specifyInput add check for missing

[Kyoshido]

Hello,

so I was tasked with creating synthetic data for a dataset with out hhid and hhsize. And I stumbled upon two errors related to hhid.

Let's use example with eusilc data.

The code below is working properly.

library("simPop")
data("eusilcS")
seed <- 1234

inp <- specifyInput(data=eusilcS,
                    hhid="db030",
                    hhsize="hsize",
                    strata="db040",
                    weight="db090"
                    )
inp

 -------------- 
survey sample of size 11725 x 19 

 Selected important variables: 

 household ID: db030
 personal ID: pid
 variable household size: hsize
 sampling weight: db090
 strata: db040
 -------------- 

1. Now let’s take a look at a similar code. But hhid and hhsize are absent.

inp <- specifyInput(data=eusilcS,
                    strata="db040",
                    weight="db090"
                    )
inp

Error in specifyInput(data = eusilcS, strata = "db040", weight = "db090") :    argument "hhid" is missing, with no default

Ok, so there is an error argument "hhid" is missing, with no default

The function definition of specifyInput says:

specifyInput <- function(data, hhid=NULL, hhsize=NULL, pid=NULL, weight=NULL, strata=NULL, population=FALSE)

This suggests that hhid is optional, because it has a default value NULL.

Inside the function, there's this block:

if(is.null(hhid)){
  # initialize hhid
  hhid <- "hhid.simPop"
  data[, hhid.simPop := .I]
}

So why do we get this error "hhid" is missing, with no default then?

Problem here is not hhid = NULL, but that hhid is missing altogether. Which is exactly what the error message explicitly says.

When a parameter is not passed at all, inside the function, hhid is not yet evaluated — it is not NULL — it is missing.

So inside the function, the condition is.null(hhid) is evaluted with error argument "hhid" is missing, with no default, because hhid is missing — not provided yet at all.

On an object that is not defined → hence the error.

So I recommend adding these checks at the beginning of the function:

if (missing(hhid)) hhid <- NULL
if (missing(hhsize)) hhsize <- NULL
if (missing(pid)) pid <- NULL

This way, the function will work as expected and will not display an error message when hhid or hhsize are not provided.

So I added it to my fork:

devtools::install_github("kyoshido/simPop")
library(simPop)
inp <- specifyInput(data=eusilcS,
                    strata="db040",
                    weight="db090"
                    )
inp

 -------------- 
survey sample of size 11725 x 21 

 Selected important variables: 

 household ID: hhid.simPop
 personal ID: pid.simPop
 variable household size: hhsize.simPop
 sampling weight: db090
 strata: db040
 -------------- 

Now it works as expected.

2. However, there was another error when the parameters were included, but set to NULL.

inp <- specifyInput(data=eusilcS,
                    hhid=NULL,
                    hhsize=NULL,
                    strata="db040",
                    weight="db090"
                    )
inp

Error in if (!inherits(hhid, "character") | length(hhid) != 1 | is.na(match(hhid,  : 
  argument is of length zero

the error message says argument is of length zero.

When we explicitly write hhid = NULL
The function see:

hhid  # is NULL
inherits(hhid, "character")  # FALSE
length(hhid)  # 0
match(hhid, colnames(data))  # integer(0)
is.na(match(hhid, colnames(data)))  # logical(0)

Logical test is not properly evaluated → hence the error.

But all this was properly handled by adding the check for missing(hhid) at the beginning of the function.

[matthias-da]

We discussed this in our meetings that simPop already can do this. There was an update 6 months ago.
Please see the Commit f76e4c1

You just need to install the github version

library(devtools)
install_github("statistikat/simPop")

and everything works. We still did not submit it to CRAN, because we need to clean simPop to not receive any notes.