TRAINING KEY 277: Logit ANOVA

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

TRAINING KEY 277: Logit ANOVA



INTRODUCTION

In this Training Key we will practice design-based multivariate model fitting for categorical data in the OHC survey setting. In point A, the aim is to model the variation of domain proportions of the binary response variable PSYCH2 (psychic strain; 1 present, 0 otherwise) across the eight domains formed by a cross-classification of SEX, a two-category variable AGE (-44 years, 45- years) and the variable PHYS (physical health hazards of work; 1 present, 0 otherwise) by fitting the model in stepwise manner. In point B, the aim is to describe and practice the GWLS method from more detailed and theoretical perspective. Point C is the option for interactive analysis.



A) EXERCISE ON LOGIT ANOVA MODEL FITTING WITH THE PML METHOD

In this part we will exercise how to fit a logit ANOVA model in the OHC survey setting described above. We use pseudolikelihood (PML) estimation in this exercise. Detailed instructions of this exercise will be given once you start.
 

SUDAAN / RLOGIST
SAS / SURVEYLOGISTIC


B) GWLS METHOD FOR LOGIT MODELLING ON DOMAIN PROPORTIONS IN THE OHC SURVEY SETTING

This option is related to Example 8.1 and is for those who wish to gain deeper understanding on how the GWLS method functions in logit ANOVA. SAS/IML macro is provided with the description of each step of the model fitting procedure. Full design-based (DES) option and model-assisted BIN option (assuming simple random sampling) are compared. Further instructions are given once you start.
 

 


C) INTERACTIVE SAS USE

  1. Please download the reduced version of the OHC data set (only relevant variables are included) and save the data in the disk.
  2. Download the SAS code for your own further training in logit ANOVA model fitting (NOTE! you need to have access to SUDAAN (a SAS callable version)).
  3. Download SAS/IML macro for practicing GWLS method and model building from more theoretical point of view (NOTE! You need to have access to SAS/IML).
Further instructions are given in the code once you download.