[R] Computing a Mantel-Haenszel chi-square using a continuous variable as the matching criterion

David Winsemius dwinsemius at comcast.net
Thu Oct 7 23:11:21 CEST 2010


On Oct 7, 2010, at 4:41 PM, Peter Dalgaard wrote:

> On 10/07/2010 08:24 PM, David Winsemius wrote:
>>
>> On Oct 7, 2010, at 2:21 PM, Barth B. Riley wrote:
>>
>>> Dear list
>>>
>>> I would like to compute a Mantel-Haenszel chi-square in which the
>>> matching variable is a continuous variable. The MH chi-square is
>>> used to assess the relationship between two categorical variables at
>>> each level or strata defined by a third variable. Specifically I
>>> would like to know if there is a straightforward way to divide the
>>> matching variable into levels, in which each level has a minimum of
>>> 20 cases. Any information would be greatly appreciated.
>>
>> Why? What makes you think matching would be valuable? (...or even
>> valid, for that matter.)
>
> What makes you think he has a choice?

He was asking for advice about how to construct matching strata using  
a continuous variable.

> Matching is generally part of the
> design and data may already have been collected...

But it sounded as though the matching was not a "done deal" (since the  
strata had not yet been defined) and I was attempting to clarify the  
purposes.

>
> For small-strata analyses look at clogit() in survival package. For
> cutting a continuous variable into roughly equal-sized strata use a
> combination og  cut() and quantile() or (AFAIR) cut2() from one of  
> Frank
> Harrell's packages.

I replied offlist with such advice.

> Finally, Breslow+Day in their classic book on
> case-control studies suggest replacing matched analysis by ordinary
> logistic regression with the effect of the matching variable modeled  
> by
> a suitably high-order polynomial.

Exactly. Breslow and Day is my preferred citation for pitfalls of  
matching. Neither matching nor covariate adjustment should be  
performed when there is a causal ( or other ) dependence of the  
adjustment covariate and the predictor of interest. The classic  
example of an erroneous use of matching in the medical field was  
publication that described adjustment for menopausal bleeding in  
analyses of potential causes of uterine cancer.

-- 
David.

>
> -- 
> Peter Dalgaard
> Center for Statistics, Copenhagen Business School
> Phone: (+45)38153501
> Email: pd.mes at cbs.dk  Priv: PDalgd at gmail.com

David Winsemius, MD
West Hartford, CT



More information about the R-help mailing list