Health, poverty and employment effects of cutting income replacement benefits for the disabled: a difference-in-differences analysis of the 2017 welfare reforms

Abstract

Background

Health-related economic inactivity in England has risen, with more working-age adults claiming disability and incapacity benefits. The 2016 Welfare Reform and Work Act reduced weekly payments for Employment and Support Allowance recipients deemed capable of some work from April 2017, potentially affecting poverty, mental health and employment among people with long-term conditions (LTCs). Evidence on its causal impact is limited.

Data and methods

Using UK Household Longitudinal Survey data (2010–2019), we evaluated the reform’s impact on working-age individuals with LTCs. Triple-difference and difference-in-differences models assessed transitions in employment, poverty risk (income <40% median) and incident poor mental health (defined as SF-12 Mental Component Score ≤45.6 among those without prior poor mental health), controlling for age, gender, children and education.

Results

The 2017 reform was associated with a 5.5 percentage point increase in poverty risk among individuals with LTCs leaving employment (95% CI 1.4% to 9.6%, p=0.008) and an 8.9 percentage point increase in the incidence of poor mental health in this group (95% CI 1.1% to 16.7%, p=0.025). No significant effects on transitions into or out of employment were observed at the 95% confidence level.

Applying these estimates to national employment figures suggests that the reform may have resulted in approximately 37254 additional people entering severe poverty and 60117 additional cases of poor mental health annually.

Conclusions

These findings suggest that reductions in disability-related benefits may exacerbate financial and mental health vulnerability among affected individuals but not affect employment. This highlights the need for policy measures that balance fiscal objectives with social protection.

WHAT IS ALREADY KNOWN ON THIS TOPIC.

WHAT THIS STUDY ADDS

• Using difference-in-differences analysis of UK Household Longitudinal Survey data (2010–2019), this study provides the first causal evaluation of the 2017 ESA payment reduction. The reform was associated with a 5.5 percentage point increase in severe poverty risk (95% CI 1.4% to 9.6%) and an 8.9 percentage point increase in incident poor mental health (95% CI 1.1% to 16.7%) among working-age individuals with long-term conditions (LTCs) leaving employment. Extrapolating these estimates nationally suggests approximately 37254 additional people entered severe poverty and 60117 developed poor mental health annually as a result of the reform. No significant effects on transitions into or out of employment were observed at the 95% confidence level.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY.

• Reducing incapacity benefit generosity is unlikely to achieve stated employment objectives for people with LTCs and may instead increase poverty and mental ill-health, creating additional costs to the National Health Service and social care that offset any fiscal savings. As similar reforms are currently being proposed or implemented in the UK, policymakers should ensure independent health impact monitoring is built into any future welfare reform and consider reorienting policy from benefit restrictions towards coordinated, health-enabling and work-enabling support. Future research should examine longer-term impacts, intensive-margin employment outcomes and the cost-effectiveness of alternative approaches to supporting labour market participation among people with disabilities.

Introduction

The UK welfare system has undergone repeated reforms over recent decades, with important implications for employment, poverty and population health. In particular, rising levels of health-related economic inactivity have increasingly shaped debates about labour market participation and the role of the welfare system in supporting people whose health limits their ability to work. As health-related economic inactivity has remained persistently high, welfare policy has increasingly focused on health-related benefits as a key policy lever for addressing both fiscal pressures and labour market outcomes.

Health-related economic inactivity in England has increased substantially since the COVID-19 pandemic. Around 4 million working-age people—approximately 1 in 10—are now claiming health-related benefits, compared with 2.8 million in 2019 (around 1 in 13).¹ In the UK, health-related benefits comprise two distinct components: incapacity benefits and disability benefits. Incapacity benefits provide income replacement for people whose health conditions limit the work they can do, whereas disability benefits are intended to help cover the additional living costs associated with disability, irrespective of employment status. This study focuses specifically on incapacity benefits, which are directly linked to labour market participation; disability benefits are discussed only where relevant to the wider policy context.

