Add prop_diff_stratnc() calculating the stratified Newcombe CI

[bahatsky]

For a study we are working on a shiny app and one of the outputs should look like below. To enable this we would need to have stratified Newcombe confidence interval (as there are strata).

see https://documentation.sas.com/doc/en/pgmsascdc/9.4_3.5/procstat/procstat_freq_details63.htm and references therein for the method

To do:

• design
• implementation
• examples
• tests

[shajoezhu]

From @bahatsky ,

https://blogs.sas.com/content/iml/2022/04/18/mcnemar-test-sas.html could be a good starting point.

[danielinteractive]

@shajoezhu I am doing this now since we need it for the Crova dry run

[shajoezhu]

Thanks so much @danielinteractive

[Melkiades]

I am adding also prop_strat_wilson (with and without continuity correction) and Yan and Su (2010) heuristic which is used to estimate weights. In Daniel's design, the stratified Newcombe is calculated with weights that come from a modified CMH which now outputs the found weights. As I implemented now the heuristic, it outputs too the found weights (I might add the switcher for that in stratnc later. Here I add some points for me to keep track of the latest developments:

• Add directly tests for estimate_proportion when used with rtables (not related to current design, it was missing)
• Implement strat_wilson from Daniel's design
• Add heuristic for weights
• Documentation
• Examples
• Tests
• Adapt s_proportion to work with strata
• Adapt estimate_proportion to work with df and therefore strata for this specific case
• Add more tests and examples for the latter specific case (behavior with rtables)

[danielinteractive]

thanks @Melkiades for working on this!

[Melkiades]

Thanks to you for the great initial design! I just completed the first part (s_proportion and prop_strat_wilson). At the beginning, I took out the suppressWarnings() around stats::prop.test(), but when I added the real data test is complaining a lot about an approximation done. Do you know what is this behavior precisely?

I have already asked for your review of the code, and I really tried my best to make s_proportion fit the new structure but I am sure there is space for improvement there as I took inspiration from prop_diff directly. Any suggestion is super welcome!

Note: it is still missing the stratified Newcombe. I will work on that tomorrow ;)

[danielinteractive]

Thanks @Melkiades - yeah we should suppress those warnings as we are not interested in them here. Due to the stratification there is small numbers everywhere, but the method is as it is, so it is supposed to use the small numbers.