Add precision (standard deviation) parameter posterior plots

athowes / multi-agyw

Spatio-temporal estimates of HIV risk group proportions for AGYW across 13 priority countries in sub-Saharan Africa

https://athowes.github.io/multi-agyw

MIT License

2 stars 2 forks source link

Add precision (standard deviation) parameter posterior plots #20

Closed athowes closed 3 years ago

athowes commented 3 years ago

Aim to show what proportion of variance is attributable to different factors
Might be a paper of Jon Wakefield's that does this nicely
Not saving the standard deviation posteriors at the moment so this would be included in the aaa_fit scripts

athowes commented 3 years ago

The paper Jeff meant is this one https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0210645 (thanks Oli!)

athowes commented 3 years ago

Need to add this to the single DHS scripts
Stacked barplot by country showing the proportions of variance "explained" by each random effect
Could also not normalise the variance such that it's not variance_i / total_variance since perhaps the magnitude of the variance in each country is of interest too. Risk of overcomplicating

athowes commented 3 years ago

Directly output a .tex or .txt file for the table?

athowes commented 3 years ago

Could add cloropleths like Li, Hsiao, Godwin et al. Figure 1. but unsure if they add much to the stacked barplots

athowes commented 3 years ago

Actually this plot should probably be horizontal.

athowes commented 3 years ago

I think this is close to the best version of this plot to convey the message about which random effects are most important in the model and how that varies over country. Closing this issue now, may reopen if there is coauthor feedback to warrant further changes.

athowes commented 2 years ago

Note that these are sometimes called Sobol indices. Might also want to investigate Shapley values.