Specifying a multilevel model in MCMCglmm (R), that is heteroskedastic at level one I am considering MCMCglmm as an alternative to MLwiN. The former package works perfectly fine, but I cannot figure out how to model heteroskedasticity at level one. For instance, if I have the following model, individuals within countries
Fixed: y = 1 + x1 + x2 
Random Level 2: 1 + x1 (country level)
Random Level 1: 1 + x1 (individual level)

For level 2 I can specify the option random=~country+x1, how do I specify the level one portion? My guess would be that this has something to do with the rcov option, but specifying the formula as rcov=~indiv.id+x1 doesn't work. 
Or am I entirely wrong and these type of models are just not fitted in MCMCglmm? 
Thanks for any tips. 
EDIT: to provide a clearer idea of what the model is, here is how the model looks in MLwiN. At the top there are fixed effects, at the bottom you have two variance matrices: Ou is variance on the country level, and Oe is variance on the individual level. Ou is easily modeled in MCMCglmm, using the random option. I cannot get MCMCglmm to estimate the Oe. 
 A: If I get it correctly - and you're interested in modeling random slope and intercept on both the level of random effects - then the only thing that I don't know is whether individuals are measured multiple times? If yes then then they are not confounded with the residual variance and I can be modeled in the random part of the MCMCglmm model. I'm not sure whether your x1 variable is categorical or continous (from you screenshot I'm guessing the former, but the "1+x1" part looks more like a random slope-intercept model, and hence continous). A cannot tell it for sure from your MLwiN screenshot - so I'm not sure if the below code is what you want.
Assuming a categorical x1 variable and multiple data for individuals it could look like this:
prior <- list(R=list(V=diag(2), nu=0.002),
         G=list(G1=list(V=diag(2), nu=0.002, alpha.mu=c(0,0), alpha.V=1000*diag(2)),
                G2=list(V=diag(2), nu=0.002, alpha.mu=c(0,0), alpha.V=1000*diag(2)))

model <- MCMCglmm(y~x1, random=~idh(x1):country + idh(x1):individual,
                  rcov=~idh(x1):units, data=your_data,
                  prior=prior)

The model assumes that you don't want the covariance between the levels of x1 to be estimates (hence idh) - but if you want you should change it to us and change the prior appropriately (bu changing 'nu' of such random effects to 1.002 - assuming 2 levels in x1; see Jarrod Hadfield Course Notes for more details). The rcov part ensures that possible heterogeneity in residual variance won't inflate estimates of differences in country/individual-level variances. Be careful with the control parameters such as nitt, burnin and thin in MCMCglmm - I've lft them to their defaults above but probably you should increase the number of iterations in your case - ?MCMCglmm will tell you how.
