Of primary interest, we attach measures of income inequality in the respondent’s state of residence.
17
We created these measures using microdata from the 1980, 1990, and 2000 Censuses along with the 2006–08 American Community Surveys on house-
hold income. These data are available from IPUMS- USA Ruggles 2010. Using these data, we estimated income cutoffs at the 10
th
and 50
th
percentiles for each state and survey year and then generate ratios of these measures like the 50 10 ratio as our
baseline measure of inequality. We then compute the average within states to generate our measure of long- term lower- tail inequality.
18
V. Results
A. Analysis of NSFG Data
Before presenting our formal econometric results, we conduct a descriptive analysis of the NSFG data that is comparable in spirit to our regression models. We compare
fertility outcomes across three categories of states—high, medium, and low inequality states—for women in three SES categories: women whose mothers were high school
dropouts, high school graduates, and those who had any college education. This com- parison is essentially a stylized version of the interaction of SES and state inequality
measures, previewing the regressions to follow. Figure 5 presents these unadjusted
17. We have considered whether the state is the right level of aggregation for this analysis. First, there are conceptual issues. A potentially important consequence of high levels of income inequality is greater resi-
dential and institutional segregation. To focus on a smaller geographic unit would thus lose this dimension of inequality. This leads us to prefer the state to something smaller. However, in very populous diverse states,
such as Texas or California, the state might be too disparate a measure, and the appropriate geographical unit might be something smaller. Mayer’s 2001 study of education outcomes and state- level income inequality
also grapples with the issue of what is the appropriate unit of aggregation in a study of the consequences of income inequality, and her paper includes a thoughtful discussion of this issue. Second, in terms of data and
empirical methodology, there are challenges to using a geographic unit smaller than the state. One empirical concern is mobility. The geographic identifi er in the NSFG is the location of current residence and mobility
is a much greater concern at the local level than at a broader level. A second data concern is limited sample size. The only substate identifi er available even in restricted NSFG data is the county of residence. This limits
the sample size available for our analysis because public use census data only identifi es county of residence for 40 percent of the population. Nevertheless, we have conducted a county- level analysis identical in form
to that reported here using the subset of NSFG respondents who live in counties for which Census income data are available. The pattern of results are similar than those reported here, but smaller in size. Rather than
a coeffi cient of 0.053 0.015 in Column 1 of Table 1, the comparable coeffi cient with county data is 0.031 0.012. This is consistent with the attenuation bias that would result from cross- county migration. It is also
consistent with the idea that some of the effects of inequality are not captured when one looks at a smaller level of geography.
18. The full list of states with associated measure of average household income inequality is provided in Table 1 of the earlier NBER working paper version of this paper, Kearney and Levine 2011. The highest
inequality states are DC, LA, NY, AL, MA, MS, GA, KY, SC, RI, TN, IL, and TX, in descending order. The 50 10 ratio is 5.88 in DC a clear outlier, 4.92 in LA, and 4.30 in TX. The lowest quartile of states ranked
by income inequality includes AZ, ME, AK, MT, IA, NE, WI, WY, VT, ID, NH, NV, and UT, in descending order. The 50 10 ratio is 3.81 in AZ and 3.41 in UT. The 50 10 ratio is 3.81 in AZ and 3.41 in UT. We classify
the remaining 25 states as “middle- inequality” states when we break things out in this way. As a rough gauge, moving from a low- inequality state to a high- inequality state increases the 50 10 ratio by around one. We
use this number below when interpreting the magnitude of our results.
differences. They reveal that high SES women defi ned as having a college- educated mother exhibit little variation in early nonmarital childbearing across states that dif-
fer in their level of income inequality. Low SES women defi ned as having a mother who dropped out of high school are more likely to give birth to a child at a young age
outside of marriage if they live in a state with a larger lower- tail income gap. Moving from a low inequality state to a high- inequality state which represents roughly a one
point increase in the 50 10 ratio appears to increase the rate of nonmarital childbear- ing by age 20 by around fi ve percentage points.
We conduct a similar exercise for conception, pregnancy failure, and shotgun mar- riage. These comparisons show no evidence of differential patterns by SES by inequal-
ity ranking for conception or shotgun marriage by age 20. In contrast, we show in Figure 6 that rates of pregnancy failure are a little over four percentage points lower
among low SES women in high inequality states, as compared to low SES women in low inequality states. This means that the differences in early nonmarital childbearing
can be attributed almost entirely to differences in the rate of pregnancy failure likely due to differences in the use of abortion. We proceed to a more rigorous examination
of these comparisons using the econometric model described above.
