[R] Regression on stratified count data

[peter dalgaard]

On Apr 24, 2013, at 06:15 , meng wrote:

> Hi all:
> For stratified count data,how to perform regression analysis?
> 
> My data:
> age case oc count
> 1       1     1    21
> 1       1     2    26
> 1       2     1    17
> 1       2     2    59
> 2       1     1    18
> 2       1     2    88
> 2       2     1     7
> 2       2     2   95
> 
> age:
> 1:<40y
> 2:>40y
> 
> case:
> 1:patient
> 2:health
> 
> oc:
> 1:use drug
> 2:not use drug
> 
> My purpose:
> Anaysis whether case and oc are correlated, and age is a stratified variable.
> 
> My solution:
> 1,Mantel-Haenszel test by using function "mantelhaen.test"
> 2,loglinear regression by using function glm(count~case*oc,family=poisson).But I don't know how to handle variable "age",which is the stratified variable.

The canonical way is to fit the model without 2nd order interaction: 

count~case*oc*age-case:oc:case  . 

(It may take the back of an envelope or two to realize that this is equivalent to the common OR assumption of the MH test.) 

Alternatively, use logistic regression 

glm(case ~ oc + age, family=binomial, weight=count, data=dd)

(NB: it is important that case is a factor here!)

-- 
Peter Dalgaard, Professor,
Center for Statistics, Copenhagen Business School
Solbjerg Plads 3, 2000 Frederiksberg, Denmark
Phone: (+45)38153501
Email: pd.mes at cbs.dk  Priv: PDalgd at gmail.com