adjust beta SE and CI for overdispersion in RMark

[henrik]

Dear RMark users,

I have fitted a CJS model with overdispersion and I wish to adjust the SE and CI for the betas.

Is it done as described on ?compute.real, i.e. "adjusting the estimated standard error of the beta parameters by multiplying it by the square root of chat to adjust for over-dispersion" and "A normal 95% confidence interval is computed for the link estimate [which I assume is the beta parameter referred to in the previous sentence] (estimate +/- 1.96*se)"? Can the adjusted SE/CI for betas be extracted from the compute.real object?

I find the help text for adjust.chat slightly inconsistent. In ?adjust.chat/Description: "Adjust value of over-dispersion constant [...] which modifies [...] estimated standard errors". In Details: ""standard errors [...] in result$beta [...] always assume chat=1". In the RMark manual C-34: [adjust.chat] does not adjust standard error [...] but that is done with functions [...] e.g. get.real". While get.real adjusts real parameters, I find no corresponding function for betas. Maybe the adjust.chat function could be extended to include adjustment for overdispersion for other relevant variables such as SE, CI for beta and real parameters?

Thanks in advance for your help!

Best regards,

Henrik

[jlaake]

I'll look at the documentation and provide some clarification. The reason the values for results$beta are not changed is because I keep them for any further adjustments for chat. Also, none of the results are changed in the output file which is static obviously. You can get the chat-adjusted se and ci for beta from summary(model). Below is an example with the dipper data. Note that there is a problem with the chat adjustment with the sin link that was reported to me and I've not fixed yet. As far as I know it works with all of the other link functions.

Dipper example--
> data(dipper)
> mod=mark(dipper)

Output summary for CJS model
Name : Phi(~1)p(~1)

Npar : 2
-2lnL: 666.8377
AICc : 670.866

Beta
estimate se lcl ucl
Phi:(Intercept) 0.2421484 0.1020127 0.0422035 0.4420933
p:(Intercept) 2.2262658 0.3251093 1.5890516 2.8634801

> mod$chat=2
> summary(mod)
Output summary for CJS model
Name : Phi(~1)p(~1)

Npar : 2
-2lnL: 666.8377
AICc : 670.866
chat : 2
QAICc: 337.4472

Beta
estimate se lcl ucl
Phi:(Intercept) 0.2421484 0.1442677 -0.04061638 0.5249132
p:(Intercept) 2.2262658 0.4597740 1.32510880 3.1274228

[henrik]

Thanks a lot for your rapid answer!

/Henrik