The Public Policy Preference Calculator (TriplePC): Developing a comprehensive welfare policy microsimulation

In this article, we describe the Public Policy Preference Calculator (TriplePC), a new microsimulation model that seeks to extend the microsimulation art in two ways.

First, as well as modelling the outcomes of a policy in the conventional way, our model uses Conjoint Analysis of public acceptability data to give an indication of the policy’s popularity. This is novel and important. There are measures that might actually be popular, but which policymakers have been unwilling to touch because of uncertainty about their electoral consequences. Perhaps the best example from UK history is the SDP/Liberal Democrat’s ‘Dead Parrot’ merger manifesto of January 1988 (Gourley, 2023; Crewe and King, 1995), which proposed the abolition of Child Benefit and the imposition of a uniform rate of Value Added Tax (VAT) to raise money for an anti-poverty program. Although this had been modelled in detail, fear of the electoral consequences among Members of Parliament meant the manifesto was abandoned within a day. The resulting confusion and indecision arguably caused long-lasting damage to centrist politics in the UK (Crewe and King, 1995). The UK’s zero-rating for food and children’s clothing remains politically untouchable to this day despite the orthodox economic arguments in favour of a uniform rate (Crawford et al., 2010). But would VAT extension really be unpopular, especially if it was part of a package that used the money raised for poverty reduction or other appealing policies? Our approach allows us to address questions like this.

Second, we integrate health outcomes into the model. There is strong evidence that welfare policies can have a substantial effect on population health (Johnson et al., 2022). A stark reminder of the real impact of worsening population health can be seen in the proportion of the UK population with a long-standing illness, disability or impairment which causes substantial difficulty with day-to-day activities. This is estimated to have risen from 19% in 2011/12 to 24% in 2021/22, an increase of 3.9 million people (Department for Work and Pensions, 2023). Indeed, the estimate increased from 14.1 million in 2019/20 to 16.0 million in 2021/22 (Department for Work and Pensions, 2023). Interestingly, the proportion among state pension age adults has remained the same between 2011/12 and 2021/22 at 45%, whereas for working-age adults it has increased from 16% to 23% and for children the figures are 6% to 11%. This suggests that increases in prevalence are not simply the effect of an ageing population (Department for Work and Pensions, 2023).

In that context, it is essential that policymakers invest real thought in realising the Government’s prevention agenda (Department of Health and Social Care, 2018), which was incorporated into the 2019 NHS England Long Term Plan (NHS England, 2019). Forty-three years on from the Black Report which highlighted the role of material circumstances on health inequalities (Working Group on Inequalities in Health, 1980), 13 since the Marmot Review (Marmot et al., 2010) and three since its 10-years-on update (Marmot et al., 2020) which highlighted worsening trends in inequalities, there is good reason to examine and tackle social determinants of health.

Some of the authors of this article (Johnson et al., 2022) have called for trials of cash transfers, in particular Basic Income, as an upstream intervention to mitigate poverty, inequality and insecurity as social determinants of mental and physical ill-health. Systematic reviews of cash transfer schemes that resemble Basic Income, such as Gibson et al. (2020), have indicated positive impacts on mental and physical health, hospital attendance and health related behaviour, such as alcohol and drug use. In contrast, conditional, means- and needs-based welfare systems in high-income countries are associated with below average health outcomes (Shahidi et al., 2019) and increased psychological distress prevalence (Wickham et al., 2020). We have suggested several explanations (Johnson et al., 2022): schemes are ‘insufficient to offset the negative health consequences of severe socioeconomic disadvantage’ (Shahidi et al., 2019); conditionality and assessment inflicts stress (Dwyer et al., 2020) and creates perverse incentives for health-diminishing behaviour (Johnson et al., 2022); and focusing on the poorest fails to mitigate broader determinants that affect society as a whole (Marmot et al., 2010).

There are existing health simulations for the UK. Public Health Scotland’s Informing Interventions to reduce health Inequalities (Triple I) tools focus on comparing the effects and costs of a range of tax-benefit (including a Basic Income) changes as well as non-economic programmes such as a lifestyle weight management service, 20-mile-per-hour speed limits or Alcohol Brief Interventions. Effects and costs modelled are based on premature deaths, years of life lost and hospital stays, with changes population health and inequalities as the key measures. It does not model the economic impacts on household types nor public preferences.

