EquityStack · Inequality measurement and survey estimation in Python

Inequality

inequality/ · 24 measures · survey-weighted

Concentration index

Is the outcome concentrated among the poor?

The concentration index on a wealth quintile or a continuous wealth measure. Negative means concentrated among the poor. Erreygers and Wagstaff corrections for comparing places or years with different prevalence.

Read the module →

Indices

How unequal is it?

Gini, Theil T and L, Atkinson, the generalised entropy family, Palma and the 80/20 ratio, quantile shares, Lorenz curves.

Read the module →

Decomposition

Where does it sit?

Theil decomposition splits national inequality exactly into within-state and between-state parts. Blinder-Oaxaca splits a group gap into endowments and returns.

Read the module →

Benefit incidence

Who receives the budget?

Allocates public spending across quintiles by who uses the service, overall and by level (primary, secondary, tertiary).

Read the module →

54 tests, each checking an identity the measure must satisfy or a value worked out by hand from the definition. Guide, sources and a bootstrap recipe for standard errors in inequality/README.md.

Survey estimation

survey_estimation/ · stratified, clustered designs

Estimator

Design-based standard errors

Weighted proportions and means with Taylor-linearised standard errors, the same estimator as Stata's svy: and R's survey. Subgroups as domains, logit intervals for proportions, t on clusters minus strata.

Read the module →

Worked example

NFHS-5 stunting by wealth quintile

Reproduces the published table from the raw children's recode: 35.5 per cent nationally, 46.1 in the poorest fifth, 22.9 in the richest. Reads the CSV written by InsightStack's DHS loader.

Read the script →

Agrees with R's survey 4.2.1 to twelve significant figures on the same data.

Modules

Ten, plus notebooks and sample data

InequalityPython Gini, Theil, Atkinson, Palma, concentration indices, decompositions, benefit incidence. Survey estimationPython Design-based proportions and means for NFHS, DHS or PLFS. Impact evaluationPython Difference-in-differences, propensity score matching, regression discontinuity. CleaningPython Column names, types, missing values, outlier flags, with a log of changes. ExplorationPython Summaries and profiling for a new dataset. ModellingPython Multicollinearity checks (VIF). ValidationPydantic Input validation for columns and types. Input and outputPython Stata and SPSS import, chunked CSV reading, formatted Excel export. Social sectorPython Composite access and vulnerability indices, with a check on how ranks change under a different normalisation. VisualisationPython District maps and annotated charts. NotebooksJupyter Gender-disaggregated outcomes; women's time use. Sample dataCSV Small datasets to try the code on.

Links open on GitHub.

Status

Stable

The code works and is not under active development. Bug reports are welcome; new features are unlikely, and replies take weeks. Dependencies are pinned so a clone still runs later. See the maintenance policy.

Companion repositories: InsightStack (calculators and the DHS and PLFS loaders) and FieldStack (the same survey work in R).