Analyze effective take-up rates across all benefit programs

[MaxGhenis]

Background

We currently seed benefit take-up rates in policyengine-uk-data based on prior studies. However, these rates change as a result of:

• Reweighting processes
• Integrating SPI (Survey of Personal Incomes) data

Current Analysis

We have an initial analysis for Universal Credit and Child Tax Credit:
https://gist.github.com/MaxGhenis/763db9278ddecdf310f160a73e138c8a

Request

We need a comprehensive analysis of effective take-up rates across all benefit programs in the UK model, including but not limited to:

• Universal Credit (UC)
• Child Tax Credit (CTC)
• Working Tax Credit (WTC)
• Pension Credit
• Housing Benefit
• Council Tax Support/Reduction
• Child Benefit
• Income Support
• Jobseeker's Allowance (JSA)
• Employment and Support Allowance (ESA)
• Personal Independence Payment (PIP)
• Disability Living Allowance (DLA)
• Attendance Allowance
• Carer's Allowance
• State Pension

Deliverables

1. Documentation of initial seeded take-up rates (from prior studies)
2. Calculation of effective take-up rates after reweighting and SPI integration
3. Comparison between seeded vs. effective rates
4. Analysis of how data processing steps affect take-up assumptions
5. Recommendations for any adjustments needed

This will help us understand how our data processing pipeline affects benefit modeling and ensure our simulations reflect realistic take-up patterns.

[MaxGhenis]

Converting to discussion instead

[nwoodruff-co]

@PolicyEngine could you try doing this

[policyengine]

I ran into an issue:

Claude Code timed out after 600s

[nwoodruff-co]

@PolicyEngine try doing this (you have a timeout of 10m so do it quickly)

[policyengine]

Benefit Take-Up Rates Analysis

I analyzed the take-up rate parameters in policyengine-uk. Here's what I found:

Current Seeded Take-Up Rates

Benefit	Take-Up Rate	Source/Notes
Universal Credit	55%	Since 2015
Pension Credit	70%	DWP estimates 2019-20
Child Benefit	89%	2022 release (97% in 2012)
JSA (Income-based)	56%	DWP 2015-16 (was 69% in 2009)
Marriage Allowance	100%	Default

Legacy Benefits (100% - closed to new claims)

• Child Tax Credit
• Working Tax Credit
• Housing Benefit
• Income Support

These are set to 100% because only existing claimants are tracked (no new claims allowed).

Benefits WITHOUT explicit take-up rates

The following use reported values from survey data without applying a separate take-up model:

• ESA (Income/Contributory)
• PIP (Daily Living/Mobility)
• DLA (Self-care/Mobility)
• Attendance Allowance
• Carer's Allowance
• State Pension
• Council Tax Support/Benefit

How Take-Up Works in the Model

Take-up is applied via would_claim_* variables that use random sampling:

takes_up = random(benunit) < parameters(period).gov.dwp.universal_credit.takeup_rate

For Pension Credit specifically, it's more sophisticated - if someone reports receiving PC or newly becomes eligible, take-up is applied probabilistically.

Recommendations for Full Analysis

To calculate effective take-up rates after reweighting/SPI integration, you'd need to:

1. Compare reported vs eligible: For each benefit, calculate the ratio of people reporting receipt to those who would be eligible at 100% take-up
2. Run against the enhanced FRS dataset: Using policyengine-uk-data, compare:
   • Sum of [benefit]_reported (weighted)
   • Sum of [benefit]_entitlement at 100% take-up (weighted)
3. Use official totals for validation: DWP publishes aggregate expenditure figures that can be compared to model outputs

This would require running simulations against the actual dataset (not just analyzing parameters), which I can't do within the time constraint. The referenced gist by @MaxGhenis appears to do this for UC and CTC specifically.

Key Files

• policyengine_uk/parameters/gov/dwp/*/takeup*.yaml - Take-up parameters
• policyengine_uk/variables/gov/dwp/would_claim_*.py - Take-up logic
• policyengine_uk/variables/gov/dwp/*_reported.py - Survey-reported benefit receipt