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

02/06/2004

Development and Well-Being of Control Group Children,
and Program Impacts on Children

We first describe findings at the two-year point for children in the control groups of the Child Outcomes Study, as a background for understanding child impact results that follow.

A.  Children in the Control Groups

How were children in the control groups of the three study sites faring at the time of the two-year follow-up?

In the absence of JOBS program influences, the children in the sample were at risk of poor developmental outcomes, especially in the area of cognitive development.

For example, as can be seen in Figure SR-2, children in the control groups were significantly less cognitively ready for school than a national sample of children of the same age, as measured by standardized scores on an assessment of cognitive school readiness, the School Readiness Composite of the Bracken Basic Concept Scale (Bracken, 1984).

  Figure SR-2. Mean Standardized Score on Bracken School Rediness Composite (0-10).

Children in the control groups also had higher average scores on a measure of behavior problems than age-mates in a national sample (Figure SR-3). On the Behavior Problems Index (Peterson and Zill, 1986; Zill, 1985), they tended to have more frequent behavior problems overall, and more frequent externalizing (aggressive/acting out) behavior problems in particular, than five- to seven-year-olds in the National Longitudinal Survey of Youth-Child Supplement. Differences here are not as marked, however, as the differences in cognitive school readiness, though they are statistically significant.

  Figure SR-3. Mean Frequency of Total Behavior Problems (0-2).

At the same time, as can be seen in Figure SR-4, mothers rated the overall health of focal children in the control groups as favorably, or more favorably, than mothers of children aged five- to-seventeen in a national health survey.(8) However, as we have noted, the Family Support Act exempted mothers who were needed in the home to care for a sick or incapacitated family member, including a child, from mandated participation in JOBS. Thus, it is not surprising that mothers of children in the control groups of the Child Outcomes Study sample rated their children's overall health in this way.

  Figure SR-4. Percent of Mothers Rating the Focal Child's General health as Very Good or Excellent.

B.  Program Impacts on Children

Were there program impacts on measures of children's development and well-being at the time of the two-year follow-up?

Did mothers' assignment to a JOBS program exacerbate this developmental risk in children? Alternately, did it diminish risk? Was it the case that a program directed primarily to mothers had little or no impact on children? Or were there impacts that varied in direction (favorable and unfavorable) according to the type of child outcome, the research site, program approach, or the characteristics of the families? We summarize impact findings first "overall" or in the "aggregate." That is, within each study site, we ask whether the measures of child well-being and development differed for the human capital development group as a whole as opposed to the control group as a whole, and for the labor force attachment group as a whole as opposed to the control group as a whole. In subsequent sections we turn to the question of whether children in families who entered the evaluation at higher and lower risk showed different patterns of program impacts. Impact analyses control for key background characteristics, including family race/ethnicity and child gender. Detailed reporting on child impacts by race/ethnicity and child gender are planned in the future.

In analyses at the aggregate level there were relatively few statistically significant program impacts on children. Further, those impacts that did occur tended to be small in magnitude. In fact, only one of the impacts reached the criterion of policy relevance. The findings indicate that the welfare-to-work strategies implemented under JOBS did indeed have impacts on selected child outcomes, though it is important to underscore that these aggregate impacts were relatively few and small in magnitude.

What were the program impacts in different aspects of children's development?

Table SR-1 provides a brief summary of the number and direction ("+" indicating favorable, and "-" indicating unfavorable) of statistically significant program impacts for each of the aspects of children's development studied (cognitive development and academic achievement, behavioral and emotional adjustment, and health and safety) and for each of the six programs studied (the human capital development and labor force attachment program approaches in each of the three sites). This table provides an overview, summarizing impact findings across discrete measures for each aspect of development. Some of the specific measures summarized in this table pertain only to the focal child (for example, the assessment of the focal child's cognitive school readiness, the measure of behavior problems, and the rating of overall health described above). However, some of the specific measures summarized in the table pertain to all of the children in the family. For example, mothers were asked whether any of the children in the family had been suspended or expelled from school, and whether any of the children in the family had had an accident, injury, or poisoning requiring emergency medical attention. Thus, the table includes both "focal child" measures and "any child" measures.(9)

