study, which represent a spectrum of reversible and hormonal methods. I later interact these methods, alongside self-identified race and survey wave, with disadvantage to identify different contraceptive practice pathways.

A second limitation is that the measures of socioeconomic status in these studies may not capture the complexity of how social disadvantage is experienced. When socioeconomic variables are included in contraceptive studies, they often consist of educational attainment, insurance status, or respondent’s poverty level (Dehlendorf et al., 2010; Kavanaugh, Jerman, and Finer 2015; Kusunoki et al., 2016; Littlejohn, 2012). Although these individual components of disadvantage are important to measure and analyze, they fail to adequately characterize how disadvantage is a complex experience that is defined not only by household income, education, or formal employment, but also draws on individual and household’s abilities to provide basic needs, how these abilities exist in relation to other strata in society, and household members’ time poverty. By creating an index of disadvantage for this study, I attempt to better capture the complex experience of disadvantage to demonstrate how structural disadvantage can shape individual contraceptive trajectories.

A final limitation of existing studies is based on how switching and discontinuation are measured. Studies trying to identify correlates with discontinuation and switching are often required to estimate associations over long periods without accounting for exposure time and temporal ordering. When models estimate non-continuation without accounting for the amount of exposure to method use they miss nuanced patterns where non-continuation events can coagulate at different points over a long period for different sub-groups of respondents. This study accounts not only for the survey wave at which a respondent continued or did not continue but also for intervals between individual measurements, which ensures that respondents are being compared to others net of length of exposure time to specific methods. Another way the present study addresses these issues is by accounting for the temporality of predictor and outcome variables: the main predictor variables—self-identified race, and socioeconomic status—are measures of the respondent’s state of being prior to selecting a contraceptive method, which is measured prior to continuing, switching, or discontinuing a method at discretely timed intervals.

METHODS AND DATA

Data

These data come from a longitudinal panel study of contraceptive users in Salt Lake City, Utah (the HER Salt Lake Contraceptive Initiative). Participants for this study were recruited from March 2016 to March 2017 in an arm of the study where they received their contraceptive method free of cost. Individuals could enroll if they were aged 18-45, spoke English or Spanish, were receiving a method at no cost at one of four participating, publicly funded, family planning clinics in Salt Lake County, Utah, and did not want to become pregnant for at least one year. All participants could switch or discontinue their method at any time, for any reason, at no cost.

Sample

At baseline, all respondents in this sample either 1) had never used any contraceptive method and started one upon enrolling in the study or 2) had chosen a new type of method compared to what they were using in the 4 weeks prior to enrolling in the study. Respondents’ information on socioeconomic status was collected at enrollment, and so existed a priori to both the selection of a new type of method and to their contraceptive practices going forward from the beginning of the study. I exclude participants who did not select one of the six most common methods—the levonorgestrel intrauterine device (LNg IUD), the copper IUD, the pill, the contraceptive implant, the injectable, and vaginal rings—respondents choosing a method other than these six constituted less than 1% of the overall participants.

My final analytic sample contains 2,499 respondents for the baseline analyses and 8,683 individual observations for the longitudinal analyses from 1-12 months following study enrollment. Seventy-five percent of respondents included in the follow-up sample were retained at the 12-month follow-up survey; 457 respondents were censored after they were lost to follow up, withdrew from the study, or became pregnant. Of these observations, 197 were removed due to missing independent variables, yielding an analytic sample of 8,683 observations. This sample is specific to Salt Lake City, Utah, and results from this data cannot be extended to urban centers that have a distinctly different racial and socioeconomic makeup. However, this sample is representative of people seeking contraceptive services from public facilities in Salt Lake City, Utah.

In this study population, every enrolled respondent expressed a desire to prevent pregnancy for at least one year following enrollment. The malleability of pregnancy intentions is well documented (see Higgins, Popkin, and Santelli, 2012), suggesting that some respondents may have changed their pregnancy intentions over the following year. Because pregnancy intentions are not measured after the first survey, I assume that the study criteria of intending to prevent a pregnancy for at least one year should be taken as a face value representation of respondents’ pregnancy intentions.

Variables

The primary independent variable of interest in this study is an index of disadvantage. This index is constructed from ten items fixed at baseline that measure respondents’ education, employment, income, receipt of public assistance, and household scarcity (see Table 1.b). Each of the ten variables is dichotomously coded, with a value of one indicating more socioeconomic disadvantage than a value of zero. These are then added together for respondents with full information on each of the ten items. The average score on this index is 3.2, with a range of 0-10, where a higher score indicates greater disadvantage (SD: 2.2). This variable is right-skewed, and few respondents report being disadvantaged on all ten measures.

Table 1.b.

