Comparing PolicyEngine and official UK poverty rates

This analysis compares PolicyEngine's poverty estimates against official figures for 2023/24 using the Family Resources Survey (FRS). To understand the differences, we first describe how official statistics are produced.

The Department for Work and Pensions (DWP) produces official poverty statistics through the Households Below Average Income (HBAI) series. The methodology uses the FRS with enhancements including replacing high earners' incomes with HMRC tax data, applying CALMAR calibration weighting to match Census totals, and adjusting benefit amounts using administrative records. Poverty is measured as relative (60% of median income) and absolute (2010/11 threshold uprated by inflation), both before and after housing costs.

PolicyEngine uses two versions of the Family Resources Survey (FRS) microdata. The standard FRS uses data as collected by the Office for National Statistics, applying the original FRS household weights without additional calibration, with income components extracted directly from survey responses and benefits used as reported without administrative data adjustments. We also construct the Enhanced FRS, which incorporates additional data sources and adjustments, reweights survey responses to match external benchmarks from HMRC tax records and DWP benefit statistics, and replaces or supplements survey-reported income with administrative data where available, particularly for benefits, tax credits, and pension income.

Both approaches calculate relative and absolute poverty thresholds using the same definitions as official statistics, measuring household income both before and after housing costs. The following sections compare poverty rates for the total population and by age group, and describe methodological factors explaining the differences.¹

The table below compares PolicyEngine's estimates with official DWP figures for 2023/24:

Measure	DWP	PE Standard FRS	PE Enhanced FRS
Relative poverty (BHC)	17.2%	18.0% (+0.8pp)	16.6% (-0.6pp)
Relative poverty (AHC)	21.1%	22.9% (+1.8pp)	20.7% (-0.4pp)
Absolute poverty (BHC)	15.0%	17.2% (+2.2pp)	12.8% (-2.2pp)
Absolute poverty (AHC)	18.3%	20.5% (+2.2pp)	17.9% (-0.4pp)

BHC = Before Housing Costs, AHC = After Housing Costs

The total population is 67.5 million in the DWP figures, 67.5 million in the standard FRS, and 67.3 million in the Enhanced FRS. The standard FRS estimates run +0.8pp to +2.2pp higher than official figures, with AHC poverty showing larger deviations than BHC. The Enhanced FRS matches official relative poverty closely but diverges on absolute poverty.

The tables below break down poverty rates by age group.

The table below shows poverty rates for children:

Measure	DWP	PE Standard FRS	PE Enhanced FRS
Relative poverty (BHC)	23.2%	24.6% (+1.4pp)	22.7% (-0.5pp)
Relative poverty (AHC)	30.5%	33.5% (+3.0pp)	29.5% (-1.0pp)
Absolute poverty (BHC)	20.0%	23.6% (+3.6pp)	17.9% (-2.1pp)
Absolute poverty (AHC)	26.4%	30.2% (+3.8pp)	25.4% (-1.0pp)

The children population is 14.6 million in the DWP figures, 14.4 million in the standard FRS, and 14.2 million in the Enhanced FRS. Children show the highest poverty rates and largest deviations, with the standard FRS estimates +1.4pp to +3.8pp higher than official figures. The Enhanced FRS reduces these deviations to -0.5pp to -2.1pp.

The table below shows poverty rates for working-age adults:

Measure	DWP	PE Standard FRS	PE Enhanced FRS
Relative poverty (BHC)	14.7%	15.7% (+1.0pp)	16.1% (+1.4pp)
Relative poverty (AHC)	19.4%	21.2% (+1.9pp)	21.0% (+1.6pp)
Absolute poverty (BHC)	12.9%	15.1% (+2.1pp)	12.2% (-0.8pp)
Absolute poverty (AHC)	16.9%	19.2% (+2.4pp)	18.6% (+1.7pp)

The working-age adult population is 40.9 million in the DWP figures, 41.8 million in the standard FRS, and 43.2 million in the Enhanced FRS. The standard FRS deviations range from +1.0pp to +2.4pp, with larger AHC gaps reflecting Housing Benefit under-reporting among renters. The Enhanced FRS deviations range from -0.8pp to +1.7pp.

