Regression using MARK estimates as dependent variables

[Chris Jones]

I'm running an experiment where I expose prey to varying densities of predators to look at the relationship between prey survival and predator density. Prey survival will be estimated using the robust design. My initial thoughts were to regress the estimates of survival obtained using MARK against the appropriate predator density value (known) and other variables such as study site and alternative prey availability. A comment on page 7-1 of the most recent users' guide suggests that this is " not valid statistically". Why is this, and could someone suggest an appropriate strategy please?

[dief]

The short answer to why the approach you described is not appropriate is because you are then doing statistics on statistics. A preferred alternative is to develop a model within MARK in which survival is a function of predator density - you could then compare this model to one in which survival is independent of predator density. If the former is the better model given your data then you can use that model to provide estimates of survival under different predator densities.

[cooch]

I'm running an experiment where I expose prey to varying densities of predators to look at the relationship between prey survival and predator density. Prey survival will be estimated using the robust design. My initial thoughts were to regress the estimates of survival obtained using MARK against the appropriate predator density value (known) and other variables such as study site and alternative prey availability. A comment on page 7-1 of the most recent users' guide suggests that this is " not valid statistically". Why is this, and could someone suggest an appropriate strategy please?

In addition to an earlier answer, because for many (but not all) models, the estimates of survival are not independent, and thus not appropriately treated as 'response' variables in a regression.