Baseline: Index of Disadvantage, by Self-Identified Race/Ethnicity

Characteristics			White %	Hispanic %	Black %	Asian %	Multiracial %	Baseline %
	During the last 12 months, did you have trouble:	paying for transportation?	Yes	17.2	17.6	39.0	8.9	22.9	17.7
paying for housing?		Yes	21.3	19.8	26.8	11.1	22.9	20.8
paying for healthcare?		Yes	28.3	23.6	43.9	20.0	31.8	27.6
paying for food?		Yes	18.5	19.6	36.6	10.0	25.1	19.2
During the last 4 weeks, did you have enough money all of the time to meet your basic needs?		No	59.3	70.9	78.1	58.9	63.7	62.2
Do you currently receive any public assistance? (includes food stamps, WIC, welfare and unemployment benefits)		Yes	6.2	13.0	6.5	5.8	5.1	9.5
What type of medical insurance do you currently have?		None	40.4	64.0	57.1	35.9	46.0	47.5
What best describes the highest level of education you have completed so far?		High school or less	34.6	53.1	40.4	27.9	45.0	39.2
What best describes your current employment status?		Not full- or part-time employed	30.7	33.7	38.5	43.3	39.8	33.8
Respondent’s income as percent of federal poverty level		Less than or equal to 100% of the federal poverty level	32.3	52.5	43.1	45.6	36.1	38.7
Total Sample, Baseline			1,694	495	41	90	179	2,499

I include a second independent variable that measures the following self-identified racial categories: “White,” “Hispanic,” “Black,” “Asian,” and “Multiracial” (note that Multiracial precludes being in any other category). These categories condense meaningful cultural and social variation into geographic or social groupings that may not be related in meaningful ways; these groups are included as they are because they had sufficient sample sizes for analyses. The majority of “Multiracial” respondents reported identifying with only 2 racial categories (156 respondents out of 180). Most of these respondents reported being “White” and “Hispanic” (n=92), “White” and “Asian” (n=31), or “White” and “Alaskan Native or Native American” (n=18), with the rest relatively evenly distributed across groups.

There are two outcome variables of interest for the present analyses. The first is the contraceptive method selected at baseline. For this outcome, over 99% of the sample chose one of six methods at enrollment in the study: the LNg IUD, the Copper IUD, the implant, “the pill”, injectable contraception, and vaginal rings. The second outcome of interest is contraceptive practice over a 12-month period, consisting of contraceptive continuation, method switching, and discontinuation (pregnancy as a competing risk outcome was too underpowered in this sample to be included). At the 1-, 3-, 6-, and 12-month follow-up surveys, participants were asked to report on all current contraceptive methods being used by selecting from a list of over fifteen methods. To construct the contraceptive practice variable, I condense responses at each survey into a single response reflecting the most effective method reported by the participant (Trussell, 2013). I use the method reported at baseline as the index method, and then calculate method continuation over discrete intervals by determining whether the most effective method reported at survey time, t + 1, matched the most effective method reported at the previous survey, at time t, starting with the baseline survey. For these analyses, respondents could continue their method from 0-1 months, 0-3 months, 0-6 months, or 0-12-months, but not, for example, from 6-12 months. If the respondent reported that they were using a different method, or that they had stopped using their method altogether, they were considered to have switched or discontinued their baseline method, respectively. Continuation, switching, and discontinuation were coded into a non-ordered categorical variable. Respondents’ observations are treated as exiting the sample after they have experienced their first method switch or first method discontinuation; they are censored across all waves for pregnancy, loss to follow up, or withdrawal from the study.

I include a set of “study” control variables, which consist of the experimental period during which respondents enrolled in the study, the study site at which they enrolled, survey wave, and respondent interval (in days) between surveys. I also include a set of “individual” control variables, including respondent age, and time-varying measures of respondent physical and emotional side effects measured using a validated questionnaire (Chesney & Tasto, 1975).

Analytic Methods

Descriptive tables demonstrate the characteristics of respondents, respondents’ distributions for the index of disadvantage, and the 0-1, 1-3, 3-6, 6-12, and 0-12-month continuation, switching, and discontinuation rates by each of the six included methods (see Tables 1.a-1.c).

Table 1.a.

Baseline: Characteristics of Respondents

Characteristics		Baseline %
Age Group	18-19	8.5
	20-24	44.5
	25-29	27.4
	30-34	12.2
	35+	7.6
Race/Ethnicity	White	67.8
	Hispanic	19.8
	Black	1.6
	Asian	3.6
	Multiracial	7.2
Education	Less than High School	4.4
	Completed High School	34.8
	Some College	42.3
	College Graduate	18.5
Relationship Status	Married	12.0
	Cohabiting	48.4
	Single	32.5
	Other (includes divorced, widowed and other)	7.2
Sexual Orientation	Entirely or Mostly Heterosexual	87.2
	Bisexual	11.7
	Entirely or Mostly Homosexual	1.2
Religion	Non-Religious	59.3
	Protestant Christian	11.0
	Catholic	8.4
	Mormon	11.0
	Other	5.9
Method Selected	The Pill	20.6
	LNg IUD	28.8
	Copper IUD	14.2
	Implant	21.9
	Injectable	9.4
	Vaginal Ring	5.2
Total Sample, Baseline		2,499

Table 1.c.

Baseline: Contraceptive Practice,0-12 Months^±ǂ

		0-1 Months, %	1-3 Months, %	3-6 Months, %	6-12 Months, %	0-12 Months*‡, %
The Pill	Continued	85.4	84.6	80.8	76.5	44.6
	Switched	9.6	6.9	15.4	16.5	36.3
	Discontinued	2.2	3.9	4.6	3.8	11.1
LNg IUD	Continued	96.7	94.8	95.2	89.2	77.9
	Switched	1.7	2.5	2.5	5.2	10.8
	Discontinued	0.3	0.7	1.4	1.2	3.3
Copper IUD	Continued	95.7	95.2	94.7	86.2	74.3
	Switched	3.2	3.4	4.7	8.5	18.1
	Discontinued	0.3	0.3	0.9	1.6	2.7
Implant	Continued	98.1	95.1	92.0	85.9	73.7
	Switched	1.6	1.2	4.8	7.6	13.8
	Discontinued	0.0	0.4	2.6	2.2	4.7
Injectable	Continued	96.1	89.4	77.3	67.1	44.5
	Switched	2.0	4.5	13.2	15.3	27.7
	Discontinued	0.4	2.4	8.6	8.8	16.0
Vaginal Ring	Continued	88.1	87.4	84.6	64.8	42.2
	Switched	7.4	8.4	8.7	15.9	31.9
	Discontinued	2.2	1.7	3.8	11.4	14.1
Total Sample, 0-12 Months		11,182

