«Abstract This study analyzes how parental investment responds to a low birth weight (LBW) outcome and finds important differences in investment ...»
significant at conventional levels. Each LBW sibling present in the household is estimated to lead to statistically insignificant increases in cognitive stimulation (1% of a SD) and emotional support (8% of a SD). The sign of the estimated effects of siblings’ endowments also suggest reinforcing investment behavior, but I cannot reject the null hypothesis that on average— conditional on a child’s own birth endowment—the number of LBW siblings in the household has no effect on investment in a child.11 The average estimated impacts of child endowments on investment behavior mask a substantial amount of heterogeneity by maternal education and income. Together with the main effects, the estimates shown in Panel B of Table 2 indicate that while low-income and loweducated mothers strongly reinforce a LBW outcome, higher-income and better-educated mothers either reinforce to a lesser extent or compensate. Each year of maternal education and every $10,000 increase in family income raises the response of overall investment in a LBW child by about 1.9% of a SD and 0.3% of a SD, respectively. The small investment differences by income are driven by the cognitive stimulation HOME score, while the larger investment differences by education are driven by the emotional support HOME score.12 The largest estimated investment response to a LBW outcome is between low-educated and high-educated mothers.
The estimated effect of the number of LBW siblings in the home on overall investment is positive, but the strength of this positive investment response decreases in magnitude with education and income—and this pattern appears to be driven by the cognitive stimulation HOME score. In particular, the analysis indicates that mothers with 16 years of education provide 40% of a SD less cognitive stimulation to her child for every LBW sibling in the home at the time of the investment (1.237 + -0.103*16 = -0.411), while mothers with 10 years of education provide 21% of a SD more cognitive stimulation to her child for the same increase in LBW siblings in the home (1.237 + -0.103*10 = 0.207). These results suggest that—conditional on a child’s own endowment—the presence of LBW siblings causes mothers at the low end of the education distribution to shift more cognitive resources toward a child while mothers at the high end shift less resources.
Many studies in the literature do not estimate the impact of siblings’ endowment on investment in a child. In an unreported analysis, I omitted the measure of siblings’ endowments from regression models, which tended to slightly attenuate the estimated impact of a child’s own endowment on investment measures. This suggests that omitting a measure of siblings’ endowments from investment demand regression equations may cause a small bias in the estimated own-endowment effects away from zero.
A possible explanation for the small estimated investment differences by income is measurement error in income, which tends to be exacerbated in specifications that identify parameters using within-family differences. In an alternate specification, I use average family income over all available observations for a child within a family to address the potential attenuation bias due to measurement error in income. The coefficient estimates of the interaction between LBW and average family income are similar in magnitude to those of the interaction between LBW and current family income, suggesting that measurement error in income is not causing a substantial amount of attenuation bias. And to the extent that maternal education is a proxy for “permanent income”, this unreported analysis also suggests that the interactive effects of LBW and maternal education on investment operate through channels in addition to the permanent-income channel.
4.2. Robustness Checks
Before further examination of the heterogeneity in the impact of child endowments by education and income, I conduct three robustness checks.13 First, an important endogeneity concern is that, since birth weight is influenced by decisions of the mother while pregnant, the estimated effects of birth endowments on the HOME scores may reflect correlation between pre-birth and post-birth investments. To reduce the risk of this endogeneity concern—instead of allowing the impact on post-birth investment of pre-birth investment differences to operate through differences in birth weight—I control for a host of prenatal investment measures including a mother’s pre-pregnancy weight, weight gain during pregnancy, and indicator variables for whether a mother smoked, drank, and obtained early prenatal care during pregnancy. A comparison of the results summarized in Panel A of Table 3 with the main results (Panel B of Table 2) reveals that the estimated investment differences by income and education are similar, suggesting that omitting pre-birth investments from the model is not a problem for the analysis.
Second, an ideal analysis would involve comparisons of within-family investments in sameage children. Unfortunately, the sample of twins captured in the C-NLSY is too small for reliable estimates of the parameters of interest. Age-standardized HOME scores and flexible controls for child age alleviate concerns that within-family comparisons of investments in children of different ages may confound age-related and endowment-related factors. To further alleviate this concern, however, I show results from an analysis that includes in the sample only children who were born less than 2 years apart from each other (Panel B of Table 3) and at least 2 years apart (Panel C of Table 3). The results summarized in both panels are similar to the main results (Panel B of Table 2), although the effects are generally less precisely estimated in the smaller latter sample. These exercises suggest that the main analysis adequately accounts for concerns related to child age effects.