Table 1 reports the main regression estimates. The fi rst column in the upper panel of this table explores nonmarital births by age 20 as the outcome. The interaction
between the 50 10 ratio and having a high school dropout mother represents β
1
in our econometric model. It indicates that low SES girls in a state with a higher lower- tail
income gap are more likely to have a nonmarital birth by age 20, all else equal. As
Figure 5 Rate of Nonmarital Childbearing by Age 20, by Mother’s Level of Education and
State Level of Income Inequality
we saw earlier, a one point change in the 50 10 ratio roughly captures the movement from a low inequality state to a high inequality state. In this case, that movement is
predicted to increase early nonmarital childbearing by 5.3 percentage points for those women whose mothers were high school dropouts. This point estimate is very similar
to the unadjusted difference that we observed in Figure 5, suggesting that the other covariates in the model have very little correlation with the interaction term of interest
and or early nonmarital childbearing.
The remainder of the top panel of the table focuses on other nonmarital outcomes and all marital outcomes by age 20. In Column 3, we see that much of the reason
why nonmarital childbearing among low SES women rises with inequality is that abortion rates, as captured by pregnancy failures, fall. The magnitude of this estimate
indicates that moving from a low- inequality state to a high- inequality state reduces the likelihood of a pregnancy failure by 4.2 percentage points. We cannot statistically
distinguish this estimate from the 5.3 percentage point increase in early nonmarital childbearing. This suggests that the rise in early nonmarital births associated with
greater inequality is mainly attributable to fewer abortions.
Column 2 provides no evidence that the likelihood of conception is affected. We fi nd little support for an impact of inequality on marital fertility. In Column 7, we
focus on shotgun marriage. We fi nd a statistically signifi cant reduction in the likeli- hood of shotgun marriage among low- SES women as lower- tail income inequality
increases.
Interestingly, when we focus on moderate SES women as captured by daughters of
Figure 6 Rate of Nonmarital Pregnancy Failure by Age 20, by Mother’s Level of Education
and State Level of Income Inequality
T he
J ourna
l of H um
an Re sourc
es 16
Table 1 Impact of Long- Term Inequality on Marital and Nonmarital Fertility Outcomes by Ages 20 and 25, by Socioeconomic Status
Nonmarital Outcomes Marital Outcomes
Shotgun Marriage
Birth 1
Conception 2
Pregnancy Failure
3 Birth
4 Conception
5 Pregnancy
Failure 6
7 By Age 20
50 10 ratio 0.053
–0.006 –0.042
–0.026 –0.006
0.003 –0.018
Mom high school dropout 0.015
0.018 0.015
0.017 0.016
0.004 0.009
50 10 ratio 0.021
–0.013 –0.013
–0.027 –0.004
0.002 –0.013
Mom high school graduate 0.012
0.018 0.015
0.010 0.006
0.003 0.007
By Age 25 50 10 ratio
0.040 –0.039
–0.022 –0.041
0.008 –0.010
–0.050 Mom high school dropout
0.013 0.026
0.025 0.025
0.029 0.006
0.014 50 10 ratio
0.003 –0.049
–0.020 –0.026
0.002 –0.003
–0.021 Mom high school graduate
0.014 0.018
0.017 0.012
0.016 0.004
0.010
Notes: reported standard errors are adjusted for clustering at the state level. Additional explanatory variables in each regression include maternal educational attainment, current age and age squared, race ethnicity, an indicator variable for living with a single parent at age 14, the state unemployment rate at age 19, state welfare policies
family cap and maximum AFDC TANF benefi t for a family of three, state abortion policies Medicaid funding, parental notifi cation consent, and mandatory delay laws, and an indicator variable for SCHIP implementation, along with state and cohort fi xed effects. The sample sizes are 24,720 and 23,037 in the models by age 20 and age 25,
respectively.
high school graduates, we fi nd that the point estimate for the increase in nonmarital births is attenuated, and now there is a statistically signifi cant reduction in marital
births, of nearly the same opposite- signed magnitude.
19
Presumably these women are at some, albeit reduced, risk of poor economic outcomes that may be exaggerated
in high inequality states. For them, greater inequality is associated with fewer marital births. This suggests that a reduced prevalence of shotgun marriage is the pathway,
as confi rmed in Column 7. These results are similar to the results for women in their young 20s, described below.
The lower panel of Table 1 replicates this analysis, focusing on childbearing mari- tal outcomes by age 25 rather than age 20. Including somewhat older young adults
increases the relevance of shotgun marriages in these fi ndings. We continue to fi nd that nonmarital childbearing increases for low SES women when they live in high-
inequality states, but we fi nd a similar drop in marital childbearing although the latter is not quite signifi cant. Changes in the likelihood of a shotgun marriage explain the
divergent pattern.
20
B. An Investigation of Alternative Mechanisms