Most other health microsimulations tend to model the economic effects of health, rather than the other way round (Schofield et al., 2017). There are some that model the effects of, potentially economic, interventions on health, particularly with regard to extrapolating from childhood and adolescence, for example, the University of York’s LifeSim (Skarda et al., 2021).

We therefore decided to create microsimulation with a public-facing user interface – the Public Policy Preference Calculator (TriplePC) – that would enable automated assessment of economic and health impacts as well as public preferences between different welfare and tax policies. The TriplePC model project therefore had three strands:

1. Estimating the likely electoral popularity of possible policies

2. Deriving relationships between income and health, in a form suitable for use in a microsimulation

3. The integration of strands 1 and 2 into a microsimulation tax-benefit model

We discuss these in turn.

TriplePC was developed over approximately four months as part of a project on income and health funded by National Institute for Health and Care Research. We view TriplePC as a highly promising prototype rather than the finished article; in this paper we will describe it ‘warts and all’, highlighting areas of particular uncertainty and the lessons learned we hope to apply to future, fuller developed versions.

Conjoint analysis (Hainmueller et al., 2013) is a survey-based technique originally developed as a market research tool, to examine how consumers value characteristics (sweetness, colour, alcohol content, etc.) of goods. Recently, the technique has become popular as a method for discovering the public’s relative valuations of competing economic or social policies (Bremer and Bürgisser, 2023). Research comparing conjoint survey experiments to actual votes has shown that the conjoint results are good predictors of voting outcomes (Bansak et al., 2023).

Our study is described in full in (Nettle et al., 2024), with the data available as (Johnson et al., 2023). The authors of that study recruited 800 UK resident adults through the Prolific online platform. The sample was UK representative for age categories 18-34, 35-49, 50-65 and 65+ (mean age 49.02, s.d. 16.42, higher than in the 2021 England and Wales Census (Office for National Statistics, 2022)), with equal numbers of men and women in each (43% men, 43% women, 14% other/no response) to match the age structure of the UK population. Participants were asked repeatedly to choose their preferred welfare policy from sets of two. Each time, the two policies contained the same input (design) and outcome (health and distributional) attributes but with randomised levels in each (e.g. payment sizes of £63 per child, £145 per adult and £190 per pensioner; poverty decreased by 25% etc.). Each participant completed 15 choice tasks. Each option within each task was defined by 10 attributes. Each attribute had three to nine possible levels. Table 1 shows the full list of 10 attributes with between three and nine levels each. Attributes were chosen on the basis of previous research conducted by the programme team examining aspects of welfare policy of relevance to public opinion as well as critical aspects of policy design (Johnson et al., 2022; Johnson et al., 2023).

Relative to the UK population, the sample contained an over-representation of people who voted for the left to centre-left Labour party at the 2019 general election (44.3% of those in our sample who voted, vs. 32.1% election result); and an under-representation of those who voted for the right to centre-right Conservative party (31.8% versus 43.6% election result). Although we did create survey weights to correct for this, TriplePC currently uses unweighted data.

All options were fully randomly generated from the possible combinations. Instructions in the survey explained that participants might prefer some features in one policy and some in the other, but they needed to consider which policy they preferred overall. The attributes on which the policies varied were explained in greater depth prior to the first choice task, and then described just with brief phrases during the choice tasks themselves. Note that many of these randomly generated pairs are totally implausible. This is integral to how a fully randomized conjoint analysis works: by generating all the possible combinations, things that are actually correlated in real life (e.g. tax rates and poverty rates) are rendered orthogonal to one another. It is this that allows identification of their independent marginal effects.

We simultaneously estimated the average impact of preference or dis-preference for particular feature attribute-value on preference for policies overall using comparable scales through computation of Average Marginal Component Effects (AMCEs) (Hainmueller et al., 2013) from linear probability models. The AMCE for a given level of an attribute can be interpreted as the marginal effect on the probability of choice of the attribute being at that level compared to the reference level, averaging across the possible levels of all other attributes. Through randomization and a high number of pairwise comparisons, this allows us to quantify the causal effect of including specific levels of individual reform elements on the support for the entire reform package, compared with the support for a reform package that contains the baseline level (status quo) of this particular reform element (Nettle et al., 2023; Nettle et al., 2024). For sub-group analysis, we used ANOVA to test formally whether including interactions between the subgroup identity and the attribute improved model fit. We then compared the marginal mean probabilities of choice for each attribute level between subgroups and used z-tests to establish where the significant differences resided. All analyses were carried out using the cregg R package.(Leeper, 2020). Data and code for the underpinning analysis can be found at https://doi.org/10.17605/OSF.IO/HTSQC.

