Skip to contents

Stratum sizes, source against synthetic

Source: R/compare.R
compare_pmx_strata_sizes.Rd

One row per declared stratum level, with the source size, the synthetic size, and the size the generator was aiming for. Reported for each strata column on its own, and – when more than one is declared – for the joint cells too, because the joint cell is what the generator balances on.

Usage

compare_pmx_strata_sizes(source, synthetic, roles)

Arguments

source

Source PMX data.

synthetic

Generated synthetic PMX data.

roles

Explicit roles from pmx_roles().

Value

A pmx_strata_sizes data frame with column, level, source_patients, synthetic_patients, expected and balanced. Zero rows when the roles declare no strata.

Details

expected is the source share of the cohort carried to the synthetic cohort size, which is the generator's own target rule. It matters whenever n_subjects differs from the source size: every stratum then changes size by design, and comparing the raw counts says "everything moved" when nothing is wrong.

balanced is whether the cell had enough source patients (3) to be held exactly. Cells below that floor are deliberately left to vary by seed – reproducing the size of a two-patient arm on every run discloses that size. A cell above the floor landing far from expected is not that mechanism, and is worth chasing.

With more than one strata column, balanced is NA on the single-column rows: the floor applies to the joint cell, so a column with 19 patients can still be split into joint cells that all fall below it.

This reads real patient data on both sides and is marked "restricted_not_releasable".

See also

synpmx_scorecard(), which reports this as row C1.

Examples

data <- pmx_simulated_fixture(20)
data$ARM <- ifelse(as.integer(data$ID) %% 2L == 0L, "A", "B")
roles <- pmx_roles(
  id = "ID", time = "TIME", dv = "DV", amt = "AMT", evid = "EVID",
  cmt = "CMT", dvid = "DVID", covariates = "WT", strata = "ARM"
)
synthetic <- suppressWarnings(synpmx_avatar(data, roles, seed = 1))
#> synpmx_avatar(): dropped 9 undeclared column(s): NTIME, TAD, OCC, RATE, MDV, CENS, LIMIT, AGE, SEX.
#>   Declare a column in `keep` to carry it through verbatim.
compare_pmx_strata_sizes(data, synthetic, roles)
#> PMX stratum sizes (source against synthetic)
#> 
#> 20 source patients -> 20 synthetic patients.
#> 
#> ARM:
#>   A
#>     10 source -> 10 synthetic (expected 10.0)
#>   B
#>     10 source -> 10 synthetic (expected 10.0)
#> 
#> Cells under the floor of 3 source patients are left to vary by seed on purpose.
#> Source-derived; not releasable.