[R-sig-ME] MCMCglmm covariance matrix question

[Ben Bolker]

   That seems quite difficult to set up in a single analysis.
   Would it work to analyze 8 of the traits with one model, and then 
analyze the other trait (which you want to treat independently) with a 
separate, univariate model?

On 12/2/20 10:56 AM, Srivats Chari wrote:
> Greetings,
> 
> I'm trying to run a multivariate MCMCglmm with 9 traits. My traits are
> measured mostly at the same time except for 1 trait which is measured only
> once or twice per individual. After reading a some literature on this I
> have a basic idea on how to structure my dataset. But the problem I am
> facing is that I need my covariance matrix to be different. I want my
> covariance matrix to exclude the 1 trait so that all trait can COvary
> together except for a particular one! So there won't be any
> within-individual COvariation between the particular trait and others.
> 
> creating a sample dataset-
> 
> df<- data.frame(ani_id = as.factor(1:10),
> sex=c("male","female","male","female","male","female","male","female","male","female"),age=c("young","adult","young","adult","adult","young","young",
> "adult","adult","adult"),value=runif(200,min=1,
> max=5),year=ceiling(runif(200,min=2010, max=2019)), PC1=runif(200, min=0.1,
> max=0.9))
> df$value[5:9]<- NA
> df$trait_id<- as.factor(paste("T",rep(1:10, each=20), sep="_"))
> 
> ## My Prior
> prior1 <- list(R = list(V =diag(10), nu = 0.002),
>                        G = list(G1 = list(V = diag(10), nu = 0.002,
>                                           alpha.mu = rep(0, 10),
>                                           alpha.V  = diag(10)*25^2)))
> ## MCMC model
> mcmc_trial1<-MCMCglmm(scale(value) ~ factor(sex)+
>               scale(year) + scale(year^2)+
>               scale(PC1)+ scale(PC1^2)+
>               factor(age),
>             random =~ us(trait_id):ani_id,
>             rcov =~ idh(trait_id):units,
>             family = c("gaussian"),
>             prior = prior1,
>             nitt=10000,
>             burnin=1000,
>             thin=10,
>             verbose = TRUE,
>             pr=TRUE,
>             data = df)
> 
> So when I see a covariance matrix I want the trait T1 to not covary with
> any other trait.
> 
> T1 T2 T3 T4 T5 T6 T7 T8 T9 T10
> T1 0,1 0 0 0 0 0 0 0 0 0
> T2 0 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0,1
> T3 0 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0,1
> T4 0 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0,1
> T5 0 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0,1
> T6 0 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0,1
> T7 0 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0,1
> T8 0 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0,1
> T9 0 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0,1
> T10 0 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0,1
> 
> Where I am stuck is I do not know how to structure the covariance matrix to
> exclude T1.
> 
> Any suggestions or help is much appreciated. :)
> 
> Regards,
> Srivats.
> 
> 	[[alternative HTML version deleted]]
> 
> _______________________________________________
> R-sig-mixed-models using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>