Note that only one option from each group could be chosen. This is a problem for the ‘other funding’ group, where a respondent might prefer a mix of policies, but can only choose one. Note also that tax and benefit levels are grouped as single choices, limiting the customisation of policies possible within the TriplePC.

Further detail on the design and analysis of the conjoint experiment is available in the working paper and subsequent article (Nettle et al., 2023; Nettle et al., 2024). That study found:

• Preference for more generous payments than less generous ones.

• Strong preference for decreases in poverty (compared to the status quo).

• Preferences on tax rates depended on the broad effects of the policy package. Increasing personal income tax rates were popular if the package they were part of also decreased poverty, and unpopular otherwise.

• Preference for a wealth tax, carbon taxes, and increased corporation tax, as opposed to increased government borrowing.

• Significant positive effect of a large reduction in inequality, and a significant negative effect of a large increase in inequality. However, the effects for inequality were weaker than for poverty.

• Other health and wellbeing consequences also had some significant marginal effects above and beyond those of poverty and inequality. An increase in life expectancy of five years was significantly preferred to the status quo, and a decrease in life expectancy of five years significantly dis-preferred.

• Dis-preference for increased rates of anxiety and depression relative to the status quo, and there was a slight preference for policies that decreased them sharply.

• No strong preference for or against means-testing or other restrictions on eligibility.

• Mild differences between left- and right- supporting participants, in the expected directions.

• No significant variation by gender or between rich and poor.

• Older people were significantly less keen on high income tax schemes and, curiously, less concerned with heath consequences.

For the TriplePC, we extract the AMCEs for each component of any welfare policy the user wishes to specify through the interface from the data of Nettle et al. (Nettle et al., 2024). In the conjoint methodology, these are assumed to have an additive effect on popularity. That is, for example, the popularity of a policy combining high income taxes, carbon tax, and a decrease in poverty will be the sum of the negative marginal effect on popularity of the high-income tax, the positive marginal effect on popularity of the carbon tax, the positive marginal effect of the poverty reduction, and so on. This summation is the popularity outcome returned by the model.

The conjoint analysis was conducted ahead of the construction of the microsimulation model. As we discuss below, some of the measures in Table 1 (income tax, payment sizes, poverty and inequality rates) are reasonably straightforward to model (though there are issues around definitions). Others, such as the ‘other funding’ options and the relationship between income and health, are harder.

We intend to model two health measures: mental health and life expectancy, but only the former has so far been implemented. We build a model relating SF-12 scores (Ware, 2002) to income and demographic characteristics. SF-12 is a widely used measure of an individual’s health-related quality of life, with two summary scores: the Physical Component Summary (PCS-12) and the Mental Component Summary (MCS-12). The model is estimated over 12 waves (2009/11-2020/22) of Understanding Society: The UK Household Longitudinal Study (UKHLS) (Institute for Social and Economic Research, 2023) panel data (Reed et al., 2024).¹ Another companion article (Reed et al., 2023) discusses this modelling in detail.

3.4. Responsiveness to Income Changes

These estimates are then integrated directly into the microsimulation model. In Figure 1) below, the SF-12 Mental Health outputs are shown in the bottom left row. As modelled, mental health appears rather unresponsive to changes in income. For example, choosing the most generous benefit increases from the options from the Conjoint Analysis – a package costing £430bn – would reduce the number of adults with critically low SF-12 scores (below 45.6) by about 3%.

Figure 1

Download asset Open asset

As discussed in Section 2, respondents have preferences over inputs (income tax rates, payment sizes, etc.) and outcomes (poverty and inequality levels, numbers of mental health cases, etc.). We use microsimulation to bridge between them

The analysis uses a heavily adapted version of Scotben (Stark, 2023), an open-source microsimulation model of Scotland written in the Julia programming language. Scotben is a conventionally structured static tax-benefit model, in the family of models branching out from the Institute for Fiscal Studies’ TAXBEN (Johnson et al., 1990) of which two of this article’s authors (Reed and Stark) were developers. For this project, we extended the scope of the model to Great Britain³ using a single 2021/22 wave of Family Resources Survey (FRS) data (DWP, 2019). The model covers the whole of the UK personal tax and benefit system, including local taxes, with some minor exceptions such as Scottish Best Starts grants. As a base, we use the tax and benefit system as of Q3 2023, and all incomes, wealth and consumption are also uprated to 2023 Q3. We use the 2021/22 FRS sample weights as-is. The entire model (minus, because of licence restrictions, the main dataset), is available as a GitHub repository (Stark, 2023).

