PLEASE READ if you are using MLOGIT links

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

PLEASE READ if you are using MLOGIT links

Postby jlaake » Sat Jan 29, 2022 2:09 pm

In working with Chris Solomon on his problem it appears that I may have a bug lurking in the code when Psi parameters are fixed to 0. What Chris noticed was that the real parameter value didn't match what you would expect with the beta value. At this point I'm not entirely sure what it is but it appears that it can be fixed by setting mlogit0=TRUE as an argument to mark or mark.wrapper functions. That argument uses the logit link for the fixed 0 Psi parameters. Alternatively you can use logit link for Psi in general as Chris noted. Sorry if this has caused anyone problems. More to follow as I diagnose the problem.
jlaake
 
Posts: 1417
Joined: Fri May 12, 2006 12:50 pm
Location: Escondido, CA

Re: PLEASE READ if you are using MLOGIT links

Postby jlaake » Tue Feb 01, 2022 6:48 pm

Good news: So it turns out that it wasn't a bug in the code but both RMark and MARK have been allowing fixed parameter values within an mlogit set and this should probably have not been allowed with the coding in MARK for mlogit and other links that are constrained to sum to 1. So in the newest version of MARK that is being distributed now, this is no longer allowed which means that mlogit with fixed 0 values can ONLY be run with mlogit0=TRUE which uses a logit link for the fixed values and since they are 0 they don't affect the mlogit sum. I'm also going to make a similar change in RMark (Gary is faster than I am).

Bad news: if you have fixed 0 mlogit values in past analyses without setting mlogit0=TRUE, then your results are incorrect unfortunately. I sincerely apologize for letting this sneak through and hope there aren't too many of you out there that have been affected by this issue.

--jeff
jlaake
 
Posts: 1417
Joined: Fri May 12, 2006 12:50 pm
Location: Escondido, CA

Re: PLEASE READ if you are using MLOGIT links

Postby X33277 » Sun Apr 10, 2022 10:45 pm

Does this mean this applies to MARK users as well? I did a bunch of analyses where I had MLOGIT set to 0 for one of the transitions. I re-ran the top model removing the constraint on the MLOGIT and the estimates appear to be same. The transition was biologically impossible which is why I had imposed the 0. Should I be worried about the candidate model set?
X33277
 
Posts: 10
Joined: Tue Mar 12, 2019 9:02 pm

Re: PLEASE READ if you are using MLOGIT links

Postby jlaake » Mon Apr 11, 2022 9:23 am

Thanks for posting this. I have been meaning to circle back to this thread. Short answer is yes because MARK does all the work and RMark only prepares an input file for MARK. BUT, the case you mention in which a transition is fixed to 0 always because it is not possible, seems to have always worked. Fortunately, that is the typical use of fixing to 0, so in most cases, all is good. The situation which identified a problem was one where the transition was only fixed to 0 for certain time intervals. If you want to be certain, download the newest version of MARK and rerun with the fixed 0 values in the mlogit set.
jlaake
 
Posts: 1417
Joined: Fri May 12, 2006 12:50 pm
Location: Escondido, CA


Return to RMark

Who is online

Users browsing this forum: No registered users and 13 guests