Abstract

Beginning August, 2012, the U.S. Patient Protection and Affordable Care Act (ACA) required new private health insurance plans to cover contraceptive methods and counseling without requiring an insured’s copay. The ACA represents the first instance of federally mandated contraception insurance coverage, but 30 U.S. states had already mandated contraceptive insurance coverage through state-level legislation prior to the ACA. This study examines whether mandated insurance coverage of contraception affects contraception use, abortions, and births. I find that mandates increase the likelihood of contraception use by 2.1 percentage points, decrease the abortion rate by 3 %, and have an insignificant impact on the birth rate. The results imply a lower-bound estimate that the ACA will result in approximately 25,000 fewer abortions.

Introduction

Although oral contraceptives (birth control pills) were approved in 1960 by the U.S. Food and Drug Association (FDA), insurance coverage of contraception has not necessarily been widespread in the United States. The Kaiser Family Foundation (2003) found that 58 % to 80 % of small firms and 58 % to 78 % of large firms had insurance plans that covered the five leading reversible contraceptives.1,2 In a study of 492 health care insurers across the United States, Sonfield et al. (2004) found that in 1993, 32 % to 59 % of typical employer-based insurance plans covered some form of contraception, and coverage increased to 78 % to 97 % in 2002.3 The importance of providing access to all forms of contraception is highlighted by the fact that one-half of all pregnancies in the United States are unintended (Finer and Henshaw 2006), which suggests a high incidence of non-use or improper use of contraception.

Beginning in August, 2012, the Patient Protection and Affordable Care Act (ACA) required new private health insurance plans to cover contraceptive methods and counseling without requiring an insured’s copay. The ACA represents the first instance of federally mandated contraceptive insurance coverage. Prior to the ACA, some states passed Equity in Prescription Insurance and Contraception Coverage (EPICC) legislation, which effectively mandated insurance coverage of contraception for employer-based plans. The first mandate was enacted in 1998 by Maryland. A total of 28 states had contraception insurance mandates prior to the ACA, and two others (Michigan and Montana) required contraception coverage as a result of Attorney General opinions. Between 1998 and 2009, 61 % to 69 % of U.S. individuals obtained insurance coverage through their employers, such that contraceptive insurance mandates impact a significant proportion of the population.

The purpose of this study is to estimate the causal effect of mandated insurance coverage of contraceptives on contraception use, abortions, and births. The estimation strategy takes advantage of variation in state-level mandate timing and uses a flexible time specification in order to estimate the effect for both the year of mandate introduction and effects in years subsequent to the mandates. State-level analysis is conducted using National Vital Statistics birth data and abortion data from the Centers for Disease Control and Prevention (CDC). Additionally, individual-level analysis is performed using the Current Population Survey (CPS) Fertility Supplement, which has the benefit of being able to link birth outcomes to other individual characteristics. Finally, the Behavioral Risk Factor Surveillance System (BRFSS) is used to estimate the effect of mandates on contraception use and contraception type.

Most research on health insurance mandates has focused on the potential labor market costs of the mandates, measured in terms of decreases in wages, employment, or the probability of insurance being offered. For example, Gruber (1994) found that regulations mandating benefits actually have very little effect on whether firms drop coverage for their employees. He further postulated that this might be the case because most firms already provided the mandated benefits prior to legislation, and therefore mandates are not binding. Kaestner and Simon (2002) found little evidence that state health insurance regulation—specifically, mandated benefits—impacts labor market outcomes, such as wages or hours worked.

A somewhat related branch of research has examined the impact of state-mandated insurance coverage of infertility treatments on treatment utilization and birth rates. Bitler and Schmidt (2006) found no evidence that the mandates increase treatment utilization for the overall population of women aged 15–44, but Henne and Bundorf (2008) and Bitler and Schmidt (2012) found evidence to the contrary. Using data from the National Survey of Family Growth (NSFG), Bitler and Schmidt (2012) found that infertility treatment mandates increase utilization, with the largest effects occurring among older, more-educated women. Hamilton and McManus (2012) found that mandates increase in vitro fertilization (IVF) usage. Schmidt (2007) found that infertility coverage mandates lead to an 8 % increase in birth rates among women over age 35 but that the mandates do not reduce racial disparities in access to treatment.

Although contraception coverage mandates have not previously been studied in the demography literature, several studies have examined the effect of insurance coverage on contraception use and fertility outcomes. Although their results are not causal, Culwell and Feinglass (2007) used the 2002 BRFSS and found that insured women are more likely to use prescription contraceptives compared with uninsured women. Kearney and Levine (2009) found that Medicaid waivers that increase income limits for eligibility increase the probability of contraception use among sexually active women. Moreover, they found that waivers decrease nonteen births by 2 % and teen births by 4 %. Postlethwaite et al. (2007) used a change in insurance coverage for members of the Kaiser Foundation Health Plan in California to determine the impact of providing coverage without copays for the most effective contraceptive types.4 The change in coverage was found to be associated with a 132 % increase in couple-year protection for intrauterine devices (IUDs) and a 32 % increase in couple-year protection for injectables.