^±Denominator for contraceptive practice in time t is the continuers from time t-1

^*Numerator for contraceptive switching and discontinuation 0-12 months is the sum of all switchers and discontinuers from 1-12 months

^‡Denominator for contraceptive practice, 0-12 months, is number of respondents choosing the specific method at baseline

^ǂ0-12 month proportions may not add to 100 due to missing responses, loss to follow up, and pregnancies

Model 1 is estimated using a multinomial logistic regression for the baseline survey only, to estimate the hazard ratio of selecting each contraceptive method, relative to the pill. I include the index of disadvantage and the self-identified race measure as the main predictor variables and I control for study and individual characteristics. In the results section, I present the hazard ratios for selecting each method relative to selecting the pill. The rationale for utilizing the pill as a reference group is that it is one of the most frequently used methods in the U.S. (Mosher and Jones, 2010).

I then examine the associations between the index of disadvantage, self-identified race, and the hazard of switching or discontinuing a method (relative to continuing) in Models 2 and 3 using discrete-time competing risk event history multinomial logit models and controlling for study and individual characteristics. Model 2 estimates the association between the index of disadvantage and self-identified race on the contraceptive practices of switching and discontinuing, while Model 3 builds on Model 2 by interacting the index of disadvantage with self-identified race, survey wave, and contraceptive method. I present the results of these analyses in Tables 3 and 4 in the results section, showing the hazard ratios for switching and discontinuation, relative to continuation, by each of the predictor variables of interest. Additionally, Figures 1.a through 2.b demonstrate the difference by each of these groups for respondents who are less disadvantaged compared to those who are more disadvantaged. All models account for nested observations by correcting for clustering within respondents and by including an indicator term for the survey wave. These models are designed to explicitly account for the relationship between the contraceptive method selected and switching or discontinuing a method, first net of the method selected, then net of the method selected interacted with disadvantage. These models net out the effect of the method selected at enrollment, which mechanically forces the estimates for the independent variables to be among respondents who chose the same enrollment method.

Table 3.

Model 2—Hazard Ratio Coefficients Predicting Association between Index of Disadvantage, Self-identified Race, and Contraceptive Practice

		HR(CI)
		Continued	Switched	Discontinued
Index of Disadvantage		-	1.05** (1.005-1.097)	1.05 (.975-1.126)
Race/Ethnicity
	White (Reference)
	Hispanic	-	1.37** (1.080-1.730)	1.75** (1.178-2.588)
	Black	-	2.18** (1.053-4.507)	3.63** (1.378-9.548)
	Asian	-	.98 (.572-1.690)	1.12 (.406-3.105)
	Multiracial	-	1.49** (1.054-2.094)	1.71 (.987-2.950)
Study Controls		✓
Individual Controls		✓
N		8,683

P-values are for a significance test of whether the slope for the index of disadvantage coefficient (accounting for the interacted terms) is significantly different from zero.

^*p<.05

^**p<.01

^***p<.001

Table 4.

Model 3—Hazard Ratio Coefficients Predicting Association between Index of Disadvantage and Contraceptive Practice, by Self-identified Race, Method, and Survey Wave, Relative to Continuation

