Model selection uncertainty in longitudinal data analysis is often much more serious than that in simpler regression settings, which challenges the validity of drawing conclusions based on a single selected model when model selection uncertainty is high. We advocate the use of appropriate model selection diagnostics to formally assess the degree of uncertainty in variable/model selection as well as in estimating a quantity of interest. We propose a model combining method with its theoretical properties examined. Simulations and real data examples demonstrate its advantage over popular model selection methods.
|Original language||English (US)|
|Number of pages||22|
|Journal||Annals of the Institute of Statistical Mathematics|
|State||Published - Apr 2012|
Bibliographical noteFunding Information:
The work of the second author is partially supported by NSF grant DMS-0706850.
- Adaptive regression by mixing
- Longitudinal data
- Model combining
- Model selection
- Model selection diagnostics
- Model selection uncertainty