Misclassification Bias and the Estimated Effect of Parental Involvement Laws on Adolescents' Reproductive Outcomes

Abstract

Objectives. We evaluated the presence of misclassification bias in the estimated effect of parental involvement laws on minors’ reproductive outcomes when subjection to such laws was measured by age at the time of pregnancy resolution.

Methods. Using data from abortion and birth certificates, we evaluated the effect of Texas's parental notification law on the abortion, birth, and pregnancy rates of adolescents aged 17 years compared with those aged 18 years on the basis of age at the time of pregnancy resolution and age at conception.

Results. On the basis of age at the time of the abortion or birth, the law was associated with a fall of 26%, 7%, and 11% in the abortion, birth, and pregnancy rates, respectively, of 17- relative to 18-year-olds. Based on age at the time of conception, the abortion rate fell 15%, the birth rate rose 2%, and the pregnancy rate remained unchanged.

Conclusions. Previous studies of parental involvement laws should be interpreted with caution because their methodological limitations have resulted in an overestimation of the fall in abortions and underestimation of the rise in births, possibly leading to the erroneous conclusion that pregnancies decline in response to such laws.

Over the past 2 decades, 22 states have passed laws restricting minors’ access to abortion, and 36 states now enforce laws requiring parental involvement in minors’ decisions to obtain abortions.¹ The popularity of such laws stems in part from the belief that these laws reduce abortion and pregnancy rates among adolescents. This belief is supported by previous studies that have reported that after the passage of a parental involvement law, there was a fall in abortion rates among minors that was unaccompanied by a rise in births.^2–6

However, most previous analyses have used the pregnant adolescent's age at the time of the birth or abortion to classify adolescents as either affected or unaffected by the parental involvement law, and this may have biased estimates of the effect of such laws on the reduction in minors’ abortion rates and the lack of increase in their birth rates. These biases have significant implications for policy. If after a parental involvement law, abortion rates fall less than legislators expect and birth rates rise, then the increase in unintended childbearing by minors may dampen enthusiasm for such statutes.

A recent analysis of Texas's parental notification law found that many 17-year-olds who are contemplating having an abortion will, if feasible, wait to do so until their 18th birthday to avoid having to notify their parents.⁷ Some of them are willing to wait even if it leads to a riskier second-trimester abortion. Such behavior introduces a misclassification bias into the estimated impact of the law on minors’ abortion rates.

The canonical research design for the evaluation of parental involvement laws is a pre–post analysis with a comparison group, in which changes in abortion or birth rates among minors are compared with changes among older adolescents or changes among minors in a similar state without a parental involvement law. The delay in the timing of abortion by older 17-year-olds causes a decline in the number of abortions obtained at age 17 years and a rise in the number of abortions obtained at age 18 years, which leads researchers to overestimate the effect of the law on the abortion rate of 17-year-olds, if exposure to the law is determined based on the adolescent's age at the time of abortion rather than age at the time of conception.

Misclassification bias will be more notable if the comparison group is older adolescents in the same state, because the delay of abortions by 17-year-olds will spuriously reduce the number of abortions performed on 17-year-olds and at the same time increase the number of abortions performed on 18-year-olds. If minors in states without a parental involvement law serve as the counterfactual, the bias results from excluding the delayed abortions from the analysis altogether.

The way exposure to the law is determined also affects inferences about the effect of the law on minors’ birth rates. By exposure we mean adolescents are subject to the law on the basis of their age. After the introduction of a parental involvement law, adolescents younger than 18 years are subject (exposed) to the law, regardless of whether or not they are aware of the law before they become pregnant. Adolescents 18 years and older are not subject to the law (are not exposed), because the parental involvement requirement does not apply to them.

Approximately three fourths of all minors who conceive as 17-year-olds give birth when they are 18 years of age. Thus, a pregnant 17-year-old who carries to term because of a parental involvement law will most likely give birth when she is aged 18 years. In all but 1 previous analysis of parental involvement laws, the mother's age has been measured at the time of delivery. Thus, births to 18-year-olds who may have been affected by the law during pregnancy will not be counted if comparisons are between 17-year-olds in different states, or such births will be included among the controls if changes among 18-year-olds within the state serve as the counterfactual. This form of misclassification bias drives estimates of the law's impact on birth rates toward the null and may even lead to the erroneous inference that birth rates have declined or remained unchanged in response to the law.