The table below shows poverty rates for pensioners:

Measure	DWP	PE Standard FRS	PE Enhanced FRS
Relative poverty (BHC)	18.6%	19.1% (+0.5pp)	12.1% (-6.6pp)
Relative poverty (AHC)	15.7%	17.8% (+2.2pp)	10.5% (-5.2pp)
Absolute poverty (BHC)	16.0%	18.0% (+2.0pp)	9.4% (-6.6pp)
Absolute poverty (AHC)	13.2%	14.6% (+1.5pp)	7.4% (-5.8pp)

The pensioner population is 12.1 million in the DWP figures, 12.0 million in the standard FRS, and 10.8 million in the Enhanced FRS. The standard FRS deviations range from +0.5pp to +2.2pp. The Enhanced FRS deviations range from -6.6pp to -5.2pp, with all measures showing lower poverty rates than official statistics.

The DWP and PolicyEngine use the same FRS data, but process it differently. The following sections explain the key methodological factors that produce the observed differences.

The DWP replaces high earners' incomes with HMRC tax data through the Survey of Personal Incomes (SPI) adjustment (above £361,200 for working-age adults or £107,800 for pensioners). PolicyEngine incorporates SPI data differently: rather than directly replacing individual incomes, it augments the FRS with synthetic individuals trained on SPI data and reweights the combined dataset to match HMRC income distribution targets across multiple income bands and sources.

The DWP uses CALMAR calibration weighting to match Census population totals across demographic dimensions including age-sex-region combinations, benefit unit composition, council tax bands, and tenure. PolicyEngine's standard FRS uses original FRS weights without calibration, producing a population estimate of 67.5 million matching official figures. The Enhanced FRS reweights to match hundreds of targets from multiple administrative data sources—including OBR tax and benefit aggregates, ONS demographic statistics, HMRC income distributions, DWP benefit caseloads, and VOA council tax records—at both national and constituency levels. Different weights alter which household types are weighted up or down.

The DWP adjusts benefit amounts using administrative records, though the methodology is not publicly documented. PolicyEngine's standard FRS uses benefit amounts as reported in the survey. The Enhanced FRS incorporates HMRC and DWP administrative data. These different approaches produce different household income estimates, affecting measured poverty rates for households receiving means-tested benefits like Universal Credit and Housing Benefit.

Both methods define housing costs identically but face Housing Benefit under-reporting in the FRS. The DWP adjusts Housing Benefit using administrative records. PolicyEngine's standard FRS uses reported Housing Benefit without adjustment, while the Enhanced FRS reweights to match official Housing Benefit expenditure and claimant count targets. Under-reporting in the standard FRS means net housing costs appear higher, lowering income after housing costs and raising AHC poverty.

The DWP assigns cash-equivalent values to in-kind benefits like free school meals and childcare, and may adjust Child Benefit using administrative data. PolicyEngine's standard FRS calculates benefits from survey data using legislative rules without administrative adjustments.

Administrative data adjustments may better capture pension income from multiple sources where survey self-reporting is less accurate. The Enhanced FRS pensioner population is 10.8 million versus 12.0 million in the standard FRS and 12.1 million official. The Enhanced FRS targets older age bands (60-70, 70-80, 80-90) by region along with pension income and benefit aggregates, but the reweighting optimization produces this lower pensioner count as a trade-off to better match other targets.

This analysis compared PolicyEngine's poverty estimates against the DWP statistics for 2023/24 using both standard and Enhanced FRS data. The standard FRS deviations range from +0.5pp to +3.8pp across all measures and demographic groups, with AHC poverty showing larger deviations than BHC. The largest differences are for child poverty AHC measures.