Table SR-1:
Summary of the Number of Aggregate Impacts,
by Developmental Domain, Site, and Program Approach
DEVELOPMENTAL DOMAIN: ATLANTA GRAND RAPIDS RIVERSIDE TOTAL IMPACTS TOTAL IMPACTS POSSIBLE
HCD LFA HCD LFA HCD LFA FAVORABLE UNFAVORABLE
COGNITIVE/ ACADEMIC 2+ 3+ 1+ 0 1+ 0 7+ 0 30
BEHAVIORAL/ EMOTIONAL 0 2+/1- 0 2- 1+ 1- 3+ 4- 78
PHYSICAL HEALTH/SAFETY 0 0 0 0 2- 2- (1) 0 4- (1) 24
TOTAL NUMBER OF IMPACTS FOUND 2+ 5+/1- 1+ 2- 2+/2- 3- (1) 10+ 8- (1) 132
NOTES:  Numbers represent statistically significant impacts (at p < .10); numbers in parentheses represent the number of statistically significant impacts that meet our criterion for policy relevance (i.e., > .33 SD).

"+" indicates a favorable aggregate impact;
"-" indicates an unfavorable aggregate impact.

The table notes the number of possible impacts (that is, the total number of measures of each aspect of development considered, across all six programs), as well as the number of statistically significant impacts actually found. The contrast of actual and potential number of impacts underscores our conclusion that there were relatively few impacts overall, though more than one would expect to occur on the basis of chance. Overall, there were 18 statistically significant treatment-control group differences in measures of child development and behavior, out of 132 impact analyses conducted across all child outcomes and programs. While the number of significant impacts is relatively small, nevertheless, it is a greater number of findings than we would expect by chance alone(10), indicating that JOBS welfare-to-work programs did indeed have impacts on selected child outcomes.

As can be seen, child impact findings at the aggregate level differed in direction according to aspect ("domain") of development. All of the impacts in the domain of children's cognitive development and academic achievement were favorable. That is, when there were statistically significant differences, it was always the case that children whose mothers had been assigned to a JOBS program scored better on measures of this aspect of development than children whose mothers had been assigned to a control group. Moreover, at least one favorable program impact on a measure of cognitive development and academic achievement occurred in four of the six programs examined here.

By contrast, all of the statistically significant impacts in the area of child health and safety were unfavorable. That is, when there were statistically significant impacts, they always occurred in a direction indicating that children of mothers assigned to a JOBS program had less favorable scores than children of mothers assigned to a control group.

All of the unfavorable health impacts occurred in the Riverside site, and unfavorable health impacts occurred in this site in both programs (labor force attachment and human capital development). Mean scores indicated that program group mothers in the Riverside site rated the focal children as less healthy overall. In addition, a smaller proportion of mothers in the program groups than in the control groups rated the focal children as in excellent or very good health (as opposed to good, fair or poor) on this overall health rating. The unfavorable impact of Riverside's labor force attachment program on the proportion of focal children rated as in very good or excellent health was large enough to be considered "policy relevant."

These unfavorable impacts indicate that a larger proportion of program than control group mothers in the two Riverside programs rated focal children in the range of good, fair, or poor health (rather than very good or excellent). We examined the origin of this impact more closely, asking especially whether the higher proportion of experimental group mothers rating their children in this range was largely attributable to the "good" rating, and thus did not indicate a shift toward fair or poor health. However, the findings in fact reflect a higher proportion of mothers rating their children as in poor or fair health in the two programs.

No statistically significant differences were found at the aggregate level in any of the research sites on the proportion of mothers reporting that an accident or injury had occurred to any of their children that was severe enough to require emergency medical attention. Thus, differences in the area of health and safety occurred in terms of mothers' ratings of focal children's overall health, but not on outcomes related to emergencies for any of the children in the family.

