Add ability to make n-way tables

[mbcann01]

Overview

Currently, I freqtables will only create one- and two-way tables. It will not create n-way tables. We want to add the ability to create n-way tables.

What I had in mind was something like:

demo_nih %>% 
    freqtables::freq_table(ethnicity_nih, race_nih, sex_nih)

However, what I've been doing in the meantime is:

make_table_section <- function(cat) {
  demo_nih %>% 
    filter(ethnicity_nih == cat) %>% 
    freqtables::freq_table(race_nih, sex_nih) 
}

And then:

purrr::map_dfc(
  .x = c("Not Hispanic or Latino", "Hispanic or Latino", "Unknown/Not Reported"),
  .f = make_table_section
)

Obviously, this is more verbose, but it gets the job done and is very versitile (e.g., user can return a list instead of a data frame). However, the spirit of freqtables isn't really to be the most "versitile" package. It's to be the easiest to use "out of the box" for 85%+ of normal use. Give this some thought.

The suggestion from a user on RStudio Community could also be useful:

mtcars %>% 
  gather(variable,category,cyl,vs,am,factor_key = TRUE)%>%
  group_by(variable,category)%>%
  summarize(n=n())

Left off at

2023-03-17: Working on test.Rmd as part of #40.

• I created two data files for comparing freqtables with Stata and SAS.
• The data files are called /inst/extdata/freq_study.dta and /inst/extdata/freq_study.xpt.
• These data files are created using data-raw/study.R.
• I also created a do file - /inst/extdata/compare_freqtables.do - and a SAS script - /inst/extdata/compare_freqtables.sas.
• I added all of these files to buildignore.

2020-06-11: Created test.Rmd on the plane to Minnesota to test out different ways of doing this. test.Rmd is git ignored and build ignored.

Tasks

• Complete one, two, and n-way tables in Stata (/inst/extdata/compare_freqtables.do). Use them for comparison.
• Complete one, two, and n-way tables in SAS (/inst/extdata/compare_freqtables.sas). Use them for comparison.
• Figure out how you want freq_tbl to treat n-way tables.
• Figure out how you want freq_table to treat n-way tables.
• Figure out how you want freq_test to calculated stats for n-way tables.

[mbcann01]

Here's how Stata does it:

[mbcann01]

2020-06-15:
We may want to distinguish between an n-way freq_table (shows overall n, prop, ci like Stata) and a grouped_by n-way freq_table that uses row n's and percents instead.

[mbcann01]

2020-06-11 - Notes from while I was on the plane:

• If you allow group_by to work again, then you may need to change the descriptive analysis vignette.
• If you allow 3-way tables then you will have to change some of the wording in the descriptive analysis vignette. Specifically, in Bivariate percentages and 95% log transformed confidence intervals.
• If you change the row/column terminology to group/subgroup terminology then you will have to change some of the wording in the descriptive analysis vignette.

Row/column labels

• What is the best way to create a contingency table? Then row/column makes sense.
• group level 1, group level 2, group level 3
• group, subgroup 1, subgroup 3

[mbcann01]

Do this in a new branch

[mbcann01]

So, meantables uses group_by. Then, the output is labeled "response_var" and "group_var". It might be worth considering keeping this consistent.

One var

mtcars %>% freq_table(am)

Two+ vars

mtcars %>% group_by(mpg) %>% freq_table(am)

Could even have "response_var" (or something similar) and "group_var" in table of results.

[mbcann01]

Needed it on the Sun Study for this (as an example):

map_student %>% 
  filter(!is.na(ss_application_f)) %>% 
  freq_table(period_f, teacher_f, ss_application_f)

Tested this in stata with: by period_f, sort : tabulate teacher_f ss_application_f, chi2. It returns this:

Tried it in SAS using :

proc freq data=map_student;
	tables period_f * teacher_f * ss_application_f;
run;

Which returned

I also tried proc surveyfreq, but that won't return chisq for three-way tables.