---
title: "Medicare Part D: Are Insurers Gaming the Low Income Subsidy Design?"
format:
  html:
    theme: cosmo
    toc: true
    embed-resources: true
    include-in-header:
      text: |
        <style>main{overflow-wrap:anywhere;max-width:1100px!important}table{display:block;overflow:auto;font-size:.85rem}td,th{padding:6px}img{max-width:100%}p code,li code{overflow-wrap:anywhere;white-space:normal}</style>
---

[Francesco Decarolis (2015)](https://doi.org/10.1257/aer.20130903). *American Economic Review*.

**Under development — community extension.** This is a provisional source-design assessment, not a new successful replication or a final eligibility ruling.

[Extension index](development.html) · [Provisional eligibility log](development.html#eligibility-log) · [Assessment JSON](aeraej-277.json) · [QMD](aeraej-277.qmd) · [Reproduction code](collection-code.zip)

## Source design and template applicability

Table 5 relates premiums to continuous current/lagged low-income-subsidy enrollment shares; the paper also uses market-level IV specifications. This mixed design is not a binary-adoption ATT.

No DiD estimates, uncertainty intervals or homogeneity-test decision were computed in this tranche. No point is added to the calibration plots. This is not a zero effect or a failed statistical test.

## Recorded treatment-column measurement

**13,794 distinct nonmissing values** in `past_lis_tot_enrol_ratio`.

Observed support in the indicated source frame; missing values excluded. N=27,861; missing=11,485. Frame: dt. Source: AER/Decarolis_AER_2015_112933/R/277_decarolis_replicate.R.

These are the existing data-measured counts, not counts inferred from design labels. A binary instrument, group indicator or interaction component need not be the treatment of interest; the source-design assessment above takes precedence for template applicability.

## Replication status

**This tranche:** unsupported design; source code/documentation inspected, no estimator run.

**Supplied library:** verified; verification: verified; preparation: runs_with_shim. These are the library’s labels, not new verification claims. Missing data and unsupported designs are separate issues.

## Evidence and provenance

The following files in the supplied library were inspected. Their SHA-256 digests are recorded in the downloadable assessment JSON. Source archives and microdata are not redistributed.

- `AER/Decarolis_AER_2015_112933/R/277_decarolis_figure.R` (101 lines)
- `AER/Decarolis_AER_2015_112933/R/277_decarolis_replicate.R` (303 lines)
- `AER/Decarolis_AER_2015_112933/verdict.md` (57 lines)
- `AER/Decarolis_AER_2015_112933/aeraej-277.md` (120 lines)