[R-sig-ME] Manipulating matrices from MCMCglmm

[Jarrod Hadfield]

Hi Adam,

Try this:

beta<-function(x, dimension=4, response=1){
V<-matrix(x,dimension, dimension)
solve(V[-response,-response],V[-response,response])
}

post.beta<-apply(m1$VCV[,1:16] 1, beta)

# you have to change 1:16 to the appropriate indices.

This has also been done in:  Phillimore, A. B., S. Stålhandske, R. J.  
Smithers, and R. Bernard. 2012. Dissecting the contributions of  
plasticity and local adaptation to the phenology of a butterfly and  
its host plants. American Naturalist 180: 655-670

Cheers,

Jarrod

Quoting Adam Hayward <a.hayward at sheffield.ac.uk> on Fri, 1 Nov 2013  
15:47:02 +0000:

> Hi all,
>
> I'mm running a 4-trait model in MCMCglmm in order to estimate selection on
> three traits (the first trait being lifetime reproductive success). The
> model gives me 1000 estimates of each of the variance components. I'm
> primarily interested in the first 16 components, which make up the 4x4
> individual-level VCV matrix, but there are other variance components, some
> of which are not fitted to all traits. model$VCV spits this out as an
> object with dimensions of 1000xN, where N is the number of VCV components.
> Primarily, I would like to convert this into the thousand estimates of my
> 4x4 individual-level matrix, which I should then be able to use to
> calculate selection gradients accounting for correlated selection and have
> some measure of error around these estimates. For example, on a single
> estimate of a 4x4 VCV matrix:
>
> # matrix of VCV estimates
>> I<-as.matrix(cbind(
> + c(1, 1, 0.1, 0.5),
> + c(1, 1, 0.1, 0.1),
> + c(0.1, 0.1, 1, 0.1),
> + c(0.5, 0.1, 0.1, 1)))
>
> # get matrix, minus the trait we're estimating selection through, i.e. LRS
>> Ipred<-I[2:4,2:4]
>> Ipred
>      [,1] [,2] [,3]
> [1,]  1.0  0.1  0.1
> [2,]  0.1  1.0  0.1
> [3,]  0.1  0.1  1.0
>
> # get a vector of covariances between predictors and fitness
>> COV<-(I[1,])[2:4]
>
> # calculate selection gradients
>> Beta<-ginv(Ipred)%*%COV
>> Beta
>             [,1]
> [1,] -0.03703704
> [2,]  0.40740741
> [3,]  0.96296296
>
> An added complication is that two of these "traits" are random slopes, but
> that will probably have to wait. I think for now my question is a
> *relatively* straightforward one- how would I go about obtaining my 1000
> 4x4 matrices to plug into the above formulas in order to get selection
> coefficients with some measure of error? I'd be very grateful if anyone has
> any advice.
>
> Best wishes,
> Adam
>
>
>
> --
> Adam Hayward
> Post-Doctoral Research Associate
> Department of Animal and Plant Sciences
> Alfred Denny Building
> University of Sheffield
> Western Bank
> Sheffield S10 2TN
> UK
> http://www.huli.group.shef.ac.uk/adam-personal.html
> http://adhayward.wordpress.com/
> https://twitter.com/adhayward18
>
> 	[[alternative HTML version deleted]]
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
>



-- 
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.