handle more than one factor variables (How to simulate missings)

[rabea-a]

Hello @imkemayer,
it's me again, sorry! I think your temporary fix to handle factors fails for more than one factor variable. Below you see a minimal example which causes an error:

set.seed(1)
x1 <- as.factor(sample(c("A","B"), 20, replace = TRUE, prob = c(0.75, 0.25)))
x2 <- as.factor(sample(c("A","B"), 20, replace = TRUE, prob = c(0.75, 0.25)))
x3 <- rnorm(20, mean = 5)
x4 <- rnorm(20, mean = -5)

x <- data.frame(x1,x2,x3,x4)
data_NA <- produce_NA(data = x, mechanism = "MAR", perc.missing = 0.2, 
                      idx.incomplete = c(0, 1, 0, 0), idx.covariates = c(1, 0, 1, 0))

I think the case distinction of the two if queries is not necessary and should be removed. Instead, insert

orig.data <- data
vars_factor <- colnames(data)[!sapply(data, is.numeric)]
# if (length(vars_factor)==1){
#   levels_factor <- list(gdata::mapLevels(x=data[,vars_factor]))
# }
# if (length(vars_factor)>1){
#   levels_factor <- sapply(data[,vars_factor], FUN = function(x) gdata::mapLevels(x=x))
# }
levels_factor <- list(gdata::mapLevels(x=data[,vars_factor]))

for all length of vars_factor.
Furthermore, the correct indexing needs be used in lines 466 and 467 afterwards (also you have to check in case of by.patterns==TRUE):

#gdata::mapLevels(x=tmp$data.init[,vars_factor[[i]]]) <- levels_factor[[i]]
#gdata::mapLevels(x=tmp$data.incomp[,vars_factor[[i]]]) <- levels_factor[[i]]
gdata::mapLevels(x=tmp$data.init[,vars_factor[[i]]]) <- levels_factor[[1]][[i]]
gdata::mapLevels(x=tmp$data.incomp[,vars_factor[[i]]]) <- levels_factor[[1]][[i]]

I look forward to hearing your opinion of my thoughts.

Best,
Rabea.

[rabea-a]

I refer to

website/static/how-to/generate/amputation.R

Lines 165 to 174 in aedd584

	# temporary fix to handle factors
	# (transform them to numeric and revert the conversion in the end)
	orig.data <- data
	vars_factor <- colnames(data)[!sapply(data, is.numeric)]
	if (length(vars_factor)==1){
	levels_factor <- list(gdata::mapLevels(x=data[,vars_factor]))
	}
	if (length(vars_factor)>1){
	levels_factor <- sapply(data[,vars_factor], FUN = function(x) gdata::mapLevels(x=x))
	}

. Sorry, in the previous comment the lines weren't assigned as I expected.

[rabea-a]

Hello,
I have to correct my suggestion,- sorry!!
Unfortunately, I haven't noticed that my solution does not work for data with only one factor variable. I would suggest to do a case distinction in lines 466 and 467:

if (length(vars_factor) > 0){
          for (i in 1:length(vars_factor)){
            tmp$data.init[,vars_factor[[i]]] <- as.factor(tmp$data.init[,vars_factor[[i]]])
            tmp$data.incomp[,vars_factor[[i]]] <- as.factor(tmp$data.incomp[,vars_factor[[i]]])
            
            if(length(vars_factor) == 1){
              gdata::mapLevels(x=tmp$data.init[,vars_factor[[i]]]) <- levels_factor[[i]]
              gdata::mapLevels(x=tmp$data.incomp[,vars_factor[[i]]]) <- levels_factor[[i]]
            }else{
              gdata::mapLevels(x=tmp$data.init[,vars_factor[[i]]]) <- levels_factor[[1]][[i]]
              gdata::mapLevels(x=tmp$data.incomp[,vars_factor[[i]]]) <- levels_factor[[1]][[i]]
            }
            
          }
        }

Greetings,
Rabea