Add WID pre-tax national income and standardized inequality summary stats

[MaxGhenis]

Summary

Create variables for standard income concepts used by major inequality databases (WID, LIS, OECD), starting with WID's pre-tax national income. Also add standardized summary statistic computation (Gini, percentile shares, etc.) that matches the methodology of these external sources.

Motivation

PolicyEngine computes household_net_income and household_market_income, but these don't map directly to the income concepts used by the World Inequality Database (WID), OECD, or LIS. This makes it hard to:

1. Validate against external benchmarks — e.g., OWID reports US Gini of 0.587 (2023) using WID pre-tax national income, but PE's net income Gini is 0.565 and market income Gini is 0.643. Neither matches.
2. Compare cross-country — WID's methodology is consistent across countries, so implementing it in PE would enable apples-to-apples comparisons.
3. Forecast for prediction markets — There are active prediction markets on OWID Gini values that PE microsim could inform, but only if we compute the same income concept.

WID pre-tax national income definition

From WID methodology:

• Pre-tax national income = factor income (labor + capital) + pensions + unemployment insurance, before taxes and transfers
• Unit: Equal-split among adults (20+) in the same household — i.e., total household pre-tax national income divided equally among adult members
• Population: Adults aged 20+

Key differences from PE's current variables:

• Includes pension/UI income (unlike pure market income)
• Excludes means-tested transfers (unlike net income)
• Equal-split among adults only (not per-capita, not equivalized)

Proposed implementation

New variables in policyengine-us

# Person-level
wid_pretax_national_income_person  # = (labor + capital + pensions + UI) / n_adults_in_household

Components:

• employment_income + self_employment_income (labor)
• capital_gains + dividend_income + interest_income + rental_income (capital)
• social_security + pension_income (pensions)
• unemployment_compensation (UI)
• Divided by number of adults (20+) in the tax unit/household

Consider for policyengine-core or policyengine.py

If the income concept definition is common across countries (which it is for WID), the variable template or summary stat functions could live in a shared package:

• wid_pretax_national_income as a cross-country variable pattern
• Standardized gini(variable, population_filter, weighting) that matches WID's methodology
• Percentile share computation matching WID conventions

Acceptance criteria

• wid_pretax_national_income_person variable computes correctly
• Equal-split among adults 20+ in the household
• Gini coefficient on this variable is close to OWID's reported 0.587 for 2023
• Document methodology and any remaining discrepancies

References

• WID methodology
• OWID Income Inequality page
• OWID US Gini 2023: 0.5868841 (WID pre-tax national income, equal-split adults)
• PE microsim comparison: net income Gini=0.565, market income Gini=0.643 (2025)