ITROW Research Projects - Neighborhoods and Gendered Child
Development
Methodological Hurdles
Why Do We Not Know More?
Omitted-context
variables problem.
As suggested
in the reviews of neighborhood effects mentioned above, one of the most
serious problems of community context research is the omission of
important variables (Duncan and Raudenbush 1998). Regressing a child’s
behavior or achievement on his or her family and contextual
characteristics will bias estimates of the relationship between those
contextual characteristics and the outcome variables if important
qualities of the context are omitted. This is a problem in studies that
use only a single neighborhood measure such as poverty. Instead, a
number of different kinds of measures are needed. To minimize this
problem, we incorporate into the models a number of different kinds of
census characteristics, and also we incorporate a number of measures
from different sources – census data, police records on crime, and
family involvement data from parental reports -- to predict outcomes
measured by school records, teacher observations, and child
self-reports.
Absence of
multilevel designs.
Furstenberg and Hughes (1997) suggest that neighborhood research needs
to be multilevel in the sense that it includes and analyzes data on both
neighborhoods and individual children within these neighborhoods. The
Jencks and Mayer (1990) review located few studies that investigated
relationships between neighborhood conditions and individual outcomes
while controlling for family-level economic status. Most focused
instead on the economic health of families rather than communities, or
on effects of community without controlling for family economic health.
Data on family processes should be collected as well (Sampson 1992).
Thus, children may be perceived as nested within families and families
within neighborhoods. Such an approach mimics the theoretical
proposition that communities affect families and children (Furstenberg
and Hughes 1997; Smith 1989).
Specification of
neighborhood constructs and measures.
As Furstenburg and Hughes indicate, delineating geographic neighborhoods
is difficult, because the concept of "neighborhood" is not precise (Chaskin
1994). There is growing evidence that residents in close geographical
proximity often do not agree on the boundaries of their neighborhood.
Including several overlapping definitions of neighborhood in one study
would enable comparison of relationships under different definitions
(see, for example, Brooks-Gunn et al. 1993). A subproject of
this proposed research will compare the qualities of neighborhoods as
defined by census tracts to those defined by traditional community
boundaries.
Selection bias.
Some argue that it is not the poverty of neighborhoods that is
detrimental to child development, but the poverty of families that live
in poor neighborhoods. Selection bias (also referred to as “endogenous
membership”) occurs when community context variables become proxies for
unmeasured parental characteristics that affect children’s outcomes and
that lead parents to move into particular kinds of neighborhoods
(Corcoran 1995). Perhaps the propensity of children to live in better
or worse neighborhoods or attend better or worse schools depends on
parent background characteristics, or parental choice in choosing a
better neighborhood. Presumed neighborhood effects might in that case
reflect parental characteristics which vary by neighborhood rather than
neighborhood characteristics. Or, a third hypothesis would be that both
the poverty of neighborhoods and the poverty of families are
detrimental. Our latent growth modeling analytic strategy will examine
the timing of exposure to neighborhood poverty in relation to changes in
family poverty and child behavior, thus producing evidence pertinent to
these alternative hypotheses.
Longitudinal
research. Studies
which look at only one point in time are of limited utility in
understanding the complex pathways of influence among neighborhoods,
families, and children. This makes it difficult to assess whether
observed differences in outcomes across neighborhoods are due to
differential selection into neighborhoods (Tienda 1991). Furstenberg and
Hughes (1997) state that all of the published research on the
relationship of neighborhoods to children’s development has employed
cross-sectional data, which limits the causal inferences that may be
drawn. Neighborhood research should attend to the slope of life-course
trajectories, not simply to the level of outcomes or statuses at a point
in time. This requires tracking the same individuals using repeated
measures of different domains of development as they encounter
distinctive neighborhood environments over time. Alternatively, one
could demonstrate neighborhood influence on children’s development by
linking the length of exposure to neighborhood conditions to specific
paths of development (Furstenberg and Hughes 1997; Tienda 1991).
Our proposed research employs longitudinal data on
children with repeated measures for the primary dependent variables.
Data on mobility from one neighborhood to another will be used both to
look at how child outcomes change as their families move from one
neighborhood to another, and at how length of exposure to neighborhood
conditions may affect child outcomes. (For details, see the analytic
section below.)
|
