Use of AICc
5 posts
• Page 1 of 1
Use of AICc
by pscha119 » Sat Oct 04, 2008 3:17 pm
Is it appropriate to calculate AICc based on the AIC results in PRESENCE when the sample size is relatively small compared to the number of parameters? There is a small reference to this in the book, saying that due to the "effective sample size" in capture-recapture and occupancy modeling not always being obvious, AIC was used throughout the book. I've also noticed mixed use of AICc in published studies using PRESENCE when it appears its use would be appropriate based on sample size. Any input would help. Thanks. Paul.
- pscha119
- Posts: 1
- Joined: Sat Dec 01, 2007 7:55 pm
by darryl » Sun Oct 05, 2008 4:37 pm
Completely appropriate, you just have to determine what your effective sample size is for these models.
We can come up with a number of reasonable definitions based on deep thought and comparison with other models. At the time of writing the book, we couldn't decide (and still really haven't) what the 'right' number might be, so took the ostrich approach and kept our heads in the sand over the issue.
Options are:
1. Number of sampling units
2. Number of detections (equivalent to number of releases in CJS)
3. Number of surveys at 'occupied' units
4. Something else
I've ranked these in order from conservative -> liberal, meaning that with option 1 you're going to get the greatest penalty for more complicated models resulting in the simpler models getting ranked too high if the real ESS is actually something bigger. Probably not a bad property oftentimes, but not always. I tend to suggest that people try a couple of options if they're concerned and see how that effects final inferences.
Darryl
We can come up with a number of reasonable definitions based on deep thought and comparison with other models. At the time of writing the book, we couldn't decide (and still really haven't) what the 'right' number might be, so took the ostrich approach and kept our heads in the sand over the issue.
Options are:
1. Number of sampling units
2. Number of detections (equivalent to number of releases in CJS)
3. Number of surveys at 'occupied' units
4. Something else
I've ranked these in order from conservative -> liberal, meaning that with option 1 you're going to get the greatest penalty for more complicated models resulting in the simpler models getting ranked too high if the real ESS is actually something bigger. Probably not a bad property oftentimes, but not always. I tend to suggest that people try a couple of options if they're concerned and see how that effects final inferences.
Darryl
- darryl
- Posts: 498
- Joined: Thu Jun 12, 2003 3:04 pm
- Location: Dunedin, New Zealand
AICc in Presence
by jasonmartin13 » Fri Aug 21, 2009 11:10 am
Burnham and Anderson (2004) suggest that AICc always be used because AICc values converge with AIC values as the number of observations becomes large. Is there general agreement on this? Are there plans to include AICc values in the Presence output in the future?
Jason
Jason
- jasonmartin13
- Posts: 3
- Joined: Thu Aug 13, 2009 3:54 pm
AICc in Presence
by jhines » Fri Aug 21, 2009 11:26 am
Jason,
If you change the effective sample size (under 'Tools' menu in PRESENCE), the results table will change to AICc.
Cheers,
Jim
If you change the effective sample size (under 'Tools' menu in PRESENCE), the results table will change to AICc.
Cheers,
Jim
- jhines
- Posts: 632
- Joined: Fri May 16, 2003 9:24 am
- Location: Laurel, MD, USA
5 posts
• Page 1 of 1
Who is online
Users browsing this forum: No registered users and 1 guest