The conceptual foundation of this research draws heavily on the "investment-in-children" theoretical framework described by Haveman and Wolfe (1994.) The effect of government policies on children have never been more relevant than in this period of transition from pre to post welfare reform era. Parental choices have always been of utmost importance to children's well-being. In this framework, parents' decisions that shape the child's environment include the number and timing of the children they decide to have, whether and how much to work, where to live, where to send their children to school, stability of their union, and others.
Children have a direct impact on their own well-being as well. Among many other decisions, they may choose to put more effort into school or not, or participate in extracurricular activities. This framework allows the effect of poverty and program participation to be incorporated at both the societal and familial level. To this framework, we also add an input for household stressors which are included in Haveman and Wolfe's framework as family circumstances. The primary indicators of household stress that we include in this category are presence of disabled adults in the household, absence of employment, and disabilities of the child relating to school. The latter also falls into each of the other categories for investment as well. While Haveman and Wolfe use longitudinal data from the PSID to examine the determinants of childhood success, we are bound by the constraints imposed by a cross-sectional analysis, and as such are examining correlations with child success/well-being.
The process of investing in children and promoting their successful development is intertwined with the concepts of social, financial, and human capital. Coleman (1988) describes the concept of social capital as existing in the relationships people have within families and communities. This is a resource on which members of the family and community, especially children, can draw. It is a resource that is embedded in these relationships, and is believed to be important in children's successful development (Coleman, 1988). Financial capital is generally defined as income and/or wealth, and human capital is usually defined as education and skills. These are central resources from the family that are believed necessary for a child's successful development, and resources which will be valuable for the child's success in the future.
For this analysis we attempt to operationalize as many aspects of capital as possible for the child and family within the context of Haveman and Wolfe's framework. Limitations within the data prevent us from including the depth and breadth of covariates that are indicated by previous research. Further, confinement to a cross-sectional perspective limits the analysis to a measure of associations, rather than using the data longitudinally to conduct a causal analysis.
The data for this analysis come from a combined sample from the 1992 and 1993 panels of the Survey of Income and Program Participation. These two panels were designed to overlap and provide a more substantial sample for estimation of cross-sectional statistics. The panels were designed to produce nationally representative samples of households. The 1992 panel consisted of an initial 19,600 households, and the 1993 sample was started with 19,900 households. The 9th wave of the 1992 panel and the 6th wave of the 1993 panel use the same instrument and were conducted between October 1994 and January 1995. In addition to the income, assets, program, and basic demographic data contained in the core instrument, topical modules are included on a variety of specialized areas through the life of the panel. The waves used in this study include topical modules on work history, adult and child disability, adult and child health care utilization, child care, child support, and child well-being.
Combining wave 9 of the 1992 panel with wave 6 of the 1993 panel yield 40,141 unweighted households including households created during the life of the panel and 100,939 unweighted persons. For this analysis we examine the children from the child well-being topical module attached to the core instrument. Respondents identified as designated parents, usually the mother, responded to a variety of items for each of their children under 18 years old. There were 24,994 children from age 0 through age 17 for whom the child well-being topical module was administered.
The subsample selected for this analysis consists of children who were between the ages of 6 and 17, not missing information on their reported grade or enrollment, not systematically missing for a number of covariates, and whose designated parents are age 18 or over. The weighted and unweighted sample sizes are presented in table 1. Frequency distributions for the population excluded as a result of missing grade distributions, interview fallout, and young designated parents are included in appendix table 1. Although exclusions based on missing information are large and significant, the remaining sample reflects the grade for age distribution based on the 1994 October CPS estimate almost exactly (Bruno and Curry 1996).
Table 1. Weighted and Unweighted Sample Sizes.
|All Children Age 0 through Age 17
|All Children Age 6 through Age 17
|Children Age 6 through Age 17 not
Limitations in this measure of child well-being are inherent in the measure itself. The retention indicator employed (modal age for grade statistics, responses to survey questions, and institutional records) yields differing rates. For example, modal age for grade is often used as a proxy for retention by computing a comparison of the modal grade with the child's age. However, this rate may be inflated due to the method of calculation used since students who delay entry into kindergarten are erroneously tabulated as being held back in school. In comparison, other studies ask students and parents if they (or their child) have repeated a grade. Yet, the negative association tied to being held back may deter respondents from honestly responding to sensitive questions, such as grade retention. Research using institutional records is less common due to the confidentiality of school records. An additional limitation is that studies tend to use an ever-retained measure because the timing of the retention (i.e. the age that the child repeated the grade or was held back) is not known.
