updated abundance estimate after adjusting c-hat

[eronje]

Hi,

I am using Huggins p and c closed capture models (M0, Mt, Mb) and don't understand how to get an updated abundance estimate once I've adjusted c-hat to match what was estimated by the median c-hat test in MARK. I see that real parameter estimates are updated, but get "NOTE: Derived Estimates not corrected for c-hat" when I check abundance. Can someone point me in the right direction?

[cooch]

Hi,

I am using Huggins p and c closed capture models (M0, Mt, Mb) and don't understand how to get an updated abundance estimate once I've adjusted c-hat to match what was estimated by the median c-hat test in MARK. I see that real parameter estimates are updated, but get "NOTE: Derived Estimates not corrected for c-hat" when I check abundance. Can someone point me in the right direction?

First, don't use median c-hat for closed population abundance modeling. Use Fletcher's c-hat (see chapter 14).

Second, c-hat influences AIC calculations, and SE of the reals, but nothing more (if you think about it, if the reals don't change, then neither will the derived estimate of N). But, reasonable to wonder why the SE for derived N doesn't change. My guess -- there was a fair bit of uncertainty about GOF testing for closed population abundance models. In fact, application of Fletcher's c-hat to same is a fairly new progression of things. I'll ask Gary, but I'm fairly sure that we simply don't have a good correction for the log-normal CI for derived N when c-hat > 1.

[eronje]

Thank you, your response is very helpful.

[ehileman]

First, don't use median c-hat for closed population abundance modeling. Use Fletcher's c-hat (see chapter 14).

Second, c-hat influences AIC calculations, and SE of the reals, but nothing more (if you think about it, if the reals don't change, then neither will the derived estimate of N). But, reasonable to wonder why the SE for derived N doesn't change. My guess -- there was a fair bit of uncertainty about GOF testing for closed population abundance models. In fact, application of Fletcher's c-hat to same is a fairly new progression of things. I'll ask Gary, but I'm fairly sure that we simply don't have a good correction for the log-normal CI for derived N when c-hat > 1.

I'm reviving this thread as I am curious to know why log-normal confidence intervals based on c-hat inflated variance-covariance matrices might not be appropriate for derived estimates of N, which Evan seems to be suggesting in this thread. When I calculate them by hand, they seem to produce reasonably inflated confidence intervals. What am I missing? Any insights appreciated.