The overall health rating reflects the mothers' perceptions of the focal child's health. We do not have doctors' assessments of the child's overall health or the incidence and/or severity of concrete illnesses, such as ear infections. Maternal perceptions may, in part, reflect the salience of this issue. For example, employed mothers may be concerned about the possibility of missing work because of a child's illness, and, thus, any indications of the child's compromised health (e.g., the common cold) may lead employed mothers more so than non-employed mothers to rate the focal child's overall health less favorably. At the same time, lower health ratings could, in fact, reflect poorer child health, which could result, for example, from exposure in child care settings to other children with health symptoms, from diminished attention to such family routines as bed time, or from mothers' having less time to bring children to health care providers for diagnosis and treatment. In either case, it is possible that lower health ratings by program mothers are, to some extent, derivative of maternal employment. (See Section VII below for results suggesting the degree to which maternal employment and/or child care did, in fact, help to explain unfavorable impacts on this rating of focal children's health.)

Impacts on measures of the children's behavioral development and emotional adjustment, in contrast with measures of cognitive development and health, included both favorable and unfavorable impacts. Unfavorable behavioral impacts occurred in each site's labor force attachment program, whereas there does not appear to be a pattern (by site and/or program approach) to the occurrence of favorable behavioral impacts.

Should we have expected child impact findings to go in the same direction across all aspects of development?

Is it problematic that findings went in different directions for different aspects of development, with impacts on cognitive development and academic achievement favorable; health and safety unfavorable; and behavioral and emotional adjustment including both favorable and unfavorable impacts? In fact, there are precedents in other bodies of research on children's development for such complex patterns. Child care research provides an example. Findings in this literature show that children who have participated in formal child care settings tend to have better scores on measures of cognitive development (e.g., Zaslow, Oldham, Magenheim and Moore, 1998), mixed outcomes on measures of behavioral development (e.g., Barnett, 1995; Yoshikawa, 1995), and some unfavorable outcomes in terms of health (such as more ear and intestinal infections, e.g., Hirsh-Pasek, 1998). In giving this example, our intent is not to suggest that impacts of JOBS programs on children are transmitted primarily through child care influences. Rather, our intent is to point out that different aspects of development can be affected in differing directions by a developmental context (like child care, or mothers' assignment to a JOBS program).

What patterns were observed with respect to the three sites and six programs?

Child impact findings at the aggregate level tended to differ according to site, as shown in Table SR-1.

Cognitive development: Favorable impacts were found in Atlanta's labor force attachment program and in all three human capital development programs.  Favorable program impacts on children's cognitive development and academic achievement occurred especially in Atlanta's labor force attachment program. In this program, children had significantly higher mean cognitive school readiness scores, as well as a more favorable distribution of scores on the assessment of cognitive school readiness (i.e., the proportion of children scoring at the high end of the distribution was greater, and the proportion of children scoring at the low end was smaller).

In addition, at least one favorable impact in the domain of cognitive development occurred in each site's human capital development program. Specifically, in Atlanta and Grand Rapids, children in the human capital development program had scores indicating a more favorable distribution on the Bracken assessment (i.e., the proportion of children scoring at the high end of the distribution was greater). In the Atlanta and Riverside sites, a smaller proportion of mothers in the human capital development program than in the control group reported that any of the children in the family had had an academic problem since the start of the evaluation.

It is possible that programs that improve mothers' educational outcomes provide the mothers themselves with a source of cognitive stimulation. This in turn may influence how much and how mothers interact with their children, with implications for the children's cognitive development. Children may also be influenced by observing their mothers pursuing educational goals. Mothers who themselves participate in educational activities may place more emphasis on school work, and/or may feel more effective in assisting their children with homework.

