Design-based supervised learning correction (T step of CRAFT)

Wraps dsl::dsl() so misclassification uncertainty in the LLM-generated labels propagates into the downstream regression estimates.

Usage

dsl_fit(
  data,
  formula,
  predicted_var,
  prediction,
  sample_prob,
  model = "felm",
  fixed_effect = NULL,
  index = NULL,
  cluster = NULL,
  ...
)

Arguments

data: A data frame containing the predicted label column, the gold-label column (for the audited subset), the sample-inclusion probability column, and any covariates / fixed-effect indices / cluster identifiers used in formula.
formula: A formula expression for the outcome model (e.g. sup ~ ideology + female + senate + education).
predicted_var: Character name of the outcome column in data.
prediction: Character name of the LLM-prediction column.
sample_prob: Character name of the inclusion-probability column.
model: Underlying regression model passed to dsl::dsl(). Default "felm" for fixed-effect linear models.
fixed_effect, index, cluster: Passed through to dsl::dsl().
...: Additional arguments forwarded to dsl::dsl().

Value

The object returned by dsl::dsl().

Examples

if (FALSE) { # \dontrun{
  dsl_out <- dsl_fit(
    data    = legis_df,
    formula = sup ~ shor_ideo + per_mining + female + senate + perc_bchhigherE,
    predicted_var = "sup",
    prediction    = "pred_sup",
    sample_prob   = "cand_incl_prob_all",
    fixed_effect  = "oneway",
    index   = "state",
    cluster = "state"
  )
} # }