This article finds evidence that mandates impact contraception use and abortions, but do not significantly impact births. Specifically, the BRFSS results indicate that mandates increase the likelihood that women use contraception by 2.1 percentage points, with a positive and significant increase in hormonal contraception and a positive but insignificant change in nonhormonal contraception. Results from state-level abortion and birth data indicate that mandates lead to a 3 % decrease in abortions and an insignificant (0.2 %) decrease in births. The state-level results are used to calculate back-of-the-envelope estimates of the impact of the ACA, which effectively extends contraceptive insurance mandates to all states. As a lower bound estimate, the ACA is expected to result in approximately 25,000 fewer abortions and 5,000 fewer births (in a single year) in the states that did not have contraceptive insurance mandates prior to the ACA. These findings are particularly important given that the contraceptive mandates of the ACA were recently challenged in the Supreme Court. In the case Burwell v. Hobby Lobby Stores, Inc., the owners of Hobby Lobby and Conestoga Wood Specialties argued that their religious beliefs prohibit them from providing health coverage for contraceptive drugs and devices that end human life after conception. In June 2014, the Supreme Court ruled that the ACA requirement to provide contraceptive coverage could not be applied to “closely held corporations” because it violates the Religious Freedom Restoration Act (RFRA). Although the long-term impact of the ruling is unclear, it has the potential to decrease the mandate’s effectiveness with respect to providing the full range of contraception.

The outline of the article is as follows. The next section provides additional background on contraception and contraceptive insurance mandates. Then I describe the data used in the analysis: descriptive statistics for state-level abortions and births, the CPS sample, and the BRFSS sample are provided. The fourth section presents the estimation strategy: the effect of mandates on fertility outcomes is initially modeled as a structural break and then extended to a more flexible dynamic framework. I then present the results and conclude.

Background

Contraception

The availability of a wide range of contraception options allows women to choose the method that best suits their needs. However, when the full range of options is not available or affordable, some women might choose less-effective methods or forgo birth control altogether, which increases the probability of an unplanned pregnancy. The estimated out-of-pocket costs for birth control are substantially less than pregnancy itself, but they are not trivial. For example, Planned Parenthood lists the following as cost estimates: oral contraceptives (the Pill) cost between $15 and $50 per month; contraceptive injections cost between $35 and $75 per injection and last for three months; hormonal implants cost between $400 and $800 up front and last for three years; and IUDs cost between $500 and $1,000 up front and last for 12 years (Planned Parenthood n.d.).5 None of these costs include the exam fees needed to obtain contraception.

Ensuring that prescription contraceptives are available and reasonably priced for women is important because these methods of contraceptives provide the lowest failure rates under typical usage. Trussell (2011) estimated the percentage of women experiencing unintended pregnancy under typical use of various types of contraception6 and found that unintended pregnancy percentages during the first year of use are lowest for prescription contraceptives: 0.05 % of women using implants experience an unintended pregnancy, with 0.8 % for IUDs; 6 % for injectables; and 9 % for the Pill, patch, and ring.7 Nonprescription contraceptive use results in a lower percentage of unintended pregnancy than using no method (85 %) but is less effective than prescription methods: 18 % of women using a condom experience an unintended pregnancy, with a 24 % pregnancy rate for fertility-awareness methods, and 28 % for spermicide use. Despite lower failure rates for prescription contraception use, approximately 14 % of all women use nonprescription contraceptives. Data from the 2006–2008 NSFG indicate that approximately 5 % of women who discontinued use of contraceptives did so because it was too expensive or their insurance did not cover it, which is similar to estimates found by Jones et al. (2002).

Contraception usage estimates from the NSFG indicate that overall contraception use has not changed much between 1995 and 2006–2008, although it has decreased slightly among 20- to 29-year-olds. The percentage of individuals using prescription contraception has increased slightly between 1995 and 2006–2008 among all age groups, with the exception of women between the ages of 20 and 24. The increase (decrease) in prescription (nonprescription) use among teenagers was considerably larger compared with other age groups between 1995 and 2002 (Mosher and Jones 2010).8

Contraceptive Insurance Mandates

Contraceptive insurance mandate information was obtained from the National Conference of State Legislatures (NCSL) as well as the Guttmacher Institute. Maryland enacted the first contraceptive insurance mandate in 1998, and 27 states have subsequently enacted mandates. Additionally, Michigan and Montana require insurance coverage of contraception as a result of Attorney General opinions. Table 1 gives full details for year of mandate introduction and exemptions by state.

The state-level mandates apply only to employer-based health plans, but employers who self-insure are exempt from the legislation because they fall under federal jurisdiction. As a conservative estimate, given that 61 % of individuals receive insurance through their employer (DeNavas-Walt et al. 2010), and 52 % of those individuals are covered by plans that are not self-insured (Kaiser Family Foundation 2003), the mandates have the potential to impact at least 31 % of the population. Twenty states have some form of an exemption clause (either allowing the employer to refuse to offer coverage, or the insurer to refuse to write a plan that includes coverage), most of which are religion-based. Although these mandates increase access to coverage, the insurance plans might still employ some form of cost sharing, which may result in inadequate decreases in out-of-pocket costs for some individuals. As an example, a 2011 study found that privately insured women using oral contraceptives whose plan covered prescription drugs paid approximately one-half (53 %) of the cost of the medication (Liang et al. 2011).

