Build a per-replicate refit that recomputes visibility at the report level

Internal. The expensive bootstrap path, for an estimated rule whose visibility is not constant within an ego X cell row. There the frame-split identity does not apply, and the only correct route is to refit the rule and re-predict for every report inside each replicate, then re-aggregate.

Usage

make_vis_refit_esc(
  rule,
  donor.dat,
  boot.weights,
  esc.dat,
  ec.dat,
  cell.vars,
  ego.id = ".ego.id"
)

Arguments

rule

the visibility_rule

donor.dat

the donor frame

boot.weights

data frame of replicate weights

esc.dat

the ego X alter X cell reports, carrying ind_vis's inputs

ec.dat

the ego X cell data the estimate is computed from

cell.vars

the columns defining a cell

ego.id

name of the ego id column

Value

function(r) returning a two-column data frame of num and denom, one row per row of ec.dat, or NULL if the rule is not estimated

Details

Costs roughly M times a point estimate. That is the price of not freezing a sample quantity; falling back to the cheap path here would silently reinstate the very bug the refit exists to prevent.