External hire premium

Audit

The paper trail. Rules set in advance, a sample that cannot move, and every null reported with its uncertainty.

Why this page exists

The result is only as good as the discipline behind it.

A null finding is easy to fake by accident. Tune the sample, pick the window, drop the awkward cases, and any effect can be made to appear or vanish. This page records the guardrails that keep that from happening here.

Decision log

Every ruling, dated

More than one hundred decisions are logged, D-001 onward, each with the date, the choice, and the rule applied. Classification calls, exclusions, parser fixes, and analysis choices are all on the record.

Pre-committed rules

Set before the data

The classification rules and exclusions were written before any pay, price, or accounting data was pulled. They could not be adjusted once a result was in view.

Frozen sample

611, and it does not move

453 internal, 158 external, 149 excluded by rule. The sample is fixed. Coverage gaps in pay or returns are reported, not closed by quietly adding cases.

Nulls reported plainly

No effect is still a result

The premium-to-return slope is flat, and that is stated as the headline, with its p-values, not buried. Directional results that miss significance are labeled as such, not dressed up.

What was checked and set aside

Two examples of the discipline

  • Comp gap-filling. Recovering more pay records was tried. The recovered cases loosened the premium estimate rather than tightening it, which showed the imprecision is structural. The clean estimate was kept. The sample was not padded.
  • Naming individuals. A per-CEO overpay or discount callout was rejected. With a null finding, a selection confound, and noisy per-person pay, naming people would be unfair and unsupported. Only name-free aggregates are shown.

Full detail is in the decision log and the methods commitments in the repository.