Child poverty modelling: update

5. Methodology

5.1 Cumulative impact assessment

The policy package modelled in the CIA includes the following policies: Free School Meals; School Clothing Grant; Council Tax Reduction (including water and sewerage discount); Discretionary Housing Payments (bedroom tax and benefit cap mitigation); Carers Allowance Supplement; Best Start Grant; Best Start Foods; Scottish Child Payment; employability services; and mitigation of the two-child limit. The methodology underpinning the CIA, which is based on the microsimulation model UKMOD, has been detailed in the original publication and subsequent updates. This section focuses on two changes that have been made since the previous update, namely the introduction of the policy to mitigate the two-child limit and improvements to our modelling of FSM.

Our modelling of two-child limit mitigation is consistent with the assumptions set out by the SFC in their costing published in January 2025.^[21] That is, we assume that mitigation will be paid from April 2026, acknowledging that Ministers have committed to making payments earlier if feasible to do safely and securely; that the payment will not be treated as income in the reserved social security system, including the benefit cap; and that it will be paid to all affected families who are in receipt of UC, but not to those who would become newly eligible for UC by means of the taper if the two-child limit had been removed at source. Unlike SFC, however, we do not incorporate behavioural effects as this is beyond the current capabilities of our model.

Our modelling of FSM has been improved in several ways. Firstly, whereas FSM were previously modelled with full take-up, UKMOD now assumes partial take-up using data from the School Healthy Living Survey.^[22] This is assumed to occur through intermittency of take-up rather than categorical non-take-up, meaning the annualised value per child is adjusted rather than the caseload. Secondly, we have revised the assumed value per meal to be consistent with the FRS. We previously based this value on the cost of providing a meal in primary and secondary schools as reported in the Local Government Finance Statistics.^[23] However, the cost to local authorities is likely higher than the value obtained by families, as approximated for example by the price that is charged to families who pay for school meals. We have therefore used the same value per meal as the FRS – specifically, £2.53 in 2023-24 – projected forwards with CPI. Finally, the measure of earnings that is used to determine FSM eligibility for households receiving UC has been amended to be consistent with the definition of earned income under UC.

5.2 Welfare reform

5.2.1 The two-child limit

The estimated numbers of children affected over time are based on official UK Government statistics, which show the rounded number of households affected by the two-child limit in Scotland as of April each year.^[24] The statistics also show, at the GB/UK level only, the number of affected households with three, four, and five or more children.^[25] Separate statistics are provided on the number of households and children in those households that are notionally affected by the two-child limit – that is, before any exemptions are applied – and the number that have CTC or the UC child element withheld for at least one child in the household. However, the child breakdown is only available for the second category, which is our main interest, from 2021-22. We fill in these gaps by applying GB/UK proportions to Scotland, and by applying 2021-22 proportions to previous years.

Our estimate of the cumulative amount of income withheld from Scottish households as a result of the two-child limit is based on running a set of scenarios in UKMOD in which the two-child limit was removed in every year between 2017-18 and 2024-25 and then calculating the additional income from UC and CTC that Scottish households receive in those scenarios.

In most cases, where input data from the Family Resources Survey is available for the year in question – that is from 2017-18 to 2022-23 – that year of data is used to run the simulation. The exception is the 2020-21 system: as the survey sample for this year was undermined by Covid-19 lockdowns, we use pooled data from 2017-18 to 2019-20 instead. As the sample size for 2021-22 was also reduced, pooled data from 2021-22 and 2019-20 is used to simulate this year. From 2023-24 onwards, a pooled sample of 2021-22 and 2022-23 is projected forwards to each year.

The estimate takes into account interactions within the benefit system, notably by excluding additional support from removing the two-child limit that would be withheld by the benefit cap. It also takes into account the impact on eligibility – that is, it includes additional support that would be received by households that become eligible for UC or CTC as a result of the two-child limit being removed – but does not count the increase in support that households may consequently receive through passported benefits.

The estimates for each year are calibrated by scaling the cost according to the percentage difference between the number of households affected in UKMOD and the number of households affected in the statistics, estimated as a simple mean between the April at the beginning of the year and the April immediately after. The relevant households in UKMOD are those who were already receiving UC or CTC before the two-child limit is removed – that is, not brought into eligibility as a result of the removal – regardless of whether they are affected by the benefit cap. The relevant households in the statistics are those that were not receiving a child element for at least one child, noting that some of those households may still have been receiving exceptions for other children. Since data was not yet available for April 2025 at the time of writing, the number of affected households at that point is extrapolated from the time series.

