Feature request: have descrip index binary variables starting with zero

[krpaulson]

For binary variables, we often want our summary statistics to be such that we can equate "mean" with the proportion belonging to one category. In the descrip function, it appears that the default behavior is to convert categorical variables to numbers, starting with one.

For example, this summary gives us a mean of 1.5:

> mri %>% select(sex) %>% descrip()
       N     Msng  Mean      Std Dev    Min       25%       Mdn       75%       Max     
sex:     735     0   1.498     0.5003    1.000     1.000     1.000     2.000     2.000 

It would be a nice feature if descrip indexed starting at zero. Even better if the output then indicated the reference group (ex: sexMale). Happy to answer any questions about this idea. Thank you!

[adw96]

Some example behaviour

mri <- read_csv("data/mri.csv")
mri %>% select(sex, race) %>% descrip()

mri.csv

[svteichman]

@adw96 what is the desired behavior here? Do we want for binary covariates to show the above stats with the baseline level set to 0 and the other level 1? For factors with >2 levels, do we also want to start at 0 or to keep these starting at 1?

Or, do we want to produce a fully different summary vector for categorical covariates, i.e. N, Msng, and then the proportion of samples in each covariate level?

[adw96]

Thanks for resurfacing this, @svteichman ! Could we convert categories to indicators variablenameCategoryname (eg "sexFemale", "sexMale", "sexNeither"...), and perhaps just summarise those indicators?

It's a bit weird to give Q1 etc for this, but then there's no missing data, at least. Plus, all categories are represented.

@krpaulson We welcome suggestions on what you think users would benefit from the most.

[krpaulson]

I agree that Q1 etc aren't that useful here.

I suppose you could make the goal of this function to pull together the descriptive statistics you might find in a Table 1. For binary or categorical variables that's usually the number and/or proportion in each category, including a missing category if applicable.

Would there be any downstream consequences of having output which is formatted differently depending on the variable type?

[svteichman]

I think there is a problem with having differently formatted output because if someone is summarizing a categorical variable and a continuous variable at the same time, for the output to be nicely formatted it needs to be the same. However, just setting the quantiles to NA for categorical covariates (their binary indicators, specifically) could be a reasonable way to only report relevant information. Then the output would look something like this:

mri %>% dplyr::select(sex, weight) %>% descrip()