Validate a VR long-format dataset — vrc_validate

Validate that a dataset follows the vrcmort canonical VR long format. This function is intentionally opinionated: it aims to catch common failure modes early (duplicate cells, missing denominators, negative counts, and pipelines where missing values are silently converted to zeros).

The function does not modify data unless repair = TRUE.

Usage

vrc_validate_data(
  data,
  id_cols = c("region", "time", "age", "sex", "cause"),
  y_col = "y",
  exposure_col = NULL,
  conflict_col = "conflict",
  allow_na_y = TRUE,
  duplicates = c("error", "warn"),
  check_missing_as_zero = TRUE,
  repair = FALSE
)

Arguments

data: A data.frame.
id_cols: Character vector of identifier column names. Defaults to c("region","time","age","sex","cause").
y_col: Name of the observed count column. Defaults to "y".
exposure_col: Name of the exposure column. If NULL (default), uses "exposure" if present; otherwise uses "pop" if present.
conflict_col: Name of the conflict intensity column. Defaults to "conflict".
allow_na_y: Logical. If TRUE (default), y may contain NA.
duplicates: How to handle duplicated identifier rows. One of "error" (default) or "warn".
check_missing_as_zero: Logical. If TRUE (default), warns when there is evidence that missing rows may have been converted to zeros.
repair: Logical. If TRUE, creates an exposure column from pop when exposure is absent.

Value

Invisibly returns the input data.frame. If repair = TRUE and exposure is absent but pop is present, an exposure column is added.