Method, Race/Ethnicity		Switching HR(CI)				Discontinuing HR(CI)
		1-month	3-months	6-months	12-months	1-month	3-months	6-months	12-months
ALL METHODSǂ	White	1.21** (1.060-1.374)	1.08 (.942-1.234)	1.17** (1.041-1.309)	1.10 (.977-1.248)	0.89 (.685-1.180)	1.07 (.834-1.381)	1.11 (.917-1.335)	1.19 (.967-1.458)
	Hispanic	131*** (1.114-1.530)	1.16 (.991-1.372)	1.26*** (1.095-1.456)	1.19* (1.027-1.389)	0.94 (.703-1.261)	1.12 (.852-1.481)	1.16 (.919-1.461)	1.24 (.995 -1.553)
	Black	1.29 (.942-1.760)	1.15 (.839-1.576)	1.25 (.916-1.693)	1.18 (.863-1.608)	0.89 (.547-1.443)	1.06 (.662-1.700)	1.09 (.688-1.738)	1.17 (.686-1.895)
	Asian	1.56** (1.176-2.068)	1.39* (1.154-1.971)	1.51** (1.083-1.878)	1.43* (1.048-1.803)	1.09 (.686-1.741)	1.30 (.823-2.067)	1.35 (.871-2.077)	1.44 (.940-2.213)
	Multiracial	1.29** (.973-1.432)	1.15 (.958-1.388)	1.25** (1.056-1.478)	1.18 (.989-1.410)	1.12 (.822-1.524)	1.34 (.997-1.791)	1.38* (1.063-1.786)	1.48** (1.150-1.898)
The Pill	White	1.10 (.983-1.221)	0.97 (.862-1.090)	1.05 (.952-1.158)	0.99 (.897-1.097)	0.82 (.551-1.044)	1.00 (.823-1.218)	1.03 (.896-1.183)	1.12 (.961-1.308)
	Hispanic	1.22** (1.060-1.403)	1.08 (.932-1.249)	1.17* (1.029-1.327)	1.10 (.965-1.262)	0.91 (.702-1.178)	1.10 (.880-1.386)	1.14 (.927-1.393)	1.24* (1.028-1.488)
	Black	1.24 (.922-1.659)	1.09 (.811-1.476)	1.18 (.887-1.582)	1.12 (.835-1.500)	0.87 (.533-1.407)	1.05 (.657-1.682)	1.08 (.672-1.739)	1.18 (.731-1.895)
	Asian	1.52** (1.154-1.997)	1.34* (1.022 -1.766)	1.45** (1.116-1.897)	1.37* (1.048-1.803)	1.16 (.698-1.936)	1.41 (.853-2.335)	1.45 (.889-2.373)	1.58 (.981-2.55)
	Multiracial	1.18 (.972-1.432)	1.04 (.863-1.262)	1.13 (.948-1.349)	1.07 (.890-1.282)	1.06 (.779-1.445)	1.29 (.970-1.709)	1.32* (1.015-1.729)	1.44** (1.119-1.859)
LNg IUD	White	1.21** (1.064-1.381)	1.07 (.927-1.242)	1.16* (1.033-1.307)	1.10 (.968-1.244)	0.81 (.614-1.055)	0.98 (.765-1.250)	1.01 (.818-1.237)	1.09 (.873-1.372)
	Hispanic	1.35*** (1.162-1.567)	1.19* (1.014-1.406)	1.249*** (1.131-1.478)	1.22** (1.057-1.412)	0.89 (.677-1.165)	1.08 (.839-1.386)	1.11 (.879-1.401)	1.21 (.964-1.513)
	Black	1.37* (1.007-1.861)	1.21 (.883-1.661)	1.31 (.971-1.771)	1.24 (.911-1.684)	.85 (.519-1.377)	1.03 (.637-1.655)	1.06 (.651-1.713)	1.15 (.705-1.875)
	Asia	1.68*** (1.276-2.214)	1.49** (1.125-1.966)	1.61*** (1.235-2.100)	1.52** (1.157-2.000)	1.14 (.679-1.898)	1.38 (.827-2.299)	1.42 (.859-2.341)	1.54 (.944-2.525)
	Multiracial	1.31** (1.073-1.591)	1.16 (.946-1.411)	1.25* (1.048-1.494)	1.18 (.980-1.426)	1.04 (.761-1.411)	1.26 (.941-1.682)	1.29 (.981-1.707)	1.41* (1.073-1.849)
Copper IUD	White	1.03 (.882-1.194)	0.91 (.780-1.056)	0.98 (.860-1.124)	0.93 (.816-1.056)	0.82 (.613-1.084)	0.99 (.761-1.287)	1.02 (.808-1.284)	1.11 (.872-1.409)
	Hispanic	1.14 (.957-1.362)	1.01 (.848-1.204)	1.09 (.935-1.280)	1.03 (.883-1.209)	0.90 (.664-1.220)	1.09 (.800-1.455)	1.12 (.853-1.481)	1.22 (.942-1.588)
	Black	1.16 (.854-1.569)	1.02 (.755-1.391)	1.11 (.952-1.289)	1.05 (.780-1.408)	0.86 (.525-1.397)	1.04 (.644-1.679)	1.07 (.656-1.743)	1.16 (.714-1.898)
	Asian	1.42* (1.082-1.869)	1.26 (.961-1.646)	1.36* (1.051-1.765)	1.287 (.991-1.672)	1.15 (.666-1.987)	1.40 (.810-2.408)	1.44 (.839-2.460)	1.56 (.925-2.643)
	Multiracial	1.11 (.897-1.361)	.98 (.801-1.194)	1.06 (.880-1.274)	1.00 (.830-1.205)	1.05 (.743-1.480)	1.27 (.917-1.769)	1.31 (.951-1.806)	1.43* (1.047-1.943)
Implant	White	1.11 (.975-1.273)	.99 (.857-1.134)	1.07 (.938-1.215)	1.01 (.887-1.145)	0.77* (.592-.996)	0.93 (.746-1.165)	0.96 (.771-1.194)	1.04 (.842-1.295)
	Hispanic	1.24** (1.073-1.432)	1.10 (.945-1.273)	1.19* (1.037-1.360)	1.12 (.977-1.288)	0.85 (.644-1.116)	1.03 (.806-1.314)	1.06 (.817-1.371)	1.15 (.914-1.453)
	Black	1.26 (.935-1.689)	1.11 (.824-1.502)	1.11 (.827-1.489)	1.14 (.847-1.528)	0.81 (.489-1.329)	0.98 (.603-1.589)	1.01 (.607-1.672)	1.10 (.664-1.811)
	Asian	1.54** (1.174-2.029)	1.37* (1.040-1.793)	1.48** (1.133-1.931)	1.40* (1.066-1.831)	1.08 (.659-1.781)	1.32 (.806-2.146)	1.35 (.824-2.221)	1.47 (.915-2.371)
	Multiracial	1.20 (.971-1.481)	1.06 (.863-1.306)	1.15 (.943-1.401)	1.09 (.888-1.328)	0.99 (.714-1.368)	1.20 (.890-1.617)	1.23 (.903-1.686)	1.34* (1.004-1.798)
Injectable	White	0.97 (.851-1.108)	0.86* (.739-.999)	0.93 (.806-1.073)	0.88 (.766-1.007)	0.77* (.603-.975)	0.93 (.766-1.132)	0.96 (.808-1.135)	1.04 (.853-1.274)
	Hispanic	1.08 (.924-1.263)	0.96 (.805-1.135)	1.04 (.881-1.217)	0.98 (.833-1.148)	0.85 (.665-1.076)	1.03 (.841-1.255)	1.06 (.866-1.290)	1.15 (.944-1.403)
	Black	1.10 (.820-1.464)	0.97 (.717-1.310)	1.20 (.899-1.613)	.99 (.739-1.331)	0.81 (.515-1.259)	0.98 (.638-1.499)	1.01 (.645-1.568)	1.10 (.696-1.722)
	Asian	1.35* (1.012-1.788)	1.19 (.891-1.590)	1.29 (.969-1.714)	1.22 (.914-1.622)	1.08 (.643-1.818)	1.31 (.788-2.188)	1.35 (.813-2.247)	1.47 (.890-2.431)
	Multiracial	1.05 (.861-1.270)	.93 (.758-1.130)	1.00 (.827-1.213)	.95 (.781-1.146)	0.99 (.718-1.356)	1.20 (.898-1.599)	1.23 (.922-1.647)	1.34* (1.004-1.793)
Vaginal Ring	White	0.97 (.799-1.173)	0.86 (.702-1.045)	0.93 (.771-1.116)	0.88 (.727-1.056)	0.73* (.556-.964)	0.89 (.704-1.121)	0.91 (.717-1.166)	1.00 (.778-1.273)
	Hispanic	1.08 (.870-1.333)	0.95 (.766-1.186)	1.03 (.842-1.265)	.98 (.792-1.201)	0.81 (.595-1.096)	0.98 (.747-1.287)	1.01 (.750-1.357)	1.10 (.831-1.451)
	Black	1.09 (.787-1.516)	.97 (.692-1.350)	1.05 (.782-1.408)	.989 (.713-1.371)	0.77 (.464-1.273)	0.93 (.575-1.516)	0.96 (.574-1.605)	1.05 (.626-1.745)
	Asian	1.34 (.978-1.840)	1.19 (.865-1.629)	1.29 (.944-1.749)	1.21 (.888-1.661)	1.03 (.609-1.751)	1.25 (.747-2.103)	1.29 (.760-2.189)	1.40 (.840-2.347)
	Multiracial	1.04 (.811-1.340)	.92 (.718-1.184)	1.00 (.788-1.266)	.94 (.741-1.202)	0.94 (.677-1.311)	1.14 (.848-1.543)	1.18 (.851-1.626)	1.28 (.941-1.744)
	Study Controls	✓
	Individual Controls	✓
	N	8,683

