First draft of paper and appendix

[athowes]

Paper

• Better description of ART attendance
• Collected together version
• Table of why each ELGM reason is important. And which section of the model it effects. Perhaps what problems I would expect it to be likely to cause.
• Connection of Laplace to AGHQ with k = 1
• Complete mathematical details of PCA-AGHQ
• Description of Laplace marginals (in conclusion / future work)
• Alter Naomi results plot to have larger text (as is the case for BioInference poster)
• Check posterior contraction results Posterior contraction assessment #45
• Group posterior contraction plot by model component (and probably move to appendix)
• Add KS plot and text from .csv
• Add Anderson-Darling (plot and text) Use e.g. Anderson-Darling test rather than Kolmogorov-Smirnov #51
• Add PSIS (plot and text) Comparison report using PSIS #41
• Further discussion of MMD -- what does this reduction practically speaking mean? Is there a way to interpret it? Comparison report using MMD #42
  • Make the numbers inline!
• Further look into standard deviation implications: what does the fact that TMB underestimates SD mean for HIV policy? Select the nodes which are the most underestimated
  • Something to do here: colour the points by parameter in the mean and SD point estimates latent plot. Could just do the parameters which are outlying
• Is there a third exceedance probability case-study to add?

Appendix

• Use bold for vectors and matrices throughout
• $51 + 14n$ versus $51 + 13n$ confusion
• All many termed equations into tables
• All observations as $y$ (and generally align notation with main text)
• Create DAG
• Complete dependence input table
• Any more MCMC diagnostics to add?
• Normalising constant AGHQ and PCA-AGHQ section
• Further inference comparison results detailed tables
  • Point estimates latent and output
  • Add output to KS

[athowes]

Add mathematics of how the Choleksy and spectral decomposition AGHQ interact with the number of points in the marginal dimensions!