National Survey of Adolescent Well-Being (NSCAW): A
Comparison of Model and Design Based Analyses of Cognitive
Marianne Bertolet, Howard Seltman, Joel Greeenhouse and Kelly Kelleher
Understanding and protecting vulnerable children is
key to helping them become productive members of society. The
Department of Health and Human Services sponsored the National Survey
of Child and Adolescent Well-Being (NSCAW) to better understand the
lives of children who come into contact with the child welfare system.
This paper uses the NSCAW data to investigate the role of maternal
depression and maternal substance abuse on a child's cognitive
stimulation scores for a subset of the children. An investigation of
the survey methodology and the actual data led to some manipulation of
the data and assumptions for the analysis. The differences between
design and model based analyses of survey data were explored.
The two analyses were applied to the cognitive
stimulation scores. For the
model based analysis, two models were developed, one which used the
weights at each stage of model building and one which did not
use the weights at each stage. The models had conflicting results.
Two recommendations were made to improve
upon the model based analysis. First was to obtain information about
the weights after each stage of adjustment. The second was to model
the different strata separately. The
design based analysis (design based-model assisted) used the two models
created in the model based analysis and re-examined them using design
based techniques. The design based analysis had similar results to
the model based analysis. The two models in the
design based setting also provided conflicting results regarding the
question of interest. To determine which is better, the question of
which ``regression line'' provides better data reductive properties
needs to be addressed.