Merge pull request #660 from PolicyEngine/two-child-limit-post

Add two-child limit post

src/posts/posts.json

@@ -592,5 +592,14 @@
     "tags": ["us", "us-nm"],
     "image": "new-mexico-income-tax-launch.png",
     "authors": ["max-ghenis"]
+  },
+  {
+    "title": "Repealing the Universal Credit two-child limit",
+    "description": "How removing the limit of some benefits to two children would affect the UK in 2023.",
+    "date": "2023-08-07 16:00:00",
+    "filename": "repealing_the_two_child_limit.md",
+    "tags": ["uk"],
+    "image": "two_child_limit.png",
+    "authors": ["nikhil-woodruff"]
   }
 ]

src/posts/repealing_the_two_child_limit.md

@@ -0,0 +1,251 @@
+_[See the full, customisable impacts on PolicyEngine
+here.](https://policyengine.org/uk/policy?focus=policyOutput.netIncome&reform=18304&region=uk&timePeriod=2023&baseline=1)_
+
+## Summary
+
+Introduced in 2017, the two-child limit in Universal Credit restricted
+parents from receiving financial support for more than two children
+(with children born before 2017 exempt). Using the PolicyEngine
+microsimulation model, we estimate that [repealing it would cost £1.8
+billion in
+2023](https://policyengine.org/uk/policy?focus=policyOutput.netIncome&reform=18304&region=uk&timePeriod=2023&baseline=1),
+rising to £2.8 billion in 2025 as non-exempt age cohorts replace exempt
+children. The repeal would also lower absolute child poverty before
+housing costs by 310,000 this year.
+
+While the Cameron-led Conservative government introduced the two-child
+limit, last month, Labour Party leader Sir Keir Starmer [confirmed that
+a Labour government would keep the
+policy](https://www.theguardian.com/politics/2023/jul/16/labour-keep-two-child-benefit-cap-says-keir-starmer)
+due to the cost of repealing it.
+
+In this analysis, we examine how the policy currently affects
+households, and use PolicyEngine to estimate the distributional and
+poverty impacts of both full and partial repeals proposed by the Child
+Poverty Action Group and the Fabian Society. We also attempt to
+reproduce claims made using other microsimulation models. The
+[open-source code powering the custom analyses in this report can be
+found
+here](https://gist.github.com/nikhilwoodruff/447d86032ff481bdeb1e35aeeb3ea18c).
+
+[**_See how repealing the two-child limit would affect your household
+here._**](https://policyengine.org/uk/household?focus=intro&reform=7226&region=uk&timePeriod=2023&baseline=1)
+
+# How the policy affects households
+
+Households claiming Universal Credit, or the legacy Child Tax Credit,
+currently do not receive additional benefits for their third or
+subsequent children. With benefit levels recently uprated as of April
+2023, the two-child limit reduces benefits by up to £2,935 per child
+(lowered with increased parental earnings as the family reduces their
+Universal Credit entitlement). The cap applies regardless of whether a
+household is in or out of work, but its impact changes with earnings due
+to the normal Universal Credit taper.
+
+Figure 1 shows the impact of removing the two-child limit under a range
+of employment incomes for a single parent family. The gains are highest
+(and equal) for families under around £30,000 in employment income,
+after which point the taper rate begins to reclaim the extra benefit
+value. For a single parent with 5 children, this value will not be
+reduced to zero before more than £60,000 in earnings.
+
+![A graph with numbers and lines Description automatically
+generated](/images/posts/repealing_the_two_child_limit/hh_impact.png)
+
+_Figure 1: the impact of removing the UC child limit on net income for a
+single parent by number of children and employment income. Note: this
+does not incorporate housing costs, which would extend the phase-out
+region to the right._
+
+## Repealing the cap
+
+The most straightforward way to repeal the cap is to remove it entirely,
+at a cost of £1.8bn in 2023. [^1] This net cost rises over time due to
+age cohort transitions: with each year, the minimum age required to be
+exempt under the 2017 transitional protection rule increases.[^2]
+
+Under a full abolition, we estimate that the number of children in
+absolute poverty before housing costs would fall by 255,000 in 2023
+(relative, after housing costs child poverty would fall by 162,000). The
+overall absolute, before housing costs poverty rate for all individuals
+would fall by 0.7 percentage points, bringing just under 450,000 people
+out of absolute poverty.
+
+Table 1 shows a range of other estimates made by the Child Poverty
+Action Group using the UKMOD microsimulation model, and PolicyEngine's
+replications for comparison. The appendix contains more details about