^ǂTo estimate the switching and discontinuation hazard ratio coefficients across all methods, the model must be specified without enrollment method as a categorical variable; the model used to determine the coefficients in the “All Methods” section differ from the subsequent model in this way

P-values are for a significance test of whether the slope for the index of disadvantage coefficient (accounting for the interacted terms) is significantly different from zero.

^*p<.05

^**p<.01

^***p<.001

Note that significance tests are not possible in a multinomial logit event history model for the hazards of the reference event (continuation).

Figure 1.a:

Predicted Probability of Switching Across All Methods

Figure 2.b:

Predicted Probability of Discontinuing

RESULTS

Respondent Characteristics

The HER Salt Lake respondents in this study were primarily young (over 50% are 24 years or younger), white (67.8%), married or cohabiting (60.4%), heterosexual (87.2%) and non-religious (59.3%) (Table 1.a). Some sub-groups of respondents, for example, Black women, represent a small portion of this study sample. Overall, this leads to larger standard errors in the estimates for these groups, but not to systematic bias in the sample because it is an accurate representation of the sampling frame (but not the general U.S. population). This study population is similar to other urban areas by levels of education and age; it differs by race compared to the average urban area in the United States (Parker et al., 2018).

Table 1.b shows the component measures for the index of disadvantage at baseline for study participants, broken out by self-identified race. Across each measure, White, Asian, or Multiracial respondents were the least likely to report disadvantage, while Black and Hispanic respondents were more likely to report disadvantage. Black respondents reported higher levels of disadvantage on measures of the daily availability of resources (such as the ability to pay for transportation) while Hispanic respondents were more likely to report disadvantage in areas of public assistance, insurance, and percentage of the federal poverty level. Model 3 specifically accounts for any interaction between self-identified race and disadvantage, as these variables can be correlated in the U.S. context. Approximately 10% of all participants currently received any public assistance, and 33.8% were at or below 100% of the federal poverty level. One-sixth to one-third of respondents had trouble paying for housing, health, food, or transportation, and 62.2% reported not having enough money to meet their daily expenses. One-third of respondents were not employed in a full or part-time job (about one-half of these non-employed participants were students), and 39.2% had completed high school or had less than a high school education.

The largest proportion of respondents in this sample chose the LNg IUD (28.8%) at enrollment in the study, followed by the implant (22.0%), the pill (20.6%), the copper IUD (14.2%), the injectable (9.4%) and the vaginal ring (5.2%). Method-specific 12-month continuation rates, given in the final column of Table 1.c, range from a high of 77.9% for the LNg IUD to a low of 42.2% for vaginal ring users. Pooling all methods, yields an overall 12-month continuation rate of 65%.