Third, one may be concerned that the generalizability of the results may be somewhat limited by the fact that the C-NLSY may not be representative of the current U.S. population of mothers, especially in light of the heterogeneity in parental investment behavior by income and education.
For example, the main analysis includes children born to mothers who made it into the sample through an effort to oversample black, Hispanic, and economically disadvantaged white youth.
The results of an analysis of a sample that excludes these children are presented in Panel D of I also explored the importance of several interactive effects that may drive the heterogeneity in investment responses to a LBW outcome by education and income. First, various aspects of the home environment may affect the ability of some mothers to make compensatory investments. For example, single motherhood or divorce, the lack of a child’s father in the home, and high or tightly spaced fertility could make mothers less able to compensate for their child being born with a LBW, since these children may require more time or resources than NBW children.
Second, a mother’s cognitive and noncognitive skills could influence their investment response to a LBW outcome.
For instance, high-educated mothers may be more productive at investing in poorly endowed children or may be better able to overcome the difficulties associated with caring for such children. In an unreported analysis, I examined these issues by adding interactions between birth endowments and the following variables to the main empirical specification: indicator for whether a mother is married, indicator for whether the child’s father is present in the home, the total number of siblings in the home, birth order, birth spacing, and a mother’s performance on the AFQT, Rotter Locus of Control, Pearlin Mastery, and Rosenberg Self-Esteem Scales. I found no evidence that these factors drive the investment differences I observe by mother’s education and family income.
Table 3, which are similar to the main results (Panel B of Table 2), suggesting that the results are not driven by these oversamples.
4.3. Nonlinear Effects of Education and Income in the Response of Investment to Child Endowments The impacts of education and income on parental investment behavior at the low ends of the education and income distributions may be different from the corresponding impacts at the high ends. For this reason, I relax the assumption of linearity in the effects of maternal education and family income on investment in LBW children relative to NBW children. Table 4 presents estimates of specifications with interactions between the child endowment measures and indicator variables for maternal education at different points of the education distribution (high school degree, some college or more but no 4-year college degree, and 4-year college degree or more) and for quartiles of the family income distribution. Focusing on the effect of a LBW outcome first, there is no evidence of strong nonlinear effects of family income on relative investment in LBW children. The estimates of the interactive effects of LBW and income quartile indicators are small in magnitude and statistically insignificant—as well as jointly insignificant (p-value 0.382)—suggesting that low-income and higher-income mothers invest about the same amount in LBW and NBW children.
In contrast, the results indicate that high school dropouts reinforce a LBW outcome by investing less in their LBW children relative to their NBW children, while higher-educated mothers compensate for a LBW outcome by investing more in their LBW children.14 Much of the heterogeneity in the overall investment response appears to be driven by differences in the amount of emotional support provided to LBW children by education. High school dropouts are estimated to provide about 12% of a SD less emotional support to their LBW children, while higher-educated mothers compensate for a LBW outcome by providing 7-12% of a SD more emotional support to their LBW children. I cannot reject the null hypothesis that the investment responses by the higher-educated mothers to a LBW outcome are equal to each other (p-value = 0. 741).
The patterns for cognitive stimulation are similar. The estimates suggest that high school dropouts provide 6% of a SD less cognitive stimulation to their LBW children whereas mothers with a high school degree or some college provide about the same amount of cognitive stimulation to their LBW and NBW children, but estimates of the main effect and these interactive effects are statistically insignificant. The analysis indicates that there is a more economically important and statistically significant response at the high end of the education distribution. Mothers with at least a 4-year college degree are estimated to provide 16.5% of a SD more cognitive stimulation to their LBW children. The estimated response of cognitive stimulation to LBW children among mothers with at least 4 years of college is marginally significantly different from that of mothers with a high school degree (p-value = 0.059) and of mothers with some college (p-value = 0.057).
In an unreported analysis, I find that the differential investment response to a very low birth weight (VLBW)— defined as a birth weight below 1,500 grams—outcome by education is even larger. I estimate that high school dropouts invest about 17% of a SD less in their VLBW children while mothers with at least 4 years of college invest 16% of a SD more in their VLBW children.