RMark: dot notation for missing data

posts related to the RMark library, which may not be of general interest to users of 'classic' MARK

RMark: dot notation for missing data

Postby violetblue » Tue Jun 29, 2021 6:23 pm

Hi, wondering if anyone can shed some light on how RMark deals with dot notation for missing data?

I am running MSORD models with missing data for the 3rd secondary occasion of the second year
and I would like yearly abundance.

When I give the missing occasion a '.' in the .inp and run in RMark,
I get an abundance of 1493 for that year, but the SE is unreasonable large.

However, when I give the missing occasion a '0' in the .inp, fix that occasion to 0 and run in RMark,
I get an abundance of 1565 for that year with SE = 21.

I realise the latter is the best approach, but I would like to explain the difference between the estimates of 1493 and the 1565.

If anyone can explain the difference in their calculation I'd be very grateful! :)
violetblue
 
Posts: 23
Joined: Tue Nov 06, 2018 11:16 pm

Re: RMark: dot notation for missing data

Postby jlaake » Tue Jun 29, 2021 10:08 pm

So this has nothing to do with RMark as it is MARK that does the calculation and handles dot notation. I would expect the results to be the same as the dot in MARK should be equivalent to a p of 0. That said there is nothing wrong about filling in with 0 and setting p to 0. Now recognize that multistate models can have multiple modes so you may want to try simulated annealing. How did deviances and AIC compare?
jlaake
 
Posts: 1417
Joined: Fri May 12, 2006 12:50 pm
Location: Escondido, CA

Re: RMark: dot notation for missing data

Postby violetblue » Wed Jun 30, 2021 5:51 pm

Thank you so much for your help.

They have exactly the same AICc and Deviance.
AICc = 180046.1, Deviance = 179591.4

I'm a bit hesitant to try simulated annealing because the model already takes about 8 hours to run.

Thanks:)
violetblue
 
Posts: 23
Joined: Tue Nov 06, 2018 11:16 pm

Re: RMark: dot notation for missing data

Postby violetblue » Mon Jul 05, 2021 9:10 pm

Thank you very much to all who helped resolve this issue in emails external to this forum.
I'm posting the essence of the response from Gary White for others who are interested:

"..using dots to specify a missing occasion in one state in a multi-state model where other states are not missing leads to a lack of identifiability. You have to fix the p to zero for the missing occasion.

Here’s why. Assume just a simple multi-state model with 2 states, A & B, for this example. The encounter history A0A requires that you consider the probability the animal moved to state B during the second occasion with probability psiAB and then back to state A with psiBA, plus the probability that the animal stayed in state A with probability psiAA and then again psiAA. The 0 value specified in the history specifies the animal was not seen, but if p in state B was zero because of no surveys, the model can’t know this without you actually fixing p to zero.

Hope this helps you understand why using dots for a missing occasion in one state is inappropriate. You have to fix the p value to zero."
violetblue
 
Posts: 23
Joined: Tue Nov 06, 2018 11:16 pm


Return to RMark

Who is online

Users browsing this forum: No registered users and 16 guests

cron