Find a weight threshold below which observations "do not contribute"

Find a weight threshold below which removing observations does not change the point estimate or the standard error of the estimate of interest by more than a given proportion.

Usage

idid_contrib_threshold(
  reg,
  var_interest,
  threshold_change = 0.05,
  search_step = 0.05,
  ...
)

Arguments

reg

A regression object.

var_interest

A string. The name of the main variable of interest.

threshold_change

A double (between 0 and 1). The change threshold in estimate and s.e. when dropping observations.

search_step

A double (between 0 and 1). The additional proportion of observations to drop in each iteration of the loop.

...

Additional elements to pass to the regression function when partialling out controls.

Value

A numeric. The largest weight below which one can drop observations without altering the estimate by more than a proportion of threshold_change.

Details

This function applies ididvar::idid_drop_change for a series of proportions of observations removed, increasing this proportion by search_step while the variation in the estimate and s.e. as compared to the full sample is less than threshold_change.

Examples

reg_ex <- ggplot2::economics |>
  lm(formula = unemploy ~ pce + uempmed + psavert + pop)

idid_contrib_threshold(reg_ex, "pce")
#> [1] 0.0005871943