Selection into Contraceptive Methods

Table 2 shows the hazard ratio coefficients for the associations between the index of disadvantage, self-identified race, and method selection at enrollment in the study. These results show that a one-point increase in the index of disadvantage is associated with a 6.6% (p<.05) and a 9.9% (p<.01) reduction in the hazard of selecting the LNg IUD and the copper IUD respectively. Identifying as Black (compared to White) is associated with an increased chance of selecting the injectable relative to the pill at baseline (HR: 4.52, p<.01), and identifying as Multiracial is associated with a 2.16 increased chance of selecting the vaginal ring (p<.05). These findings suggest that socioeconomic disadvantage and self-identified race have significant independent associations with the selection of some types of contraceptive methods. Models 2 and 3 intrinsically account for the selection of specific groups into more or less effective methods by forcing estimates to compare similar respondents along lines of disadvantage, race, and contraceptive method.

Table 2:

Model 1—Hazard Ratio Coefficients Predicting the Association of Index of Disadvantage and Self-identified Race with Contraceptive Method Choice at Baseline

		HR (CI)
		The Pill	LNg IUD	Copper IUD	Implant	Injectable	Vaginal Ring
Index of Disadvantage		-	0.93* (.886-.985)	0.91** (.852-.968)	0.98 (.917-1.037)	1.07 (.998-1.140)	1.01 (.930-1.101)
Race/Ethnicity
	White (Reference)
	Hispanic	-	0.76 (.551-1.037)	0.86 (.597-1.246)	1.21 (.887-1.640)	1.15 (.775-1.704)	0.75 (.433-1.296)
	Black	-	1.10 (.362-3.344)	1.40 (.398-4.907)	1.61 (.558-4.628)	4.52** (1.593-12.816)	0.70 (.080-6.090)
	Asian	-	0.82 (.454-1.475)	0.88 (.433-1.786)	0.62 (.314-1.205)	1.02 (.454-2.270)	0.82 (.272-2.461)
	Multiracial	-	0.92 (.564-1.490)	1.13 (.651-1.966)	1.13 (.701-1.833)	1.19 (.641-2.202)	2.16* (1.142-4.087)
Study Controls		✓
Individual Controls		✓
N		2,499

P-values are for a significance test of whether the slope for the index of disadvantage coefficient (accounting for the interacted terms) is significantly different from zero.

^*p<.05

^**p<.01

^***p<.001

Note that significance tests are not possible in a multinomial logit model for the hazards of the reference group (the pill).

Selection into Contraceptive Practices

The results in Table 3 demonstrate that socioeconomic disadvantage has a significant relationship with contraceptive switching even when method, race, and side effects are controlled for. However, there is no significant relationship between disadvantage and discontinuation. A one-point increase in the index of disadvantage is associated with a 5.0% increase in the hazard of switching from 1-12 months (p<.01). Self-identified race is significantly associated with both switching and discontinuation across multiple groups, net of the enrollment contraceptive method, side effects, and level of disadvantage: Hispanic, Black, and Multiracial identification are associated with 37% to 363% increased chances of switching or discontinuing relative to White respondents. For Hispanic and Black respondents, the magnitudes of these associations are stronger for discontinuation than for switching. Because this is a competing risk model, this indicates that Black and Hispanic participants may be more likely to go straight to discontinuing a contraceptive method rather than switching. Multiracial respondents are more likely to experience switching, but not discontinuation, relative to White respondents.

Interacted Contraceptive Practice Model

Table 4 shows the coefficients (as hazard ratios) for the index of disadvantage for Model 3, which interacts disadvantage with race, method, and survey wave to better understand the dynamic interplay of these factors for contraceptive practice over time. The results in Table 4, alongside those shown in Figures 1.a through 2.b, are presented this way to facilitate the reader’s interpretation of these coefficients stratified by the various interactions, as these are often difficult to interpret.

Table 4 demonstrates significant positive associations between the index of disadvantage and switching at 1-month and 6-months across all methods used: more disadvantaged White and Multiracial respondents had significantly increased chances of switching at 1- and 6-months; more disadvantaged Hispanic respondents had increased chances of switching at 1-, 6-, and 12-months, while more disadvantaged Asian respondents had increased chances of switching at 1-, 3-, 6-, and 12-months across all methods. When all contraceptive methods are grouped, only more disadvantaged Multiracial respondents had an increased hazard for discontinuing at 6- and 12-months.

At 1-month, having a 1-point increase in the index of disadvantage is associated with increased chances of switching among all racial groups for the LNg IUD; Hispanic and Asian respondents for the pill and the contraceptive implant; and among Asian respondents for the Copper IUD and the injectable, on a magnitude of 21-68% increased hazards. Recall that this model stratifies respondents by self-identified race, contraceptive method, and survey wave, as well as by their interactions with the index of disadvantage. This means that, for example, White pill users at 1-month are compared to White pill users at 1-month—what varies is their score on the index of disadvantage. These findings suggest that at 1-month, there is a significant intersectional association between socioeconomic disadvantage and switching across methods used.

At 3-months, these significant associations attenuate, and then reappear at 6-months, followed by an attenuation again by 12-months. This pattern suggests that for some racial-method disadvantaged combinations, i.e. Hispanic pill users, White and Multiracial LNg IUD users, Hispanic implant users, and Asian Copper IUD users, the contact with contraceptive professionals prior to 1- and 6-months of use may be the most important in assuring that contraceptive users have methods that are meeting their varied needs.

