[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: Adjusting for clustering within groups

From   Ronan Conroy <>
Subject   Re: st: Adjusting for clustering within groups
Date   Tue, 8 Apr 2008 12:31:50 +0100

On 8 Apr 2008, at 03:11, Ashwin Ananthakrishnan wrote:
I am currently using Stata 9.2. My analysis involves
logistic regression of a dichotomous outcome (dead /
alive) across various predictors. My main predictor of
interest is hospital type (Small, medium, big). How do
I adjust for clustering within each type of hospital?
There are two things to consider. The first is that hospital type may influence prognosis, so that the death rate is different in each type. You can stratify your analysis by hospital, or you can include hospital size as a predictor variable.

If you have a variable coded 1 to 3 for small, medium and large hosptals, and a variable that identifies each hospital

xi: logistic dead i.hospital_size p1 p2 p3, cluster(hospital_id)

looks at the effects of three predictors (p1 to p3) and uses -xi- to create dummy variables to represent hospital size. The cluster option allows for clustering within hospitals.

The largest of the 'big' hospitals may contribute more
patients to the analysis than the smaller of the 'big'
hospitals. Do I need to adjust for this phenomenon?
i.e. if we hypothesize that larger hospitals have
better outcomes, isn't it appropriate that the larger
of the big hospitals contribute more patients and
hence have a larger weight?
Larger hospitals contribute more patients and so their death rate can be estimated with greater precision. But downweighting them for this reason makes no sense.

Or do I need to adjust for this? and how do I do it?

P    Before printing, think about the environment
Ronan Conroy
Royal College of Surgeons in Ireland
Epidemiology Department,
120 St Stephen's Green, Dublin 2, Ireland
+353 (0)1 402 2431
+353 (0)87 799 97 95

*   For searches and help try:

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