Given that mandates operate through employer-provided insurance, they may lead to greater disparities within the population. The most obvious disparity would occur between the employed and insured relative to the unemployed and uninsured, meaning that disparities between certain demographic groups could arise due to their correlation with insurance status. Based on estimates from the Kaiser Family Foundation, white individuals are more likely to have employer-based insurance than black individuals (68.5 % to 74 % vs. 51.4 % to 58 %), which implies that mandates might have a larger impact on whites. Similarly, individuals with more education are also more likely to have employer-based insurance: 30.9 % of individuals with less than a high school diploma receive insurance through their employer, compared with 57.2 % of high school graduates and 80.6 % of college graduates. Individuals aged 35–54 are more likely to have employer-provided coverage; however, they are also outside their prime childbearing years, so it is unclear whether fertility levels or insurance levels will have the predominate effect for various age groups.

Higher-income individuals are also more likely to receive insurance through their employer: 86.6 % of individuals with incomes greater than 300 % of the poverty level have employer insurance, compared with 70.7 % of individuals with incomes between 200 % and 299 % of the poverty level (Kaiser Family Foundation 2010). Individuals with incomes below 200 % of the poverty level are more likely to be covered by Medicaid, which is required to provide free contraception services.9 Although higher-income individuals are more likely to receive insurance through their employer, they are also more likely to be able to pay for contraception out of pocket. Therefore, it is possible that the mandates may not have a large effect on higher-income individuals, particularly on the extensive margin. However, given the high up-front cost of the most-effective forms of contraception—such as implants and IUDs—perhaps high-income individuals are willing to pay for less-effective forms of contraception out of pocket, but the mandates allow these individuals to switch to more effective contraception.

The potential effect of the mandates could be small under three scenarios. First, mandates might not be binding if a majority of firms claim religious exemptions. The findings by Sonfield et al. (2004) indicated that religious exemptions might not be common because they found a large disparity in contraception coverage in mandate and nonmandate states for plans determined at the local level.10 Specifically, they found that 47 % to 61 % of plans in nonmandate states cover the five leading contraception methods, compared with 87 % to 92 % of plans in mandate states. The high level of coverage in mandate states would not be present if a substantial amount of religious exemptions were claimed. Additionally, the wording of the exemption in the legislation in a majority of states is narrow enough such that many religious-based organizations, such as schools and hospitals, are not eligible for exemptions.

Second, the mandates might not be binding if a majority of insurance plans provided contraception coverage prior to the mandates. Based on 1999 estimates, 32 % to 57 % of small firms and 42 % to 68 % of large firms provided coverage for reversible contraception (Kaiser Family Foundation 1999). These numbers trended downward in 2001 and increased in subsequent years. Slightly more than one-half of the mandates were enacted prior to 2001; therefore, it is unlikely that coverage levels were high at the time of mandate implementation.

The final scenario under which mandates might not be binding is if a majority of individuals receive insurance through employers that self-insure. Although a large percentage of insured workers (48 %) are covered through self-insured plans, a majority are not (Kaiser Family Foundation 2003).

Data

Behavioral Risk Factor Surveillance System

This article uses the BRFSS to determine whether contraception insurance coverage mandates impact contraception usage. The BRFSS asks questions about contraception usage and type in 1998, 1999, 2000, 2002, 2004, 2006, 2010, and 2011. The sample is limited to women aged 18–44, and women who are currently pregnant or report having a same-sex partner are dropped from the sample. Summary statistics are presented in Table 2.

The analysis focuses on whether individuals use any contraception and the contraception type. Contraception type is first broken into two broad categories: hormonal and nonhormonal. I expect mandates to increase hormonal birth control usage. The effect of mandates on nonhormonal contraception will be negative if individuals are switching from nonhormonal to hormonal. However, if the increase in hormonal birth control is driven by individuals who did not use contraception prior to mandates, then nonhormonal contraception use could be unchanged as a result of the mandates.

Finally, I would like to examine whether individuals are switching to the most effective types of hormonal contraception. Three subgroups of hormonal contraception are used to analyze this issue. Short-term contraception methods comprise the Pill, patch, or ring; medium-term contraception methods comprise shots and implants; and long-term contraception is provided via IUDs. Among the subsample that uses contraception, 7 % use short-term contraception, 38 % use medium-term contraception, and 43 % use long-term contraception. It is unclear whether individuals will change their contraception type in response to mandates. If individuals are already using their preferred type, or if cost sharing under mandates is high, then the mandates may have little impact. However, if individuals were using the least-expensive type prior to the mandates, there may be some switching to more effective types, particularly if cost sharing is low.

State-Level Abortion and Birth Rates

Aggregate data represent state-level data for 1993–2011.11 The two dependent variables considered are the abortion rate and birth rate. Birth rates are collected from the CDC National Vital Statistics System, and can be separated by various age categories. The birth rate is given as the number of births per 1,000 women aged 15–44.12 The abortion rate represents the number of abortions performed in a state per 1,000 women aged 15–44 who reside in that state. The CDC collects these data annually; however, some states do not report abortion statistics to the government.13 As with the birth data, abortion data can be separated by various age categories. Population estimates from the U.S. Census Bureau are used to calculate rates for subgroups because these data are reported in totals rather than rates. Population-weighted averages for abortion and birth rates for various mandate definition subsamples are given in Table 3.

