[Mediation-information] handling treatment for mediation R package

Teppei Yamamoto teppei at MIT.EDU
Mon Nov 7 03:57:55 CET 2011


Hi Hiroko,

The package unfortunately does not support a categorical treatment at 
the moment. It currently assumes either a binary treatment (factor, 
logical TRUE/FALSE or numeric 0/1) or a continuous numeric treatment, so 
the output you obtained is probably nonsensical.

We are planning to include support for such a case in the next major 
update, but one thing you could try now is to drop one of the genotypes 
from the sample and estimate the mediation effects with respect to the 
remaining two types. Assuming that the covariates in the model are well 
balanced across treatment groups and your sample size is not too small, 
you should get an estimate similar to the full analysis. Alternatively, 
you could code up the algorithm for a three-category treatment on your 
own or modifying our function.

Best,
Teppei


Hiroko Solvang wrote:
> Hi,
> I am interested in mediation structure applying to genomic data analysis.
> I think your developed software is very useful for such analysis.
> I have one question for 'treatment' that you described as 'binary' or
> 'continous' variables.
> In our case, genotype variation, e.g. AA, AB and BB, three categorical
> variables would be used as 'treatment'.
> Is this package available for such three categorical data?
> I have just tried running this package using the three categorical data
> (using 'as.factor'), but I didn't get any error message. I just wonder
> if such data can be correctly analyzed on this package or not.
> I am looking forward to receive your answer.
> Best regards,
> Hiroko Solvang
> Oslo University Hospital, Radiumhospitalet, Norway

-- 
====================================
Teppei Yamamoto
Assistant Professor
Department of Political Science
Massachusetts Institute of Technology
http://web.mit.edu/teppei/www/
====================================


More information about the Mediation-information mailing list