Stratified Analysis of Clustered Binary Data in Maternal and Child Health

In some studies randomization is done at the Block/village/community/school/work site level, but the inference are drawn as if the randomization is carried out at individual level. This invalidates the application of usually adopted statistical method. There are new methods available to take into account of dependency among subjects within the cluster to analyze data under stratified design. In this paper, an attempt is made to compare these methods using real life data set. To demonstrate these methods, the data is adopted from the maternal and child health study. It is observed that there is a wide variation between the unadjusted Mantel- Haenszal chi-square value and the other methods chi-square values. It is suggested that the Zhang and Boos chi-square method is appropriate in analysing the clustered binary data in stratified design