Table 4 also demonstrates that there are several racial-method-disadvantage groups for which there is an increased chance of switching over the entire first year of use. These groups include more disadvantaged Asian pill users, more disadvantaged Hispanic and Asian LNg IUD users, and more disadvantaged Asian implant users. These groups have an increased chance of switching, relative to continuing their initial method, at each follow-up point after enrollment on a magnitude of 19% to 68% increased hazards.

These associations are illustrated by Figures 1.a through 2.b, which show the predicted probability of switching and discontinuing over 1-12 months by method, race, and by the first and fifth quintile of the index of disadvantage (where the first quintile represents the groups labeled “Low Disadvantage” and the fifth quintile represents the groups labeled “High Disadvantage”). In these figures, the solid lines indicate the groups with less disadvantage, while the dashed lines indicate groups who are more disadvantaged. Figure 1.a shows that the smallest within-racial switching gap is between White respondents. Asian respondents with fewer disadvantages have the lowest probability of switching over the 12-month period; however, the gap in switching between the least and most disadvantaged Asian respondents is the largest across racial groups. This is a new finding in the contraceptive literature and speaks to the complexities of how “model minority” status can intersect with socioeconomic disadvantage. Figure 1.a also demonstrates that the gaps between least and most disadvantaged Black and Hispanic respondents are of a similar magnitude, however, Hispanic respondents start at a lower level of switching.

In Figure 1.b, which breaks out the results in 1.a by contraceptive method, we can see that the gap between high and low levels of disadvantage are moderated by the type of contraceptive method used. Across all methods, Asian respondents from low levels of disadvantage have the lowest probability of switching. Black women from high levels of disadvantage experience some of the highest probabilities of switching across each method, while Multiracial, Asian, and Hispanic women from high levels of disadvantage experience similar switching trajectories across methods. White women from more disadvantaged backgrounds experience a protective effect for switching relative to less disadvantaged White women for the injectable at 3-months. Injectable contraception lasts for 3 months, suggesting that disadvantaged White women experience an increased ability to access their 3-month re-injection. Figure 1.b also shows that advantaged Hispanic women may have similar switching experiences as advantaged White women, however, there is a larger gap between advantaged and disadvantaged Hispanic women than within the White respondent group.

Figure 1.b:

Predicted Probability of Switching

When aggregated (see Table 3), the index of disadvantage is not significantly related to discontinuation. However, the inclusion of the interaction terms clues us in to some potential trends. At 1-month, more disadvantaged White implant and injectable users experience a protective effect against discontinuation relative to less disadvantaged White implant and injectable users (see Table 4). There are no other significant relationships at 1- and 3-months for discontinuation, but by the time we reach 12-months, respondents who identify as Multiracial and who are more disadvantaged are significantly more likely to discontinue their method by 12-months compared to continuation across all methods except for the vaginal ring. Additionally, disadvantaged Hispanic pill users have a 24% increased chance of discontinuing relative to continuing by 12 months. Figure 2.a shows that more disadvantaged Multiracial, Asian, and Hispanic women follow similar discontinuation patterns, but have different magnitudes of gaps from their more advantaged counterparts, with Asian women having the largest gap, then Multiracial women, then Hispanic women. Across the different methods (Figure 2.b), we can see that the least disadvantaged women discontinue earlier on, and then their levels of discontinuation fall, while more disadvantaged women’s discontinuation increases from 1-3 months (although not significantly).

Figure 2.a:

Predicted Probability of Discontinuing Across All Methods

DISCUSSION

The findings presented here suggest that contraceptive patients with more disadvantage are first sorted into specific methods when they are choosing a new type of method free of cost, and then—through pre-existing structural factors, cultural norms, or personal preferences—they experience differential non-continuation of contraceptive methods by socioeconomic disadvantage and self-identified race. The relationship between disadvantage and contraceptive switching and discontinuation is distinct compared to the initial contraceptive selection process, suggesting that pre-existing disadvantage has related but unique relationships with method selection and with contraceptive practices over time.

The distinct patterns of contraceptive selection and practice associated with socioeconomic disadvantage and self-identified race coincide with decreased protection against pregnancy. For example, more disadvantaged women were less likely to select IUDs and more likely to select injectables at enrollment in the study (IUDs offer better protection against pregnancy over one year of use compared to injectables among typical users, Trussell, 2013). The associations between disadvantage and switching or discontinuing suggest that disadvantaged respondents are more likely to have gaps in their contraceptive use, which in turn leads to a higher risk of pregnancy. The process of switching methods is related to disadvantage in ways that suggest that there are distinct experiences over race and time. For example, disadvantaged White respondents—a racial (but not socioeconomic) group which often receives increased educational and occupational prospects in the U.S. context—experienced protection against switching for certain methods, while disadvantaged non-White respondents experienced increased chances of switching methods.