Until its replacement for new claimants by Universal Credit, incapacity benefits in the UK were primarily delivered through Employment and Support Allowance (ESA). ESA was payable to individuals whose health conditions limited their capacity for work, regardless of their employment status, including those who were unemployed, self-employed or employed but unable to work.² Claimants were assessed and allocated to one of two groups: the Support Group, for individuals deemed unable to undertake work-related activity and the Work-Related Activity Group (WRAG), for individuals assessed as having limited capability for work but considered able to undertake work-related activity in the future.³ ESA has different payment levels depending on assessed work capability.

The share of the working-age population receiving incapacity benefits has been increasing since around 2017, following earlier declines since the early 2000s.⁴ In contrast, the number of people claiming disability benefits has risen steadily since 2000.⁴ Both trends accelerated during and after the pandemic.⁵ As a result, spending on health-related benefits has grown rapidly, reaching £48 billion in 2023–24, equivalent to 1.7% of GDP.⁵ These trends have renewed policy concern about the sustainability of health-related benefit expenditure and the effectiveness of existing arrangements in supporting people with health conditions into work.

In response, recent policy proposals have focused on reforms to incapacity benefits, particularly reductions in benefit generosity for new claimants assessed as having limited capability for work.⁶ These proposals closely resemble changes introduced under the 2016 Welfare Reform and Work Act, which reduced weekly ESA payments for new claimants placed in the WRAG by around 28% from April 2017, aligning benefit rates with Jobseeker’s Allowance—the main unemployment benefit. Existing claimants were protected, and the reduction applied only to new claimants. Similar mechanisms underpin current proposals affecting the Universal Credit Health Element, including reduced benefit generosity for new claimants and restrictions on eligibility through links to disability benefit receipt.⁷

A central question underpinning these reforms is whether reducing the generosity of incapacity benefits is an effective means of increasing employment and reducing economic inactivity, or whether such reductions instead exacerbate poverty and worsen health outcomes without delivering meaningful labour market gains. Robust empirical evidence on the causal impacts of past reforms is therefore essential to inform current policy debates.

While there is evidence that reducing the duration and generosity of unemployment benefits increases entrances into employment,⁸ previous evidence on incapacity benefit reforms suggests that employment effects are often modest or inconsistent.^8–11 Adverse impacts on mental health have been more consistently observed.^{12 13} Previous UK reforms to eligibility assessments and conditionality have been linked to increases in mental health problems and suicides,^{14 15} with little evidence of sustained improvements in employment outcomes among people with long-term conditions (LTC).^{16 17} However, there has been no causal evaluation of the 2017 ESA payment reduction, and limited evidence examining its combined effects on poverty, mental health and employment transitions.

This study addresses this gap by using longitudinal data from the UK Household Longitudinal Survey (UKHLS) to investigate the impact of the 2017 reform on poverty and incident poor mental health among people moving out of employment with an LTC, and the impact of the reforms on transitions into and out of employment. By analysing a past reform that closely resembles current policy proposals, this study provides timely and policy-relevant evidence on the likely consequences of reducing incapacity benefit generosity for people with health-related limitations.

Method

Policy change

We investigated the impact of changes to ESA that were introduced in the 2017 reform. At that time, ESA was the main ‘income replacement’ benefit for people who have a health condition or disability which limits their ability to work. Eligibility and group assignment were determined through the Work Capability Assessment (WCA), under which individuals were either found fit for work and ineligible, placed in the WRAG, or placed in the Support Group.

The standard rate of ESA for people in the ‘WRAG’ before the reform was £102 per week. The reform reduced this to £73 per week, aligning it with the amount paid for the main unemployment benefit—Jobseeker’s Allowance. Existing claimants were not affected, and the rule change only applied to new claimants.¹⁸

Data source and measures

The UKHLS is a nationally representative longitudinal survey of approximately 40 000 households, collected since 2009. The survey asks a range of questions about health, economic circumstances and household composition.¹⁹ We limited the analysis to individuals of working age (18–64 inclusive) between years 2010 and 2019, to avoid the effects of the COVID-19 pandemic.