We sought to evaluate the extent of misclassification bias in estimates of the effect of parental involvement laws if exposure to the law was determined on the basis of the pregnant adolescent's age at the time of abortion or birth rather than her age at the time of conception. We were able to accomplish this objective because of unique data from abortion and birth certificates from Texas, which enacted a parental notification law in January 2000. The Texas data provide precise information that allowed us to determine the adolescent's age at the time of conception as well as her age at the time of pregnancy resolution (abortion or birth). Initially, we followed previous work and used age at the time of abortion or birth to classify adolescents as affected and unaffected by the law. We evaluated the effect of the Texas parental notification law on the abortion, birth, and pregnancy rates of 17-year-olds using 18-year-olds as the comparison group. We then repeated the analysis but used age at conception to define exposure and contrasted the 2 sets of estimates.

METHODS

Data

We obtained all individual abortion and birth certificates from the Texas Department of State Health Services for the years 1998 to 2001. Data on abortions in Texas are noteworthy in that Texas includes information on patients’ exact date of birth on the abortion certificates. With the date of the patient's birth, the exact date of the termination, and the clinician's estimate of gestational age in weeks, we were able to estimate the date of conception as well as the adolescents’ age at conception. We did the same with birth certificates. Thus, we knew the adolescents’ age at 2 points in time, conception and pregnancy resolution, allowing us to evaluate the extent of the misclassification bias if exposure to the law was determined on the basis of the adolescent's age at the time of abortion or birth.

We limited our analyses to 1 year before and year after implementation of the law for 2 reasons. First, reporting of the patients’ age on Texas's abortion certificates was incomplete before 1999; second, an analysis over a longer period could be affected by differences in underlying trends between the treatment and comparison groups.

The reporting of the patient's date of birth on abortion certificates improved over time; by 1999 reporting was relatively complete. About 5.2% of abortion certificates lacked the patient's exact date of birth in the pre-law period, defined as the conception period from August 1, 1998, to July 31, 1999, and about 2.3% in the after-law year, defined as conception year 2000. (For simplicity, hereafter the 12-month period between August 1, 1998, and July 31, 1999, will be referred to as the year 1999.)

By contrast, the mother's date of birth was well recorded on birth certificates; less than 1% lacked this information each year. We excluded all cases in which data on the exact date of the pregnant adolescent's birth were missing.

Statistical Analyses

We calculated abortion and birth rates as the number of induced abortions or live births per 1000 age-specific population. Pregnancy rates are the sum of induced abortions and live births also divided by the age-specific population. To assess the change in the reproductive outcomes of 17-year-olds associated with the law, we subtracted the pre- to postlaw change in the abortion and birth rates among 18-year-olds from the pre- to postlaw change in the abortion and birth rates among 17-year-olds. This approach assumed that the change among 18-year-olds between 1999 and 2000 (the pre- and postperiod) served as the counterfactual, or the change in abortion and birth rates among 17-year-olds that would have been expected had the law not gone into effect.

We calculated the “difference-in-differences” in levels and in natural logarithms, where the difference in logs is equivalent to the relative rate ratio. We used the delta method to compute the standard error of the difference-in-differences in logs.⁸

RESULTS

Table 1 shows the change in the abortion rate of 17-year-olds between 1999, the year before the law was implemented, and 2000, the year afterward, with 18-year-olds as the comparison group. Abortion rates based on the patient's age at the time of abortion and abortion rates based on the patient's age at the time of conception are shown.

TABLE 1.

Abortion Rates for 17- and 18-Year-Olds, by Age at the Time of Abortion Versus Age at the Time of Conception and by Year of Conception: Texas, 1999 and 2000

	1999,a No./1000	2000, No./1000	Difference Between 1999 and 2000	Log Difference	Difference-in-Differences (SE)	Log Difference-in-Differences (SE)
Age at abortion, y
17	17.5	13.1	−4.4	−0.29	−3.53*** (0.72)	−0.256*** (0.037)
18	27.0	26.1	−0.9	−0.03
Age at conception, y
17	18.7	15.3	−3.4	−0.20	−1.91** (0.75)	−0.146*** (0.035)
18	28.3	26.9	−1.5	−0.05

Note. In columns 1 and 2, the rates are defined as the number of abortions per 1000 age-specific female population. Column 4 contains the difference in the natural logarithm of the rates, and column 5 shows the difference-in-differences in levels (e.g., −4.4 − [−0.9]). Column 6 shows the difference-in-differences in logs (e.g., −0.29 − [0.00]).

^aRefers to the period from August 1, 1998, to July 31, 1999.

**P < .05; ***P < .01.

The abortion rate of 17-year-olds fell by 4.4 per 1000 when measured by age at the time of abortion (Table 1, column 3). This represents a decline of 29% as measured in logarithms (Table 1, column 4). The abortion rate of 18-year-olds fell by 0.9 per 1000 or by 3% in relative terms (Table 1, columns 3 and 4).