The outcome questions are phrased as changes (e.g. ‘50% fewer cases of anxiety and depression’, ‘Poverty increased by 50%’). A particularly tricky question arising from this is how to establish a baseline for comparison. The conjoint experiment survey had no ‘keep things as they are’ option for the tax and benefit inputs, so there were two options for the TriplePC:

1. Using a tax-benefit system some way from the current one as baseline and assuming that the outcome changes represent changes in poverty, health, etc. from that point, rather than changes from the actual current situation.

2. Using the current system as the baseline, but then the default output will have significant deviations for the outcome variables.

Neither of these choices is good but on balance we decided 1) was the least bad. I It makes the conjoint popularity output much easier to understand: with option 2) we would be applying large changes in poverty and inequality to the base conjoint results which would make those results very unintuitive. The chosen baseline therefore comprised the first (default) option from each set in Table 1 above:

• Benefits: £0p.w per Child, £63p.w. per Adult and £190p.w. Pensioner, with these replacing Child Benefit and the State Pension;

• Taxes: current non-Scottish income tax rates, bands and allowances, applied across the whole UK including Scotland;

• Net cost of the scheme met by additional borrowing.

We discuss how we interpret and model these changes these in more detail in the next section; but the upshot is that our modelling starts some distance from the actual existing system. The relatively large payments to each adult (offset somewhat by the abolition of child benefit) means that the scheme costs a net £108bn p.a.. The adult payments reduce headcount poverty by 3% but child poverty rises by 0.7%, since the Child Benefit abolition more than offsets the gains to parents from the adult payments. There are small decreases (less than 1%) in adults with depression. All the results from the model are relative to this new baseline, we appreciate that this is not ideal and this is certainly a lesson for future work.

Putting all this together, a model run has five main stages:

1. The user selects from the Payment size, Income tax, Other funding, Conditionality, Means testing, Universality options from Table 1.

2. The model then calculates net incomes for each person in the FRS households given these choices.

3. These net incomes are in turn plugged in to the equations discussed in Section 2 to give us estimates of changes to the prevalence of depressive disorders and mortality.

4. The model next calculates gainers and losers, revenues and costs, and changes in poverty and inequality.

5. Finally, the model calculates conjoint public preferences based on 1 to 4 above and displays the results.

The model has a simple single page web interface (publicly available at https://triplepc.northumbria.ac.uk/). Figure 1, below, shows this in action. The user has selected a relatively generous benefit increase (top left panel), partially paid for by income tax increases (top centre). The bottom half of the screen shows the results, relative to the base discussed in section 3 above. Health results are in the bottom row, showing small mental health improvements. The net cost of this scheme is £118bn (right middle panel). Poverty and inequality are both reduced (centre left panel). The Conjoint analysis is in the centre: the scheme is more popular by 4.7 points than the baseline due to the popularity of the poverty and inequality reductions and the benefit increases, though this is partly offset by unpopular tax increases.

One important lesson for similar future work is the need for good coordination between conjoint analysis and microsimulation modelling at the outset of the project. In our case, the conjoint analysis was conducted ahead of microsimulation modelling work. Consequently, microsimulation requirements were largely fixed by the questions in the conjoint survey. This has several consequences:

• The model could present only a very limited set of options for taxes and benefits compared to the model’s underlying capabilities. The survey system used for the survey – Qualtrics – had a hard limit on the number of attributes that could be included, which meant that it was not possible to ask about basic and higher tax rates individually. It might also have increased respondent load to an unacceptable level and therefore reduced the quality of the preference data.

• The meaning of options such as ‘Tax on wealth’ should, where possible, be made clearer in order to provide a clear direction for modelling.

• Co-development of a conjoint survey and microsimulation might have enabled respondents to see accurate consequences of their preferred policies for incomes and health.

