Missing data/Skipped sampling

questions concerning analysis/theory using program MARK

Missing data/Skipped sampling

Postby Renata » Sat Feb 09, 2008 12:00 am

This is a topic that was only briefly discussed in the forum, and a while ago, so I'll bring this up again in hopes that something new came up.

I mist-netted birds on 15 sites within a research station; sampling was carried out every 3 months, for a total of 9 sampling occasions. I'm interested in estimating survival rates and, more specifically, if survival rates change with habitat type (there are 5 sites in primary, secondary, and altered forest, respectively; so I want to compare these 3 groups).

However, I have some gaps in the data, meaning that each of the 15 sites was not sampled for at least one of the 9 occasions; the skipped occasions differ among sites.

How can I deal with these missing data? Is there a way to truncate the capture histories, or is there a way to code for missing data?

Renata
Renata
 
Posts: 3
Joined: Sun Nov 19, 2006 9:01 pm

Missing data/Skipped sampling

Postby gwhite » Sat Feb 09, 2008 12:29 pm

Renata:
I'm assuming you want to use the CJS live recaptures data type. I have modified this data type to allow dots in the encounter history to identify an occasion that was not sampled, and hence has p = 0. This approach is probably the easiest for your purposes.

The dot to identify missing occasions is also allowed in the closed captures and multi-state data types, but not in the robust design versions of these data types.

Gary
gwhite
 
Posts: 340
Joined: Fri May 16, 2003 9:05 am

dot for missing groups as well?

Postby ljuliusson » Tue Feb 19, 2008 10:57 pm

Gary, and others:

Can the dot structure be used to indicate missing groups, as well as occasions? If so, is there some additional documentation on this technique? How would it work for my data given my example below, if for example group 3 wasn't sampled for type 2? Would I replace my group indicator "1" with a dot?

/* 1 type 1 */ 1111 1 0 0 0 0 0 0;
/* 2 type 1 */ 1100 0 1 0 0 0 0 0;
/* 3 type 1 */ 1100 0 0 1 0 0 0 0;
/* 4 type 1 */ 0100 0 0 0 1 0 0 0;
/* 5 type 1 */ 1111 0 0 0 0 1 0 0;
/* 6 type 1 */ 0110 0 0 0 0 0 1 0;
/* 7 type 1 */ 1110 0 0 0 0 0 0 1;
/* 1 type 2 */ 1111 1 0 0 0 0 0 0;
/* 2 type 2 */ 1100 0 1 0 0 0 0 0;
/* 3 type 2 */ 1100 0 0 . 0 0 0 0;
...

I am using closed capture modeling.

Thank you,
Lara
ljuliusson
 
Posts: 3
Joined: Wed Sep 26, 2007 12:28 pm
Location: Denver, CO

Postby Katherine McClure » Wed Oct 22, 2008 6:08 pm

Hi there,
I am also trying to figure out what to do about missing data. I'm using a multi-strata robust design to analyze 4 years of prairie dog mark recapture data from 24 different colonies. Due to logistical constraints this data is unbalanced in several different ways.

Some colonies were not sampled every year, and thus primary encounters are not equal across all sites. Secondary encounters are not always equal either, since some sites were sampled fewer or more times than the average (4 sequential days per year at a similar time each summer). I certainly don't want to omit entire sessions to reduce the data set to what's common among all colonies. Using "." doesn't seem to be an option for missing values with the robust design. Another suggestion found here was to group individuals according to similar days sampled and constrain detection probability to zero were applicable. Is the latter the best approach?

Any other suggestions?

Thanks!
Katherine
Katherine McClure
 
Posts: 1
Joined: Wed Oct 22, 2008 5:15 pm

missing data in MS robust design

Postby Bill Kendall » Wed Oct 22, 2008 6:34 pm

Katherine,

Although 24 states is going to get complicated, from what I understand of your problem that last option you mentioned should be best. Set up you capture history file based on the largest number of samples across the 24 colonies, by primary period. Set up the primary and secondary samples in the model setup screen on that basis. Then, for whatever secondary sample you don't have sampling effort for a given colony, set p=0. This of course will cause estimability problems for some parameters under more general models, but it will be set up correctly.
Bill Kendall
 
Posts: 96
Joined: Wed Jun 04, 2003 8:58 am

Postby murray.efford » Wed Oct 22, 2008 10:16 pm

Katherine
If your interest is in abundance (density) and not inter-colony movement or survival then you may be able to use the 'session' construct in the Density software (www.otago.ac.nz/density). This allows both trap configuration and number of occasions to vary among 'sessions'. In your case, each year x colony combination would be a 'session'. The current release does not allow additive between-session models (colony effect + year effect) but this is fixed in a new release due out next week.
Murray Efford
murray.efford
 
Posts: 712
Joined: Mon Sep 29, 2008 7:11 pm
Location: Dunedin, New Zealand


Return to analysis help

Who is online

Users browsing this forum: No registered users and 2 guests

cron