The concentration of favorable cognitive impacts on children in each site's human capital development program and in Atlanta's labor force attachment program is consistent with this hypothesis. These four programs increased mothers' educational attainment (e.g., the receipt of a high school diploma or GED, the receipt of a trade degree).

Health: Unfavorable impacts were found only in Riverside. All of the unfavorable child health impacts at the aggregate level occurred in the Riverside site. Unfavorable impacts occurred in both programs in this site, and all findings pertained to the focal child's health, as rated by the mother. One of these impacts was large enough to be considered policy relevant. No unfavorable health impacts were statistically significant at the aggregate level for either of the programs in the other two study sites. As noted above, no program impacts were found at the aggregate level on the more extreme measure of accidents, injuries or poisonings requiring a visit to an emergency room or clinic. Further, as noted above, the maternal report measure of overall health likely reflects not only the actual occurrence of particular health symptoms in the focal child, but also the salience of the focal child's general health status (which may vary as a function of maternal employment status and history).

Behavior: Unfavorable impacts were found only in labor force attachment programs, though favorable impacts were also found in Atlanta's labor force attachment program. Each of the labor force attachment programs studied had at least one unfavorable program impact in the area of behavior and emotional adjustment as reported by the mother. However, the picture was mixed for the labor force attachment program in Atlanta. For this program there were also two favorable impacts on measures of behavior and emotional adjustment. (A single favorable behavioral impact also occurred in Riverside's human capital development program.)

Pattern of findings in Atlanta and Riverside. On balance, impacts in Atlanta were favorable for each program approach. Of eight statistically significant child impact findings in Atlanta, seven were favorable (with the single unfavorable impact in this site occurring on a behavioral outcome for children whose mothers were in the labor force attachment program).

By contrast, on balance, in the Riverside site, significant child impact findings were unfavorable. Of seven statistically significant child impacts in the Riverside site, five were unfavorable. The unfavorable findings in Riverside were concentrated in the health area, but one also occurred on a behavioral outcome for children whose mothers were in the labor force attachment program.

We started out with an interest in exploring possible favorable as well as unfavorable program impacts on children. What can we conclude about favorable impacts?

In the context of previous research on welfare-to-work programs, it is important to note that the aggregate impact findings in the present study include evidence of favorable impacts on children, particularly in the cognitive domain of development, and particularly in Atlanta.

Findings at the aggregate level indicate slightly more favorable than unfavorable significant impacts on child outcome measures. All of these favorable impacts may be considered small in magnitude, with none reaching the threshold for policy relevance. They are concentrated in the domain of cognitive development. Each site's human capital development program had at least one favorable impact in the cognitive domain, but the greatest concentration of favorable cognitive impacts occurred in Atlanta's labor force attachment program.

To date, only three evaluations of welfare-to-work programs with rigorous experimental designs have had an explicit focus on child impacts: the present study, the Teenage Parent Demonstration (Kisker, Rangarajan, and Boller, 1998), and the New Chance Evaluation (Quint, Bos and Polit, 1997). The Child Outcomes Study is the first noting a pattern of favorable child impacts at the aggregate level for cognitive outcomes, and a greater number of favorable than unfavorable impacts, on balance, at the aggregate level.

Favorable impacts on children, and especially boys, were recently reported in a program evaluation of a different kind, however: the New Hope Evaluation (Bos, Huston, Granger, Duncan, Brock, and McLoyd, 1999). New Hope is a demonstration program in which low-income adults, already working or willing to work 30 or more hours a week, are assured of an income above the poverty level, along with benefits such as health insurance and child care subsidies. Participants in this program, in contrast with those in the three welfare-to-work demonstrations described above, were not all receiving welfare, were not all women, and they did not all have children. Also, unlike current and recent welfare-to-work programs, participation in New Hope was voluntary. Children, and especially boys, were more likely to have been in organized after-school activities and were reported by their teachers to be doing better academically and to be showing more positive social behavior. Parents in the program group reported more positive social behavior in their sons, and the boys themselves had higher aspirations and expectations in terms of future occupations and advanced education.

