Calculate local SPM thresholds for congressional district-level analysis

[MaxGhenis]

Problem

When reweighting CPS records to congressional district targets, SPM thresholds remain attached to records from their original geography. This causes:

1. Reduced geographic variation - California records reweighted to Iowa still carry California's ~20% higher threshold
2. Biased poverty estimates - Low-cost areas get overstated poverty (thresholds too high)
3. Invalid local policy analysis - SPM-based eligibility calculations become unreliable

Background: SPM Threshold Components

The SPM threshold has three components (Census methodology):

1. Base threshold - From BLS Consumer Expenditure Survey, 83% of median spending on food, clothing, shelter, utilities, telephone, and internet (FCSUti)
2. Housing tenure adjustment - Three tenure groups: renters, owners with mortgage, owners without mortgage
3. Geographic adjustment (GEOADJ) - Based on median gross rent index from ACS 5-year data

The GEOADJ formula: GEOADJ = (local_median_rent / national_median_rent) * 0.492 + 0.508

Geographic variation ranges from ~0.84 (West Virginia) to ~1.27 (Hawaii) - roughly 50% variation that we're currently losing when cloning records.

Proposed Solution

Fully calculate SPM thresholds from source data:

1. Base threshold from CE Survey - Use rolling 5-year Consumer Expenditure Survey to calculate 33rd percentile FCSUti by tenure (follows BLS methodology, enables proper forecasting)
2. District GEOADJ from ACS - Calculate geographic adjustment for each of 435 congressional districts using ACS Table B25031 (median 2-bedroom rent)
3. Apply during district reweighting - threshold = base × equiv_scale × district_geoadj

Implementation Components

New module: policyengine_us_data/spm/

1. ce_threshold.py - Download CE Survey PUMD, calculate 33rd percentile FCSUti by tenure
2. district_geoadj.py - Generate GEOADJ lookup table from ACS median rents by congressional district
3. local_threshold.py - Calculate local thresholds: base × equiv_scale × district_geoadj

Updates to Enhanced CPS

Ensure these SPM unit variables are passed through (some already extracted but unused):

• NUMADULTS, NUMKIDS - Family composition
• TENMORTSTATUS - Housing tenure status
• EQUIVSCALE - Equivalence scale
• GEOADJ - Original geographic adjustment (for validation)

Benefits

• Proper geographic variation in poverty thresholds (~50% range from WV to HI)
• Rolling 5-year CE data enables accurate forecasting (replaces simple CPI-U uprating)
• Modular components can be updated independently
• Full control over methodology

Acceptance Criteria

• CE Survey threshold calculator matches BLS published values within 2% (2024 reference: $39,430 renter, $39,068 owner w/ mortgage, $32,586 owner w/o)
• District GEOADJ ranges from ~0.84 to ~1.27
• National aggregate SPM threshold matches Census totals
• State-level poverty rates within reasonable range of Census SPM estimates

References

• Census SPM Threshold Methodology
• BLS 2024 SPM Thresholds
• Geographic Adjustments: Fair Market Rents and SPM (Census 2024)

Related Issues

• Add congressional district targets #212 - Add congressional district targets
• Critical Data Bug: All Congressional Districts Assigned to Wyoming #435 - Critical bug: All districts assigned to Wyoming
• Add script that generates spm_threshold_agi.csv #33 - Add script that generates spm_threshold_agi.csv