RD(Huggins): any predictor inflates Nhat

[geofgivens]

Hi folks.

I don't believe the results I'm getting and I suspect that something must be wrong in my analysis or data.

I'm (trying to) attach an example dataset and DBF file. The analysis is Robust Design with Huggins model within primary sessions. Time intervals are 0, 17, 0, 0, 6, 0. However, the final occasion in the final primary session is a dummy, set to have no captures or recaptures, with p=c=0 fixed. The other components of the basic analysis are: constant survival, p(time+maybe other individual predictors), c=p, gamma"=0, gamma'=1. I have fiddled with these settings but my findings are qualitatively similar.

Below I have tabulated the 1095 histories (without the dummy occasion), their frequencies, and the mean values of the "iscore" individual predictor.

My problem is that adding *any* individual covariate increases the abundance estimate. (It is the 3rd abundance that interests me.) I don't believe it because:

• No predictor or combination of predictors seems to decrease Nhat. (My enclosed dataset only contains 3 of about 20 predictors)
• Even gaussian noise (a 4th "predictor" in the example data) increases Nhat.
• I have a gold standard, completely independent non-cap/recap dataset yielding Nhat=4880 (4400, 5410) so the changes to Nhat illustrated here are huge compared to the "right answer".
• There is a strong negative correlation between AICc and Nhat.
• Qualitatively, all of this is also true if I switch to a POPAN analysis (by deleting the trailing 0 in the catch histories and making the time intervals 1, 17, 1, 1, 6.)
• The table below seems to offer no clues.
• The same issue arises if I use RMark rather than MARK (to check for making user-errors in one or the other)

So...Have you seen this sort of behavior before? Is it a symptom of a problem with the analysis? Is it a symptom of an issue with the data?

I'm baffled and can't publish results under these circumstances. Any ideas you can provide would be greatly appreciated!

Geof

PS. I can't figure out how to attach files to a forum posting. I'll figure this out if possible but in the meantime I can email them to anyone willing to help.

Code: Select all

This is not the data but a descriptive table; see above for explanation

ch    freq mean(iscore)
000001 315 0.75
000100 219 1.00
100000 172 0.76
001000 130 0.89
010000 102 0.76
000010  66 1.02
001001  15 1.70
000101  13 1.62
110000  10 0.90
001100   8 1.56
010001   7 1.64
100001   7 1.43
101000   6 1.08
011000   5 1.50
100100   4 1.31
000011   3 1.25
010100   3 1.83
000110   2 0.25
001010   2 2.00
111000   2 1.13
001101   1 1.00
100010   1 2.50
110001   1 1.50
110100   1 2.75


[cooch]

PS. I can't figure out how to attach files to a forum posting. I'll figure this out if possible but in the meantime I can email them to anyone willing to help.

That's because, you can't. From the forum FAQ (which you should check next time):

4. - why can't I upload attachments, create an avatar, or add signatures? Why can't I change my email address after I've registered? Simple. Those features are turned off, since they will either bog down the system - and it's users (especially attachments), and since they generally take up server bandwidth and space for little general benefit. Various other restrictions are imposed (like restrictions on ability to change email) to minimize the risk of spammers using the forum.

[geofgivens]

Fair enough. I did read the FAQ and all relevant posts since 2010 before posting this message. At the time, however, I wasn't looking for advice on attaching files.

Can I email you the files?