Perhaps New Hope and the JOBS programs examined here, to a greater extent than New Chance or the Teenage Parent Demonstration, affected family economic activities and resources, child care use, or the stimulation and support available in the home environment. Such impacts on families might help to explain the presence of favorable child impact results. We will return to these possibilities for the Child Outcomes Study in examining the evidence on the processes through which the child impacts arose (Section VII).

How were children from higher- and lower-risk families affected?

As noted above, the Child Outcomes Study was initiated in order to examine four diverging possibilities: that there would be favorable program impacts on children, unfavorable program impacts on children, no net impacts overall, or impacts especially for particular subgroups. We turn now to the fourth possibility, examining child impacts occurring for families with particular characteristics.

Previous research examines whether welfare-to-work programs and other kinds of interventions have different effects on children according to whether or not families have particular risk factors and also according to the total number of risk factors they have. For example, the New Chance Demonstration (Quint et al., 1997) found unfavorable impacts on measures of child behavior problems as perceived by the mother. Unfavorable impacts occurred especially for families in which the mother had entered the evaluation at high risk of depression, in which the mother had been out of school two years or more, and in families with more risk factors overall. The Infant Health and Development Program, which aimed at improving the cognitive development of children born at low-weight, provided center-based early intervention, home visits, and parent support services. This program had positive effects for children from poor families when these families had no or few risk factors, but not when poor families had multiple risk factors (Brooks-Gunn, 1997). Thus there is evidence that children in low income families respond differently to interventions according to the presence of particular risk factors, and the total number of risk factors, in their families.

Following in this tradition, in the present study we looked at child impacts in light of the presence of particular risk factors in families and also the total number of risk factors in families. We defined risk in terms of characteristics found in previous research on welfare families to be associated both with greater difficulty for the mother in making a transition to employment and also with a less positive course of development in children, apart from any intervention.

Research on factors that impede employment among families receiving welfare ("barriers to employment"), point to the importance of demographic characteristics of the family (such as having more children in the family), human capital factors (such as limited educational attainment and low basic skills; lack of work experience and history of longer welfare receipt), and psychosocial factors (such as maternal depressive symptoms) (evidence reviewed in Zaslow, Hair, Dion, Ahluwalia, and Sargent, 1999). These same factors have been found to be associated with the well-being and development of young children in families receiving welfare (Moore et al., 1995). Thus, there is evidence that family configuration, educational background and skills, work and welfare history, and maternal psychological well-being, are associated both with employment and with children's development in samples of families receiving welfare, apart from welfare-to-work programs.

It is not clear from the existing research, however, which type or types of risk may be important to the pattern of program impacts for children in light of mothers' assignment to a JOBS welfare-to-work program. Accordingly, we examined program impacts on children in light of a number of different types of risk. Each family within the study was categorized as at higher or lower risk in four different ways:

  1. sibling constellation risk (with higher risk involving more children and/or more closely spaced children);
  2. educational risk (with higher risk involving limited maternal educational attainment or low maternal math or reading literacy scores);
  3. work risk (with higher risk involving limited maternal employment history, multiple barriers to employment, or a history of welfare receipt of five or more years);
  4. maternal psychological well-being risk (with higher risk involving mothers with more indications of psychological distress);

Families could be categorized as at higher risk on more than one category. For example, a family might fall into the higher-risk category for all four types of risk, or might be categorized as at higher risk in terms of sibling constellation and maternal psychological well-being risk (but not in terms of educational and work risk). By studying each of these types of risk, we could ask whether program impacts on children were especially linked with risk on one (or more) dimension.

In keeping with the previous research, we also created a summary variable looking at the number of these risk factors for which each family was in the higher-risk category:

  1. cumulative risk, (with higher risk involving a larger cumulative number of the preceding four categories for which the family fell into the higher-risk category).

This made it possible to ask whether program impacts on children differed according to whether the family had a greater or lesser total number of risk factors.

