I am having a hard time figuring out how to do the model averaging. Can anyone simply explain how to calculate this by hand or how to do it in Presence?
And what does a deltaAICc of -0.00 mean? Why is it negative?
How do you model average in Presence?
6 posts
• Page 1 of 1
How do you model average in Presence?
by hoarybat » Sun Mar 10, 2013 5:28 pm
- hoarybat
- Posts: 6
- Joined: Thu Mar 07, 2013 9:48 am
Re: How do you model average in Presence?
by jhines » Mon Mar 11, 2013 7:37 am
In PRESENCE, there is an option to do model averaging in the 'tools' menu.
The way to do it by hand is to simply multiply the AIC weight of each model by the parameter estimate of interest for each model. A good explanation of the procedure can be found in the following links:
warnercnr.colostate.edu/~gwhite/mark/.../modelaveraging.htm
warnercnr.colostate.edu/~gwhite/fw663/average.pdf
Jim
The way to do it by hand is to simply multiply the AIC weight of each model by the parameter estimate of interest for each model. A good explanation of the procedure can be found in the following links:
warnercnr.colostate.edu/~gwhite/mark/.../modelaveraging.htm
warnercnr.colostate.edu/~gwhite/fw663/average.pdf
Jim
- jhines
- Posts: 632
- Joined: Fri May 16, 2003 9:24 am
- Location: Laurel, MD, USA
Re: How do you model average in Presence?
by hoarybat » Mon Mar 11, 2013 3:05 pm
Jim,
Do you know why I would obtain a -0.00 deltaAICc for my best model?
It seems to be messing up all the AICc values for the remaining models. As in, the results window will list a model's AICc as 50.14, but when I go into the output file for that model, the output file says 53.73 for the AICc. I don't get it. I think the -0.00 is throwing off my outputs. I have deleted that top model and reconstructed it only to receive the same -0.00 deltaAICc value.
Thanks for your help.
-Zach
Do you know why I would obtain a -0.00 deltaAICc for my best model?
It seems to be messing up all the AICc values for the remaining models. As in, the results window will list a model's AICc as 50.14, but when I go into the output file for that model, the output file says 53.73 for the AICc. I don't get it. I think the -0.00 is throwing off my outputs. I have deleted that top model and reconstructed it only to receive the same -0.00 deltaAICc value.
Thanks for your help.
-Zach
- hoarybat
- Posts: 6
- Joined: Thu Mar 07, 2013 9:48 am
Re: How do you model average in Presence?
by jhines » Mon Mar 11, 2013 3:11 pm
Zach,
Can you send me the most recent backup zipfile in your project folder? I can't think of why you'd get -0.00.
Jim
Can you send me the most recent backup zipfile in your project folder? I can't think of why you'd get -0.00.
Jim
- jhines
- Posts: 632
- Joined: Fri May 16, 2003 9:24 am
- Location: Laurel, MD, USA
Re: How do you model average in Presence?
by hoarybat » Mon Mar 11, 2013 4:44 pm
Jim,
Can you contact me via email? I don't know how to add attachments to this forum.
zdkaiser AT gmail DOT com
Can you contact me via email? I don't know how to add attachments to this forum.
zdkaiser AT gmail DOT com
- hoarybat
- Posts: 6
- Joined: Thu Mar 07, 2013 9:48 am
Re: How do you model average in Presence?
by jhines » Tue Mar 12, 2013 10:01 am
Hi Zach,
The values in the results table are AICc values. AICc is the AIC value after a 'correction' for small sample sizes. When you entered the effective sample size (35), it changed the results table to AICc instead of AIC. The model output files only print AIC values. So, there is nothing wrong with the results table, except that the top delta-aic printed negative zero instead of zero. This is only a cosmetic problem caused by rounding error when subtracting the top AICc value from itself. It does not affect any other values.
AIC = -2*loglike + 2*k
AICc = AIC + 2*k*(k+1)/(ESS-k-1)
where k=number of parameters and ESS=effective sample size
Jim
The values in the results table are AICc values. AICc is the AIC value after a 'correction' for small sample sizes. When you entered the effective sample size (35), it changed the results table to AICc instead of AIC. The model output files only print AIC values. So, there is nothing wrong with the results table, except that the top delta-aic printed negative zero instead of zero. This is only a cosmetic problem caused by rounding error when subtracting the top AICc value from itself. It does not affect any other values.
AIC = -2*loglike + 2*k
AICc = AIC + 2*k*(k+1)/(ESS-k-1)
where k=number of parameters and ESS=effective sample size
Jim
- jhines
- Posts: 632
- Joined: Fri May 16, 2003 9:24 am
- Location: Laurel, MD, USA
6 posts
• Page 1 of 1
Who is online
Users browsing this forum: No registered users and 1 guest