• Some of the options in the survey, such as VAT increases, were quite burdensome to model in the time available.

• Careful thought must be given to the definition of the base case the model is comparing against.

There are also interesting questions about how best to present results of a model with such diverse outputs. For instance, since one is a stock and the other a flow, can we count payments by a household from a wealth tax in the same way as payments for income tax? And should we be imputing a monetary value to any health improvements?

Finally, there is the question of uncertainty. Although it is technically straightforward to quantify uncertainty by running the model multiple times and reporting ranked outcomes, we are reluctant to do so because we feel that it gives a false sense of certainty, by implying that the principal uncertainties around microsimulation models can be captured with a few extra loops. Some sources of error can be quantified by bootstrapping, such as sampling errors in the data, the regression error terms and, to some extent, uncertainty around matching. However, the critical sources of error in microsimulation models – specification and implementation errors in programming and inappropriate data – cannot be quantified in this way. And in any event a bootstrapping model would be too slow to be useful interactively, unless run on more processors than it is feasible to fund for this project. The best defences are instead a good test suite, clearly written code and openness and we are happy to have made all data freely available.

We have presented the TriplePC, a new microsimulation model with novel but important features. We have established the importance and practicality of using conjoint data to provide instant analysis of the political implications of welfare packages, but also learned some important lessons on how best to conduct integrated microsimulation and conjoint analysis. We have also estimated new measures of the relationship between income and health and shown how these, too, can be integrated into the model. The TriplePC is available online at https://triplepc.northumbria.ac.uk/ and its source code, linked on the main site, is released under an open-source licence. In describing the process of its development, ‘warts and all’, we hope that others will be able to develop similar approaches and avoid some of the challenges faced.

1. 1
   https://www.ukwealth.tax/

   1. A Advani
   2. E Chamberlain
   3. A Summers

   (2020)
2. 2
   Lectures on Public Economics

   1. A Atkinson
   2. J Stiglitz

   (2015)

   Lectures on Public Economics, 2nd, ed, https://EconPapers.repec.org/RePEc:pup:pbooks:10493.

   • Google Scholar
3. 3
   Using conjoint experiments to analyze election outcomes: the essential role of the average marginal component effect

   1. K Bansak
   2. J Hainmueller
   3. DJ Hopkins
   4. T Yamamoto

   (2023)

   Political Analysis 31:500–518.

   https://doi.org/10.1017/pan.2022.16

   • Google Scholar
4. 4
   Explaining fixed effects: random effects modeling of time-series cross-sectional and panel data

   1. A Bell
   2. K Jones

   (2015)

   Political Science Research and Methods 3:133–153.

   https://doi.org/10.1017/psrm.2014.7

   • Google Scholar
5. 5
   Fixed and random effects models: making an informed choice

   1. A Bell
   2. M Fairbrother
   3. K Jones

   (2019)

   Quality & Quantity 53:1051–1074.

   https://doi.org/10.1007/s11135-018-0802-x

   • Google Scholar
6. 6
   The estimation of a preference-based measure of health from the SF-36

   1. J Brazier
   2. J Roberts
   3. M Deverill

   (2002)

   Journal of Health Economics 21:271–292.

   https://doi.org/10.1016/s0167-6296(01)00130-8

   • PubMed
   • Google Scholar
7. 7
   Do citizens care about government debt? Evidence from survey experiments on budgetary priorities

   1. B Bremer
   2. R Bürgisser

   (2023)

   European Journal of Political Research 62:239–263.

   https://doi.org/10.1111/1475-6765.12505

   • Google Scholar
8. 8
   Using SAS to model the distributional impact of government policies

   1. W Brice

   (2015)
9. 9
   In Handbook of Microsimulation Modelling

   1. B Capéau

   (2014)

   223–273, Consumption and Indirect Tax Models, In Handbook of Microsimulation Modelling, Vol, 293, Emerald Group Publishing Limited, p, 10.1108/S0573-855520140000293007.

   • Google Scholar
10. 10
    Defining the tax base: Design issues

    1. E Chamberlain

    (2020)

    Defining the tax base: Design issues, CAGE.

