new GOF test | Fletcher c-hat

announcements (new versions, changes, bugs, installation problems...) related to program MARK

new GOF test | Fletcher c-hat

Postby cooch » Tue Jan 05, 2016 2:55 pm

Greetings --

A new goodness of fit method has been implemented in MARK -- referred to as the 'Fletcher c-hat', given the conceptual origins in an interesting paper by David Fletcher:

Fletcher, D. (2012) Estimating overdispersion when fitting a generalized linear model to sparse data. Biometrika, 99, 230-237.

Gary has implemented the Fletcher c-hat in MARK, and it seems to work very well (based on simulation experiements conducted to date), for certain types of models. Beyond its 'robustness' , the Fletcher c-hat takes only milliseconds for MARK to calculate (in fact, MARK generates the Fletcher c-hat for all models as they are optimized), which is arguably an advantage over (say) the median c-hat.

The details are in the newly revised Chapter 5, section 5.8.
cooch
 
Posts: 1628
Joined: Thu May 15, 2003 4:11 pm
Location: Cornell University

Re: new GOF test | Fletcher c-hat

Postby telensky » Wed Mar 08, 2017 10:22 am

Hmm.. I got Fletcher c-hat = 239.29231 in CJS model for a species with really good data, with handling transients with time effect and an ageclass (https://github.com/oliviergimenez/abundance_estimation). Simple CJS model with just time effects (no transients taken into account) has Fletcher c-hat = 29.70. Is it caused by missing visits in capture history?:

5574 .0001100...
5575 .0001000...

(here http://warnercnr.colostate.edu/~gwhite/mark/mark.htm it is mentioned that "problems are caused by dots in the encounter history" - see point 237).
telensky
 
Posts: 2
Joined: Wed Mar 08, 2017 8:52 am

Re: new GOF test | Fletcher c-hat

Postby cooch » Wed Mar 08, 2017 11:00 am

telensky wrote:Hmm.. I got Fletcher c-hat = 239.29231 in CJS model for a species with really good data, with handling transients with time effect and an ageclass (https://github.com/oliviergimenez/abundance_estimation). Simple CJS model with just time effects (no transients taken into account) has Fletcher c-hat = 29.70. Is it caused by missing visits in capture history?:

5574 .0001100...
5575 .0001000...

(here http://warnercnr.colostate.edu/~gwhite/mark/mark.htm it is mentioned that "problems are caused by dots in the encounter history" - see point 237).


Really good data doesn't have missing values, so perhaps isn't the best description.

And yes, missing values (dots) are the problem. As per the MARK book (that incomplete document you refer to in another post)

First, losses on capture or dots in the encounter history will create encounter histories that are not considered in the total number of possible encounter histories. That is, the total number of possible
encounter histories is based on no missing data.


Meaning, the Fletcher approach won't work. Nor will most other approaches that are based on expected frequencies of encounter histories. However, resampling approaches (say, median c-hat) might.
cooch
 
Posts: 1628
Joined: Thu May 15, 2003 4:11 pm
Location: Cornell University


Return to software problems/news

Who is online

Users browsing this forum: No registered users and 11 guests