Check a BRR-Fay target or rebuild it in the current format

pv_revalidate_brr_target() checks a target from pv_brr_target() and returns it in the current format. A target in the current format (schema_version "0.2.0") is returned unchanged; the function stops with an error if that target was changed after it was created. A target in the format of pvstackr 0.1.x (schema_version "0.1.0") is computed again from data and formula, and the new target is returned if it matches the old one. In pvstackr 0.2.x no target saved by pvstackr 0.1.x matches, because pvstackr 0.2.0 computes the design checksum design_hash differently (see the package NEWS); compute such a target again with pv_brr_target(). The target that you pass is not modified.

Usage

pv_revalidate_brr_target(
  target,
  data = NULL,
  formula = NULL,
  conf_level = NULL
)

Arguments

target

A BRR-Fay target (class pvstackr_brr_target).

data

For a target in the old format, the data frame from which it was built; default NULL. Not used for a target in the current format.

formula

For a target in the old format, the formula from which it was built; it must be the formula stored on the target. Default NULL; not used for a target in the current format.

conf_level

For a target in the old format, the confidence level to store in the new target, a number between 0 and 1. It is required when the old target does not store one (targets saved by pvstackr 0.1.x do not); when it does, give NULL (default) or the same value. Not used for a target in the current format.

Value

A target (class pvstackr_brr_target) in the current format: the input itself when it is already in that format, otherwise the new target.

Details

The old target is computed again with pv_brr_target(), using the columns and settings stored on it. The new target must match the old one: the formula, the settings (such as the numbers of plausible values and replicate weights, the Fay coefficient and the degrees-of-freedom rule), the fixed-effect names and design_hash exactly, and the estimates, covariance matrices, standard errors and degrees of freedom up to a small numerical tolerance (absolute 1e-12 plus relative 1e-10), because the last digits of least-squares results can differ between computers. Otherwise, and for a target in the old format without data and formula, the function stops with an error. The new target records the upgrade in binding_manifest$migration.