“Too much too young”: race, descriptive representation, and heterogeneous policy responses in the case of teenage childbearing

Abstract

Teenage childbearing has long been a discomforting discourse in the USA. Despite the increasing efforts in finding out what policies work to reduce teen birth rates, the literature lacks a sufficient explanation of heterogeneous policy effects on racial and ethnic groups. Focusing on race and ethnicity, we examine how descriptive representation is associated with heterogeneous policy effects for Whites, Blacks, and Latinos. Analyzing data on 50 states from 1990 to 2006, we find that descriptive representation in the healthcare and education workforce conditions the effects of publicly funded family planning services and state-mandated sex education programs. Minority healthcare professionals and teachers both contribute to reduce minority teen birth rates. Moreover, teen birth rates for White students are lower in states with more minority teachers. Our research highlights the importance of incorporating race and representation in policy studies that seek to understand heterogeneous policy effects. Key empirical findings suggest that the lack of diversity of the core policy implementation workforce is a major challenge for reducing teenage childbearing in states with diverse populations.

Keywords:

• race
• representation
• heterogeneous policy response
• teenage childbearing

Acknowledgements

We thank Danielle N. Atkins and Vicky M. Wilkins for their helpful discussion; Andrea Eckelman and Tonia Buia for their research assistance. We also thank the editors and the anonymous reviewers for making valuable suggestions. All errors remain ours.

Supplemental data

Supplemental data for this article can be accessed at http://dx.doi.org/10.1080/09593330.2013.842928

Notes

† An earlier version of this paper was presented at the 69th Midwest Political Science Association Annual Conference in Chicago, Illinois (2010).

1. Data are from Centers for Diseases Control and Prevention (CDC). Teen birth rates are measured by the number of live births to mothers between 10 and 19 years old. Although public health scholars often measure teen birth rates by the number of live births to mothers aged 15–19 years, recent research suggests the need of including the age group of 10–15 years (Phipps and Sowers 2002).

2. This variable covers public spending on contraceptive service. Ideally, we would like to include a measure for contraceptive prevalence among adolescents aged 10–19 years and by race, which captures both publicly and privately funded contraceptive use. CDC's Youth Risk Behavior Surveillance System is the best database to track youth sexual heath behavior with nationally representative samples. However, the CDC data are only available biannually and do not cover all 50 states. Data availability for minority contraceptive usage is particularly poor. Due the data limitation, we do not include contraceptive prevalence in our analysis.

3. The four eligibility items load positively on a single factor, with an eigenvalue of 2.42. The retained factor index ranges from −1.002 to 1.982. A large value means a state has inclusive Medicaid eligibility rules for family planning services, and vice versa.

4. The principle-component factor analysis renders only one significant factor, with an eigenvalue of 3.056.

5. See details in Supporting Information, Figures 1–4. The measure for Latino representation does not capture the percentage of Latino healthcare professionals based on their different ethnic (the country of origin) backgrounds. We use the aggregated measure to match the representation variable with the overall Latino teen birth rates. When disaggregating the individual-level CPS sample by state, year, and country of origin, sample sizes become unreliably small pertaining to most Latino groups by the country of origin. Therefore, we only measure the overall Latino representation, not for each ethnic group within the Latino population.

6. See details in Supporting Information, Table 3.

7. The Guttmacher Institute measure for state abortion rates has missing years: 1993, 1997, 2001, and 2002. CDC documents annul state abortion rates, but the CDC measure is biased (Meier and McFarlane 1994). Hence, we use the Guttmacher Institute abortion measure and interpolate year gaps by taking the means between year t−1 and t+1.

8. Data for income, education, and Christian adherents are drawn from the Census Bureau American Community Survey. Data for the female labor participation rate are drawn from US Bureau of Labor Statistics (see Supporting Information).

9. Alternatively, one can apply the seemingly unrelated regressions (SUR). The SUR specification improves estimation efficiency if the residuals in separately estimated regressions are highly correlated. We found very low cross-equation correlations – a near 0 and insignificant correlation between the White and Black equation, a correlation of 0.23 between the White and Latino equation, and a correlation of 0.27 between the Black and Latino equation. Because the SUR specification does not correct cross-state heterogeneity and serial autocorrelations, we choose to estimate the three models in separate equations and control for the small cross-equation correlations using the peer-group teen birth rates. The SUR specification and separate panel equations reach to similar substantive conclusions, but the separate panel models yielded more conservative results (against the hypotheses) than SUR.

10. Using the Phillips–Perron specification with a first-order lag and the demean specification to remove cross-sectional dependence, we perform tests for panel unit roots (Levin, Lin, and Chu 2002; Maddala and Wu 1999). The chi-square statistics (df = 100) for White, Black, and Latino teen birth rates are 164.031, 293.499, and 159.494, respectively, and are statistically significant. We do not find evidence for panel unit roots.

11. Because Initiative and Medicaid Eligibility are coded based on state dummies, adding state fixed-effects introduces perfect collinearity. The substantive focus of our analysis is to analyze the impact of health policies and representation across space and time. Cross-state differences in health policies and representation are important. State fixed-effects absorb cross-state data variation and makes inferences only based on within-state variation (Zhu 2013). Hence, we do not add state fixed-effects in the model specification.

12. We include many population-related variables, such as education, per capita income, religious population, etc. These variables are associated with troublesome variance inflation factor statistics. Therefore, we take a mean-centering transformation of these variables.

13. In the public health literature, “Latino Paradox” refers to the observation that Latinos have better health outcomes even though their socioeconomic status is lower than their White or Black counterparts. It is a paradoxical association because low socioeconomic status is expected to be associated with poor health outcomes (LaVeist 2005).

14. For each dependent variable, we compare results based on the full specification and the alternative specification only including control variables. For the White equation, adding policy and representation variables increases the adjusted R² from 0.57 to 0.64. For the Black equation, the adjusted R² in the full model is 0.58, and 0.48 in the model only with control variables. We observe a change of R² from 0.32 to 0.40 for the Latino equation. Using joint F-tests to compare the three pairs of models, we find that adding the key policy and representation variables adds more to the explanation of teen birth rates than the control variables. The F(df = 8, 809) statistics for comparing the White, Black, and Latino models are 21.41, 23.55, and 15.66, respectively. They are all statistically significant.