When we subtracted the change among 18-year-olds from that of 17-year-olds, we obtained the change in the abortion rate of 17-year-olds associated with the law. Thus, Texas's parental notification law was associated with a decline of 3.5 abortions per 1000 or 26% (P < .01) among 17-year-olds as measured by age at the time of abortion (Table 1, columns 5 and 6). The relative decline in the abortion rate of 17-year-olds when age was measured at the time of conception was –1.91 per 1000, or –15% (P < .01; Table 1, columns 5 and 6), which is substantially less than the estimated reduction based on age at the time of the abortion.

Minors may have responded to a parental involvement law in ways that led researchers to overestimate the magnitude of the decline in abortions given how data were collected. This is precisely the result that we show here. The greater decline in the abortion rate of 17-year-olds when exposure was measured by the patient's age at the time of the abortion was because of the delay in the timing of the abortion by older 17-year-olds.

As evidence of this delay, we show the number of abortions by the patient's age in months for 17- and 18-year-olds by her age at the time of abortion (Figure 1). One series shows the number of abortions in which conception occurred in 1999, and the other pertains to abortions conceived in 2000, the first year that the law was in effect.

FIGURE 1.

Number of abortions, by age in year and months at the time of abortion and by year of conception: Texas, 1999 and 2000.

^aRefers to the period August 1, 1998, to July 31, 1999. Source. Data from induced termination of pregnancy records from the Texas Department of State Health Services.

As expected from the results in Table 1, the number of abortions among 17-year-olds was noticeably smaller in 2000 than in 1999. But more importantly, in conception year 2000, there was a noticeable dip in the number of abortions that were obtained at age 17 years and 10 or 11 months and a jump in abortions that were obtained at exactly age 18 years (18 years and 0 months).

The difference in the number of abortions between adolescents aged 17 years and 11 months and adolescents aged 18 years and 0 months at the time of the abortion was 173 in conception year 2000. By contrast, the number of abortions in which conception occurred in 1999 steadily increased with the patient's age. There were 264 abortions to minors who were aged 17 years and 11 months at the time of abortion and 305 abortions to adolescents who were exactly aged 18 years, a change of only 41 abortions compared with 173 in 2000.

Table 2 shows differences in birth rates between 17- and 18-year-olds in the years before and after the law. Birth rates based on the mother's age at the time of delivery and birth rates based on the mother's age at the time of conception are shown. The birth rate of 17-year-olds fell by 4.7 per 1000 between 1999 and 2000, whereas the birth rate of 18-year-olds was unchanged. Thus, the birth rate among 17-year-olds declined by 7% (P < .01) more than the birth rate among 18-year-olds if exposure to the law was based on the mother's age at the time of birth (Table 2, column 6). However, when the mother's age was measured at conception, the changes were reversed. The birth rate among 17-year-olds was nearly unchanged, whereas the birth rate among 18-year-olds declined by 2.9 births per 1000. Thus, in relative terms, the birth rate among 17-year-olds rose by 2% more than that among 18-year-olds, but the change was statistically insignificant (P = .15).

TABLE 2.

Birth Rates for 17- and 18-Year-Olds, by Age at the Time of Birth Versus Age at the Time of Conception, and by Year of Conception: Texas, 1999 and 2000

	1999,a No./1000	2000, No./1000	Difference Between 1999 and 2000	Log Difference	Difference-in-Differences (SE)	Log Difference-in-Differences (SE)
Age at birth, y
17	66.7	62.0	−4.7	−0.07	−4.55*** (1.37)	−0.071*** (0.018)
18	94.1	94.0	−0.1	0.00
Age at conception, y
17	86.0	85.8	−0.1	0.00	2.76* (1.52)	0.023 (0.016)
18	116.8	113.9	−2.9	−0.03

Note. In columns 1 and 2, the rates are defined as the number of births per 1000 age-specific female population. Column 4 contains the difference in the natural logarithm of the rates, and column 5 shows the difference-in-differences in levels (e.g., −4.7 – [−0.1]). Column 6 shows the difference-in-differences in logs (e.g., −0.07 – [−0.03]).

^aRefers to the period from August 1, 1998, to July 31, 1999.

*P < .1; ***P < .01.

Table 3 shows changes in pregnancy rates. Again, associations between the law and changes in pregnancy rates varied dramatically by how exposure to the law was determined. If it was based on the pregnant adolescent's age at the time of the birth or abortion as has been done in previous work, Texas's parental notification law was associated with an 11% (P < .01) decline in the pregnancy rate of 17-year-olds (Table 3, column 6). However, we uncovered no changes in the pregnancy rate of 17-year-olds when exposure to the law was determined on the basis of the pregnant adolescent's age at the time of conception (Table 3, column 6).

