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]

st: RE: Regression analyses for matched data

From   "Hussein, Mustafa (Mustafa Hussien)" <>
To   "" <>
Subject   st: RE: Regression analyses for matched data
Date   Fri, 25 Oct 2013 16:59:47 +0000

Hi Pooja,

If you have multiple observations for your cohort over time (i.e. a longitudinal dataset) then, you could use xtgee and test for the best working correlation structure for your data. If your data are not longitudinal but otherwise clustered (e.g. within some geographic region, counties, neighborhoods, etc) then you could run just GLMs with diagnostics. For count data with multiple zeros, you could run hurdle models, which are two-part models where first you estimate the probability of a count being more than zero, and then run a count data model (Poisson or negative binomial) for those with counts more than zero.

I hope that helps

From: [] on behalf of Pooja Desai []
Sent: Friday, October 25, 2013 12:28 AM
Subject: st: Regression analyses for matched data

Hello Statalisters,

I have data which has patients on antipsychotic medications. There are two
groups: cases--on multiple antipsychotics and controls--on single
antipsychotics. The two groups are matched (1:1 match) based on their
duration of exposure to an antipsychotic. I want to run some regression
models on them to look at cost and healthcare utilization outcomes. I am
not sure if I should use xtgee (since it is correlated data) or glm with
vce(cluster clustervar).

Another followup question is that I have one count outcome which has many
zeros. If I do use xtgee, is there a way to carry out zero inflated Poisson
or zero inflated negative binomial regression using it?

Any advice will be appreciated.

Thanks in advance.

Pooja Desai
Health Outcomes and Pharmacy Practice
The University of Texas at Austin
*   For searches and help try:

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index