Cut bootstrap trend draws down to what the targets actually need

The full England grid is 79 ages x 16 years x 2 sexes x 5 quintiles x 3 states. At 1,000 iterations that is roughly 38 million rows, and Tables 7 to 10 need about a tenth of it. Thin before stacking rather than after.

The thinning is checked. A filter that quietly matched nothing would produce a raw bootstrap file that looks fine on disk and is missing a whole table.

Usage

thin_trend_draws(dt, keep_ages, keep_years, keep_states = "current")

Arguments

dt

A single trend_fit output (one bootstrap iteration).

keep_ages, keep_years

Integer vectors to retain.

keep_states

Character vector of smoking state columns to retain.