Income and Poverty: The economic situation of the family is measured in this analysis in a number of ways. First, income is averaged over the four-month reference period for these waves of the survey, and divided by the four-month average poverty level assigned to each family based on their individual composition (number of persons in the family.)1 This generated a relative income measure for both the family and the household. We have included these measures in the model in two different configurations. The first compares those families or households receiving less than 100% of poverty or more than 300% of poverty with those families or households receiving between 100% and 300% of poverty. Through preliminary analysis we found that family and household income measures did not differ with respect to being ontrack academically. As a result of this, only family income is included in this analysis. This is not completely unexpected, as it is not clear that a child in a household with extra-familial income would have access to that income, or its associated resources. Ownership of the house or apartment by the householder is included as a dummy variable coded one if the family is renting or resident without making a cash payment. The employment status of the designated parent is also included as a measure of income and poverty, and is coded one if the designated parent is employed full or part time. Variables measuring the number of working adults in the household and marital status replace an independent measure of spouse working, since this would more appropriate when all marital statuses are included in the model. That is, a variable for whether the designated parent works is inherently related to the marital status of that parent, and unemployment/non-participation in the labor force would have very different meanings by marital status.
Program participation was measured with two variables. First, public welfare recipiency was included as a dummy variable coded as '1' if the family received AFDC, WIC, Food Stamps, or General Assistance aid. Second, receipt of Supplemental Security Income (SSI) was also included as a separate variable. This item was later dropped due to colinearity with measures of disability among adults and children in the household. Participation in government assistance programs is also measured by a variable included that is marked '1' if the family is in public housing or rent subsidized housing, and '0' otherwise. Although they are not indicators of government transfer programs, receipt of alimony or child-support, and residence in a household that was not owned by the family were also included as dummy variables in this section.
Neighborhood: Measures of the designated parents perceived level of neighborhood social capital related to their children were included in the models as indices generated through factor analyses of the questions in table 2. These questions were asked on a "how much" scale with scores that range from zero through ten. Zero means "not at all" and ten means "the most". As shown in table 2, these items were broken into two sets based on the results of the factor analysis. The first set is an indication of the positive aspects of social capital relating to children, and the second set represents the negative aspects within the neighborhood or community. An index missing one or more elements was coded as missing for the index and excluded from the analysis. After removing the cases with missing data, the elements are then averaged to create the index. These indices were then divided into high, medium, and low according to their distributions. These indices were included in the analysis in a number of different arrangements. We included high trust and low danger, using medium and low trust, and medium and high danger as the reference groups. In subsequent models the indices were combined into a neighborhood assessment measure and included as a dummy variable coded one if the neighborhood assessment index had a value between zero and six, indicating a negative assessment compared to an intermediate or high neighborhood assessment.
|Table 2. Question
|People in this (neighborhood/community) help each
We watch out for each other's children in this
There are people I can count on in this
If my child were outside playing and got hurt or scared,
there are adults nearby who I trust to help my child.
|There are people in this (neighborhood/community)
who might be a bad influence on my child(ren).a
I keep my children inside my home as much as possible
because of dangers in the (neighborhood/community)a
There are safe places in the (neighborhood/community)
for children to play outside.
a The scales for these questions were reversed to make them consistent with the other items.
An item included as a neighborhood measure was the neighborhood poverty rate. This is based on information taken from the 1980 Census and updated to the date the panels were started. Large numbers of new housing (permit areas) areas were created between the 1980 Census and the sample selection for these SIPP panels. These households did not receive an estimate of their neighborhood poverty rate. After comparison with income and education, it was clear that the majority of these households were very similar to households with less than a 20% poverty rate, and were included with these for the analysis. We also include measures of urbanicity and population density. Urbanicity is included with two dummy variables identifying respondents in central city areas of metropolitan sampling areas (MSAs) and respondents in the balance of the MSA, and compares them with residents not living in an MSA. Population was divided into four categories. Residents in areas that have less than 1 million population or were not in an MSA are compared to residents in MSAs with 1 to 5 million residents, as were residents living in areas with 5 to 10 million, and areas with 10 or more million persons.