Note that UKMOD determines whether a child is subject to the two-child limit based on their age in whole years, as day and month of birth are not recorded in the input data. This leads to an imperfect alignment with the age cut-off of April 2017, particularly since different households are surveyed at different points in the financial year. As a result, we can either incorrectly exempt some children from the limit who are in fact subject to it; or we can incorrectly subject some children who are in fact exempt. The first option is preferred in UKMOD as the scale of the error is likely to be smaller.

For example, among households interviewed over the 2024-25 financial year, a child recorded as being seven years old could have been born as early as April 2016 (turning eight in April 2024, shortly after the interview) and as late as April 2018 (turning seven in April 2025, shortly before the interview). Setting the minimum age for exemption at seven therefore incorrectly exempts those younger seven-year-olds who were born between April 2017 and March 2018. However, setting the minimum age at eight in order to ensure these children were subject to the limit would, by extension, incorrectly subject those older eight-year-olds who were born before April 2017. This would be in fact be a large majority of eight-year-olds, assuming a relatively even distribution of births across the year.

A side-effect of this approach is that it implies that no children were subject to the two-child limit in 2017-18, since children from age 0 upwards are modelled as being exempt. We do not attempt to compensate for this omission, since the number of affected over 2017-18 is unknown and likely small: the official statistics begin in April 2018, at which point 3,870 households in Scotland were denied a child element for at least one child as a result of the policy. In addition, the sample in the output data that represents the affected group in our model would likely to be too small to derive meaningful results.

Note that UKMOD does not model other exceptions to the two-child limit, which can be applied in the event of multiple births, non-parental care, adoption, or non-consensual conception. Nor does it distinguish between new and existing claims. Only the former are technically subject to the limit, but will comprise an increasingly large proportion of the caseload over time. A final caveat is that the model does not incorporate demographic change, meaning that a static population is assumed for years that are projected forwards from the input data. These caveats are to some extent addressed by the calibration method outlined above.

5.2.2 The benefit cap

Our analysis of the benefit cap is based on official DWP data, available from the Stat-Xplore platform, and includes all affected households whether they are capped through UC or HB. Note that negligible figures are suppressed in the data and therefore assumed to be zero in this analysis.

For purposes of estimating the amount of income withheld by the benefit cap – estimated by multiplying the mean amount withheld by the number of households affected – the latest month of data (November 2024) is assumed to be representative of the remainder of 2024-25. HB amounts, which are given in weekly terms, are converted to monthly terms to ensure comparability with UC.

Families on UC with five or more children are assumed to have five children, while families on HB with more than five children are assumed to have six children. Prior to April 2019, the number of children in capped households is not available; we estimate this by applying the proportions in April 2019 to the total number of capped households in each previous month.

5.2.3 Cuts to Local Housing Allowances

This analysis compares the reported rent of each household in the Scottish Household Survey (SHS) who reports receiving UC or HB with the relevant LHA for that household and with the relevant 30^th percentile as calculated in the previous September from Rent Service Scotland data. This allows us to count the number of households – and children within those households – whose notional housing-related benefit would have been higher if LHAs had been linked to the 30^th percentile, as well as to quantify the difference. In the few cases where the 30^th percentile is lower than the LHA, the analysis assumes that LHAs would not be reduced.

To correct for underreporting of benefits as well as any underrepresentation of the private rented sector, the weighted results are proportionally scaled up to match administrative data on the average number of private sector tenants on UC or HB in each year, split into households with and without children. DWP data on UC recipients whose rents exceed their LHAs are used to validate this method. The validation exercise indicates that our analysis may still be somewhat underestimating the number of children who are affected, although this comparison is not definitive since the DWP data only starts in 2019 and will include some cases where the 30^th percentile is not higher than the LHA. The data also excludes HB and does not quantify the degree to which rents exceed LHAs.

The SHS is collected in calendar years. Our analysis allocates LHAs based on the financial year in which each respondent’s interview took place, but the results are presented as simple averages of two calendar years. Data collection for the years 2020 and 2021 was carried out remotely due to Covid-19 lockdown restrictions; as there may consequently be concerns around consistency of the data over time, averages that include either of these years are highlighted in the chart. The figure presented for 2011-2012 assumes that no households were negatively affected by LHAs diverging from the 30^th percentile in 2011, since LHAs tracked the 50^th percentile in 2010-11 and the 30^th percentile in 2011-12.

Service charges also count towards eligible rent, but are not included in this analysis due to a lack of data. The analysis does not capture households who earned too much to be eligible for UC, but would have been brought into UC eligibility had LHAs been higher; nor does it take into account the effects of the benefit cap, which for some households would limit any gains from increasing LHAs.