Empirical Framework Manajemen | Fakultas Ekonomi Universitas Maritim Raja Ali Haji 1.full

More recently, Edin and Kefalas 2005 contributed an infl uential ethnographic account of nonmarital childbearing among poor women. They make the following observation: “. . . the extreme loneliness, the struggles with parents and peers, the wild behavior, the depression and despair, the school failure, the drugs, and the general sense that life has spun completely out of control. Into this void comes a pregnancy and then a baby, bringing the purpose, the validation, the companionship, and the or- der that young women feel have been so sorely lacking. In some profound sense, these young women believe a baby has the power to solve everything” p. 10. Our reading of these seminal and infl uential works is that they fi nd common ground in the notion that growing up in an environment where there is little chance of social and economic advancement leads young women to bear children outside of marriage. These women perceive that they have so little chance for success in life that they see no reason to postpone having a child and may even benefi t from having one, regardless of marital status. 11

III. Empirical Framework

The social science literature reviewed above emphasizes the role of economic hopelessness and marginalization in driving the childbearing and marriage decisions of economically disadvantaged young women. We consider this to be po- tentially important in explaining why some places have so much higher rates of early nonmarital childbearing than others. Disadvantaged individuals who live in locations with more discouraging economic conditions will be more likely to have early non- marital births. We focus our attention on one potential persistent economic condition: lower- tail income inequality. Income inequality tends to be persistent within a place. Using data from the IPUMS Censuses described below we fi nd that the correlation in the 50 10 ratio between the 1980 and 2000 Census years averages 0.74 across states. 12 This indicates that states that tend to have relatively high levels of inequality remain the high- inequality states, and likewise for low inequality states. In addition, income in- equality shows a strong correlation with rates of early nonmarital childbearing. 13 11. Many of these early works focus directly on the issue of race. Now that rates of early nonmarital child- bearing are high for all women albeit still higher for black women, this is less an issue of race today. 12. While overall inequality as measured by the 90 10 ratio of income, for example has increased dramati- cally in recent decades, this increase is driven by upper- tail income inequality, as in the gap between the 90th and the 50 th . See Autor, Katz, and Kearney 2008. Lower- tail income inequality has generally plateaued or compressed since the mid- 1980s. Therefore, we would not necessarily expect changes in lower- tail income inequality over time to be of fi rst- order relevance to the decline in teen childbearing over the past 30 years. That said, we have estimated regressions using pooled data from the 1980, 1990, and 2000 Census data fi le and the 2005–09 American Community Survey that examine the relationship between individual level teen childbearing outcomes and state- year level income inequality, conditional on state fi xed effects. The results suggest that greater levels of lower- tail income inequality are associated with higher rates of teen childbear- ing among low SES girls. In other words, the cross- sectional story we see in the data in this paper seems to carry over to within- state changes in the 50 10 ratio of income. Results are available upon request. 13. As noted earlier, in their infl uential book, Wilkinson and Pickett 2009 call attention to the correlation between inequality and a broad range of “social ills,” including increased crime, drug use, mortality, and teen childbearing, among others. Wilkinson and Pickett 2009 focus on correlational relationships, which are merely the starting point of our empirical investigation. Furthermore, the robustness of their correlational An important step in establishing causation is to determine that inequality oper- ates on those who are most likely to be negatively impacted by it. With this in mind, our main empirical test is based on determining whether low SES women in places with a larger lower- tail income gap are more likely to have an early nonmarital birth compared to higher SES women in those locations. We estimate regression models for a series of fertility outcomes birth, conception, etc. controlling for year- of- birth fi xed effects and state fi xed effects. We also control for individual level demographics, public policy variables, and time- varying labor market conditions in these models. The key variable in our models is the interaction between long- term measures of inequality and a woman’s socioeconomic status. We are interested in explaining per- sistent differences in rates of early nonmarital childbearing across places. We therefore focus our analysis on “fi xed” characteristics of environments. In terms of inequality, this means we empirically consider variation in long- term averages in state and na- tional measures of inequality. More formally, we estimate regression models for outcomes by age 20 of the form: 1 Outcome isc = β + β 1 I s ⋅ LS isc + β 2 I s ⋅ MS isc + β 3 LS isc + β 4 MS isc +β 5 X isc + β 5 E sc + γ s + γ c + ε isc where I is our measure of inequality, LS and MS are indicators of low and middle SES, respectively, and the interaction terms are the main regressors of interest. They represent the differential response of low and middle SES teens to inequality relative to high SES teens. The subscripts i and s index individuals and states, and c indexes birth cohorts. The terms γ s and γ c represent state and birth cohort fi xed effects, respec- tively. The vector X consists of additional personal demographic characteristics—age, age squared, race ethnicity, and an indicator for living with a single parent at age 14. The vector E captures environmental factors including relevant public policies and labor market conditions in the state- year: the unemployment rate, an indicator for a welfare family cap, the maximum welfare benefi t for a family of three, an indicator for SCHIP implementation, an indicator for whether the state Medicaid program covers abortion, an indicator for whether state abortion regulations include parental notifi ca- tion or mandatory delay periods, and whether the state Medicaid program includes expansion policies for family planning services. See Kearney and Levine, 2009a for a discussion of these expansion policies. By including all of these individual and state level controls in the model, our estimated effect of inequality for low- SES women is net of effects driven by policies that might be correlated with inequality. 14 Our primary question of interest is whether β 1 is positive: Are low- SES women in high inequality states relatively more likely to have a nonmarital birth by age 20? We consider the multiple channels through which a difference in birth rates could be real- ized. First, an individual can take actions with regard to sexual behavior and contra- ceptive practices; low SES women in more unequal places may be more likely to get pregnant. Second, a nonmarital childbirth could be avoided through the choice to end relationships has been called into question in other contexts. See Deaton and Lubotsky 2003 on the relation- ship between inequality and mortality, for example. 14. The data sources used to create this set of environmental variables are described in detail in the data ap- pendix of the earlier NBER working paper version of this manuscript Kearney and Levine 2011. a pregnancy; low SES women in more unequal places may be relatively less likely to choose an abortion to end her pregnancy. And fi nally, nonmarital births depend upon the parents choosing to remain unmarried after a pregnancy occurs; low SES women in more unequal places may be relatively less likely to get married before the birth through a so- called “shotgun” marriage. One limitation of this modeling approach is that any other characteristic of the socioeconomic environment that might be correlated with inequality and also directly related to nonmarital early childbearing propensities among low- SES women will be captured by our long- run inequality SES interaction term. Although it is impossible to completely rule out this form of omitted variable bias, we examine whether introduc- ing additional interactions of SES with other potentially troublesome characteristics change our results. More formally, we estimate “horse race” models that take the form: 2 Outcome isc = β + β 1 I s ⋅ LS isc + β 2 I s ⋅ MS isc + β 3 A s ⋅ LS isc + β 4 A s ⋅ MS isc +β 5 LS isc + β 6 MS isc + β 7 X isc + β 8 E sc + γ s + γ c + ε isc This specifi cation is largely the same as that in Equation 1 except that we now include alternative factors A that are time invariant within states and we interact them with an indicator for low SES. We consider two categories of alternatives. The fi rst includes other features of the income and wage distribution. The second includes other social measures that are more directly focused on identifying alternative explanations—political composition of the state, a measure of religiosity of the state, the percentage of the population that is minority, the incarceration rate, poverty rate, and Putnam’s Social Capital Index. We describe all of the alternatives we consider in more detail when we report the results of our analysis.

IV. Data Description