Because mandates work to decrease the cost of contraceptives, I expect them to affect abortions and births among those with employer-based insurance, but the direction of the effect is ambiguous. Holding sexual activity fixed, pregnancy rates (and consequently abortion and birth rates) should decrease. However, the lower price of contraception is also likely to increase sexual activity, which increases the probability of a pregnancy and subsequent abortion or birth. Therefore, the overall effect on abortions and births will depend on which effect dominates.

Trends in abortion and birth rates between 1993 and 2011 are shown in Fig. 1. Population-weighted average rates are plotted for states that enacted a contraceptive insurance mandate prior to 2002 (“early adopters”), states that enacted a mandate after 2002 (“late adopters”), and states that do not have a mandate. The difference in levels between mandate and nonmandate states indicates that analysis should control for state-specific factors. Late adopters have the highest abortion rates compared with early adopters and states without mandates, but early adopters have the highest birth rates. Abortion rates were already declining in mandate states prior to mandates, but they continue to fall between 1998 (the year when the first mandate was enacted) and 2002. Therefore, the extent to which the mandates are effective will depend on whether they impact abortion rates beyond the existing pre-mandate decline. The patterns for birth rates are less clear; however, birth rates declined after 2002 in late adopter states, indicating that mandates may have a significant effect on birth rates.

CPS Fertility Supplement

As a complement to the state-level birth analysis, the June CPS Fertility Supplement allows us to examine the effect of mandates on individual birth outcomes. The June CPS is available in 1995 and even years between 1998 and 2010. The survey asks all women when their most recent child was born. This question is used to construct an indicator variable for whether the individual has given birth in the past year. The sample is limited to women between the ages of 15 and 44 for consistency across state-level and individual-level data. Individual-level analysis also uses the following controls: race/ethnicity, marital status, education, family income, employment status, and age. Summary statistics are presented in Table 4.

Estimation Strategy

State-Level Regression Models

A simple framework to determine the impact of the mandates on fertility outcomes is given by
Yst=βMandatest+Xstγ+θs+θt+εst,
(1)
where s indexes state, t indexes time, Yst represents the abortion rate or birth rate, Mandatest is an indicator equal to 1 if there is a contraceptive insurance mandate present in state s at time t, and Xst are state-level controls.14
The model in Eq. (1) implicitly assumes the effect of the mandates is captured by a structural break. A change in abortions might occur in the year of the mandate itself; however, a birth outcome would be delayed such that an impact on birth rates might not occur for at least one year following a mandate. Additionally, the model in Eq. (1) can confound preexisting trends with the response of the outcomes to contraceptive insurance mandates. To address these issues, I also estimate a flexible time specification as proposed by Wolfers (2006). Specifically, the following model is estimated:
Yst=kKβkMstk+Xstγ+θs+θt+εst,
(2)
where s indexes state, t indexes time, Yst represents the abortion rate or birth rate, Mstk is a dummy variable equal to 1 in the kth period following the contraceptive insurance mandate (k = 0 implies the mandate introduction took place that period), θt are year indicator variables, θs are state indicator variables, and Xst are state-level controls. The state indicator variables (θs) allow for time-invariant unobservable state characteristics to affect fertility outcomes, whereas the year indicator variables (θt) allow for a flexible specification of the aggregate trend in fertility outcomes. βk is interpreted as the impact of a mandate for insurance coverage for contraception in state s at time t on the abortion/birth rate k periods following the mandate (relative to states without a mandate at time t). The set K indicates the number of leads and lags included in the model. For the upcoming results, I use the set K = {−5 : −1; 0 : 1; 2 : 4; 5 : 7}.

Even with the dynamic specification in Eq. (2), there might still be concern that states experiencing downward trends in fertility outcomes are more likely to enact mandates, such that the estimates in Eq. (2) would simply be picking up the preexisting trends. To further rule out potential endogeneity of the presence or timing of mandates, I estimate models that regress whether a state ever adopts a mandate on state characteristics as well as the trends in birth, abortion, and insurance rates. Additionally, for states with mandates, I regress year of mandate on state characteristics and the average birth, abortion, and insurance rates in the years leading up to mandate introduction.

The results from the endogeneity checks are presented in Table 5. The only significant predictor of the presence of a mandate is the percentage of the population that voted Democrat in the 1996 presidential election (significant at the 10 % level), which should be captured by state-level fixed effects.15 Among states with mandates, none of the variables were significant predictors of mandate timing. The results are robust to limiting the sample to states that adopted mandates prior to 2002. Although these results indicate that preexisting trends in fertility can be ruled out as a source of endogeneity, it is not possible to rule out state-level shocks in a given year as a potential source of bias.

Individual-Level Regression Models

Given the relatively small sample sizes in any given state-year cell, the BRFSS analysis estimates only the model in Eq. (1) augmented with individual controls:
PrYi=1Xi,si,ti=βMandatesi+Xiγ+δsi+δti,
(3)
where i indexes individuals; Yi is an indicator equal to 1 if the individual uses birth control, or an indicator equal to 1 if the individual uses a specific type of birth control; Xi is a vector of individual controls, s(i) is individual i’s state, and t(i) is the year in which individual i was sampled. The CPS analysis estimates the model given in Eq. (3) where Yi is an indicator equal to 1 if the individual has given birth in the past year, Xi is a vector of individual controls, s(i) is individual i’s state, and t(i) is the year in which individual i was sampled. Because state-year sample sizes are larger for the CPS sample, flexible time models analogous to Eq. (2) are also estimated using the CPS data.

