Missing values might not treated well in simContinous

[matthias-da]

Here is an executable example:

Data:

library(simPop)
data(eusilc13puf, package = "simPop")
df <- eusilc13puf[,c(1:6, 8:9,14, 16, 46)]
df$age <- as.numeric(df$age)
df$pid <- as.factor(df$pid)
vars <- c("hhid", "hsize", "region", "age", "sex", "pid", "eco_stat", "citizenship", 
          "pgrossIncome","hgrossIncome","weight")
colnames(df) <- vars
# We are using reduced weights to obtain a reduced population by factor 100 for computation time
df$weight <- df$weight/100

inp <-
  specifyInput(
    data = df,
    hhid = "hhid",
    hhsize = "hsize",
    strata = "region",
    weight = "weight"
  )

simPop <-
  simStructure(
    data = inp,
    method = "distribution",
    basicHHvars = c("age", "sex", "region") 
  )

simPop <-
  simCategorical(
    simPop,
    additional = c("eco_stat", "citizenship"), 
    method = "multinom"
  )

And here comes the error:

simPop <- simContinuous(
  simPop,
  method = "lm",
  additional = "pgrossIncome",
  regModel = ~ sex + hsize + eco_stat + citizenship + age,
  zeros = TRUE,
  log = FALSE,
  const = 1,
  alpha = NULL, #0.05,
  residuals = TRUE
) 

Reason is most likely related missing values in variable ecoStat that causes in line 338 of simContinous

X <- model.matrix(estimationModel, data = dataS) # has about 10000 rows
y <- dataS[[name]] # has more than 13000 entries
weights <- dataS[[weight]] # has more than 13000 entries

Afterwards mod <- logitreg(X, y, weights = weights) fails.

[JohannesGuss]

@matthias-da fixed bug with commit f76e4c1 in simContinuous() when zero=TRUE was specified.
As stated in the issue missing values where not treated correctly when synthesizing 0s. Now for the 0-modelling step the missing values will be imputed with VIM::hotdeck() and a warning is raised. This code snipped was already in the function but commented out.