Good morning Jesse,
Thank you for your quick response, it is much appreciated!
I’ve a couple of queries to get my head around this approach.
- If I interpret the model graph correctly the underlying true value mu is a function of the noisy observation of x, and the parameters alpha & beta function on the noisy x. The example I’ve looked at for an ‘error-in-variables’ approach from this post is slightly, but importantly, different. alpha & beta function on the true, underlying variable mu_x. The uncertainty in x flows through to obs through the draws of mu_x taken for each sample. Is there a particular reason why the model you’ve shown is set-up for mu = f(x) rather than mu = f(mu_x)?
- From that starting point, I think I follow the process. The draws on obs being conditional on x (or, in the model structure I’m assuming, mu_x) makes sense. But, and this is where I’m struggling conceptually, if we then want to use a call to plot_hdi (such as is shown in the example ppc page, what values do we plot the hdi against? It seems to me it should be against mu_x, but I’m not sure which mu_x (an average of realisations?)
Thanks again.
Regards,
Jack