We investigated the impact of the 2017 reform on four outcomes (1) severe poverty in those with an LTC exiting employment, (2) incident poor mental health among individuals with an LTC exiting employment, (3) transitions out of employment in those with an LTC and (4) transitions into employment in those with an LTC.

Individuals were followed longitudinally across survey waves, with treatment status defined at the person–wave level rather than as a fixed individual characteristic. For analyses of employment exit, poverty and incident poor mental health, inclusion in each analytical sample was conditional on respondents being in paid employment or self-employment in the previous wave, as determined using lagged employment status. Within this sample, respondents who remained employed contributed comparison observations, while respondents who exited employment contributed intervention observations in the wave of exit. Following an employment exit, individuals were not included in subsequent waves unless they again met the inclusion criterion of being employed in the previous wave.

For analyses of transitions into employment, inclusion was instead conditional on respondents being not employed in the previous wave. Respondents who moved from non-employment in the previous wave to paid employment or self-employment in the current wave were classified as employment entries, while those who remained out of employment contributed comparison observations. This design allows individuals to contribute multiple observations over time and to switch between comparison and intervention status depending on their employment transitions, rather than assigning treatment status permanently at the individual level.²⁰

We defined severe poverty as equivalised household income below 40% of the contemporaneous, survey-weighted median household income within the UKHLS, making this a relative measure of poverty. Incident poor mental health was defined as a score of less than or equal to 45.6 on the mental health component score (MCS) of the 12-item Medical Outcomes Study Short Form Health Survey (SF-12)²¹ among respondents who in the previous wave scored >45.6 and had no previous history of depression or anxiety. We used a threshold of 45.6 as this has previously been shown to be the optimal cut-off point to screen for 30-day depressive disorders in Europe.²² Transitions out of employment were defined as respondents who described their current employment situation as not employed among people who had described their employment situation as either self-employed or in paid employment in the previous wave. Transitions into employment were defined as respondents who described their current employment situation as either self-employed or in paid employment among people who had described their employment situation as not employed in the previous wave.

Respondents were defined as having an LTC or disability if they reported a long-standing physical or mental impairment, illness or disability that had troubled them over a period of at least 12 months or that was likely to trouble them over a period of at least 12 months. This produces a single broad binary indicator capturing any LTC or disability, regardless of specific diagnosis or condition type.

Additional control variables included in the analysis were age, gender, number of children in the household and whether the respondent had a degree level education or not.

Statistical analysis

Initially, we plotted the annual trends in each of our outcomes for each of the intervention and control groups as defined below, to explore the change in outcomes before and after the policy.

We used a difference-in-differences (DiD) analysis to investigate the changes in outcomes before and after the 2017 reform, when a 28% reduction in benefit rates was introduced for new claimants of benefits for people who could not work due to health conditions or disability, but who were assessed as being capable of some work in the future. DiD methods compare the average change in an outcome over time in a group affected by a policy to the average change over time for a group or groups not affected by a policy. The difference in these changes can be taken as the effect of the policy if a set of assumptions holds, of which the parallel trends assumption is a key requirement. The parallel trends assumption requires that in the absence of the policy, the difference between the intervention and the control groups would have been constant over time. While we cannot test whether this assumption holds, we can assess whether it is likely to be true by investigating whether the trends in the pre-treatment period were approximately parallel. Although baseline outcomes may differ between the intervention and control groups, the DiD approach accounts for this by focusing on changes over time rather than baseline levels.

We estimated separate models for each of the four outcomes outlined above. To investigate the impact on poverty and mental health problems, we enhanced the simple DiD by applying a triple difference approach.²³ We had four groups—people who stayed in employment, without and with an LTC and people who left employment without and with an LTC. It is only the last of these groups that we would expect to be affected by the policy, people who left employment with an LTC. As the policy only affected new claimants, people with an LTC transitioning out of employment would be the most exposed. We therefore compared the change in outcomes for this intervention group after the 2017 reforms to the change in outcomes in the other three groups. This means we can account for potential unobserved time-varying confounding factors that affect all groups with LTCs equally and those that affect all groups leaving employment equally. We did this by using a logistic regression model including all people who were in employment in the previous wave, with severe poverty or mental health problem in the current wave as the outcome and a three-way interaction between employment exit, LTC and a dummy variable indicating if the survey year was before or after the 2017 reform. For the model with mental health problems as the outcome, we additionally condition respondents who had no mental health problems in the previous wave, so we estimated the effect on new incident cases of mental health problems.