TABLE 3.

Pregnancy Rates for 17- and 18-Year-Olds, by Age at the Time of Abortion or Birth Versus Age at the Time of Conception and by Year of Conception: Texas, 1999 and 2000

	1999,a No./1000	2000, No./1000	Difference Between 1999 and 2000	Log Difference	Difference-in-Differences (SE)	Log Difference-in-Differences (SE)
Age at abortion or birth, y
17	84.1	75.1	−9.1	−0.11	−8.08*** (1.52)	−0.106*** (0.016)
18	121.1	120.1	−1.0	−0.01
Age at conception, y
17	104.7	101.2	−3.5	−0.03	0.85 (1.66)	−0.004 (0.015)
18	145.2	140.8	−4.4	−0.03

Note. In columns 1 and 2, the rates are defined as the number of pregnancies per 1000 age-specific female population. Column 4 contains the difference in the natural logarithm of the rates, and column 5 shows the difference-in-differences in levels (e.g., −9.1 – [−1.0]). Column 6 shows the difference-in-differences in logs (e.g., −0.11 – [−0.01]).

^aRefers to the period from August 1, 1998, to July 31, 1999.

***P < .01.

DISCUSSION

Advocates for parental involvement laws often cite social science research that shows that these laws lower rates of both abortions and pregnancies among minors. We show that results from previous studies are likely biased because of the misclassification of exposure to the law on the basis of age. Texas's parental notification law was associated with a fall in the abortion rate but no change in the pregnancy rate when exposure to the law was determined by the age at conception, a finding contrary to a recent analysis of parental involvement laws in all 50 states.⁴

By contrast, when we measured exposure based on the age at the time of abortion or birth, as was done in previous studies, abortion, birth, and pregnancy rates all declined. Our results also suggest that in response to the law, a significant number of older minors delay their abortion until they are aged 18 years. Previous research has been unable to account for such behavior and has likely overestimated the impact of parental involvement laws on minors’ abortion rates.

We focused on the comparison of 17- and 18-year-olds for several reasons. First, many studies have compared changes between minors aged 15 to 17 years and older adolescents aged 18 and 19 years. However, the prelaw abortion and pregnancy rates of older adolescents are 2.5 times greater than those of minors. This implies large differences in sexual activity, education, and psychological maturity. Moreover, the decline in birth rates among minors has been much more rapid than that among older adolescents nationally since 1990, which confounds effects of the law in a pre–post analysis.⁹

In short, using older adolescents as a counterfactual for minors can yield misleading results. Other studies use the experiences of minors in states without parental involvement laws as a counterfactual. However, abortion rates among minors also vary substantially among states, and this variation is likely related to differences in other reproductive policies as well as social and economic conditions.

The second reason to focus on 17-year-olds is that they are the group of minors least likely to involve their parents in their decision to obtain an abortion and thus most likely to be affected by the law. In a survey of minors having an abortion in states without parental involvement laws, 51% of 17-year-olds, 59% of 16-year-olds, and 74% of 15-year-olds said that 1 or both of their parents knew about the abortion.¹⁰ More-recent surveys of minors in family planning clinics revealed that the proportion of parents who knew of the minor's use of family planning services fell as the adolescents’ age increased.^11,12

A significant strength of our study is that there is little evidence that a substantial number of minors went out of state for an abortion. Louisiana and Arkansas, 2 border states, both enforced a parental involvement law in 2000, and Oklahoma's statute was being contested in the courts, which may have generated confusion among minors and made providers cautious about providing abortions to minors without parental involvement.

Most studies of parental involvement laws use data from the 1980s and 1990s, when there were many fewer states that enforced such statutes. Thus, minors were more likely to go out of state for an abortion after the implementation of the law.^13–16 Other researchers had tried to analyze abortions by state of residence using abortion statistics collected by the Guttmacher Institute.⁴ However, researchers at the Guttmacher Institute who constructed these data have warned against the use of the data for analyses of parental involvement laws, because they do not account for minors who leave a state in response to a parental involvement law.¹⁷

The policy implications of our findings are significant. If parental involvement laws are associated with fewer abortions and pregnancies, then, as concluded in 2 recent studies,^4,18 minors may have increased contraceptive use or decreased sexual activity in response to the law. If, however, abortions fall but pregnancies are unchanged as our study indicates, then it is likely that some minors who would have aborted the pregnancy in the absence of the law, instead carry to term, and such births are likely unintended.

Peer Reviewed

Contributors

S. Colman and T. Joyce conceptualized and designed the study, organized and analyzed the data, and wrote the article. R. Kaestner contributed to data analyses and interpretation and to the editing of the article.