DDIwR (version 0.20)

recodeMissings: Consistent recoding of (extended) missing values

Description

A function to recode all missing values to either SPSS or Stata types, uniformly (re)using the same codes across all variables.

Usage

recodeMissings(
  dataset,
  to = c("SPSS", "Stata", "SAS"),
  dictionary = NULL,
  start = -91,
  ...
)

Value

A data frame with all missing values recoded consistently.

Arguments

dataset

A data frame

to

Software to recode missing values for

dictionary

A named vector, with corresponding Stata missing codes to SPSS missing values

start

A named vector, with corresponding Stata missing codes to SPSS missing values

...

Other internal arguments

Author

Adrian Dusa

Details

Package DDIwR uses numeric declared missing codes as its internal R representation. Recoding to "Stata" or "SAS" is therefore best viewed as an export-oriented temporary representation, not as a preferred internal storage strategy for declared vectors in R.

When no dictionary is provided, export-oriented recoding can be performed either by scanning the entire dataset for a harmonized mapping, or by recoding each variable independently. The package now defaults to the faster per-variable strategy in export preparation. Supplying a dictionary keeps the harmonized cross-dataset behavior.

When a dictionary is not provided, it is automatically constructed from the available data and metadata, using negative numbers starting from -91 and up to 27 letters starting with "a".

If the dataset contains mixed variables with SPSS and Stata style missing values, unless otherwise specified in a dictionary it uses other codes than the existing ones.

For the SPSS type of missing values, the resulting variables are coerced to a declared labelled format.

Unlike SPSS, Stata does not allow labels for character values. Both cannot be transported from SPSS to Stata, it is either one or another. If labels are more important to preserve than original values (especially the information about the missing values), the argument chartonum replaces all character values with suitable, non-overlapping numbers and adjusts the labels accordingly.

If no labels are found in the metadata, the original values are preserved.

Examples

# x <- data.frame(
#     A = declared(
#         c(1:5, -92),
#         labels = c(Good = 1, Bad = 5, NR = -92),
#         na_values = -92
#     ),
#     B = labelled(
#         c(1:5, tagged_na('a')),
#         labels = c(DK = tagged_na('a'))
#     ),
#     C = declared(
#         c(1, -91, 3:5, -92),
#         labels = c(DK = -91, NR = -92),
#         na_values = c(-91, -92)
#     )
# )

# xrec <- recodeMissings(x, to = "Stata")

# attr(xrec, "dictionary")

# Supply a dictionary to harmonize missing meanings across variables
# dictionary <- data.frame(
#     old = c(-91, -92, "a"),
#     new = c("c", "d", "c")
# )
# recodeMissings(x, to = "Stata", dictionary = dictionary)

# recodeMissings(x, to = "SPSS")

# dictionary$new <- c(-97, -98, -97)

# recodeMissings(x, to = "SPSS", dictionary = dictionary)

# recodeMissings(x, to = "SPSS", start = 991)

# recodeMissings(x, to = "SPSS", start = -8)