sae.fay_herriot_mse

Per-area EBLUP estimates and their mean squared error (Prasad-Rao) from a fitted Fay-Herriot model -- the model output of sae.fay_herriot or sae.all_subsets_fh. Adds the standard error ratio, the relative efficiency versus the direct estimate, residuals, and the synthetic (regression-only) prediction. Nothing is refitted: the model carries the coefficients, their covariance, and the between-area variance with its asymptotic variance and bias, which is everything an area's MSE needs beyond its own sampling variance and covariates.

Areas without a direct estimate (a missing target or sampling variance, or a table with neither column) get the synthetic prediction as FH_est and no MSE.

  • Kind: standard
  • Network: no

Inputs🔗

NameTypeDescription
datafile (Parquet or CSV)One row per area, with the model's predictor columns and, where available, the direct estimate and its sampling variance under the names the model was fitted with.
modelfile (SpadeModel)A fay_herriot model.
id_columnsstringColumn selector for identifier columns to carry through unchanged, e.g. STATECD,COUNTYCD,COUNTYNM,surveyunit. Default empty.
na_actionstringAn area with a missing predictor: fail (default), drop, or keep it with missing estimates.

Outputs🔗

NameTypeDescription
estimatesfile (Parquet or CSV, as the input)One row per area: the id columns, then direct_est, direct_var, FH_est (EBLUP), FH_mse, RE (= FH_mse / direct_var), SER (= sqrt(RE)), resid (= FH_est - direct_est), and syn_pred.
summaryjsonmethod, refvar, n_areas, n_with_direct, mean_ser, mean_re.

Example🔗

- id: "@estimates"
  name: sae.fay_herriot_mse
  inputs:
    - block: "@long_data"
      output: result
      as: data
    - block: "@selection"
      output: model
      as: model
  args:
    id_columns: "STATECD,COUNTYCD,COUNTYNM,surveyunit"