Enhancement: Report missing values

[Kudusch]

Add a function to report the occurrence of missing values across variables. I could imagine a table with the number/share of missing values and a plot that shows common combinations of missing values.

I imagine something like the functionality of the VIM package.

If that sounds like a good addition to volker, I could prepare a PR with the implementation of that feature.

[strohne]

That's a very much needed feature :) So far, we have ignored missings to keep things simple. But it's something we should add. If you try it, I'll buy you a beer or two, that's for sure.

I guess the appropriate way to implement this is in data_rm_missings(). Create the report here as a tibble and use .attr_insert() to add it to the attributes. Then tweak the print.vlkr_tbl() and knit_table() functions to output the missings report. There may be other things to check in the volker architecture, but I'm glad to check them later. That being said, if you have other ideas how to implement it, just go ahead.

Please make sure to work on the devel branch. Out git workflow is: new branch from devel, once it is finished merge into devel, once it shall be deployed merge into main.

[strohne]

A note to myself as soon as we have the first prototype of this functionality: baseline data such as missings, at the moment, are transferred from object to object using attr_transfer(). Thus, potentially, those data is printed multiple time, e.g. by vlkr_lists. We prevent this by carefully selecting when to transfer and when not. A better option might be to keep track whether an output already has been printed.

[Kudusch]

These are the kind of plots I was thinking of for this feature:

These are all meant to give users a better understanding of how missing values are distributed in their data. The marginal plots are especially interesting if you want to impute missing values.

[strohne]

I see, thanks @Kudusch .

First, we need another plot type for plot_counts_one_grouped(data, cg_adoption_social_01, cg_adoption_social_02) - output a confusion matrix instead of a bar plot. Maybe we can solve this by pimping the props parameter (by now knows "total", "rows", "cols") and add "confusion" or "heatmap". We already have an internal function .plot_heatmap() to generate the plot.

Then we should add something like a missing/marginals-parameter to plot functions where it makes sense.

I like the combinations plot a lot. It can be helpful for other types of analyses as well (not only for missings). In this case, we need to add a variation to plot_counts_items() and tab_counts_items(). Again, need to think about parameter names.

In all cases, we first need to decide about appropriate parameter names or which parameters should be enhanced. I will think about it after I come back from holidays.