Summarise Monte Carlo runs for a DGP × estimator combination

This helper takes the output of cs_run_seeds() (or any tibble with the same columns) and computes Monte Carlo summary statistics for the ATT error. The input is expected to contain one row per seed.

Usage

cs_summarise_runs(runs)

Arguments

runs

A tibble or data.frame, typically the output of cs_run_seeds(), with columns including: dgp_id, estimator_id, n, seed, oracle, supports_qst, true_att, est_att, att_error, att_abs_error.

Value

A tibble with one row and summary columns:

• dgp_id, estimator_id, n, oracle, supports_qst

• n_runs: number of seeds / runs

• mean_true_att, mean_est_att

• mean_error, sd_error

• mean_abs_error, max_abs_error

If multiple (dgp_id, estimator_id, n) combinations are present, the function summarises within each group and returns one row per group.

Moment-regime limitation

These columns are direct summaries of the supplied finite set of runs; this function does not establish that their population analogues exist. mean_error and mean_abs_error require the corresponding first moments, and sd_error requires a finite second moment. max_abs_error is specific to the supplied replicate set and is not a stable tail-performance metric. On infinite-variance or no-mean DGPs such as synth_heavytail, do not use these columns for bias, RMSE, conventional Monte Carlo standard errors, or estimator rankings. Inspect robust summaries of the per-run deviations and use QST for valid distributional comparisons instead.