Gorrell P; AcademyHealth. Meeting (2004 : San Diego, Calif.).
Abstr AcademyHealth Meet. 2004; 21: abstract no. 2021.
Social & Scientific Systems, Inc., Computer Systems & Data Analysis, 8757 Georgia Avenue, Silver Spring, MD 20910 Tel. (301)628-3237 Fax 301.628.3201
RESEARCH OBJECTIVE: SAS and SUDAAN are two software packages commonly used by healthcare researchers to analyze data from complex surveys. This paper compares SAS and SUDAAN's implementation of the Taylor linearization method for variance estimation with complex surveys. Beginning with the release of version 8, SAS procedures such as SURVEYMEANS use the Taylor expansion method for variance estimation. Although both SAS and SUDAAN make use of this method (currently it's the only option in SAS), there are important differences, such as assumptions concerning missing values (non-response).By showing concrete examples from the analysis of national healthcare surveys, this paper examines and clarifies these differences so that healthcare researchers can make informed choices when analyzing data from complex surveys. STUDY DESIGN: Not applicable. POPULATION STUDIED: The surveys used in this paper are designed to provide national estimates of healthcare utilization and expenditures in the U.S. PRINCIPAL FINDINGS: For variance estimation, both SAS and SUDAAN implement the Taylor linearization method. For each, this method produces identical results across a wide range of data. However, important difference exist in the treatment of nonresponse and the analysis of subgroups (domain analysis). CONCLUSIONS: SAS and SUDAAN each provide healthcare researchers with important analysis tools for complex surveys. Researchers need to be aware of the different assumptions SAS and SUDAAN make concerning nonresponse and subgroup analysis when choosing which analysis tool best fits their analytic requirements. IMPLICATIONS FOR POLICY, DELIVERY OR PRACTICE: Variance estimation is an essential component of healthcare survey analysis. It is important for researchers who are evaluating surveys, or survey results, to understand the sometimes differing assumptions being implemented in the software packages they use.
Publication Types:
Keywords:
- Cross-Sectional Studies
- Data Collection
- Delivery of Health Care
- Health Care Surveys
- Research Design
- methods
- hsrmtgs
UI: 103625055
From Meeting Abstracts