TRAINING KEY 298: Alternative multivariate analyses for the OHC Survey design

Practical Methods for Design and Analysis of Complex Surveys.
Risto Lehtonen and Erkki Pahkinen

TRAINING KEY 298: Alternative multivariate analyses for the OHC Survey design



INTRODUCTION

In this Training Key, accounting for the sampling design complexities is studied for multivariate survey analysis. The main concern is in the clustering effects. Design-based and model-based analysis options are compared empirically, using the appropriate SUDAAN and SAS analysis procedures, and R functions. We use the OHC Survey data set in our analyses.



A) ACCOUNTING FOR CLUSTERING EFFECTS FOR A CONTINOUS STUDY VARIABLE

For a continuous study variable, modeling takes place under a fixed-effects linear model and under a mixed linear model. For a fixed-effects linear model, GEE estimation (generalized estimating equations) is used with the SUDAAN procedure REGRESS, the SAS procedure GENMOD and the R function gee. For a mixed linear model, REML estimation is used with the SAS procedure MIXED and the R function lme. Further instructions will be given once you start.
 

 


B) ACCOUNTING FOR CLUSTERING EFFECTS FOR A BINARY STUDY VARIABLE

For a binary study variable, modeling takes place under a fixed-effects logistic model and under a mixed logistic model. For a fixed-effects logistic model, GEE estimation is used with the SUDAAN procedure LOGISTIC (RLOGIST), the SAS procedure GENMOD and the R function gee. For a mixed logistic model, REML estimation is used with the SAS macro GLIMMIX and the R function glmmPQL. Further instructions will be given once you start.
 

 


C) INTERACTIVE SAS USE

Please download the SAS code for your own further training (NOTE! you need to have access to SUDAAN (a SAS callable version)).

Further instructions are given in the code once you download.