To estimate the effect of the policy on transitions out of and into employment, we compared the probability of respondents with and without an LTC making these transitions before and after the policy, using a logistic regression model, with employment transition as the outcome and an interaction between LTC and time period. In all models, we additionally included controls for age, gender, number of children and education to account for time-varying confounders. Individual fixed effects were not included due to limited within-individual variation in the binary outcomes. We used survey weights to account for non-response and attrition. To avoid potential confounding from the two-child limit policy, we additionally controlled for an interaction between the post-2017 period and having three or more children. To test whether pre-trends were approximately parallel, we fitted the models as above but included only the pre-intervention period and an interaction between year and treatment group. We then tested whether there was a significant difference in pre-trends, at the 95% level between intervention and control groups. All analyses were carried out using R V.4.5.1. All model formulae are given in online supplemental appendix 1.

While DiD methods are often presented using linear models, they can also be implemented within non-linear models, such as logistic regression, when outcomes are binary as they are in this study. In non-linear models, the coefficient on the interaction term does not directly represent the DiD effect on the probability scale. Instead, policy effects are estimated by comparing differences in predicted probabilities across groups and time periods.^{24 25} Accordingly, we fitted logistic regression models including the relevant interaction terms and estimated policy effects as differences in predicted probabilities before and after the reform between intervention and comparison groups. We derived average marginal effects using the marginaleffects package in R,²⁶ following established methodological guidance. Standard errors for all estimated effects were obtained directly from the fitted models using the delta method.

Results

Figure 1 shows the trend in severe poverty in UKHLS respondents before and after the 2017 reform among people with and without an LTC, remaining in employment or exiting employment since the previous survey wave. As expected, those remaining in employment had a considerably lower risk of poverty compared with those who had left employment. Before the 2017 reform, rates were similar between people with and without an LTC, when stratifying by employment transition. This changed, however, after the act where there was a marked increase in risk of poverty among people with an LTC exiting employment.

Figure 1. Trend in the severe poverty rate before and after the 2017 reform among people with and without a long-term condition (LTC), remaining in employment or exiting employment since the previous survey wave. The red vertical line indicates the timing of the 2017 reform.

Figure 2 shows the trend in poor mental health among people with LTC and without an LTC remaining in or exiting employment since the previous survey wave. Levels of mental health problems were higher in people with an LTC than in those without, and in both groups, the trend was increasing prior to the policy change. Following the 2017 reform, poor mental health then increased substantially among individuals with an LTC who exited employment, while not increasing to the same extent in the other groups.

Figure 2. Trend in the incidence of poor mental health before and after the 2017 reform among people with and without a long-term condition (LTC), remaining in employment or exiting employment since the previous survey wave. The red vertical line indicates the timing of the 2017 reform.

Figures 3 and 4 show the trend in transitions into and out of employment for people with an LTC. As expected, people with an LTC were much less likely to enter employment and more likely to exit employment than people without an LTC. However, this trend did not change markedly after the 2017 reform, apart from a temporary uptick in employment exits from people with LTCs in 2017.

Figure 3. Trend in the percent of people with and without a long-term condition (LTC) entering employment since the previous survey wave before and after the 2017 reform.

Figure 4. Trend in the percent of people with and without a long-term condition (LTC) exiting employment since the previous survey wave before and after the 2017 reform.

