Doing statistics on derived parameter estimates

[shelly]

I have a theoretical question for the more experienced MARK users or more experienced statisticians than me.

I was wondering what the general consensus is about performing statistical analyses on derived parameter estimates (ie estimates obtained in MARK).

My specific query concerns determining whether disease infection rates (as estimated by transition rates from a multi-state model) vary between sites. Is it appropriate in this case to run multi-state models separately for each site and then use standard statistical tests to test that the derived estimates (either model averaged or just from the best model in each case) do not vary between sites? Or is it more appropriate to run one multi-state model with the sites as groups and compare the AIC of models where infection rates vary with group versus the AIC of models not including group and then determine site effects based on total support for models including “group”?

Also in general is performing standard statistical analyses on MARK parameter estimates good statistical practice?

[cooch]

I have a theoretical question for the more experienced MARK users or more experienced statisticians than me.

I was wondering what the general consensus is about performing statistical analyses on derived parameter estimates (ie estimates obtained in MARK).

My specific query concerns determining whether disease infection rates (as estimated by transition rates from a multi-state model) vary between sites. Is it appropriate in this case to run multi-state models separately for each site and then use standard statistical tests to test that the derived estimates (either model averaged or just from the best model in each case) do not vary between sites? Or is it more appropriate to run one multi-state model with the sites as groups and compare the AIC of models where infection rates vary with group versus the AIC of models not including group and then determine site effects based on total support for models including “group”?

In the majority of instances, the latter.

Also in general is performing standard statistical analyses on MARK parameter estimates good statistical practice?

It depends - most of the tests assume 'independence' and other such things, and ignore estimation/samling covariances. So, for example, 'doing a regression on survival estimates' would, in most instances, be invalid.