[R-sig-eco] AICcWts in modavg {AICcmodavg}

[Kristen Gorman]

Dear All, 
I am using the AICcmodavg package to calculate model averaged parameter estimates on a small candidate set of lme class models. I noticed this evening that the AICcWt values given in modavg actually appear to change from the original aictab (AICc Table). Hopefully I can explain with code:

The original aictab was:

Model selection based on AICc :

      K     AICc Delta_AICc  AICcWt  Cum.Wt        LL
mod2  5 407.4459    0.00000 0.42998 0.42998 -198.6548
mod5 11 409.2742    1.82824 0.17237 0.60234 -193.3329
mod4  7 409.3215    1.87558 0.16833 0.77068 -197.5329
mod1  3 409.4667    2.02075 0.15655 0.92723 -201.7062
mod3  5 410.9987    3.55275 0.07277 1.00000 -200.4312



Multimodel inference on each parameter followed:

Multimodel inference on " (Parameter1) " based on AICc 
 AICc table used to obtain model-averaged estimate:

      K   AICc Delta_AICc AICcWt Estimate   SE
mod1  3 409.47       2.02   0.16     8.54 0.06
mod2  5 407.45       0.00   0.43     8.38 0.08
mod3  5 411.00       3.55   0.07     8.42 0.11
mod4  7 409.32       1.88   0.17     8.28 0.12
mod5 11 409.27       1.83   0.17     8.42 0.15

Multimodel inference on " Parameter2.level2 " based on AICc 
 AICc table used to obtain model-averaged estimate:

      K   AICc Delta_AICc AICcWt Estimate   SE
mod2  5 407.45       0.00   0.56     0.23 0.16
mod4  7 409.32       1.88   0.22     0.22 0.16
mod5 11 409.27       1.83   0.22    -0.48 0.35

Multimodel inference on " Parameter2.level3 " based on AICc 
 AICc table used to obtain model-averaged estimate:

      K   AICc Delta_AICc AICcWt Estimate   SE
mod2  5 407.45       0.00   0.56     0.29 0.12
mod4  7 409.32       1.88   0.22     0.28 0.12
mod5 11 409.27       1.83   0.22     0.12 0.21
Multimodel inference on " Year.22 " based on AICc 

Multimodel inference on " Parameter3.level2 " based on AICc 
 AICc table used to obtain model-averaged estimate:

      K   AICc Delta_AICc AICcWt Estimate   SE
mod3  5 411.00       1.72   0.18     0.22 0.14
mod4  7 409.32       0.05   0.41     0.20 0.14
mod5 11 409.27       0.00   0.42     0.08 0.20

Multimodel inference on " Parameter3.level3 " based on AICc 
 AICc table used to obtain model-averaged estimate:

      K   AICc Delta_AICc AICcWt Estimate   SE
mod3  5 411.00       1.72   0.18     0.10 0.14
mod4  7 409.32       0.05   0.41     0.09 0.14
mod5 11 409.27       0.00   0.42    -0.19 0.20

Multimodel inference on " Parameter4.level2 " based on AICc 
 AICc table used to obtain model-averaged estimate:

      K   AICc Delta_AICc AICcWt Estimate   SE
mod5 11 409.27          0      1     0.96 0.43

Multimodel inference on " Parameter4.level3 " based on AICc 
 AICc table used to obtain model-averaged estimate:

      K   AICc Delta_AICc AICcWt Estimate   SE
mod5 11 409.27          0      1     0.82 0.42

Multimodel inference on " Parameter5.level2 " based on AICc 
 AICc table used to obtain model-averaged estimate:

      K   AICc Delta_AICc AICcWt Estimate   SE
mod5 11 409.27          0      1     0.03 0.29

Multimodel inference on " Parameter5.level3 " based on AICc 
 AICc table used to obtain model-averaged estimate:

      K   AICc Delta_AICc AICcWt Estimate   SE
mod5 11 409.27          0      1     0.46 0.29

--

What confuses me is that in comparison with the original aictab, the AICcWt values change for a given model in most of the multimodel results, with the exception of Parameter 1 because it is included in all models. I don't understand why the AICcWt is changing for the multimodel results? I assume it is recalculating AICcWts based on a reduced candidate set that includes the parameter of interest.

Further, I am interested in calculating parameter likelihoods and believe given the odd results for the multimodel inference I should use the AICcWts offered in the original aictab for each parameter. Is there a way with the AICcmodavg package to calculate parameter likelihoods?

Thanks in advance for any insight.

Kristen Gorman