Finally, we also explored a measure of risk that we thought might be informative, but for which there was little precedent in the research on families receiving welfare. The research on maternal employment documents differences in outcomes for children according to whether the mother's work role is concordant or discrepant with her preferred role (Zaslow, Rabinovich, and Suwalsky, 1991). Extending this to the present study, we asked whether impacts of JOBS on children would differ according to whether, at baseline, mothers felt more or less positively about working. On a more exploratory basis, then, we examined reservations about working (with higher risk involving mothers who tended to endorse statements indicating a preference for taking care of family over finding employment).

The question that the subgroup impact analyses address is that of whether risk factors that are associated with less progress toward employment in mothers, and less positive development in children (in the absence of any programs or interventions), are also associated with different patterns of impacts for children in the context of JOBS welfare-to-work programs. Given the priority in the Child Outcomes Study of providing a rigorous examination of the possibility of unfavorable impacts for children, careful scrutiny must be given to the possibility that unfavorable effects occurred for children in certain subgroups. An important possibility is that unfavorable impacts might occur especially for families at higher risk, both in terms of specific risk factors and/or on the overall number of risk factors.

While this prediction must be examined carefully, it is not the only pattern that can be anticipated. Indeed, JOBS programs were intended to address the difficulties faced by families in making a transition to employment, through a combination of supports and requirements. The possibility exists that especially for families with risk factors, one or both JOBS programs might be especially helpful, with children benefitting accordingly.

The findings for subgroups did not indicate any one approach to defining risk as more important than the others. Indeed, the pattern of findings generally held across each of the different approaches to defining risk.(11) Accordingly, we discuss below the pattern of findings for higher-risk families overall, and for lower-risk families overall.

As for the aggregate impact findings, the impact findings for subgroups were not pervasive; that is, there were relatively few statistically significant program impacts for particular higher- or lower-risk subgroups. Yet when looking separately at findings for each particular type of risk (sibling constellation risk, educational risk, work risk, maternal psychological well-being risk, cumulative risk, or reservations about working), the number of statistically significant differences exceeded what would be expected to occur simply on the basis of chance.(12) We begin with the question of whether there was a pattern of unfavorable impacts for children in higher-risk subgroups. However, we also explore the possibility of a pattern of favorable impacts for children in higher-risk subgroups, and we summarize findings for lower risk subgroups as well.

Higher-risk subgroups. The findings for higher-risk subgroups do not support the simple and straightforward hypothesis that there would be a concentration of unfavorable impacts for children in higher-risk families. Rather, the findings are in accord with a more complex pattern. We underscore that as for the aggregate impacts, statistically significant findings were relatively few.

Findings for children in higher-risk families were generally favorable when the family had been assigned to a human capital development group or to Atlanta's labor force attachment program. The only exceptions relate to unfavorable health impacts of Riverside's human capital development program (though all other findings for higher-risk families in this program were favorable), a single unfavorable health impact of Atlanta's labor force attachment program, and a single unfavorable behavioral impact of each of Atlanta's JOBS programs.By contrast, impacts for children in higher-risk families assigned to either Grand Rapids' or Riverside's labor force attachment program tended to be unfavorable.

Impacts for children in higher-risk subgroups tended to be small in magnitude. Yet several of the findings, all of these unfavorable and occurring in labor force attachment programs, were of sufficient magnitude to be called policy relevant.

That largely favorable impacts were found for children from higher-risk families when mothers had been assigned to human capital development programs or to Atlanta's labor force attachment program may be of particular interest to policy makers. In the new policy context, as caseloads drop, there are concerns that perhaps families remaining on welfare, like the higher-risk families studied here, face barriers to making the transition from welfare-to-work. Programs that have favorable impacts on mothers' education or training may help address barriers, and in doing so, also benefit children.

