Weight truncation at a percentile of the weight distribution

Potter's simplest procedure: pick a cut point from the empirical weight distribution and pull every larger weight down onto it. Truncation trades bias for variance, so the removed mass is reported and a rescaled copy restoring the original total is returned too. The cut is the type-7 quantile.

Usage

Trmwgt(weights, quantile = 0.99)

Arguments

weights: Non-negative design weights.
quantile: Upper percentile in (0, 1], default 0.99.

Value

List with estimate (cut point), weights, rescaled, n_trimmed, mass_removed, sumw, sumw_trimmed, cv_before, cv_after, n, method.

Details

Formula: w_i' = min(w_i, Q_q(w)).

References

Potter, F. J. (1990). A study of procedures to identify and trim extreme sampling weights. Proceedings of the Section on Survey Research Methods, American Statistical Association, 225-230.

Examples

Trmwgt(c(1, 1, 1, 100), 0.75)
#> $estimate
#> [1] 25.75
#> 
#> $weights
#> [1]  1.00  1.00  1.00 25.75
#> 
#> $rescaled
#> [1]  3.582609  3.582609  3.582609 92.252174
#> 
#> $n_trimmed
#> [1] 1
#> 
#> $mass_removed
#> [1] 74.25
#> 
#> $sumw
#> [1] 103
#> 
#> $sumw_trimmed
#> [1] 28.75
#> 
#> $cv_before
#> [1] 1.92233
#> 
#> $cv_after
#> [1] 1.721739
#> 
#> $n
#> [1] 4
#> 
#> $method
#> [1] "weight truncation at the type-7 q-th percentile [Potter 1990]"
#>