2006 HSR&D National Meeting Abstract
2001 — Three Ways to Improve the Validity of Statistical Results
Sox-Harris A (Center for Health Care Evaluation)
This workshop describes three common practices in data analysis that can lead to misleading results: (a) Assuming independence of observations when dependencies exist, (b) Using ad hoc methods to address missing data (e.g., mean imputation, case-wise deletion), and (c) Selecting tests that are inappropriate for the underlying distribution.
Each of the three problematic practices, and the strategies that represent the “best practice” to address them, will be presented using concrete examples from health services data sets. Brief introductions will be given to basic logic and mechanics of the recommended practices: mixed-effects regression models, model-based multiple imputation procedures, and non-normal and distribution free statistical tools. Participants will be provided handouts that summarize workshop material. They will also have access to on-line electronic files that contain the example data sets and tutorials to run the examples themselves. Suggested readings for further study of these topics will be provided.
The workshop is intended for health service researchers who conduct and/or interpret statistical analyses.
Workshop participants should have basic familiarity with statistical concepts and methods, especially regression models.