Results

The mandate indicator Mandatest in both the aggregate and individual-level regressions is defined as equal to 1 if state s has any contraceptive insurance mandate at time t. This broad definition of mandates leads to the inclusion of both Michigan and Montana as states with mandates, even though their mandated coverage was enacted through Attorney General opinions. Several definitions of mandates were used as a robustness check, but the results do not change much by how broadly mandates are defined.

BRFSS Results: Contraception Use

The results from the BRFSS estimation are given in Table 6.16 Although the BRFSS asks whether individuals have health insurance, respondents do not indicate whether their insurance is obtained through their employer or the government. Therefore, it is not possible to determine the exact effect of mandate on individuals who obtain insurance through their employer. For the full sample (women aged 18–44), mandates increase the probability of using any type of contraception by 2.1 percentage points.

The next set of results breaks contraception use into different types: hormonal, nonhormonal, and various subgroups within hormonal contraception. Hormonal contraception includes the Pill, patch, ring, shots, implants, and IUDs. Nonhormonal contraception includes condoms; diaphragms; natural family planning; withdrawal method; and foam, jelly, or cream spermicides. For the full sample, the effect of mandates on hormonal birth control usage has the expected sign. Specifically, mandates increase the probability of using hormonal birth control by 1.8 percentage points. The effect of mandates on nonhormonal contraception use is positive but not statistically significant, which indicates that individuals are not switching from nonhormonal to hormonal contraception as a result of the mandates.

Hormonal contraception is broken into three subgroups: the Pill, patch, or ring (short-term contraception); shots and implants (medium-term contraception); and IUDs (long-term contraception). Mandates lead to an increase in usage for all three contraception subgroups; however, the effect for IUD use is insignificant. The effect of mandates on short-term contraception use is statistically significant, but the point estimate is quite small (0.002). Mandates lead to increases in shot and implant usage of 1.5 percentage points for the full sample and 3.6 for women in their 20s. This finding is important because one of the arguments for providing insurance coverage of contraception is to induce women to use more effective types of birth control. Although there is no evidence that women are using the most effective type (IUDs) as a result of the mandates, there is some evidence that women are more likely to use shots and implants, which is an improvement over the Pill, patches, and rings.

State-Level Results: Abortion and Birth Rates

The state-level estimation results for the single mandate indicator specification are presented in Table 7.17 In the single mandate indicator specification, contraceptive mandates lead to 1.035 fewer abortions per 1,000 women aged 15–44. This translates to a 6 % decrease in abortion rates relative to baseline.18 The birth rate per 1,000 women aged 15–44 decreases by 1.95, which corresponds to approximately a 2 % decline relative to baseline.

To determine whether contraceptive insurance mandates have differential effects on various age subgroups, I split the sample into three categories: teenagers (between age 15 and 19), adults in their 20s, and adults in their 30s. Mandates decrease abortion rates across all age subgroups. Although the point estimates are largest for women in their 20s, the implied percentage decrease in abortions is larger for teenagers: mandates result in a 6 % decrease in the abortion rate for women in their 20s compared with an 11 % decrease in the abortion rate for teenagers. Similarly for births, although the point estimates are larger for women in their 20s (3.2) compared with teenagers (0.9), both translate to an approximately 3 % decline in the birth rate for each group.

For the flexible time specification, there is no evidence of significant pre-mandate effects, and the mandate effects appear 0–1 years after mandate introduction.19 There is evidence that mandates have an effect on fertility outcomes beyond the initial year of introduction and remain stable in the long run.

Robustness Checks and Placebo Regressions

The results for three sets of robustness checks are presented in Table 8. The first checks for whether the results are sensitive to the inclusion of both linear and quadratic state-specific time trends. The second robustness check restricts the sample to states that ever adopt a mandate. Finally, there may be concerns that the results in this article are driven by legislation related to fertility outcomes that was passed during the relevant sample period. To check for this possibility, I estimate regressions with a set of controls for other state legislation related to fertility outcomes. Specifically, I include controls for whether the state has mandated insurance coverage for infertility treatments, whether a state has early postpartum discharge laws, and whether a state has expanded Medicaid eligibility for pregnant women.

The abortion results are still statistically significant but are attenuated after state-specific time trends are added to the model.20 Accounting for state-specific time trends, mandates decrease abortions by approximately 0.561 per 1,000 women aged 15–44, which implies a 3 % decrease. Controlling for state-specific time trends, mandates still negatively impact birth rates, but this is no longer statistically significant.

The results for both the simple and flexible time specifications are similar if the sample is restricted to states that pass a mandate. Adding controls for other fertility-related legislation to the model does not change the results much. In general, the main results along with the robustness checks indicate that the effect of contraceptive insurance mandates on fertility outcomes is dependent on the inclusion of state specific trends, and this is particularly true for the flexible specification and births. The range of estimates indicates that both abortions and births either remain unchanged or fall in response to contraceptive insurance mandates.