Patterns and associations with non-continuation of a contraceptive method are important patient-centered outcomes to examine because they represent dynamic and complex socio-medical processes through which embodied users engage technologies in order to interact with the institutions of sex and reproduction. These processes are socially patterned in raced, heteronormative, and classed ways that affect equity and the ability to achieve individual bodily autonomy. Some of the results in this paper regarding switching contraceptive methods will require more investigation, in particular the identified gaps between the least and most disadvantaged respondents for switching (and sometimes discontinuing) across racial groups. Asian respondents have the largest gap, suggesting that there may be important research to be done identifying differences between disadvantaged and advantaged Asian contraceptive users. Advantaged Hispanic women have a similar switching trajectory as White women (see Figure 1.a), but have a larger gap between themselves and disadvantaged Hispanic women. This suggests that some race-socioeconomic combinations may experience similar benefits to more advantaged White groups, but the more pronounced difference for non-White disadvantaged respondents leads to increases in switching across a number of methods. Further research is needed to understand whether and how Multiracial identification can exacerbate or temper socioeconomic disadvantage and its relationship with contraceptive use. These findings offer an important insight for implementation in contraceptive programs: eliminating financial barriers to access contraceptive services does not eliminate the socioeconomic contexts that influence method selection and use that occur as part of everyday lived experiences.

This study is subject to some limitations. First and foremost, this dataset is not representative of the national U.S. population. The diversity of the sample offers enough variance for the findings to be robust, but expansion to a more diverse urban setting may yield different effects than what are demonstrated in this manuscript. While this suggests that these findings could be non-transferrable to other populations, the exact magnitude of the results are irrelevant to the point that this paper is making. This point is that the finding—that race/ethnicity interacts with disadvantage to differentially shape contraceptive use, net of costs—is a significant finding. This suggests that populations with as much or more variation than the population examined here may retain the complex relationships between socioeconomic disadvantage, race and ethnicity, and contraceptive use, which is important for how we provide reproductive justice to contraceptive users. Second, it deals with reversible contraceptive methods and focuses on the relationship between socioeconomic disadvantage, self-identified race, and selection of one of six hormonal contraceptive methods at enrollment. Other work suggests that one of the main contraceptive methods in the U.S. is sterilization, although its pre-eminence as a contraceptive method is on the decline (Daniels and Abma, 2018). Because this study examines not only method selection, but also method use over time, it was important that the methods included not be permanent. Additionally, the 12-month continuation rates for LARC methods in the HER Salt Lake study are slightly lower than those demonstrated in a systematic review of continuation rates, while those for the pill and injectables are slightly higher (Usinger, 2016). However, other studies that sample different populations or examine different durations of use have estimations of continuation rates that vary on the order of up to 40 percentage points from Usinger et al.’s calculations (Moreau, Cleland, and Trussell, 2007; Vaughn et al., 2008; Peipart et al., 2011). I am also unable to identify the exact time and reason for contraceptive switching or discontinuation due to how the data is set up, rather I must infer that switching or discontinuing occurred over discrete intervals. Finally, I am unable to identify specific causal pathways for why the index of disadvantage is associated with contraceptive selection and practice. However, there are several possible pathways through which disadvantage could be acting on contraceptive selection and practice: these could include user preferences for features of contraceptive methods (Jackson et al., 2016), provider bias in counseling on methods (Higgins, Kramer, and Ryder, 2016; Stevens, 2018), awareness of historical injustices relating to reproductive health in the U.S. (Roberts, 1999), distrust or discomfort with the current health systems (Bell et al., 2018; Guzzo and Hayford, 2020), barriers related to access to services (Frost, Lindberg, and Finer, 2012; Guzzo and Hayford, 2020), different levels of sexual and reproductive health knowledge (Guzzo and Hayford, 2020), and differences in the perceived consequences of childbearing and differences in ambivalence towards unintended pregnancy (Guzzo and Hayford, 2020).

My findings suggest that the platform that some policymakers and researchers use to promote the contraception-as-poverty reduction logic is weaker than is suggested in existing literatures. Continuing to espouse this logic is an onerous path to walk—it puts the burden of poverty reduction on women’s reproductive bodies, attributes social inequalities to the reproductive practices of individuals, and legitimizes the regulation of fertility and bodies (Higgins, 2014). Additionally, the goal of reducing unintended pregnancy through contraceptive provision as a “solution” to poverty reinforces the concept that unregulated or misregulated reproduction is the cause, rather than the consequence, of social disadvantage or poverty. It does this by attributing the cause or timing of an unintended pregnancy to individuals’ characteristics, rather than to constraints they may face. People’s choices about what contraceptive method to use and whether they are able to use them consistently and effectively transmutes the right to bodily autonomy into a moral statement about who is a responsible reproducer.

CONCLUSION

This study focuses on the associations that pre-existing socioeconomic disadvantage has with contraceptive method selection and use. It explicitly does not argue that programs that offer contraceptive supplies and services to low-income women are in any way negative—they have proven health benefits for women and their families (Hanson et al., 2015; Kost and Lindberg, 2015). Instead, I posit that the argument that contraceptive provision offers a silver bullet solution to reducing socioeconomic disadvantage offers a potential problematic to both scientific and policy communities. For scientific communities, I have demonstrated here that socioeconomic disadvantage is endogenous to method selection and use, which is often unaccounted for in existing work. From a policy perspective this argument does a disservice to women, who are disproportionately affected by poverty, by not supporting programs that offer women and their families a basic minimum standard of living, and by using the provision of contraceptive methods in the service of poverty reduction, rather than as a way to achieve women’s reproductive autonomy and agency as a goal in itself.

• Disadvantage structures selection of new contraceptive methods and use of methods

• Social disadvantage acts in unique ways over time to structure contraceptive use

• Switching among disadvantaged patients suggests initial preferences may not be met

• SES and racial components of continuation offer evidence for stratified reproduction

• Results suggest need for increased focus on reproductive justice approach