    • Google Scholar
11. 11
    Quantifying the mental health and economic impacts of prospective Universal Basic Income schemes among young people in the UK: A microsimulation modelling study

    1. T Chen
    2. H Reed
    3. F Parra-Mujica
    4. EA Johnson
    5. M Johnson
    6. M O’Flaherty
    7. B Collins
    8. C Kypridemos

    (2023)

    BMJ Open 13:e075831.

    https://doi.org/10.1136/bmjopen-2023-075831

    • PubMed
    • Google Scholar
12. 12
    Welfare Benefits & Tax Credits Handbook, 2023/24. CPAG

    1. Child Poverty Action Group

    (2022)
13. 13
    Dimensions of Tax Design: Value Added Tax and Excises

    1. I Crawford
    2. M Keen
    3. S Smith

    (2010)

    Institute for Fiscal Studies.

    • Google Scholar
14. 14
    SDP: The Birth, Life and Death of the Social Democratic Party

    1. I Crewe
    2. A King

    (1995)

    Of Merger and a Dead Parrot, SDP: The Birth, Life and Death of the Social Democratic Party, Oxford University Press, 10.1093/oso/9780198280507.003.0021.

    • Google Scholar
15. 15
    Family Resources Survey 2021/22: Disability DataTables.GOV.UK

    1. Department for Work and Pensions

    (2023)
16. 16
    https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/753688/Prevention_is_better_than_cure_5-11.pdf

    1. Department of Health and Social Care

    (2018)

    Prevention is better than cure.
17. 17
    Methods for the Economic Evaluation of Health Care Programmes (Fourth edition)

    1. MF Drummond
    2. MJ Sculpher
    3. K Claxton
    4. GL Stoddart
    5. GW Torrance

    (2015)

    Oxford University Press.

    • Google Scholar
18. 18
    http://research.dwp.gov.uk/asd/frs/

    1. DWP

    (2019)

    The Family Resources Survey.
19. 19
    Work, welfare, and wellbeing: The impacts of welfare conditionality on people with mental health impairments in the UK

    1. P Dwyer
    2. L Scullion
    3. K Jones
    4. J McNeill
    5. ABR Stewart

    (2020)

    Social Policy & Administration 54:311–326.

    https://doi.org/10.1111/spol.12560

    • Google Scholar
20. 20
    The public health effects of interventions similar to basic income: A scoping review

    1. M Gibson
    2. W Hearty
    3. P Craig

    (2020)

    The Lancet. Public Health 5:e165–e176.

    https://doi.org/10.1016/S2468-2667(20)30005-0

    • PubMed
    • Google Scholar
21. 21
    The ‘Dead Parrot’ document

    1. D Gourley

    (2023)

    Journal of Liberal History, https://liberalhistory.org.uk/history/dead-parrot-document/.

    • Google Scholar
22. 22
    Causal Inference in Conjoint Analysis: Understanding Multidimensional Choices via Stated Preference Experiments

    1. J Hainmueller
    2. DJ Hopkins
    3. T Yamamoto

    (2013)

    SSRN.

    • Google Scholar
23. 23
    The Incidence of the Corporation Income Tax

    1. AC Harberger

    (1962)

    Journal of Political Economy 70:215–240.

    https://doi.org/10.1086/258636

    • Google Scholar
24. 24
    VAT rates on different goods and services. GOV.UK

    1. HM Revenue & Customs (HMRC)

    (2022)
25. 25
    https://commonslibrary.parliament.uk/social-policy/welfare-pensions/benefits/constituency-data-universal-credit-roll-out/

    1. House of Commons Library

    (2020)

    Constituency data: Universal Credit rollout.
26. 26
    Survey timeline. UnderstandingSociety

    1. Institute for Social and Economic Research

    (2023)
27. 27
    Taxben2: The New IFS Tax Benefit Model (IFS Working Papers)

    1. P Johnson
    2. G Stark
    3. S Webb

    (1990)

    Institute for Fiscal Studies.

