
Blindly benchmark age-splitting methods against a single-age schedule
Source:R/age.R
benchmark_age_split.RdAggregate an independent single-age population schedule to source_breaks,
restore it with each method, and compare the restored schedule with the
held-out source. This provides a reproducible validation workflow for an
official single-age population series without allowing the target series to
be used as a "reference" schedule.
Arguments
- population
Held-out single-age population schedule. Supply a named numeric vector or a data frame accepted by
expand_age_pop(). Ages at or above the finalsource_breaksvalue are treated as the open group.- source_breaks
Lower bounds used to blind the held-out schedule.
- methods
Character vector of methods passed to
expand_age_pop().- reference
Independent reference schedule used only by the
"reference"method.- cases
Optional held-out single-age case schedule. If supplied, age-standardized-rate bias is calculated using the restored population.
- stdpop
Optional single-age standard-population weights for ASR. Equal weights are used when
casesis supplied withoutstdpop.- mp
Multiplier for the ASR, defaulting to 100,000.
- high_age
High-age configuration passed to
expand_age_pop().- pclm_control
PCLM control settings passed to
expand_age_pop().
Value
A data frame with MAE, RMSE, total variation distance, ASR bias,
maximum source-group error, number of negative values, and available
uncertainty-interval diagnostics. The restored schedules are stored in
the estimates attribute; the error for every effective source group is
stored in the group_errors attribute.