Software for the longitudinal analysis of survey data

DESCRIPTION (provided by applicant): Third Way Statistics will develop and market software for the analysis of complex survey data to researchers in health sciences, economics, social sciences, market research firms, and statistics. The software will be able to perform longitudinal analyses for a broad range of generalized estimating equation (GEE) models, multilevel models, and survival models, using estimators appropriate for complex survey designs. In Phase I, research will be conducted into the currently outstanding theoretical issues surrounding design-based estimators for GEE and multilevel models (such as nonconstant weights within panels) and appropriate variance estimators for complex survey designs will be developed. At the end of Phase II, the software will have a full implementation of GEE for complex survey designs (currently only one software package has GEE for complex survey designs, but only supports a limited number of models). The software will be able to estimate a large number of single-level and multilevel models for survey data (currently no software has survey design-based estimators for these models). The software will have a comprehensive set of survival analysis tools for complex survey data (currently only one survival procedure in one software package exists for complex survey data). The software will be built with modem software tools and be user friendly with a flexible graphical user interface and data I/O standards that are compatible with other software (e.g., handle input from databases, and produce HTML and RTF output). Phase I of this development project focuses on building a core set of software tools and statistical routines: (1) memory management, (2) data structures, (3) interface system, (4) optimizers, (5) general routines for the computation of Taylor linearization, balanced repeated replication, and jackknife variance estimators, and (6) simulation of complex survey samples to support statistical research on the properties of estimators. To demonstrate these core routines, Phase 1 will have implementations for GEE, selected fixed- and random effects (single-level) models, and the Cox proportional hazards model.