Family and household: We consider the child's family situation in the analysis in several ways. A variable is included to distinguish the previously married and never married from the currently married designated parents. Although this is a non-optimal way to consider marital status and family structure, it does represent a significant distinction in the resources available to the children, in terms of social, human, and financial capital, as well as the occurrence of marital instability. We are not, however, able to clarify the timing of the disruption, with respect to the children in the household, or their educational derailment.
There are some shortcomings that limit completeness of the measurement of the children's family situation. First, only one parent, usually the mother, is identified for the child. The relationship between the spouse of the designated parent and the child cannot be fully determined, i.e. there is no way to discriminate between biologic and step children. Second, the marital history, fertility history, and household relationship matrix for the adults in the household, which could illuminate the family situation, is only available in the second wave of each panel. Linking the files creates additional problems, not the least of which is that the two panels differ with regard to the length of time between wave two and the waves containing the child well-being topical modules. The 1992 panel has an additional year of time between the two waves.
As social capital adheres in the relationships between people and human capital in the characteristics of those people, identifying the number and qualities of the adults in the household should assist in capturing the social, human, and financial capital that may be available to the children. Characteristics of the household's composition included are; the number of adults (persons > 17 years old) in the household, the number of adults having attended college, and the number of adults employed. The number of adults in the household was included in a single dummy variable, coded one if there were three or more adults in the household. The reference group is households with one or two adults. Similarly the number of adults in the household who were working full or part time is included as a dummy variable comparing households with two or more working adults with households having zero or one working adults. Education of the adults in the household is also a dummy variable, and is coded one if there are two or more adult in the household with at least some college. This is compared to households with zero or one adult with college experience.
Enrichment and Rules: Social interaction and childhood enrichment of the child is measured with three items identifying the children's participation in extracurricular activities. The first is a dummy variable coded one if the child plays sports either in or out of school, and zero otherwise. A second variable is created to represent participation in clubs or organizations after school or on weekends. Scouts, religious groups, and boys and girls clubs are the examples given for the types of clubs referenced by the questionnaire item. Finally, a dummy variable is coded one if the child participates in lessons after school or on weekends in subjects like music, dance, language, or computers, and zero otherwise. Parental rules are included in reference to rules for television watching. We create a variable coded one if the parent has rules about either the number of hours, time of day, or programs that this child may watch on television.
Stressors: Household and child level stress is an important component within the conceptual framework. Presence of stressors may affect parents' ability to invest time and resources in their children, and may also affect children's ability or motivation to invest in themselves. We include three measures of household and child related stress. The first of these is a dummy variable indicating whether or not the child has changed schools since the first grade other than the transitions from elementary to junior high school or middle school, or into high school. This variable is coded one if a change occurred and zero otherwise.
The next two of these indicators of household and childhood stress are related to disability. If there is a disabled adult in the household a dummy variable is coded one. This information is from self-reported status of the adults in the household. An adult was identified as disabled if disabled is marked on the interview control card for that person. Childhood disability is also included, which indicates the presence of a current disability for the child. Stress would certainly be introduced if a sibling were disabled, however, not all of the respondents have siblings in the household. The disability being reported is specifically related to school activities, though not specifically identified. The questionnaire item asks, "Because of a physical, learning, or mental condition, does (child) have any limitations in his/her ability to do regular school work?" If the designated parent says yes to this, a dummy variable for educational limitations is coded one.
Background Socio-demographics: A number of other variables were included in the analysis to control for differences in background. For the designated parent, we include age, race/ethnicity, education, marital status, and region of residence. Age is broken into three groups, those less than 35 years old, parents 35 to 40, and parents 40 years old and over. Thirty-five to forty year old parents are the reference group. Background characteristics of the child are also included. These are the child's gender, age in single years from six through seventeen, and the number of siblings under age 18 in the household.
We use logistic regression to measure the correlates of child well-being as measured by the children's academic progress. Three models are presented. The first model, the base relationships, presents the results of regressions for each dummy or categorical variable entered separately. This shows relationships that would be observed in a typical tabular analysis. It is followed by model 2 which includes all of the covariates in the multivariate regression. Finally, a more parsimonious model is presented in model 3. It includes only those variables related to the outcome, with the exception of employment status of the designated parent. The later variable is retained, although it fails to show a significant relationship to educational progression, because it serves as both a control variable and an income and poverty indicator. Table 3 presents percent distributions of the principal study variables for the sample analyzed. The distributions of these variables for the population age 6 through 17 that was excluded from the analysis is presented in appendix table 1.