Create R Notebook for model analysis

[ld-archer]

Would be very useful to have a single document that collects lots of information from the model for debugging/comparing outputs after changing something.

This document would need:

• [x] T-tests from both minimal and cross-validation runs
  • Collect same categories together (i.e. demographics, binecon)
  • [x] Look into a visual way to present high/low p-value?
    • Done, found Formattable package
  • Calculate differences between previous run and current? Or improvement? Present this improvement with colour?
  • Generate Boxplots comparing means? Would need original data I think and would have to generate T-tests again
• [x] Transition models
  • Might not need this all the time but would still be useful to have transition models, especially if we could compare to previous
  • Maybe only needs a couple of the important ones? Mortality etc.
• Think of more!

Run the document from a Make rule, so we can automatically generate it at the end of every run. Can also then collect all the info from a specific run in a new directory, so we can come back to every run and get full information about it.

[ld-archer]

Formattable is an R package for special formatting of tables, including the ability to add graphs, bars and other complex structures inside table columns. We will use this to add colour to the P-value columns, for visually representing the p-values. The coloured bar will represent the size of the p-value, from 0 to 1.

[ld-archer]

Changed to use reactable R package instead of formattable as more powerful.

Update model_analysis.nb.html is now produced after running the debug make rule, which runs the baseline, both cross-validations, and the minimal models, then produces the debug document. This is a good start, and only one or 2 things to do to make it very useful.

• [x] Add the code to visualise the CV2 'handover' type plots
• [x] Visualise the difference in T-tests between previous and current
  • Do this for 1 group of vars at a time (function arg)
  • table1 - table2? That simple?
  • Can use very similar formatting to the current P value columns
  • Not worth the time, can easily see the difference when outputs are alongside eachother

Github doesn't support the filetype or I would upload one as an example