Stage 6: honest reject inference through a sensitivity band

Does not ship parceling as the default behavior: instead of inventing a single multiplier and reweighting, it declares the population scope of the scorecard, measures the coverage per band (where an observed outcome exists, and in what volume) and presents a sensitivity band: the event rate each band would have if the population without an outcome were 2, 4 or 8 times worse than the observed one, with the effect on the total. The analyst reads the band; no single number is fabricated.

Usage

scr_reject(
  x,
  population = NULL,
  accepted = NULL,
  multipliers = NULL,
  sample = "holdout"
)

Arguments

x

An object from scr_scorecard().

population

Optional: a table of the full population (accepted and rejected, without outcome), scored by scr_apply(). NULL restricts the scope to the population with an outcome.

accepted

Optional: a logical vector, of the length of population, marking the rows with an observed outcome. NULL treats the whole population as without an outcome beyond the development sample.

multipliers

Sensitivity band. NULL uses the configuration.

sample

Reference sample of the observed outcomes.

Value

An scr_reject object with scope, coverage (per band) and sensitivity (per band and multiplier, plus the TOTAL row).

See also

Other stages: scr_align(), scr_bin(), scr_cutoff(), scr_model(), scr_scorecard(), scr_select(), scr_split(), scr_strategy(), scr_triage()

Examples

cfg <- scr_config(verbose = FALSE, nthread = 1, use_ranger = FALSE,
                  xgb_rounds = 60, n_boot = 20)
res <- scr_select(scr_demo, "default", config = cfg, drop = "id",
                  date_col = "ref_date")
sc <- scr_scorecard(res)
scr_reject(sc)
#> <scr_reject> target "default" | multipliers 2x, 4x, 8x
#>   The scorecard describes the population WITH an observed outcome. No extrapolation to rejects was made; the sensitivity band shows the effect of declared assumptions, not an inferred number.
#>   observed event rate: 14.50%
#>   implied rate if the population without outcome is 2x worse: 14.50%
#>   implied rate if the population without outcome is 4x worse: 14.50%
#>   implied rate if the population without outcome is 8x worse: 14.50%
#>   bands with weak coverage: (590, Inf] (few_events), (577,590] (few_events), (567,577] (few_events), (558,567] (few_events), (550,558] (few_events), (542,550] (few_events), (533,542] (few_events)
# with a through-the-door population: rows with an outcome are the hold-out
acc <- seq_len(nrow(scr_demo)) %in% res$split$holdout_idx
scr_reject(sc, population = scr_demo, accepted = acc)
#> <scr_reject> target "default" | multipliers 2x, 4x, 8x
#>   The full population has 4,200 rows, of which 1,400 (33.3%) have an observed outcome. The rest enter only the sensitivity band, under declared multipliers.
#>   observed event rate: 14.50%
#>   implied rate if the population without outcome is 2x worse: 24.56%
#>   implied rate if the population without outcome is 4x worse: 40.45%
#>   implied rate if the population without outcome is 8x worse: 53.43%
#>   bands with weak coverage: (590, Inf] (few_events), (577,590] (few_events), (567,577] (few_events), (558,567] (few_events), (550,558] (few_events), (542,550] (few_events), (533,542] (few_events)