Negative deviances

questions concerning analysis/theory using program MARK

Negative deviances

Postby alexiscerezo » Tue Jun 29, 2021 3:40 pm

Hello all,

I´ve been running standard CJS analyses on capture-recapture bird data from tropical Guatemala, and when looking at the results, I noticed that the deviances were negative, for all models and different species. I'm not sure how to interpret these, since I thought that deviances were always positive. The AICc´s and model results, i.e., parameter and point estimates looked fine though (no ridiculous values for estimates and standard errors). The only thing that is noticeable for my analysis is the capture histories file: data comes from different sites that were sampled on different years in the 1992-2019 interval, so the capture histories files has a lot of "."'s (for years with no data). Otherwise, everything else looks fine.

Many thanks for advice and suggestions.

Alexis
alexiscerezo
 
Posts: 4
Joined: Mon Jun 28, 2021 10:19 pm

Re: Negative deviances

Postby cooch » Tue Jun 29, 2021 8:21 pm

alexiscerezo wrote:...so the capture histories files has a lot of "."'s (for years with no data).


When you say a lot, what do you mean? [As, say, a percentage of occasions?]

Send me the .fpt and .dbf files and I can have a look.
cooch
 
Posts: 1628
Joined: Thu May 15, 2003 4:11 pm
Location: Cornell University

Re: Negative deviances

Postby alexiscerezo » Thu Jul 01, 2021 1:21 pm

Dear Evan,

Many thanks for your reply. 73% of observations where 0's or 1's, the rest where ".". Could you please tell me how to attach .fpt and .dbf files in my next post? I´ll be sending the files for one species, Red-capped Manakin (Ceratopipra mentalis), but I've run analyses on 13 species. Deviances where negative for all species and all models.

Best,

Alexis
alexiscerezo
 
Posts: 4
Joined: Mon Jun 28, 2021 10:19 pm

Re: Negative deviances

Postby cooch » Thu Jul 01, 2021 3:19 pm

alexiscerezo wrote:Dear Evan,

Many thanks for your reply. 73% of observations where 0's or 1's, the rest where ".". Could you please tell me how to attach .fpt and .dbf files in my next post?


You can't attach or post them to the forum. You'll need to send them to me offline. [i.e., via email].
cooch
 
Posts: 1628
Joined: Thu May 15, 2003 4:11 pm
Location: Cornell University

Re: Negative deviances

Postby cooch » Fri Jul 02, 2021 11:13 am

Turns out the negative deviances were an artifact of incorrect use of the 'dot' notation, to code for missing sampling occasions (see relevant sections of chapter 2 and chapter 4). Basically, the encounter data were coded such that 'dots' were used as de facto individul covariates (in a fashion). This might arise if for some reason, you did not attempt to search for (encounter) some marked individuals, but did search for others (although I have trouble imaginging how such a scenario might arise -- maybe a telemetry study where you run out of time and are unable to search for a sub-sample of the tagged individuals?).

So, for example, consider the following 2 histories

Code: Select all
10.001
1.11.1


In the first history, the individual has a 'dot' in the 3rd poisiton in the encounter string, which would indicate that the individual was not 'sampled' (looked for) on that occasion. For the second individual, it was not 'sampled' (looked for) on the second and fifth occasions.

This is incorrect. The 'dot' is used to indicate the situation where *all* individuals were not 'looked for' (say, beacause you were unable to go into the field). At which point, there would be a 'dot' in that position in the encounter file for *all* individuals in the file.
cooch
 
Posts: 1628
Joined: Thu May 15, 2003 4:11 pm
Location: Cornell University


Return to analysis help

Who is online

Users browsing this forum: No registered users and 14 guests