Extending Missing Values Options

[lilyclements]

Written in 2018 in #3711

A package I have come across which has some functions on dealing with missing data is finalfit.
It has some crossover in terms of functions as naniar and visdat.

For example
vis_miss in visdat is equivalent to missing_plot in finalfit, just flipped on the co-ordinates. (Although I prefer vis_miss aesthetically and in its parameters. However, missing_plot can be nice with factor labels)

However there are some nice functions in finalfit that I haven't (yet) seen in naniar or visdat so could be useful to consider.

I'm running through finalfit briefly this afternoon so can let you know further if you are interested.
I have a R file on some of naniar and visdat too.

So here's an example of something I have not found how to visualise in the naniar or visdat packages (although may be possible!) to quickly look at missingness

Using the airquality data:

airquality %>% mutate(Month = factor(Month)) %>% missing_pairs(c("Ozone", "Solar.R", "Wind", "Month"))

• So the columns are the four variables, and the rows have split each of these variables into "missing or observed"
• e.g. Column 1, row 2 shows a boxplot for Ozone levels split by if Solar.R is missing (grey) or observed (blue). These values can be run with the bind_shadow function in the naniar package.
  airquality %>% bind_shadow() %>% group_by(Solar.R_NA) %>% summarise_at(.vars = "Ozone", .funs = c("mean", "sd", "var", "min", "max"), na.rm = TRUE)
• Or in column 4, row 1 we can see there are a lot of missing values for the ozone variable in month six.

Or, with Roger's Malawi rainfall data if we just consider the month, and years 2015 and 2016 we can quickly see that most missing values occur between September and December.

The other columns aren't really relevant. I also think the plot could look a lot nicer than this, however, as a start it's quite a nice way to look at missing data. Perhaps it's not useful or relevant for R-Instat, but, I thought I'd mention it just in case.