+the reasons PolicyEngine's microsimulation modelling differs from other
+estimates.
+
+| Estimate                                                      | CPAG/UKMOD | PolicyEngine | Difference | Relative |
+| ------------------------------------------------------------- | ---------- | ------------ | ---------- | -------- |
+| Net cost                                                      | £1.3bn     | £1.8bn       | +£500m     | +38%     |
+| Child poverty reduction (relative, AHC)                       | 250,000    | 162,000      | -88,000    | -35%     |
+| Children in affected households                               | 1,500,000  | 1,780,000    | +280,000   | +19%     |
+| Children benefitting but remaining in poverty (relative, AHC) | 850,000    | 1,320,000    | +470,000   | +55%     |
+
+_Table 1: comparisons between CPAG's UKMOD-based modelling results and
+PolicyEngine's replications_ _for 2023._
+
+## Alternative repeal methods
+
+The Fabian Society's 2021 report _"Going with the grain"_ proposes
+instead repealing the two-child limit for families with a parent who
+meets one of the following conditions:
+
+- Having employment or self-employment income
+
+- Receiving disability benefits
+
+- Having a child aged two or under
+
+We estimate this partial abolition would cost £1.3 billion in 2023,
+saving £500m compared to the full repeal. The Fabian Society modelling,
+based on the IPPR tax-transfer model, projected that the number of
+households hit by the two-child limit would fall by 94% from 790,000 to
+75,000. This projection assumes a full roll-out of the policy as it
+would be in 2035, where the transitional protection rule does not occur.
+
+PolicyEngine estimates in 2023 that 333,000 households are affected by
+the two-child limit (close to [estimates by the House of Commons Library
+in April 2021 that 317,500 families were
+affected](https://commonslibrary.parliament.uk/research-briefings/cbp-9301/)),
+and that this would fall by 302,000 to 31,000 (a reduction of 93%, close
+to the Fabian Society estimate).
+
+## Conclusion
+
+PolicyEngine's modelling largely aligns with the findings from other
+microsimulation models but finds around 30% lower net costs and poverty
+impacts than CPAG.
+
+Funding proposals also feature in the discussion around the two-child
+limit. Tom Clark, fellow at the Joseph Rowntree Foundation, [pointed to
+a 1p increase in the higher rate of income tax as a possible
+option](https://twitter.com/prospect_clark/status/1680875250542575618?s=20).
+Given our higher net cost estimate for the two-child limit repeal, we
+estimate this would leave £470m deficit (distributional impacts shown in
+Figure 2).
+
+![A graph of a graph showing the amount of income decile Description
+automatically generated](/images/posts/repealing_the_two_child_limit/decile_chart.png)
+
+_Figure 2: [the distributional impacts of funding the two-child limit
+repeal with a 1p higher rate
+increase.](https://policyengine.org/uk/policy?focus=policyOutput.decileRelativeImpact&reform=18067&region=uk&timePeriod=2023&baseline=1)_
+
+PolicyEngine's microsimulation modelling is completely open-source. [See
+the full analysis for the repeal on PolicyEngine
+here.](https://policyengine.org/uk/policy?focus=policyOutput.decileRelativeImpact&reform=7226&region=uk&timePeriod=2023&baseline=1)
+If you have feedback or questions on the results in this report, please
+do [get in touch](mailto:hello@policyengine.org).
+
+## Appendix: reconciling administrative and survey data
+
+When first estimating the impacts of this reform, we used our Family Resources Survey-based
+microdata and reached impacts of lower magnitude. However,
+administrative statistics suggested that the true impacts may in fact be
+larger, so we extended our data enhancement process to repair the
+specific biases remaining present in the FRS under the guidance of the
+administrative statistics.
+
+Microsimulation models based on survey microdata often reach estimates
+differing from administrative data, due to sampling or measurement bias
+in the data collection process. PolicyEngine applies a calibration
+process to adjust for this as much as possible by both imputing missing
+income data and reweighting households to repair consistency with
+administrative totals.
+
+To illustrate, take the following example (based on completely
+hypothetical characteristics).
+
+| Household       | Total income | Original weight | Calibrated weight |
+| --------------- | ------------ | --------------- | ----------------- |
+| High-earner     | £80,000      | 2m              | 5m                |
+| Low-earner      | £25,000      | 25m             | 22m               |
