“Warning: In norm::prelim.norm(data) : NAs introduced by coercion to integer range” while using mcar_test in naniar package

Question

I am trying to run an MCAR test using the naniar package in R. I have a dataset with 52 variables (most of these responses to items on likert type scales). I am getting the warning "Warning: In norm::prelim.norm(data) : NAs introduced by coercion to integer range". I tried changing the data types, and running the test on subsets of data frames. I am not getting the warning when I run the test on a smaller number of variables but when I increase it to about 31 columns is when the error shows up. This lead me to conclude that it's likely not a specific column that is causing the problem. Is this warning ignorable? Can i proceed with the interpreting the results despite it? or is there any other way to address it?

Answer 1

This is because the naniar::mcar_test() function is based on the prelim.norm function in the norm package which can handle about 30 variables.

There is an extensive description of this issue in the misty::na.test() help file (which is also based on prelim.norm.

Note that the mcar_test function in the naniar package is also based on the prelim.norm function which results are not trustworthy whenever the following warning message is pronted on the console: In norm::prelim.norm(data) : NAs introduced by coercion to integer range

With more than 30 variables specified in the argument x, the prelim.norm function might run into numerical problems leading to results that are not trustworthy. In this case it is recommended to reduce the number of variables specified in the argument x.

The recommend to use the LittleMCAR function in the BaylorEdPsych package which can deal with up to 50 variables. But you need to install this package from the CRAN Archive, since it was archived some time ago. Which mean you have to install it manually. Maybe the MissMech package also works - but did not check this one.