[R-sig-eco] MuMIn package, Inquiry on summary output _ Conditional or Full average models?
Cade, Brian
cadeb at usgs.gov
Tue Feb 16 16:41:45 CET 2016
You might want to reconsider whether it make any sense to model average the
individual regression coefficients. See Cade (2015. Model averaging and
muddled multimodel inferences. Ecology 96: 2370-2382).
Brian
Brian S. Cade, PhD
U. S. Geological Survey
Fort Collins Science Center
2150 Centre Ave., Bldg. C
Fort Collins, CO 80526-8818
email: cadeb at usgs.gov <brian_cade at usgs.gov>
tel: 970 226-9326
On Sun, Feb 14, 2016 at 9:07 AM, Daniel Gruner <dsgruner at umd.edu> wrote:
> Dear Laura,
>
> Grueber et al. (2011) discusses the distinctions and the rationale for
> making this choice (p 705-706), citing Burnham & Anderson (2002) and
> Nakagawa and Freckleton (2011).
>
>
> Burnham KP, and DR Anderson (2002). Model Selection and Multimodel
> Inference: a Practical Information-Theoretic Approach. 2nd edition.
> Springer, New York.
>
> Grueber CE, S Nakagawa, RJ Laws, and IG Jamieson (2011). Multimodel
> inference in ecology and evolution: challenges and solutions. Journal of
> Evolutionary Biology 24:699-711.
>
> Nakagawa S, and RP Freckleton (2011). Model averaging, missing data and
> multiple imputation: a case study for behavioural ecology. Behavioral
> Ecology and Sociobiology 65:103-116.
>
>
>
>
>
> On 2/14/2016 10:02 AM, Laura Riggi wrote:
>
>> Dear all,
>>
>> I have a question regarding the output for model averaging in R with
>> MuMin package. In the summary for model averaging two models of coefficient
>> calculations come out: the "full average" and the "conditional (or subset)
>> average" model (example of output below).
>>
>> As explained on the MuMin package pdf:
>> "The 'subset' (or 'conditional') average only averages over the models
>> where the parameter appears. An alternative, the 'full' average assumes
>> that a variable is included in every model, but in some models the
>> corresponding coefficient (and its respective variance) is set to zero.
>> Unlike the 'subset average', it does not have a tendency of biasing the
>> value away from zero. The 'full' average is a type of shrinkage estimator
>> and for variables with a weak relationship to the response they are smaller
>> than 'subset' estimators."
>>
>> However, I cannot find information online concerning the theory behind
>> these different outputs. I am not sure what is the point of having a
>> "conditional" model as it seems to go against the idea of doing a model
>> averaging analysis.
>> Do you know of articles / books that discuss this? When should we use one
>> or the other?
>> Any advice would be appreciated.
>>
>> summary(model.avg(dd, subset = delta < 2))
>>>
>> Call:
>> model.avg.model.selection(object = dd, subset = delta < 2)
>>
>> Component model call:
>> lme.formula(fixed = log(Parasitoi_S1.S2 + 1) ~ <8 unique rhs>, data =
>> data, random = ~1 | Field.x/Site.x, method
>> = ML, na.action = na.fail)
>>
>> Component models:
>> df logLik AICc delta weight
>> 1345 8 -161.74 340.52 0.00 0.22
>> 345 7 -162.97 340.74 0.22 0.19
>> 12345 9 -161.26 341.82 1.31 0.11
>> 13456 9 -161.36 342.03 1.51 0.10
>> 2345 8 -162.53 342.10 1.58 0.10
>> 3456 8 -162.54 342.11 1.60 0.10
>> 35 6 -164.76 342.12 1.60 0.10
>> 145 7 -163.84 342.47 1.96 0.08
>>
>> Term codes:
>> L OSR2012_X500 OSR2013_X500
>> Weed.cover Wood_X500 Weed.cover:Wood_X500
>> 1 2 3
>> 4 5 6
>>
>> Model-averaged coefficients:
>> (full average)
>> Estimate Std. Error Adjusted SE z value Pr(>|z|)
>> (Intercept) 2.5693356 0.5295081 0.5337822 4.813 1.5e-06
>> ***
>> L -0.0005893 0.0007720 0.0007756 0.760 0.447
>> OSR2013_X500 -3.7932641 2.1558940 2.3307509 1.627 0.104
>> Weed.cover 0.1331237 0.0915813 0.0922877 1.442 0.149
>> Wood_X500 -5.5516524 3.2502461 3.5300659 1.573 0.116
>> OSR2012_X500 0.4326628 1.3007077 1.3966718 0.310 0.757
>> Weed.cover:Wood_X500 0.1922066 0.6215019 0.6255589 0.307 0.759
>>
>> (conditional average)
>> Estimate Std. Error Adjusted SE z value Pr(>|z|)
>> (Intercept) 2.5693356 0.5295081 0.5337822 4.813 1.5e-06
>> ***
>> L -0.0011489 0.0007205 0.0007279 1.578 0.1145
>> OSR2013_X500 -4.1301091 1.9155698 2.1268744 1.942 0.0522 .
>> Weed.cover 0.1474601 0.0847131 0.0855581 1.724 0.0848 .
>> Wood_X500 -5.5516524 3.2502461 3.5300659 1.573 0.1158
>> OSR2012_X500 2.0508163 2.1681256 2.4346913 0.842 0.3996
>> Weed.cover:Wood_X500 0.9639767 1.0923693 1.1039224 0.873 0.3825
>> ---
>> Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
>>
>> Relative variable importance:
>> Wood_X500 OSR2013_X500 Weed.cover L OSR2012_X500
>> Weed.cover:Wood_X500
>> Importance: 1.00 0.92 0.90 0.51 0.21
>> 0.20
>> N containing models: 8 7 7 4 2
>> 2
>>
>> Thank you for your help.
>> Kind Regards,
>> Laura
>>
>>
>> [[alternative HTML version deleted]]
>>
>> _______________________________________________
>> R-sig-ecology mailing list
>> R-sig-ecology at r-project.org
>> https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
>> .
>>
>>
> --
>
> Daniel S. Gruner, Associate Professor
> Department of Entomology
> 4112 Plant Sciences Bldg
> University of Maryland
> College Park, MD 20742 U.S.A.
> (o) 301-405-3957 (f) 301-314-9290
> dsgruner at umd.edu
>
> http://grunerlab.umd.edu
> https://twitter.com/GrunerDaniel
>
> _______________________________________________
> R-sig-ecology mailing list
> R-sig-ecology at r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
>
[[alternative HTML version deleted]]
More information about the R-sig-ecology
mailing list