Understanding Associations Between Low-Income Mothers' Participation in Education and Parenting

ABSTRACT

Maternal education is one of the strongest predictors of children's academic outcomes. One possible explanation for this is that more highly educated mothers more frequently engage in parenting practices that may promote children's later cognitive development; however, most of this evidence is correlational. This study uses Head Start Impact Study data (N = 1,953) to explore whether low-income mothers' participation in education affects their parenting practices and beliefs. Principal scores, which predict maternal educational participation based on covariates, were used for analysis. Principal score matching was used to identify mothers who we predicted participated in education because their children were randomly assigned to Head Start. We compared these mothers' outcomes to those of mothers we predicted would have participated in education if they were assigned to Head Start. For these mothers, participation in maternal education was associated with children watching fewer hours of TV, having more types of printed media at home, and more frequent participation in cultural activities, but it was not associated with a host of other parenting outcomes. Changing parenting is one potential pathway by which maternal educational participation may influence children's later academic outcomes.

KEYWORDS:

• parenting
• maternal education
• low-income
• intergenerational
• parenting practices

Funding

The research reported here was supported by the American Psychological Foundation Elizabeth Munsterberg Koppitz Graduate Student Fellowship and the New York University Predoctoral Interdisciplinary Research Training Fellowship. The opinions expressed are those of the authors.

Notes

¹ The small proportion of control families who received Head Start at the center they applied to (5.6%) and the treatment families who did not receive Head Start at the center they applied to (7.4%) were included within their originally assigned groups to retain the intent-to-treat component of the analysis.

² Analyses were also conducted in the sample with complete information available (N = 1,282). Although some of the coefficients were of a similar magnitude to the primary results, most of the results were not statistically significant (see Appendix A in the online supplemental material). This is likely because the sample of predicted compliers was 66% of the size of the imputed sample (N = 96–104).

³ This analysis assumes there are no “defiers”—mothers who would not take up education when assigned to Head Start, but who would take up education when assigned to the control group. It is difficult to think of a situation where random assignment would cause defiance systematically, although there are plausible individual cases where this could occur.

⁴ Although language from instrumental variables (IV) analysis was used to describe the different groups of mothers, IV was not used because this case does not meet the exclusion restriction, which assumes that random assignment to Head Start only affects parenting through maternal education (Moulton, Peck, & Bell, 2014). There are a number of other ways that Head Start could affect parenting given that many Head Start centers offer parents' workshops and opportunities to be involved. Moreover, the bias introduced by not meeting the exclusion restriction is larger when there are low odds that someone is a complier, which is a concern in the present case because the effect of random assignment on participation in maternal education is small (i.e., the first-stage F statistic was less than 10). IV analyses with site-by-treatment interactions have been used to identify multiple mediators when the exclusion restriction is implausible, but these models introduce additional strong assumptions (Reardon & Raudenbush, 2013).

⁵ One criticism of principal-score approaches is that the model that generates the principal scores may provide a better fit among the group on which it was modeled (Hill, Weiss, & Zhai, 2011; Moulton et al., 2014; Peck, 2003). To address this concern, a similar method to the current approach, the Analysis of Symmetrically Predicted Endogenous Subgroups method uses a modeling sample or a cross-validation approach to predict subgroup membership (Moulton et al., 2014; Peck, 2003). We do not use these techniques in the current application because of the complexity of trying to identify multiple subgroups. As such, we use observed educational stratum membership when it is available; we compare observed control always takers to predicted treatment always takers and observed treatment never takers to predicted control never takers. However, membership in our key comparison group—compliers—was predicted for both the treatment and control groups, lessening concerns about mismatch in the accuracy of subgroup prediction.

⁶ The matching methods in Step 1c (caliper matching, without replacement) and Step 2c (one-to-one matching, with replacement) were selected in order to create proportionally sized groups across the Head Start treatment and control conditions. Other matching and adjustment techniques were used, including inverse probability of treatment weighting in Step 2c, and results were consistent with the primary analyses (results available from the first author). See Stuart (2010) for a description of matching methods.

⁷ Interestingly, mothers in the Head Start treatment group who had low probabilities of educational participation were more likely to participate in education than similar mothers in the control group. It may be that these are the mothers who are being induced into education because of the opportunities provided by random assignment to Head Start. Alternatively, the prediction model may provide a better fit for the control group given that it was modeled on this group (Peck, 2003). Unfortunately, we are unable to adjudicate between these possibilities, but it seems plausible that mothers with a lower predicted probability of educational participation are those who most need the additional supports provided by Head Start to participate in education.

⁸ The primary results do not control for baseline parenting because baseline interviews occurred between 1–5 months after random assignment to Head Start and random assignment was significantly associated with many of the “baseline” parenting variables. Additional principal-score matched regression analyses were conducted controlling for the parenting measures available at baseline (frequency of cognitively stimulating activities, types of printed media in the home, frequency of participation in cultural activities, positive parenting, and perceived support). Results for the compliers showed the same pattern of significant results as the primary analyses, except that the result for types of printed media was no longer statistically significant (results available from the first author).

⁹ The multiply imputed results for the never and always takers groups were almost exactly the same as the single imputation estimates presented in Table 4, so they are not presented (results available from the first author).

¹⁰ Although treatment families could still participate in Head Start in year two, a far smaller proportion of the sample did participate (66%), and this proportion was more similar to the proportion of control group families who attended Head Start who could also experience the potential benefits to parenting (49%).