TRAINING KEY 288: Logistic ANCOVA

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

TRAINING KEY 288: Logistic ANCOVA



INTRODUCTION

Example 8.2: In this Training Key we concentrate on performing design-based logistic ANCOVA (Analysis of Covariance) modelling with the pseudolikelihood or  PML method. We fit a multivariate logistic ANCOVA model by entering in the model some of the predictors as continuous measurements and some as discrete variables. We will also demonstrate the effect of interaction terms in the interpretation of the results. The data is again from the OHC Survey.



A) LOGISTIC ANCOVA

In this part we will study how to perform design-based logistic ANCOVA modelling with the PML method for a binary response variable in the case of three continuous predictors and one discrete predictor. We will build the model by removing interaction terms from a model including all possible interaction terms of the discrete predictor with the continuous predictors. Graphical displays are used to show the effect of presence or absence of an interaction term on predicted proportions calculated for a given logistic ANCOVA model. The results can be compared with those from the logit ANOVA exercise (key 277).

Instructions will be given once you start.
 

SAS/SUDAAN
SAS/SURVEYLOGISTIC


B) INTERACTIVE SAS USE

Please first download the reduced version of the OHC data set (only relevant variables are included) and save it to the disk.

Then download the enclosed SAS code for your own further training of model fitting by logistic ANCOVA.

NOTE! You need to have access to SAS and SUDAAN in your computer.