Results from three sets of placebo checks are presented in Table 9. The placebo regressions include all the controls listed in Table 7, and are robust to inclusion of state-specific time trends. The first placebo check estimates the effect of contraception insurance mandates on the percentage of adult smokers. The mandates should not affect the number of individuals who smoke; as expected, none of the results are significant. The second placebo check uses the full sample and randomly assigns mandates in years when there was no change in coverage. Results from 5,000 replications indicate that the placebo mandates do not have any significant effect in either the simple or flexible framework. The third placebo check limits the sample to nonmandate states and randomly assigns “placebo” mandates. As expected, the placebo mandates do not have a significant effect on outcomes.

CPS Fertility Supplement Results: Births

The CPS regressions include all the controls listed in Table 4. Although the coefficients on these controls are not of direct interest and therefore not reported, it is worth noting that the probability of giving birth in the past year is lower among women with higher levels of education; with higher levels of income; and who are single, employed, or white. Given these results, models are estimated with interaction terms to determine whether contraceptive insurance mandates impact certain demographic groups disproportionately. Specifically, the models estimate differences between whites and nonwhites; non–college graduates and college graduates; married and single individuals; individuals less than 30 years old and individuals greater than or equal to 30; and households with low income (those earning less than $50,000 per year) and high income (greater than $50,000 per year). The results for the CPS estimation are presented in Table 10.

The effects for the CPS results are marginal effects multiplied by 1,000, which can be interpreted as the effect on the birth rate. For example, in the first cell of Table 10, the effect would be interpreted as the mandate leading to 2.77 fewer births per 1,000 women. For the flexible specification, the effect of mandates on the birth rate does not become significant until two to three years following the mandate introduction, and remains relatively stable until it becomes insignificant eight years following the mandate. Given that the birth rate in the full CPS sample is 92.2, the effect two to three years following the mandate represents a 3 % decrease in the birth rate, which is consistent with the results found using the aggregate data.

Columns 2–6 in Table 10 present differences in effects for various demographic subgroups. The estimates are coefficients on interaction terms, with the included group defined as the first group listed. For example, in column 2, 3.03 indicates the effect of mandates on birth rates for white individuals, relative to the base group of nonwhites. For the most part, there are not significant differences by subgroup. Consistent with the aggregate results, the difference between women under the age of 30 and women 30 or older is marginally significant, but this effect does not persist in the long run.

Policy Extension: Affordable Care Act

Under the ACA, beginning August 1, 2012, women’s preventive health care services are covered with no cost sharing, and this includes prescription contraceptive drugs and devices.21 This legislation effectively extends existing contraceptive insurance mandates to the 20 states that did not previously have mandates, and also extends coverage to all employer-based insurance plans (including employers that self-insure).

The estimates from the mandate indicator specification with quadratic state-specific time trends are used to make back-of-the-envelope calculations of the effect of the ACA in the 20 states that did not have mandates prior to the ACA. Based on the single mandate indicator aggregate estimates, mandates decrease abortions (births) by approximately 3 % (0.2 %). This implies that the effects of ACA implementation will result in approximately 25,000 fewer abortions and 5,000 fewer births in a single year in the 20 states that did not have contraception insurance mandates prior to the ACA.

These estimates are a lower bound for three reasons. First, the state mandates allow for cost sharing, but the ACA will remove cost sharing from plans.22 Second, employers that fully insure are impacted by the state mandates, whereas both employers that fully insure and employers that self-insure will be impacted by the ACA. Finally, as of January, 2014, individuals who did not have employer-provided insurance are able to obtain insurance through health insurance exchanges. Plans purchased through exchanges will provide contraception coverage without cost sharing. As with the state mandates, the ACA has limited religious exemptions, but the language is similar to current exemptions, such that exemptions should not have an increased impact under the new legislation.23

Conclusion

This article estimates the effect of contraceptive coverage insurance mandates on contraception use, abortions, and births. Results using data from the 1998–2011 BRFSS indicate that mandates increase the likelihood that women use contraception by 2.1 percentage points, with significant increases in hormonal contraception use, particularly for shots and implants. The results from state-level abortion data from 1993–2011 indicate that mandates lead to a decrease in abortion rates. For the simple model with state-specific time trends, mandates decrease the abortion rate by 3 %, but based on the flexible time specification, the effects do not persist in the long run. Birth rates either decrease or do not change depending on whether state-specific trends are included in the model. These estimates may be somewhat understated due to spillover effects caused by insurance companies that determine policies at the national level. Individual-level estimates do not indicate that differential effects exist by demographic characteristics other than age.

The effects found in this study complement other studies that examine the impact of various public policies on fertility outcomes. For example, Mellor (1998) found that expanding Medicaid access to family planning services leads to a 7.2 % decrease in the probability of giving birth, and Kearney and Levine (2009) found that income-based Medicaid expansions decrease nonteen births by 2 %. Levine et al. (1996) found that adding Medicaid funding restrictions on abortion leads to about a 3 % reduction in the abortion rate but does not impact birth rates after state trends are added. This result is of particular interest from a policy standpoint because abortion-funding restrictions have been used as a primary policy tool for decreasing abortions. The decrease in the abortion rate is approximately the same under contraceptive insurance mandates when compared with abortion restrictions, which suggests that providing better contraception access may be a more effective policy tool given that a greater share of individuals receive insurance through employers compared with Medicaid.

