---
title: "Attrition and influence after a verified design"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Attrition and influence after a verified design}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(collapse = TRUE, comment = "#>")
```

Attrition and influence diagnostics can support sensitivity review, but they do
not replace the verified external target or authorize outcome-driven reweighting.

After a weight plan is locked, inspect influential units without changing the
weights:

```{r influence, eval=FALSE}
influence <- wf_influence(
  locked,
  target = target,
  sample = design$data,
  id = design$roles$id,
  top = 20
)

influence$table
```

If panel-retention weights are studied, keep their formula, stabilization, trim
settings, and balance diagnostics as a separate sensitivity analysis. Do not
select those settings to make an outcome cross a preferred threshold.

```{r impact, eval=FALSE}
impact <- wf_assess_impact(
  locked,
  analysis_data,
  id = "person_id",
  outcomes = c("score", "approved")
)

wf_report(impact, audience = "statistician")
```