Combining share and fixed?

[JeffHostetler]

I'm trying to apply the robust design to a dataset that includes some radio-collared animals, for which we want to set p and c to 1. I'm interested in including models where p & c are the same for non-collared animals (such as p.session=list(formula=~session,share=T)) or where they always differ by the same amount (such as p.c.session=list(formula=~c+session,share=T)). When I try to combine these models and fixing p & c, it either doesn't fix c or gives me an error message. I first tried fixing p using:

Code: Select all

p_collared.Y=as.numeric(row.names(ShisRDall.ddl$p[ShisRDall.ddl$p$collared=='Y', ]))
p.c.session=list(formula=~session+c, share=T, fixed=list(index=p_collared.Y,value=1))

which does not fix any c values. I then tried deleting rows from the p and/or c design data, which gave me the error "One or more formulae are invalid because the design matrix has all zero rows for the following nonfixed parameters."

It now looks like models where p & c vary independently work best for this dataset, but I'm still curious: is there a way to do this?

Thanks,

Jeff

[jlaake]

Jeff-

You don't want to delete design data because that would set p to 0 when you want to set it to 1. Also, as you found I don't think you can delete design data when you are sharing. Sharing simply combines the DM for p and c to use the same columns. It does not mean that p=c unless that is the model you specified. Thus if you want to fix p and c parameters then you need to specify the indices for c as well as p. You have only done it for p and you have left the c parameters to vary. If you add the corresponding indices for c it should work although I've not tried it.

--jeff

[JeffHostetler]

Jeff,

Thanks! Based on your suggestion, I got it working, but not while using mark.wrapper.

Code: Select all

new.model=mark(ShisRDall.proc,ddl=ShisRDall.ddl,output=F,model.parameters=list(N=list(formula=~1),c=list(fixed=list(index=c_collared.Y,value=1)),
  S=list(formula=~sex+site2+fire), GammaDoublePrime=list(formula=~1,share=T), p=list(formula=~session+c,share=T, fixed=list(index=p_collared.Y,value=1))))


works correctly, but

Code: Select all

p.c.session=list(formula=~session+c,share=T,fixed=list(index=p_collared.Y,value=1)) 
c.fixed=list(fixed=list(index=c_collared.Y,value=1))
...
ShisRDall.CML=create.model.list("Robust")
model.list=mark.wrapper(ShisRDall.CML,data=ShisRDall.proc,ddl=ShisRDall.ddl,output=F)


does not fix c and

Code: Select all

c.fixed=list(fixed=list(index=c_collared.Y,value=1)) #ORDER SWITCHED
p.c.session=list(formula=~session+c,share=T,fixed=list(index=p_collared.Y,value=1)) 
...
ShisRDall.CML=create.model.list("Robust")
model.list=mark.wrapper(ShisRDall.CML,data=ShisRDall.proc,ddl=ShisRDall.ddl,output=F)


gives a "Error in if (x4 > x2) { : argument is of length zero" message. If these particular models have to be run separately, that's no problem for us.

Thanks again,

Jeff

[jlaake]

Not sure why that is failing wih mark.wrapper but try the following

p.c.session=list(p=list(formula=~session+c,share=T,fixed=list(index=p_collared.Y,value=1)),c=list(fixed=list(index=c_collared.Y,value=1)))

This let's you pair specific model specifications in mark.wrapper rather than using all combinations of p/c models. This should work as long as you do the same with all p/c models.

--jeff

[JeffHostetler]

That works. Thanks, Jeff!