Skip to contents

IPTW Weight-Trimming Sensitivity Sweep

Source: R/sa-sensitivity.R
sa_trim_sweep.Rd

Re-computes IPTW weights across a range of winsorisation thresholds and reports how the weight distribution and effective sample size (ESS) change as trimming increases. This helps identify a trimming threshold that adequately controls extreme weights without an unacceptable loss of effective sample size.

Usage

sa_trim_sweep(
  x,
  trim_seq = seq(0, 0.1, by = 0.01),
  estimand = c("ATE", "ATT", "ATC"),
  stabilise = TRUE,
  score_col = "prob_t",
  treatment_col = "tavr",
  treated_level = NULL
)

Arguments

x

A ps_weight object (returned by ps_weight()) or a ps_match object or a plain data frame that contains a propensity score column and a binary treatment column.

trim_seq

Numeric vector of trimming values to evaluate. Each value is the lower tail quantile; weights outside the (trim, 1 - trim) range are winsorised. 0 means no trimming. Default seq(0, 0.10, by = 0.01).

estimand

One of "ATE", "ATT", or "ATC" – the causal estimand used to compute IPTW weights. Taken from x$meta$estimand when x is a ps_weight object; defaults to "ATE" otherwise.

stabilise

Logical. Whether to stabilise weights. Taken from x$meta$stabilised when available; defaults to TRUE. As in ps_weight(), stabilisation applies to the ATE only.

score_col

Name of the propensity score column. Taken from x$meta$score_col when x is a ps_data object. Default "prob_t". The score of a ps_match() object built with scale = "logit" is a logit, and is back-transformed with stats::plogis() before use.

treatment_col

Name of the binary treatment column. Taken from x$meta$treatment_col when x is a ps_data object. Default "tavr".

treated_level

The value of treatment_col that is treated. Read from the metadata of a ps_data object when it records one. NULL (default) requires a 0/1 or logical column and takes 1 or TRUE.

Value

An object of class c("sa_trim_sweep", "ps_data"):

$data

The input data frame (x$data for a ps_data object), unchanged.

$meta

score_col, treatment_col, treated_level, scale, estimand, stabilised (whether stabilisation was applied, so FALSE for the ATT and ATC), trim_seq (sorted), n_total and n_used, the patients with a non-missing treatment and score.

$tables$sweep

Data frame with one row per trim value and columns trim (the winsorisation threshold), ess_control and ess_treated (effective sample sizes), max_weight (maximum weight after trimming), sd_weight (standard deviation of the weights after trimming) and pct_trimmed (percentage of patients whose weight was winsorised).

$models

An empty list.

Details

For a ps_data input, a supplied estimand, stabilise, score_col, treatment_col or treated_level must agree with the object's metadata; a conflict aborts. Arguments not supplied are read from the metadata, then take their defaults.

See also

ps_weight(), sa_rosenbaum(), sa_evalue(), sa_overlap()

Examples

dta    <- sample_ps_data(n = 400, seed = 3)
w_obj  <- ps_weight(dta, estimand = "ATE")
sweep  <- sa_trim_sweep(w_obj)
head(sweep$tables$sweep)
#>   trim ess_control ess_treated max_weight sd_weight pct_trimmed
#> 1 0.00       303.1       305.0     4.4440    0.5578           0
#> 2 0.01       312.6       312.7     3.1193    0.5209           2
#> 3 0.02       316.2       315.5     2.8898    0.5068           4
#> 4 0.03       320.7       318.9     2.7022    0.4895           6
#> 5 0.04       328.8       325.9     2.4054    0.4558           8
#> 6 0.05       339.1       334.9     2.0931    0.4126          10

# Directly from a data frame
sweep2 <- sa_trim_sweep(dta, score_col = "prob_t", treatment_col = "tavr",
                        estimand = "ATE")