Population estimates from small data sets?

[murray.efford]

Yes, Density has trouble automatically finding initial values for the parameters when there are no recaptures in a session. The way around this is to click on 'Manual' in Options | ML SECR and give it something plausible (I'd guess 2/ha, g0=0.1, sigma=40 m in your case, assuming a halfnormal detection function). Hope this works! (Remember to pool detection function parameters across sessions - 'constant' for g0 and sigma on the between-sessions page of Options | ML SECR).

[cooch]

I beg to differ: there may be no recaptures purely by chance, especially if the treatment has lowered density (we have very little to go on in the original post). Also: different methods for different treatments will give misleading results if the methods differ in bias, as they often do. Better to apply one method that works.

Depends on your comfort level as to which heroic set of assumptions you want to make. I simply make different ones than you do. In the absence of recaptures in one treatment, you're left having to make some assumptions however you approach the problem - assumptions which appear to be not easily testable (in this case).

[jaxzwolf]

Yes, Density has trouble automatically finding initial values for the parameters when there are no recaptures in a session. The way around this is to click on 'Manual' in Options | ML SECR and give it something plausible (I'd guess 2/ha, g0=0.1, sigma=40 m in your case, assuming a halfnormal detection function). Hope this works! (Remember to pool detection function parameters across sessions - 'constant' for g0 and sigma on the between-sessions page of Options | ML SECR).

Hi again murray,

Thank you so much for this advice! It worked! I was able to get ML density estimates for each of my four sites!

I'm curious about setting an appropriate buffer width for the purpose of these analyses. The default is at 100 m, but will I need to change that since I input the initial values manually?

[murray.efford]

It's great you can make a some progress. Remember when you come to interpret the results that you've got there by assuming the same detection function applies in the two treatments. Without recaptures in both there's no hope of testing this. For good biological reasons, detection functions usually do change when density changes; you might mention this, but such issues are probably minor given the likely uncertainty in your estimates. A nonspatial (e.g. Zippin) approach does not provide a way out because you would then still have to assume equal movement.

Buffer widths in Density are not critical. unless they are too small. Having got an estimate of sigma you could safely set the buffer width to 4 x sigma or 5 x sigma and recompute. Try ML SECR | 2-D integration in the Help index if you want to get in more deeply.

[jaxzwolf]

It's great you can make a some progress. Remember when you come to interpret the results that you've got there by assuming the same detection function applies in the two treatments. Without recaptures in both there's no hope of testing this. For good biological reasons, detection functions usually do change when density changes; you might mention this, but such issues are probably minor given the likely uncertainty in your estimates. A nonspatial (e.g. Zippin) approach does not provide a way out because you would then still have to assume equal movement.

Buffer widths in Density are not critical. unless they are too small. Having got an estimate of sigma you could safely set the buffer width to 4 x sigma or 5 x sigma and recompute. Try ML SECR | 2-D integration in the Help index if you want to get in more deeply.

I want to thank you again for taking the time to walk me through this. You've been enormously helpful, and I finally feel as if I'm making some progress. I really appreciate the advice!