Add the inequality package, so the repository's name is earned

[Varnasr]

What was here

Counted before starting: 27 Python files outside tests/, 872 lines. survey_estimation (357), impact_evaluation (220) and one cleaning pipeline (155) were 732 of them. Exactly one thing in the repository was about distribution, and the file whose name promised it, public_health_access_index.py, was a five-line row mean.

What this adds

inequality/, 24 functions across five modules, ~950 lines.

Module	Contents
indices.py	Gini, the generalised entropy family with Theil T and mean log deviation named, Atkinson at any aversion, Palma, 80/20, quantile shares, Lorenz curve, weighted quantiles
concentration.py	The concentration index, Erreygers and Wagstaff corrections, concentration curve, a by-group table, Wagstaff's achievement index
decomposition.py	Theil within/between split, Blinder-Oaxaca in twofold and threefold form
incidence.py	Benefit incidence, single-service and split by level of service
ranks.py	Weighted fractional ranks, shared by everything above

The concentration index is the one that makes the name true. An average tells you the level; it cannot tell you whether stunting falls on the poor, which is the question a programme budget answers to.

Three decisions that are centralised rather than repeated

Ranks live in one file. The Gini, the concentration index and both curves share fractional_rank, so they agree by construction. Ties take the block midpoint, which is what makes a wealth quintile a legitimate ranking variable and what the World Bank's DHS equity work does.

A raw concentration index is not comparable across prevalences. Its possible range shrinks as a binary outcome's mean moves away from 0.5, so the same indicator at 8 per cent and at 60 per cent cannot go in one table uncorrected. Both fixes are here and they can disagree about the direction of change over time; the README says to pick one per table and state it.

Every function refuses rather than returning a plausible wrong number. A Gini on negative values, GE(0) on zeros, Wagstaff where everyone has the outcome, a service level with no budget, an Oaxaca on three groups.

Tests: 25 → 101

Every new test is an identity or a hand-computed figure, never previous output. A distributional index cannot be checked by eye: 0.31 and 0.34 are both plausible consumption Ginis, and a sign error gives a number in exactly the right range that says the opposite of the truth.

• Gini against brute-force mean absolute difference on fifteen random datasets
• Curve areas against their indices by trapezoid, for both Lorenz and concentration
• within + between == total for GE(0) and GE(1), with weights
• Oaxaca components summing to the gap under all four reference choices
• GE(2) equalling half the squared coefficient of variation
• A weight of 2 equalling the row appearing twice
• The mean fractional rank being exactly 0.5, which three derivations depend on

Two of my own test expectations were wrong on first run and the code was right, which is the point of pinning to closed forms. An Oaxaca test drew both groups from the same distribution and then asserted the realised endowment gap was zero; a composite-index test hardcoded arithmetic I had done in my head. Both now compute the expectation from the definition.

Also

social_sector/composite_index.py replaces the row mean with an index that makes its three decisions arguments rather than assumptions: direction, normalisation and weights. Adds goalpost normalisation so a series is comparable across years, geometric aggregation where indicators should not substitute for each other, a minimum non-missing count so a district reporting two of six indicators is not scored as though it reported all six, and rank_sensitivity to show how far a ranking moves under an equally defensible choice. compute_access_index keeps working.

Deleted workflows/ci.yml: dead configuration, pinned to Python 3.10 and running R tests in a repository with no R, duplicating the real workflow. The landing page described it as "start-to-finish routes through an analysis".

🤖 Generated with Claude Code

https://claude.ai/code/session_01DMYhECx2Dzbqad6KJxWZ7t

---

Generated by Claude Code