Inconsistent Closed Captures model results

questions concerning analysis/theory using program MARK

Inconsistent Closed Captures model results

Postby wjensen1 » Thu Nov 01, 2018 1:24 pm

We are using Closed Captures with "Full Likelihood p and c" and are getting different model ranks among different computers (all running Mark 9.0), and results are also changing among re-runs of model sets per the same computer. The number of parameters also sometimes automatically sets to the correct value but sometimes uses an apparent default value. Setting the number of parameters (Adjustments menu) doesn't seem to help. Additionally, the detail of Real Parameter estimates (p, c, and f(0)) sometimes varies among computers for selecting output from the same models. Please help explain. Thanks.
wjensen1
 
Posts: 6
Joined: Tue Jul 25, 2017 12:39 pm

Re: Inconsistent Closed Captures model results

Postby cooch » Thu Nov 01, 2018 3:49 pm

wjensen1 wrote:We are using Closed Captures with "Full Likelihood p and c" and are getting different model ranks among different computers (all running Mark 9.0), and results are also changing among re-runs of model sets per the same computer. The number of parameters also sometimes automatically sets to the correct value but sometimes uses an apparent default value. Setting the number of parameters (Adjustments menu) doesn't seem to help. Additionally, the detail of Real Parameter estimates (p, c, and f(0)) sometimes varies among computers for selecting output from the same models. Please help explain. Thanks.


Without seeing the data set, and the models you're trying to fit to those data, there is absolutely no way to answer this sort of question. If you send me the .fpt and .dbf files offline, I can take a look.
cooch
 
Posts: 1628
Joined: Thu May 15, 2003 4:11 pm
Location: Cornell University

Re: Inconsistent Closed Captures model results

Postby gwhite » Thu Nov 01, 2018 4:31 pm

I suspect you have a really flat likelihood, so that minor rounding issues become important. You can verify this by letting me see the estimates and their SE for the real estimates.

Try setting the number of threads used on each of your computers to 1, so that each computer is performing calculations in exactly the same order. With multiple threads being used, computations across the set of encounter histories is random based on Windows managing the multiple threads. You set the number of threads used under the File | Preferences menu choice.

If you get different results on different machines when threads=1, then send me the MARK DBF/FPT files!
gwhite
 
Posts: 329
Joined: Fri May 16, 2003 9:05 am

Re: Inconsistent Closed Captures model results

Postby wjensen1 » Fri Nov 02, 2018 1:50 pm

Thanks for the tip, Gary. Looks like I'm getting same results now on one machine with threads=1. However, only have access to one machine at the moment. Here are the estimates from the time-dependence model:


Real Function Parameters of {p(t), c(t), f(0)}
95% Confidence Interval
Parameter Estimate Standard Error Lower Upper
-------------------------- -------------- -------------- -------------- --------------
1:p 0.4049080 0.0384482 0.3322997 0.4819304
2:p 0.3092783 0.0469289 0.2254706 0.4078343
3:p 0.3880598 0.0595342 0.2795183 0.5089745
4:p 0.2682925 0.0691960 0.1552289 0.4225190
5:p 0.4666668 0.0910839 0.2992616 0.6419310
6:p 0.1250000 0.0826797 0.0314461 0.3859686
7:p 0.8571429 0.0935219 0.5731673 0.9640403
8:p 0.7816525E-012 0.8841117E-006 -0.1732858E-005 0.1732860E-005
9:p 1.0000000 0.8393649E-004 0.9337781E-296 1.0000000
10:p 0.1946686 168.84889 0.2383329E-304 1.0000000
11:c 0.3939393 0.0601452 0.2840341 0.5157377
12:c 0.3645833 0.0491238 0.2746528 0.4650796
13:c 0.3196723 0.0422213 0.2430947 0.4073890
14:c 0.4285715 0.0429108 0.3472586 0.5139333
15:c 0.2448980 0.0354680 0.1821369 0.3208027
16:c 0.4026845 0.0401783 0.3270205 0.4832832
17:c 0.2049690 0.0318144 0.1495446 0.2743093
18:c 0.4347827 0.0390689 0.3603224 0.5123069
19:c 0.2085889 0.0318239 0.1529953 0.2777589
20:f0 0.2141405E-007 0.3511316E-004 0.1136306E-010 0.4035544E-004
wjensen1
 
Posts: 6
Joined: Tue Jul 25, 2017 12:39 pm

Re: Inconsistent Closed Captures model results

Postby gwhite » Fri Nov 02, 2018 3:12 pm

As you can see, the estimate of f0 is zero. Unless you are putting some constraints on the p's and c's with a design matrix, the p(t) c(t) model is not identifiable. I also see that one of the p estimates equal 1, which further suggests this model is inappropriate.
gwhite
 
Posts: 329
Joined: Fri May 16, 2003 9:05 am

Re: Inconsistent Closed Captures model results

Postby cooch » Fri Nov 02, 2018 3:18 pm

gwhite wrote:As you can see, the estimate of f0 is zero. Unless you are putting some constraints on the p's and c's with a design matrix, the p(t) c(t) model is not identifiable. I also see that one of the p estimates equal 1, which further suggests this model is inappropriate.


Which is discussed in considerable detail at various points in Chapter 14 of the MARK book. See in particular section 14.3.1, and 14.11
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 10 guests

cron