Get Default Subset Tiers for Post-Hoc Optimization

Returns a structured list of tiered criteria for identifying optimal subsets of calibration results. Uses a hybrid degradation strategy with 30 hardcoded tiers.

Usage

get_default_subset_tiers(
  target_ESS_best = 500,
  target_A = 0.95,
  target_CVw = 0.7
)

Arguments

target_ESS_best

Numeric target ESS for the best subset (default: 500)

target_A

Numeric target agreement index (default: 0.95)

target_CVw

Numeric target coefficient of variation (default: 0.7)

Value

A named list where each element contains:

• name: Tier identifier

• A: Target agreement index (0-1)

• CVw: Maximum coefficient of variation for weights

• ESS_B: Target effective sample size

Details

The function generates 30 hardcoded tiers with a hybrid degradation strategy:

Tiers 1-20 (ESS-constant):

• ESS_B remains constant at target_ESS_best

• A degrades by 5% per tier (multiplied by 0.95)

• CVw increases by 5% per tier (multiplied by 1.05)

• Prioritizes statistical power while relaxing quality criteria

Tiers 21-30 (Fallback):

• All three criteria degrade by 5% per tier

• ESS_B, A, CVw all relax simultaneously

• Graceful degradation when ESS target cannot be achieved

Example with the function defaults (target_ESS_best=500, target_A=0.95, target_CVw=0.7):

Tier  1: ESS=500, A=0.950, CVw=0.700  (Stringent)
Tier  5: ESS=500, A=0.774, CVw=0.851
Tier 10: ESS=500, A=0.599, CVw=1.086
Tier 20: ESS=500, A=0.358, CVw=1.769  (Last ESS-constant)
Tier 21: ESS=475, A=0.341, CVw=1.857  (Fallback begins)
Tier 30: ESS=299, A=0.215, CVw=2.881  (Final fallback)

run_MOSAIC() calls this with its control targets (ESS_best = 100, A_best = 0.70, CVw_best = 1.0 by default), which gives:

Tier  1: ESS=100, A=0.700, CVw=1.000
Tier 20: ESS=100, A=0.264, CVw=2.527
Tier 30: ESS=60,  A=0.158, CVw=4.116

Examples

if (FALSE) { # \dontrun{
# Get default 30 tiers
tiers <- get_default_subset_tiers(
  target_ESS_best = 500,
  target_A = 0.95,
  target_CVw = 0.7
)

# Custom starting criteria
tiers_custom <- get_default_subset_tiers(
  target_ESS_best = 600,
  target_A = 0.90,
  target_CVw = 0.75
)

# Loop through tiers in calibration with grid search
for (tier_name in names(tiers)) {
  tier <- tiers[[tier_name]]
  result <- grid_search_best_subset(
    results = results,
    target_ESS = tier$ESS_B,
    target_A = tier$A,
    target_CVw = tier$CVw,
    min_size = 30,
    max_size = 1000
  )
  if (result$converged) break
}
} # }