Repository navigation
Alaska's reported dividend stays in miscellaneous_income, so the simulated dividend counts it twice #1443
Description
Problem
Populace US carries reported Alaska Permanent Fund Dividends in miscellaneous_income, and policyengine-us also computes the dividend for every Alaska resident in ak_permanent_fund_dividend. Both reach market income, so people who reported the dividend count it twice.
Evidence
The file is populace_us_2024.h5 at populace-us-2024-spm-20260909, the default for policyengine-us 2.90.2. These are unweighted person records:
| State | Share of people with miscellaneous_income above 0 |
|---|---|
| Alaska | 11.05% |
| Washington | 0.80% |
| Oregon | 0.69% |
Alaska's most common positive values are $3,284, $1,312, $3,200, $1,702 and $2,622. Several match past payments: $3,284 is the 2022 dividend with its energy relief, $2,622 is the 2022 dividend alone, $1,312 is the 2023 dividend, and $1,702 is the 2024 dividend with its energy relief. So the rule should also take out ak_energy_relief, which policyengine-us computes for some years. I could not match $3,200.
packages/microcosm-build/src/microcosm/build/us_runtime/alimony.py splits ASEC OI_VAL by OI_OFF. Code 20 goes to alimony, code 12 goes to strike benefits, and every other code goes to miscellaneous_income. I found no code that takes the dividend out, and no Census source that names a separate field for it.
scripts/ak_misc.py in ~/reviews/household-spm-recon-2026-11 reproduces the table.
Effect
household_market_incomealready counts both today.spm_unit_market_incomestarts to count both when Count the same cash payments in household and SPM income policyengine-us#10573 merges, because it adds the dividend to personmarket_income. That PR fixes the model's own lists, so it should not wait for this one, and its body states the overlap.
Proposed fix
Remove reported dividends from Alaska residents' miscellaneous_income at build time, so that the simulated dividend stands alone. The file can't identify the dividend by source, so the rule needs a tested match on amount and year, or a share taken from the difference between Alaska's rate and comparable states' rates. Either choice should appear in the build's provenance.
Done means
- Alaska's share of people with
miscellaneous_incomeabove 0 falls near that of comparable states. - A test fails on any Alaska record whose reported other income equals a past dividend amount.
- The release notes state the change in Alaska's market income, because it lands in a data release.
🤖 Generated with Claude Code
Activity
MaxGhenis commentedon Nov 9, 2026
#1551 proposes one build stage that covers this issue and also seeds Alaska's dividend take-up. It matches
OI_VALagainst the dividend and energy relief amounts for the record's source year and the prior year, within $100, for one to eight dividends.In a prototype on the default file, the match reaches 23.37% of Alaska's people, against a 0.16% median in other states and 1.70% at most. Zeroing the matching ASEC rows removes $192.5 million of
miscellaneous_incomeand lowers Alaska's unweighted share above 0 from 11.05% to 1.10% (Washington 0.80%, Oregon 0.69%). The prototype appears in no Populace release.🤖 Generated with Claude Code