There are a few limitations to the current study. Although controls for percentage of individuals with employer-based insurance are included in the models, more-detailed insurance data would allow for better identification of the groups affected by mandates. Additionally, data linking contraception use or expenditures with detailed insurance status or coverage information would provide a means of determining what specific channels are driving the results: specifically, whether individuals who had no coverage are obtaining coverage, or whether individuals are gaining coverage of additional contraception types. Although the BRFSS analysis gives some indication that women are using more effective birth control methods, a panel data set with contraception use would allow us to estimate contraception switching behavior. This would provide insight into whether the policy is successful in encouraging women to use more effective contraception methods and consequently having a better chance of avoiding an unintended pregnancy. If mandates do not induce women to use the most effective contraception methods, it might be necessary to pair mandates with a second policy tool (such as increased information about contraception effectiveness) to further decrease unintended pregnancies. Contraceptive insurance mandates have the potential to have impact beyond fertility outcomes. Future research could examine how increased access to contraception through insurance mandates impacts education, marriage, and labor market outcomes, as well as the utilization of prenatal care during pregnancy.

Acknowledgments

I am grateful to the Editor and two anonymous referees for comments that substantially improved this article. I am especially grateful to Jason Abrevaya, Daniel Hamermesh, Sandra Black, and Steve Trejo for their feedback and invaluable guidance. I also thank Mark Hayward, John Smith, and participants at the Labor lunch seminar at UT-Austin for their useful comments and feedback.

Notes

1

Small firms are defined as those with 3–199 employees; large firms have 200 or more employees.

2

The five leading methods are the diaphragm, one- and three-month injectables, IUDs, and oral contraceptives.

3

These estimates include only HMO, PPO, and POS plans, and therefore may not be representative of plans for a significant proportion of individuals.

4

These methods include intrauterine devices (IUDs), injectables, and implants. Because the implant was discontinued in 2002, it was excluded from the final analysis. Copays remained in place for oral contraceptives in an attempt to incentivize switching to contraception methods with the lowest failure rates.

5

The estimated costs are dependent on the type of insurance coverage; therefore, the upper bound of the cost represents the uncovered cost of contraception.

6

Typical usage refers to actual use, which includes both inconsistent and incorrect use.

7

Trussell’s estimates are derived using the 1995 and 2002 National Survey of Family Growth.

8

I am unaware of any specific legislation that might have been responsible for such large changes in the composition of teenage contraception use relative to adults, and determining whether changes in social norms played a role is beyond the scope of this article. The change in composition is mainly driven by a decrease in condom usage coupled with an increase in oral contraceptives and injectable usage.

9

Between 1998 and 2009, 10 % to 16 % of individuals in the United States obtained insurance coverage through Medicaid and are therefore unaffected by mandates (U.S. Census Bureau, 2010).

10

Differences for plans at the local level are used for comparison because plans determined at the national level are more likely to adhere to state mandates. More than one-half (58 %) of insurance companies report determining rates at the national level.

11

More specifically, birth rates are available from 1993–2011, but abortion rates are available only through 2009.

12

For regressions involving age subgroups, the rate is calculated as the number of births per 1,000 women within that subgroup.

13

Specifically, starting in 1998, California and New Hampshire do not report abortion statistics, and are therefore not included in the abortion analysis. Alaska, Louisiana, Oklahoma, and West Virginia do not report in all years, but they report in a majority of the years.

14

Controls include the percentage of the population that has employer insurance, median household income, whether emergency contraception is available over the counter, the unemployment rate, the average number of children per household, the average number of children less than age 5 per household, percentage of the population that is married, percentage of the population that is nonwhite, percentage of the state with high school and college diplomas, the median age of women, and percentage of the population enrolled in Medicaid.

15

This result is similar if an indicator for blue states is included instead of the percentage of the votes in the 1996 presidential election.

16

Results are from linear probability models, but results from probit estimation are similar. The regressions include all controls listed in Table 2.

17

Additionally, following Levine et al. (1996), models were estimated using the “pregnancy rate” as the outcome variable. The pregnancy rate is a constructed measure defined as the sum of the abortion rate and the birth rate, and serves as an approximation to the number of pregnancies per 1,000 women aged 15–44. The results are very similar to results using abortion and birth rates as outcome variables, and are available upon request.

18

The baseline fertility outcomes are calculated as the average in 1997, which was the year prior to the first mandate introduction. The baseline abortion (birth) rate is 17.3 (63.5).

19

The pre-mandate effect for birth rates for the age 20–29 subsample is significant. The effect in the first year following the mandate is statistically different from the pre-mandate effect (p = 0.001).

20

Robustness checks by age subgroup are available upon request.

21

Sterilization procedures and patient education and counseling are also included under the new coverage laws; the law does not include coverage of abortifacient drugs.

22

In an effort to give insurers the flexibility to control costs, certain forms of cost sharing will be allowed. Specifically, if a generic brand is available for certain contraceptives, the insurance company can impose cost sharing for the branded drug.

23

As mentioned in the introduction, the case Burwell v. Hobby Lobby Stores, Inc. muddles the issue of whether religious exemptions will have negligible effects, but this should be studied as more post-ACA data become available.