    • Google Scholar
28. 28
    Designing trials of Universal Basic Income for health impact: identifying interdisciplinary questions to address

    1. M. T. Johnson
    2. EA Johnson
    3. D Nettle
    4. KE Pickett

    (2022)

    Journal of Public Health 44:408–416.

    https://doi.org/10.1093/pubmed/fdaa255

    • Google Scholar
29. 29
    SP2.1 Conjoint experiment on welfare scheme preferences and predictive tool [Dataset]. OSF

    1. EA Johnson
    2. MT Johnson
    3. J Chrisp
    4. D Nettle

    (2023)
30. 30
    Are “red wall” constituencies really opposed to progressive policy? Examining the impact of materialist narratives for Universal Basic Income

    1. M Johnson
    2. E Johnson
    3. D Nettle

    (2023)

    British Politics 18:104–127.

    https://doi.org/10.1057/s41293-022-00220-z

    • PubMed
    • Google Scholar
31. 31
    Health-related quality of life measurement in public health

    1. RM Kaplan
    2. RD Hays

    (2022)

    Annual Review of Public Health 43:355–373.

    https://doi.org/10.1146/annurev-publhealth-052120-012811

    • PubMed
    • Google Scholar
32. 32
    Simple Conjoint Tidying, Analysis, and Visualization. Thomasleeper.Com

    1. TJ Leeper

    (2020)
33. 33
    Statistical matching: A model based approach for data integration

    1. A Leulescu
    2. M Agafitei

    (2013)

    European Commission, Eurostat.

    • Google Scholar
34. 34
    Fair society, healthy lives: The Marmot review

    1. M Marmot
    2. J Allen
    3. P Goldblatt
    4. T Boyce
    5. D McNeish
    6. M Grady
    7. I Geddes

    (2010)

    The Marmot Review, https://www.instituteofhealthequity.org/resources-reports/fair-society-healthy-lives-the-marmot-review.

    • Google Scholar
35. 35
    Health equity in England: the Marmot review 10 years on

    1. M Marmot
    2. J Allen
    3. T Boyce
    4. P Goldblatt
    5. J Morrison

    (2020)

    BMJ m693.

    https://doi.org/10.1136/bmj.m693

    • Google Scholar
36. 36
    Quality-Adjusted Life Expectancy Norms for the English Population

    1. S McNamara
    2. PP Schneider
    3. J Love-Koh
    4. T Doran
    5. N Gutacker

    (2023)

    Value in Health 26:163–169.

    https://doi.org/10.1016/j.jval.2022.07.005

    • PubMed
    • Google Scholar
37. 37
    What Do British People Want from a Welfare System? Conjoint Survey Evidence on Generosity, Conditionality, Funding, and Outcomes

    1. D Nettle
    2. J Chrisp
    3. E Johnson
    4. MT Johnson

    (2023)

    SocArXiv.

    • Google Scholar
38. 38
    Short‐term changes in financial situation have immediate mental health consequences: Implications for social policy

    1. D Nettle
    2. C Chevallier
    3. B de Courson
    4. EA Johnson
    5. MT Johnson
    6. KE Pickett

    (2024)

    Social Policy & Administration, 10.1111/spol.13065.

    • Google Scholar
39. 39
    What do British people want from a welfare system? Conjoint survey evidence on generosity, conditionality, funding, and outcomes

    1. D Nettle
    2. J Chrisp
    3. E Johnson
    4. MT Johnson

    (2024)

    Analysis of Social Issues and Policy, 10.31235/osf.io/zfnuh.

    • Google Scholar
40. 40
    The NHS Long Term Plan. NHS England

    1. NHS England

    (2019)
41. 41
    Wealth and Assets Survey

    1. Office for National Statistics

    (2019)

    Wealth and Assets Survey, UK Data Service.

    • Google Scholar
42. 42
    Living Costs and Food Survey, 2017-2018

    1. Office for National Statistics

    (2019)

    Living Costs and Food Survey, 2017-2018, UK Data Service.

    • Google Scholar
43. 43
    Population of England and Wales

    1. Office for National Statistics

    (2022)

    Population of England and Wales, Office for NationalStatistics.

    • Google Scholar
44. 44
    Understanding the relationship between income and mental health among 16- to 24-year-olds: Analysis of 10 waves (2009-2020) of Understanding Society to enable modelling of income interventions

    1. F Parra-Mujica
    2. E Johnson
    3. H Reed
    4. R Cookson
    5. M Johnson

    (2023)

    PLOS ONE 18:e0279845.

    https://doi.org/10.1371/journal.pone.0279845

    • PubMed
    • Google Scholar
45. 45
    SF-6D® HealthUtility. Quality Metric | We Measure Health

    1. QualityMetric

    (2022)
46. 46
    Universal Basic Income is affordable and feasible: evidence from UK economic microsimulation modelling1

    1. Howard Robert Reed
    2. MT Johnson
    3. S Lansley
    4. E Aidan Johnson
    5. G Stark
    6. KE Pickett

    (2023)

    Journal of Poverty and Social Justice 31:146–162.

    https://doi.org/10.1332/175982721X16702368352393

    • Google Scholar
47. 47
    Examining the relationship between income and both mental and physical health among adults in the UK: Analysis of 12 waves (2009-2022) of Understanding Society. OSF.

