Interpret Super Learner weights for LMTP nuisance fits — margot_interpret_lmtp

Generates concise prose describing which Super Learner components dominate the outcome (`m`) and density-ratio (`r`) nuisance regressions across waves and shifts. Highlights waves where a single learner receives (approximately) all the weight, which can signal limited information (e.g., after LOCF imputation).

Usage

margot_interpret_lmtp_learners(
  x,
  outcome,
  shifts = NULL,
  label_mapping = NULL,
  waves = NULL,
  remove_waves = NULL,
  component = c("both", "outcome", "treatment"),
  digits = 0,
  return = c("text", "list")
)

Arguments

x: LMTP run output (e.g., the result of [margot_lmtp()]) or any object that exposes `$density_ratios` in the same structure as the plot helpers.
outcome: Character scalar giving the outcome name to summarise.
shifts: Optional character vector of shifts to include (either full names such as `t5_pwi_z_shift_up` or cleaned suffixes such as `shift_up`). If `NULL`, all available shifts for the outcome are used.
label_mapping: Optional named list passed to [transform_label()] for readable outcome/shift labels.
waves: Optional integer vector of waves to keep (matching the wave index used by the LMTP fits).
remove_waves: Optional integer vector of waves to drop after subsetting.
component: Which nuisance models to include: `"both"` (default), `"outcome"` (only `m`), or `"treatment"` (only `r`).
digits: Integer number of decimal places to use when reporting percentages.
return: Either `"text"` (default) for a single character string or `"list"` for structured components.

Value

Either a character string or a list (depending on `return`).