References

Bitler, M., & Schmidt, L. (
2006
).
Health disparities and infertility: Impacts of state-level insurance mandates
.
Fertility and Sterility
,
85
,
858
865
. 10.1016/j.fertnstert.2005.11.038
Bitler, M., & Schmidt, L. (
2012
).
Utilization of infertility treatments: The effects of insurance mandates
.
Demography
,
49
,
125
149
. 10.1007/s13524-011-0078-4
Culwell, K., & Feinglass, J. (
2007
).
The association of health insurance with use of prescription contraceptives
.
Perspectives on Sexual and Reproductive Health
,
39
,
226
230
. 10.1363/3922607
DeNavas-Walt, C., Proctor, B. D., & Smith, J. C. (
2010
).
Income, poverty, and health insurance coverage in the United States: 2009
(Current Population Reports, No. P60-238).
Washington, DC
:
U.S. Census Bureau
.
Finer, L., & Henshaw, S. (
2006
).
Disparities in the rates of unintended pregnancy in the United States, 1994 and 2001
.
Perspectives on Sexual and Reproductive Health
,
38
,
90
96
. 10.1363/3809006
Gruber, J. (
1994
).
State-mandated benefits and employer-provided health insurance
.
Journal of Public Economics
,
55
,
433
464
. 10.1016/0047-2727(93)01407-2
Hamilton, B., & McManus, B. (
2012
).
The effects of insurance mandates on choices and outcomes in infertility treatment markets
.
Health Economics
,
21
,
994
1016
. 10.1002/hec.1776
Henne, M., & Bundorf, K. (
2008
).
Insurance mandates and trends in infertility treatments
.
Fertility and Sterility
,
89
,
66
73
. 10.1016/j.fertnstert.2007.01.167
Henry J. Kaiser Family Foundation, & Health Research and Educational Trust (HRET)
. (
1999
).
Employer health benefits: 1999 Annual Survey
.
Menlo Park, CA
:
Kaiser Family Foundation
.
Henry J. Kaiser Family Foundation, & Health Research and Educational Trust (HRET)
. (
2003
).
Employer health benefits: 2003 Annual Survey
.
Menlo Park, CA
:
Kaiser Family Foundation
.
Henry J. Kaiser Family Foundation, & Health Research and Educational Trust (HRET)
. (
2010
).
Health insurance coverage in America, 2009
.
Menlo Park, CA
:
Kaiser Family Foundation
.
Jones, R., Darroch, J., & Henshaw, S. (
2002
).
Contraceptive use among women having abortions in 2000–2001
.
Perspectives on Sexual and Reproductive Health
,
34
,
294
303
. 10.2307/3097748
Kaestner, R., & Simon, K. (
2002
).
Labor market consequences of state health insurance regulation
.
Industrial and Labor Relations Review
,
56
,
136
159
. 10.1177/001979390205600108
Kearney, M., & Levine, P. (
2009
).
Subsidized contraception, fertility, and sexual behavior
.
Review of Economics and Statistics
,
9
,
137
151
. 10.1162/rest.91.1.137
Levine, P., Trainor, A., & Zimmerman, D. (
1996
).
The effect of Medicaid abortion funding restrictions on abortions, pregnancies and births
.
Journal of Health Economics.
,
15
,
555
578
. 10.1016/S0167-6296(96)00495-X
Liang, S., Grossman, D., & Phillips, K. (
2011
).
Women’s out-of-pocket expenditures and dispensing patterns for oral contraceptive pills between 1996 and 2006
.
Contraception
,
83
,
528
536
. 10.1016/j.contraception.2010.09.013
Mellor, J. (
1998
).
The effect of family planning programs on the fertility of welfare recipients: Evidence from Medicaid claims
.
Journal of Human Resources
,
33
,
866
895
. 10.2307/146401
Mosher, W., & Jones, J. (
2010
).
Use of contraception in the United States: 1982–2008
. (Vital and Health Statistics Report - Series 23, No. 29).
Hyattsville, MD
:
National Center for Health Statistics
.
National Conference of State Legislatures
. (n.d.).
Insurance coverage for contraception laws
. Retrieved from http://www.ncsl.org/default.aspx?tabid=14384
Planned Parenthood
. (n.d.).
Birth control
. Retrieved from http://www.plannedparenthood.org/health-topics/birth-control-4211.htm
Postlethwaite, D., Trussell, J., Zoolakis, A., Shabear, R., & Petitti, D. (
2007
).
A comparison of contraceptive procurement pre- and post-benefit change
.
Contraception
,
76
,
360
365
. 10.1016/j.contraception.2007.07.006
Schmidt, L. (
2007
).
Effects of infertility treatment insurance mandates on fertility
.
Journal of Health Economics
,
26
,
431
446
. 10.1016/j.jhealeco.2006.10.012
Sonfield, A., Gold, R., Frost, J., & Darroch, J. (
2004
).
U.S. insurance coverage of contraceptives and the impact of contraceptive coverage mandates, 2002
.
Perspectives on Sexual and Reproductive Health
,
36
,
72
79
. 10.1363/3607204
Trussell, J. (
2011
).
Contraceptive failure in the United States
.
Contraception
,
83
,
397
404
. 10.1016/j.contraception.2011.01.021
Wolfers, J. (
2006
).
Did unilateral divorce laws raise divorce rates? A reconciliation and new results
.
American Economic Review
,
96
,
1802
1820
. 10.1257/aer.96.5.1802