R help - Jan 2016 - Constrained Poisson model / Bayesian Poisson model

Hi all,

I am dealing with a problem about my linear Poisson regression model  
(link function=identity).

I am using the glm()-function which results in negative coefficients,  
but a negative influence of the regressors wouldn't make sense.

(i) Is there a possibility to set constraints on the regression  
parameters in glm() such that all coefficients are positive? Or is  
there another function in R for which this is possible?

(ii) Is there a Bayesian version of the glm()-function where I can  
specify the prior distribution for my regression parameters? (e.g. a  
Dirichlet prior s.t. the parameters are positive)

All this with respect to the linear Poisson model...

Thanks in advance!
Best,
Mara

> On Jan 22, 2016, at 7:01 AM, mara.pfleiderer at uni-ulm.de wrote:
> 
> Hi all,
> 
> I am dealing with a problem about my linear Poisson regression model (link
function=identity).
> 
> I am using the glm()-function which results in negative coefficients, but a
negative influence of the regressors wouldn't make sense.
Negative coefficients merely indicate a lower relative rate. You need to be more
specific about the exactly data and model output before you can raise our
concern to a level where further comment can be made.

> 
> (i) Is there a possibility to set constraints on the regression parameters
in glm() such that all coefficients are positive? Or is there another function
in R for which this is possible?
> 
> (ii) Is there a Bayesian version of the glm()-function where I can specify
the prior distribution for my regression parameters? (e.g. a Dirichlet prior
s.t. the parameters are positive)
> 
> All this with respect to the linear Poisson model...
As I implied above, the word "linear" means something different than
"additive" when the link is log().

--

David Winsemius
Alameda, CA, USA