Impacts on Young Children and Their Families Two Years After Enrollment: Summary Report. Findings: Linking Impacts on Children to Impacts on Families

02/06/2004

Linking Impacts on Children to Impacts on Families

Through what pathways do children appear to have been affected by their mothers' assignment to a JOBS welfare-to-work program? In this section, we summarize results from statistical analyses that identify which program impacts on targeted and non-targeted outcomes appear to underlie selected impacts on children. The analyses of program impacts on children summarized above are experimental in nature. Experimental analyses provide strong causal evidence regarding the existence of program impacts on children; however, they cannot address the question of how these impacts came about. Analyses seeking to identify the pathways through which a program had its impacts are necessarily non-experimental. Because mediational analyses are non-experimental, they do not allow firm causal inferences to be made regarding the pathways through which these impacts came about.(15) In addition, the specific mediational analyses conducted in this study rely on information about mediators and child outcomes measured contemporaneously and used a relatively modest statistical approach in modeling pathways; as such, results should be considered preliminary.

To what extent were program impacts on measures of focal children's cognitive school readiness, externalizing behavior problems, and general health explained by program impacts on mothers (i.e., on targeted and non-targeted outcomes)?

In a modest attempt to understand the pathways through which specific welfare-to-work programs may affect children's developmental outcomes, five of the aggregate impacts on focal children in the Child Outcomes Study sample were examined in more detail through mediational analyses. These five impacts were selected because they illustrate the general pattern of findings at the aggregate level of favorable cognitive, unfavorable health, and mixed behavioral impacts. In particular, we examine:

A favorable impact on a measure of cognitive development:

  1. the favorable impact of Atlanta's labor force attachment program on focal children's mean cognitive school readiness scores;

Two unfavorable impacts on measures of children's health:

  1. the unfavorable impact of Riverside's human capital development program on ratings of the focal child's overall health;
  2. the unfavorable impact of Riverside's labor force attachment program on ratings of focal children's overall health;

A favorable and an unfavorable impact on behavioral and emotional adjustment:

  1. the favorable impact of Atlanta's labor force attachment program on focal children's reported externalizing behavior problems; and
  2. the unfavorable impact of Grand Rapids' labor force attachment program on focal children's reported externalizing behavior problems.

Mediational analyses are statistical analyses that attempt to identify one or more variables that appear to explain, statistically, the relationship between two other variables (see Baron and Kenny, 1986). In the present study, targeted and non-targeted outcomes that were affected by a given JOBS program (i.e., there was an experimental impact on the outcome) were examined as possible "mediators" of the program impact on children. The mediational analyses ask whether the statistical significance of the association between the JOBS program and a child outcome (the significant impact) is diminished or eliminated when the mediating variable, or set of mediating variables, is taken into account. When this happens, there is evidence that the mediator is a conduit through which the child impact is coming about. In these analyses, we find that some of the child impacts are "fully mediated" (the statistical significance of the impact on a child outcome is eliminated when considering the role of the mediators), while in other instances there was only "partial mediation" (the statistical significance of the impact on the child outcome is diminished but not eliminated when considering the mediators).

It is also possible that a mediator operates to increase (instead of decrease) the statistical significance (and, thus, the magnitude of the experimental impact) of a given JOBS program on a child outcome. In this case, the mediator is not one that helps to explain the experimental impact on children. Instead, this mediator is operating in an opposing direction, indicating that the program impact on the given child outcome would have been even more pronounced if not for this variable's buffering, or offsetting, effect.

Mediational analyses first examined for which of the 60targeted and non-targeted adult and family outcomes studied in the COS there were statistically significant impacts of any of the six JOBS programs studied (see Chapter 9, Table 9.1). Next, for each particular analysis (i.e., focusing on one of the selected impacts on children noted above), the set of variables that was included in each specific model was then narrowed to those adult and family outcomes on which there were statistically significant impacts in the program we were focusing on. Of all the adult and family outcomes that emerged as mediators in any of these five analyses (see Chapter 10, Table 10.1), three sets of adult variables consistently emerged as important in explaining child impact findings: variables related to the mothers' employment, to the mothers' parenting, and to maternal psychological well-being.

Cognitive development. The favorable impact of Atlanta's labor force attachment program on focal children's mean cognitive school readiness scores appears to be related to program mothers' greater employment and improvements in parenting.

Health. The unfavorable impact of Riverside's human capital development program on mothers' ratings of focal children's overall health appears to be related to program mothers' increased feeling of time stress. However, program impacts would have been even more unfavorable in the ratings of children's health if program mothers were not also more likely to have been sanctioned, to have participated in basic education, and obtained a high school diploma or GED. (Perhaps these factors  being sanctioned, participating in basic education, and obtaining a high school diploma or GED  go together, indicating a program in which there was both more encouragement and pressure from case workers to participate, and indeed resulting in more participation and educational progress).(16)

The unfavorable impact of Riverside's labor force attachment program on ratings of focal children's overall health appears to be related to decreases in AFDC receipt and increases in mothers' work hours, although the picture remains somewhat unclear.(17) One possibility that would require further examination is that the loss of cash welfare benefits led to a loss of Medicaid coverage which, in turn, if coverage was not replaced by employer-provided coverage or other types of coverage, led to a decline in children's overall health ratings.

