Bug: Formula interface not handling total codes when given as named list.

[jnlb]

When providing explanatory variables via the dimVar option, the user may provide a set of total codes for each variable as either a vector of strings or a named list of strings. For the formula interface, only a vector of strings works, and a named lists causes the program to crash with an uninformative error. Below is a minimal reproducible example.

library(SSBtools)
library(GaussSuppression)

dataset <- SSBtoolsData("d2s")
microdata <- MakeMicro(dataset, "freq")

GaussSuppressionFromData(
  data = microdata,
  formula = ~ region * main_income,
  freqVar = "freq",
  total = list(region ="SSS", main_income = "ALL"),
  primary = FALSE,
  protectZeros = FALSE
)

The error this produces is the following:

Error in firstROW[rep(1, NROW(ur)), , drop = FALSE] : 
  incorrect number of dimensions

Tested on an environment with:

• R 4.3.3
• SSBtools 1.7.5
• GaussSuppression 1.0.0

What I know so far

I think there are only two possibilities: 1) This is unintended behaviour (bug), or 2) instead, the fact that a named list accidentally works for the dimVar interface is unintended behaviour.

In case of 1) the failure probably stems from FormulaSums.R in the SSBtools package: https://github.com/statisticsnorway/ssb-ssbtools/blob/31d8d68692f15b6a1bd0be155cafbda561bfc2ff/R/FormulaSums.R

However, it is also viable to go for option 2), i.e. define the parameter total to always be a vector of strings, e.g. total = c(region ="SSS", main_income = "ALL") in the example above. In this case, this information could be more clearly indicated in the documentation and the GaussSuppressionFromData function could even sanitize the input before passing it on to SSBtools::ModelMatrix. Or a more informative error message could be given to the user.

[olangsrud]

Thanks for the feedback. At Statistics Norway, it's not very common to use the total parameter. Recoding is usually done afterwards. So it's perhaps not so surprising that total input leads to some unexpected behavior.

In the ModelMatrix() function, the total parameter is only vaguely documented, and the function itself is a wrapper around several others. The underlying function, Formula2ModelMatrix() (same as FormulaSums()), is actually only implemented to handle total as a single string. In that sense, the fact that it works with a vector might actually be considered a bug.

Improvements to both the documentation and code should probably be made in the SSBtools package. I'll leave the issue open here for now.