+| Original survey | £785bn       | 27m             | 27m               |
+| total           |              |                 |                   |
+| Administrative  | £950bn       | 27m             |                   |
+| total           |              |                 |                   |
+| Calibrated      | £950bn       | 27m             |                   |
+| survey total    |              |                 |                   |
+
+By adjusting survey weights, we can shift the distributions of survey
+variables to reconcile them with more trusted administrative data. This
+requires collecting as many administrative statistics as possible, and
+calibrating as closely as possible to them all, together. In the example
+above, it was possible to match exactly, but this often isn't possible
+in practice because we are operating on a survey, and not a census.
+
+A key learning point in this report was the government's [administrative
+statistics on households affected by the two-child
+limit](https://www.gov.uk/government/statistics/universal-credit-and-child-tax-credit-claimants-statistics-related-to-the-policy-to-provide-support-for-a-maximum-of-2-children-april-2023/universal-credit-and-child-tax-credit-claimants-statistics-related-to-the-policy-to-provide-support-for-a-maximum-of-two-children-april-2023).
+Using benefit administrative databases, the Department for Work and
+Pensions estimated that in April 2023 around 1.5 million children lived
+in households meeting two conditions:
+
+- The household claims Universal Credit or the Child Tax Credit.
+
+- The household does not receive a child element for at least one
+  child.
+
+Our initial model estimate was lower than this (around £1.3m children),
+largely due to biases in the Family Resources Survey not fully countered
+by our data enhancement process (although we included child counts and
+Universal Credit caseloads in our calibration function, we did not
+include this specific intersection).
+
+To correct this underestimate, we recalibrated the survey microdata,
+applying an extra penalty to the algorithm for deviating from:
+
+- The number of households affected by the UC and CTC two-child
+  limits, respectively.
+
+- The number of children living in UC- and CTC-claiming households
+  with 3, 4, and 5-plus children, respectively.
+
+Figure 3 shows how this process operates for the first of these
+parameters. Note that PolicyEngine still carries out this process for
+the other 2,000+ statistics we target, ensuring that the model weights
+do not over-calibrate towards this new set of statistical targets.
+
+![A graph of a graph Description automatically
+generated](/images/posts/repealing_the_two_child_limit/calibration.png)
+
+_Figure 3: An example of PolicyEngine's calibration process, in which
+survey weights are adjusted to shift the FRS projection of UC-claiming,
+child limit-affected households towards the administrative estimate._
+
+PolicyEngine now reproduces this 1.5m statistic. However, what we
+subsequently found suggests that the 1.5m figure is an underestimate of
+the true number of children affected by the reform. Because the
+administrative data does not include households brought into eligibility
+by reforms, it cannot include households who are not currently eligible
+for Universal Credit but would be under an abolition of the two-child
+limit.[^3] These households are represented by points on the diagonal
+slope in Figure 1.
+
+We estimate 200,000 children fall into this category, bringing the
+number of children living in households affected by the two-child limit
+to 1.7 million.
+
+[^1]:
+    Our cost estimate of £1.8bn is 38% higher than the widely-reported
+    £1.3bn [estimate by the Child Poverty Action
+    Group](https://cpag.org.uk/sites/default/files/files/policypost/CPAG_Budget_Submission_March_2023.pdf),
+    drawn from the UKMOD microsimulation model. PolicyEngine [applies
+    machine learning-based algorithms to survey
+    microdata](https://policyengine.org/uk/blog/how-machine-learning-tools-make-policyengine-more-accurate)
+    to counter measurement and sampling bias present in the input data,
+    which could explain some of the disparity.
+
+[^2]:
+    For example, in 2019 only children aged below two could be
+    affected. This age of exemption increases until 2035, at which point
+    no children can be exempt under the transitional protection rule.
+
+[^3]:
+    Administrative data might in practice include a small number of
+    these households, because households whose income fluctuations mean
+    they move along the edge of entitlement might stay on administrative
+    databases.