Behavioral and emotional adjustment. The favorable impact of Atlanta's labor force attachment program on focal children's reported externalizing behavior problems appears to be related to the program's favorable impact on parenting, despite also increasing mothers' reports of time stress and perceptions of feeling "pushed" by the welfare office.

The unfavorable impact of Grand Rapids' labor force attachment program on focal children's reported externalizing behavior problems appears to be related to the program's unfavorable impacts on maternal depressive symptomatology and parenting.

What have we learned about possible pathways through which various welfare-to-work programs can affect children?

Several themes emerge from these results. First, the mechanisms through which children can be affected by a given welfare-to-work program include both targeted outcomes (e.g., employment, AFDC receipt) as well as non-targeted outcomes (maternal psychological well-being and parenting). In particular, this study highlights the role played by intervening mechanisms more proximal to the child in the home environment  namely, maternal psychological well-being and parenting. Aspects of family life targeted by welfare-to-work programs emerged less consistently as mediators of the five child impacts selected for mediational analyses. It may be, however, that these targeted outcomes do play a role inasmuch as program impacts on these outcomes activate processes more proximal to the child. For example, changes in employment or earnings may not, in and of themselves, consistently lead to program impacts on children's behavior problems unless these lead to changes in parenting or depressive symptomatology. Testing such multi-step hypotheses with statistical methods that can model such hypotheses is warranted as a next step.

It is worth noting that, because none of the six JOBS programs studied here had aggregate impacts on total household income, this impact could not serve as a pathway through which any of the aggregate impacts on children (not just the five impacts selected for further study) came about.

In addition, it is also noteworthy that  despite pervasive program impacts on child care and more circumscribed program impacts on health insurance coverage - these impacts did not help to explain either the favorable or the unfavorable child impacts examined here.However, this does not mean that child care and health insurance coverage are not important to children's outcomes - only that these variables were not found to explain the particular impacts examined here. The reader should keep in mind that we examined the relation between mediators and child outcomes in an experimental context, that is, only in cases where there was an experimental impact both on a mediator and on a child outcome. This addresses one specific type of relation between a mediating variable and a child outcome, namely, whether a given mediating variable  such as child care  plays a role in helping to explain a particular child impact finding. However, there are likely to be associations between child care and child outcomes in this sample even where there were no program impacts on child care and/or no impacts on children. Indeed, findings from the experimental literature indicate that quality preschool experiences can have lasting benefits for low-income children (Barnett, 1995), and findings from the non-experimental literature show that formal child care arrangements bode well for low-income children's developmental outcomes (Zaslow, McGroder, Moore, and LeMenestrel, 1999; Zaslow, Oldham, Magenheim, and Moore, 1998). For example, a recent set of analyses shows that among control group families in the Riverside site of the Child Outcomes Study, five- to seven-year-old children identified to be at risk in terms of cognitive school readiness (because they were two or more years behind in basic concepts they should have mastered by school entry) were less likely to be currently enrolled in a formal child care arrangement (Zaslow, McGroder, Moore, and LeMenestrel, 1999). In short, because we restricted our focus to instances in which there was an impact on both a child outcome measure and a mediator, our analyses do not address the broader question of whether these mediators are important to children's outcomes, in general.

A second theme relates to the size of the mediating effects. Mediational analyses show that even relatively "small"(18) impacts on targeted outcomes (e.g., the favorable impact of Atlanta's labor force attachment program on mothers' current employment) and non-targeted outcomes (e.g., the favorable impact of Atlanta's labor force attachment program on measures of parenting; the unfavorable impact of Grand Rapids' labor force attachment program on a measure of parenting) can help to explain relatively small program impacts on children. Thus, program impacts on mothers need not be "large" to translate into impacts on children.

A third theme to emerge from these findings is that some welfare-to-work programs have effects in opposing directions on aspects of family life that are important to children. For example, Atlanta's labor force attachment program increased mothers' feelings of time stress (which was associated with more frequent externalizing behavior problems) but, at the same time, this program also led to improvements in parenting (which was associated with less frequent externalizing behavior problems). The net effect of these opposing influences for focal children's externalizing behavior problems was favorable in this program.

These findings suggest that, in modeling pathways through which children can be affected by welfare-to-work programs, it is imperative to consider both positive and negative pathways, regardless of whether one is trying to explain a favorable child impact or an unfavorable child impact. Influences of mediating variables going in contrasting directions may help explain the small number and size of significant impacts on child outcomes: counteracting influences may, at times, result in no or little "net" influence on children. Understanding such influences is central to strengthening pathways that yield favorable impacts on children.

As noted above, because the present analyses rely on information about mediators and child outcomes measured contemporaneously, results should be considered preliminary. Additional waves of data from the five-year follow-up, as well as statistical methods that more effectively partition effects among multiple mediators, more completely control for selection effects, and allow alternative models to be tested explicitly (e.g., structural equation modeling), are needed to provide more definitive answers regarding the multiple pathways through which program impacts on children came about. Nevertheless, given the limited state of knowledge on the pathways through which welfare-to-work programs can affect children and families, it is important to begin to address this issue, statistically, even with contemporaneous measures of mediators and child outcomes and using a relatively modest statistical approach. We expect that the mediational results reported here begin to shed light on this important topic, and will serve as the basis for model-building and testing explicit hypotheses in the future, especially when five-year data become available.