Lower-risk subgroups. For children in lower-risk subgroups, there was a concentration of unfavorable impacts in three programs: both of Riverside's programs, and the labor force attachment program in Grand Rapids. Across these three programs, a number of the statistically significant impacts were large enough to meet the criterion for policy relevance. In all but one instance, these policy relevant impacts were unfavorable. Further, many (one-third) of the unfavorable and policy relevant impacts found for children in lower-risk families substantially exceeded the threshold for policy relevance, in that effect sizes were .50 or larger (which is considered "moderate" to "large" in magnitude; Cohen, 1988).

On balance, the significant impacts for lower-risk families in Atlanta's labor force attachment program were favorable. None of these impacts was of a magnitude to be called policy relevant. Lower-risk families in human capital development programs outside of Riverside showed either no impacts or only isolated impacts, mixed as to direction.

In all three programs in which there were unfavorable and policy relevant impacts for lower-risk families, such impacts occurred both in the domain of child health and also in the domain of child behavior. In these analyses, there is also evidence of unfavorable health impacts of sufficient magnitude to be called policy relevant occurring beyond Riverside. In general, the subgroup impact analyses, in contrast with the aggregate impact analyses, provide evidence of more widespread health impacts. Significant, although not always policy relevant, impacts in an unfavorable direction went beyond Riverside in these analyses. Indeed they occurred for lower-risk subgroups in all programs and sites with the exception of Grand Rapids' human capital development program. All health impacts for children in lower-risk families were unfavorable.

In the Riverside site, children in lower-risk families assigned to the labor force attachment program also showed unfavorable (and in some instances policy relevant) program impacts in the cognitive domain. This was the only set of unfavorable impacts in the cognitive domain across all of the subgroup impact (and aggregate impact) analyses.

In sum, the findings for lower-risk subgroups are not pervasive (that is, the proportion of tested impacts to reach statistical significance was not high) nor do they appear specifically in one program approach or another (human capital development or labor force attachment). Rather, the larger unfavorable impacts are concentrated in three particular programs (Grand Rapids' labor force attachment program, and both of Riverside's programs).

What are the implications of the child impact findings for the concern about possible unfavorable impacts of JOBS welfare-to-work programs on children?

On the one hand, the dearth of policy relevant impacts at the aggregate level diminishes concern about unfavorable effects for children of mothers' assignment to a JOBS welfare-to-work program. Yet the concentration of unfavorable and policy relevant impacts, occurring especially among lower-risk subgroups in three programs suggests a need for continued vigilance.

It was not the case that the subgroup impact findings were all large in magnitude. Yet some subgroup impacts were moderate to large in magnitude (i.e.,were .50 or larger). Moreover, all of these moderate to large impacts were unfavorable, and all occurred for lower-risk subgroups. The fact that the unfavorable policy relevant impacts occurred in a concentrated manner (for lower-risk families in three specific programs), and that some of these were moderate to large in magnitude, suffices to indicate that we cannot entirely dismiss the possibility of unfavorable effects on children.

We need to continue to follow the development of children in particular lower-risk subgroups over time. In analyses of the five-year follow-up data in the future, it will be critical for understanding the impacts on children of these programs to ask whether the impacts for these subgroups are sustained, grow in magnitude, diminish, or disappear and to examine the context of such findings. For example, if unfavorable impacts are sustained, it will be important to explain whether the impacts are occurring in the context of ongoing economic struggle for families trying to sustain employment at low wages. If, on the other hand, the unfavorable impacts for these subgroups disappear over time, it will be important to ask if an initial period of economic struggle gave way to one of increased and stable economic well-being.

The findings suggest the need for continuing vigilance regarding the possibility of unfavorable impacts on children, though especially in particular programs and at particular risk levels. At the same time, the occurrence of favorable program impacts, albeit rarely large enough to call policy relevant, is encouraging and important to monitor over time. Favorable impacts were present especially for cognitive outcomes, and in Atlanta's programs. In addition, there was a tendency for favorable impacts to occur for children from higher-risk families whose mothers had been assigned to human capital development programs or to the Atlanta labor force attachment program.