Research and Analysis by Salvatore J. Gallicchio
Covariance Estimates for Regression Parameters from Complex Sample Designs: Application of the Weighted Maximum Likelihood Estimator to Linear and Logistic Regression Analysis in Which Observations Might Not be Independent
Statistical methods of variance estimation are presented in this paper for the analysis of survey data involving complex sample designs. With certain complex sample design, estimation of the covariance matrices in linear and logistic regression is not straightforward. The design may be complex because of disproportionate sampling of strata, necessitating the use of weights, or because the observations are not independent, or possibly both. Examples are given from projects at the Social Security Administration, and computer programs written in SAS (Statistical Analysis System) are provided.
This is the second of two articles on the effects of Old-Age Disability Insurance (OASDI) and Supplemental Security Income (SSI) payments on the poverty status of children. Based primarily on a data file from the 1990 SIPP matched with Social Security Administration (SSA) administrative records, the principal findings in the article are that: (1) the families of children who were entitled to survivors benefits, and in particular those families in which the surviving parent was remarried, were much less likely to have income below the poverty threshold than other families with children who received OASDI benefits; (2) families with a child eligible for SSI payments, and headed by a single adult, received considerably less income from earnings, and had less income overall, than other families with children that received SSI payments; and (3) the primary reason that some families who received OASDI and SSI benefits remained in poverty was the absence of any employed family member.
A Note on Maximum Likelihood Estimation of Discrete Choice Models from the 1978 Survey of Disability and Work
This paper demonstrates an alternative maximum likelihood procedure for estimating discrete choice models in retrospective samples, such as a model of SSA disability beneficiaries or application status in the 1978 Survey of Disability and Work.