Table 1 shows the estimated effect of the 2017 reforms on each of our outcomes estimated from the difference in differences’ models. Following the reforms, the risk of severe poverty among people with an LTC leaving employment increased by 5.5 percentage points, relative to other groups not affected by the reforms. The incidence of poor mental health in this group increased by 8.9 percentage points. Both of these increases were significant at the 95% level. There was no significant effect on either transition into or out of employment at the 95% confidence level. Results from models additionally adjusting for the two-child limit were reported in online supplemental appendix 6 and were consistent with the main results.

Table 1. Estimated effect of the 2017 reform from difference-in-differences regressions.

			95% CI
Estimated effect in percentage point increases of the 2017 reform on:	Estimate	P value	Lower	Upper
Poverty risk among people with LTC leaving employment	5.5%	0.008	1.4%	9.6%
Incidence of poor mental health among people with LTC leaving employment	8.9%	0.025	1.1%	16.7%
Transitions into employment among people with LTC	0.2%	0.686	−0.9%	1.4%
Transitions out of employment among people with LTC	1.0%	0.051	−0.1%	2.0%

Note: Sample sizes for each model were as follows: (1) 126 335; (2) 84 456; (3) 50 975; and (4) 126 402. Results of the logistic regression testing pre‑trends for all outcomes are presented in online supplemental appendix 3. Detailed information for the prevalence of long-term conditions in the analytical sample prior to the 2017 reform, as well as the definitions and sample sizes of the intervention and comparison groups, are presented in online supplemental appendix 2 and online supplemental appendix 4, respectively. Full model results are presented in online supplemental appendix 5. Estimates were from difference-in-differences regressions using pooled UK Household Longitudinal Survey panel data. Models adjusted for age, gender, educational attainment, number of children, and survey wave were weighted using longitudinal survey weights. Individual fixed effects were not included due to limited within-individual variation in the binary outcomes.

LTC, long-term conditions.

Applying these estimates to the total number of people in work in the UK, along with estimates from the UKHLS on the proportion of employed people with an LTC and the proportion exiting employment each year, implies that the policy led to around 37254 additional people entering severe poverty and 60117 people developing poor mental health each year following the reform (see online supplemental appendix 7 for details of these calculations).

Discussion

The 2016 Welfare Reform and Work Act, which reduced benefit payments for ESA from April 2017, led to an increase in poverty and mental health problems for people with LTCs leaving employment. We found no evidence that the policy improved the employment chances for people with an LTC or disability.

These findings are consistent with other studies investigating policies that reduce benefit levels or restrict access to social security benefits,^27–31 which have reported similar negative effects on mental health. However, these studies generally did not investigate the health impacts of changes to income replacement benefits for disabled people. One exception is a study of the introduction of stricter assessment processes for these benefits in the UK that similarly found adverse mental health outcomes.¹²

It is perhaps unsurprising that policies that reduce the payments to social security claimants increase risk of poverty. It could be argued, however, that these income effects are compensated by increased employment incentives, reducing poverty through higher employment. In our study, we found no evidence of an increase in employment following the reduction in benefit generosity, while the risk of severe poverty increased. This is consistent with previous evidence suggesting that reductions in benefit generosity for people with disabilities are associated with either small or modest employment effects.^{14 16 17} Taken together, this suggests that any employment responses to such policies are likely to be limited and insufficient to offset the increased risk of poverty among those affected.

Our study has several strengths. We employed DiD analysis accounting for baseline differences between treatment and control groups that could confound our results and time-varying confounders that have similar effects over time across groups defined by LTC and employment transition. Additionally, our use of triple differences controlled for group-specific shocks that affect only one of the two comparison dimensions but not the interaction. We used a large nationally representative sample, providing adequate statistical power. Moreover, our findings are consistent across four separate outcomes.

There are several limitations to consider. We are unable to identify exactly which individuals would potentially claim ESA. This is due to under-reporting of benefit receipt in survey data and lack of information on timing and duration of benefit receipt. It is also possible that the policy change could negatively affect people even if they do not eventually claim ESA. The change in the value of the benefit could adversely affect people who fear they may need to rely on the benefits in the future. We therefore used cohorts—those transitioning out of employment with an LTC—who were most likely to be exposed to the policy change. This approach, however, also means our analysis is missing other groups who may be affected by the policy. For example, people who were out of work for other reasons (eg, unemployed) whose health has deteriorated may then need to claim ESA. The results therefore probably underestimate the overall impact of the policy change.