    1. H. R. Reed
    2. D Nettle
    3. F Parra-Mujica
    4. G Stark
    5. RG Wilkinson
    6. MT Johnson
    7. EA Johnson

    (2024)
48. 48
    Taxing Our Wealth

    1. F Scheuer
    2. J Slemrod

    (2021)

    Journal of Economic Perspectives 35:207–230.

    https://doi.org/10.1257/jep.35.1.207

    • Google Scholar
49. 49
    A Brief, Global History of Microsimulation Models in Health: Past Applications, Lessons Learned and Future Directions.

    1. DJ Schofield
    2. MJB Zeppel
    3. O Tan
    4. S Lymer
    5. MM Cunich
    6. RN Shrestha

    (2017)

    International Journal of Microsimulation 11:97–142.

    https://doi.org/10.34196/ijm.00175

    • Google Scholar
50. 50
    The impact of social assistance programs on population health: a systematic review of research in high-income countries

    1. FV Shahidi
    2. C Ramraj
    3. O Sod-Erdene
    4. V Hildebrand
    5. A Siddiqi

    (2019)

    BMC Public Health 19:2.

    https://doi.org/10.1186/s12889-018-6337-1

    • PubMed
    • Google Scholar
51. 51
    LifeSim: A Lifecourse Dynamic Microsimulation Model of the Millennium Birth Cohort in England

    1. I Skarda
    2. M Asaria
    3. R Cookson

    (2021)

    International Journal of Microsimulation 14:2–42.

    https://doi.org/10.34196/IJM.00228

    • Google Scholar
52. 52
    Conjoint [Dataset], version 1ffd9d5

    1. G Stark

    (2023)

    Github.
53. 53
    How income changes affect mental health for working-age adults: A systematic review and meta-analysis

    1. RM Thomson
    2. E Igelström
    3. AK Purba
    4. M Shimonovich
    5. A Pearce
    6. A Leyland
    7. H Thomson
    8. SV Katikireddi

    (2021)

    The Lancet 398:S4.

    https://doi.org/10.1016/S0140-6736(21)02547-2

    • Google Scholar
54. 54
    The mental component of the short-form 12 health survey (SF-12) as a measure of depressive disorders in the general population: results with three alternative scoring methods

    1. G Vilagut
    2. CG Forero
    3. A Pinto-Meza
    4. JM Haro
    5. R de Graaf
    6. R Bruffaerts
    7. V Kovess
    8. G de Girolamo
    9. H Matschinger
    10. M Ferrer
    11. J Alonso
    12. ESEMeD Investigators

    (2013)

    Value in Health 16:564–573.

    https://doi.org/10.1016/j.jval.2013.01.006

    • PubMed
    • Google Scholar
55. 55
    How to Score Version 2 of the SF-12 Health Survey (with a Supplement Documenting Version 1). QualityMetric Incorporated

    1. JE Ware

    (2002)
56. 56
    Effects on mental health of A UK welfare reform, Universal Credit: A longitudinal controlled study

    1. S Wickham
    2. L Bentley
    3. T Rose
    4. M Whitehead
    5. D Taylor-Robinson
    6. B Barr

    (2020)

    The Lancet. Public Health 5:e157–e164.

    https://doi.org/10.1016/S2468-2667(20)30026-8

    • PubMed
    • Google Scholar
57. 57
    https://www.sochealth.co.uk/national-health-service/public-health-and-wellbeing/poverty-and-inequality/the-black-report-1980/black-report-foreword/

    1. Working Group on Inequalities in Health

    (1980)

    Black Report.

Author details

1. Graham Stark

   Department of Social Work, Education and Community Wellbeing, Northumbria University, Newcastle upon Tyne, United Kingdom

   "This ORCID iD identifies the author of this article:" 0000-0002-4740-8711

2. Elliot A Johnson

   Department of Social Work, Education and Community Wellbeing, Northumbria University, