Bug: SIPP tip income imputation includes allocation flag columns

[MaxGhenis]

Summary

The tip income construction in sipp.py sums all columns matching *TXAMT*, which inadvertently includes both the actual tip dollar amounts (TJB*_TXAMT) and Census allocation flags (AJB*_TXAMT). The allocation flags are small integers (0, 1, 2) indicating whether Census imputed the value, not dollar amounts.

Current code

policyengine_us_data/datasets/sipp/sipp.py line ~69-72:

df["tip_income"] = (
    df[df.columns[df.columns.str.contains("TXAMT")]].fillna(0).sum(axis=1)
    * 12
)

Fix

Filter to only the actual tip amount columns:

df["tip_income"] = (
    df[df.columns[df.columns.str.match(r"TJB\d_TXAMT")]].fillna(0).sum(axis=1)
    * 12
)

Impact

Likely minor since allocation flags are small integers vs dollar amounts, but it's incorrect and should be fixed.

Context

This was identified while comparing PolicyEngine's tip income deduction revenue estimate ($4.7B) against JCT's score ($10.0B for FY2026). See related issues for other improvements to close this gap.