One potential concern is that the reform may have been accompanied by changes in WCA outcomes, for example, if individuals were increasingly allocated to the Support Group rather than the WRAG, which was directly affected by the policy. If such reclassification occurred, the observed effects might not reflect the intended policy exposure. However, analysis of administrative statistics shows that the proportion of claimants assigned to the WRAG remained relatively stable over the policy period (online supplemental appendix 8 and online supplemental figure A1). While there is a gradual shift from ‘fit for work’ decisions towards the Support Group, this appears to reflect a longer-term trend rather than a change specifically associated with the 2017 reform. This suggests that the policy could plausibly have affected the group examined in our analysis.

Our measure of poor mental health, although well validated, is self-reported. This could introduce bias if there is a change in propensity to report adverse symptoms over time. The SF-12 is not a diagnostic tool and may overestimate mental health symptoms; therefore, our findings should be interpreted as reflecting increased self-reported distress rather than necessarily an increase in clinically diagnosed mental health conditions. Furthermore, problems may arise if unobserved time-varying factors confound the relationship between exposure and outcome. In other words, factors that increased poverty and mental distress among people with an LTC leaving employment after 2017 that didn’t increase these outcomes in other groups (people without an LTC leaving employment or people with an LTC who remained in employment). To our knowledge, however, there were no other policy changes at that time that explained these effects in this specific cohort.

Our evaluation defines exposure dynamically at the person–wave level around employment exits, rather than assigning fixed treatment/control status across the full panel. This aligns the timing of policy exposure—the reform applied to new claimants—with the timing of outcome measurement and is compatible with DiD frameworks that allow treatment to vary over time.²⁰ A trade-off is that this risk set design prioritises short run, within person changes around employment exits; it does not continuously follow individuals through prolonged non-employment spells, so longer run impacts may be conservatively estimated.

This study focuses on extensive-margin employment outcomes, specifically transitions into and out of employment, which are the outcomes most directly targeted by the 2017 reform. We did not examine intensive-margin outcomes such as hours worked, as the policy affects income replacement for individuals out of work rather than incentives for those already employed. Moreover, among people with long-term health conditions, hours worked are strongly shaped by health-related constraints, making intensive-margin effects difficult to interpret. Future research could explore such effects where appropriate.

Our results have important implications for policy. Although there is growing evidence that welfare reforms such as these can have major impacts on public health, they are introduced without any plans to monitor these impacts or modify them to minimise these effects. These adverse health consequences, alongside increased risks of poverty, are devastating for the individuals involved but also create a false economy of cost shifting. The initial welfare savings may also lead to major costs to other public services, notably the National Health Service, social care and local authority support. Our study shows that reducing payment levels for these benefits may have unintended but serious consequences for population mental health and increased poverty. These adverse effects may outweigh any cost-savings from cutting benefits. The stated aim of policymakers for these reforms is to improve the economic participation and employment of people with disabilities. Our evidence indicates that this is not the effect of these changes. Rather, by increasing mental distress and poverty, these reforms may have the opposite effect and reduce the social inclusion of people with disabilities.

At the time of writing, further reductions to these benefits are being proposed or implemented in the UK. The findings of this study suggest that such reforms are unlikely to achieve their stated employment objectives and may instead increase poverty and mental health problems among people with disabilities. To achieve a sustainable and sustained effect, the policy structure needs to shift from benefit restrictions to a model of targeted, health-enabling and work-enabling, comprehensive support. Such an approach in the long run requires coordinated action across multiple sectors addressing both demand and supply side factors affecting the employment of people with disabilities. This includes rapid access to care for the most employment-disabling conditions, integrated employment support, support for adapting work environments and aligning growth strategies to invest in jobs in the places and sectors that are likely to be accessible to those currently out of the labour market due to ill health.

Data availability statement

Data are available in a public, open access repository.