Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: comparing results from cluster and stratified analysis

From   Steven Samuels <>
Subject   Re: st: comparing results from cluster and stratified analysis
Date   Thu, 30 Dec 2010 14:25:30 -0500

You can certainly compare the results, but it won't hurt to do the non- stratified analysis on the reduced data set. A formal test of the effect of family would require a bootstrap or a test of the Gamma parameter in the random effects frailty model; but the Gamma assumption is quite restrictive.

Perhaps you misunderstand the role of clusters. The -vce()- options, including -vce(cluster ), affect only the estimation of standard errors. You could use any -vce()- option for a non-stratified analysis or stratified analysis. So, in the stratified analysis, family could be both a cluster and a stratum variable.

To drop all groups with tied values for covariates x1 x2 x3 x4:.

*********First Principles********************
bys group: gen gc1= _N
bys group x1 x2 x3 x4: drop if _N==gc1
duplicates drop group x1 x2 x3 x4


Steven J. Samuels
18 Cantine's Island
Saugerties NY 12477
Voice: 845-246-0774
Fax:    206-202-4783

On Dec 29, 2010, at 4:30 AM, Grethe Søndergaard wrote:


I am analysing my data in two steps, using cox-regression analysis in
both. My data set consists of siblings from different families.

First I am using a standard cox-regression analysis, where I include a
cluster term to take into account that some of the persons in my data
are related.
Next, I am using using a stratified cox-regression analysis, where I
use a “sibling” stratum variable, to stratify my analysis on persons
who are siblings.

I want to compare the results from the cluster analysis with the
result from the stratified analysis – and know I am having doubts of
whether I am doing this the right way. In the stratified analysis,
only siblings who are discordant on both exposure and outcome will
contribute to the analysis. However, in the cluster analysis all
siblings in the data set will contribute (as far I understand). But
can you actually compare these results?

If no - Is there a way to delete all the individuals, who are not
discordant on exposure and outcome with at least one of his/her
siblings, before running the cluster analysis?


*   For searches and help try:

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index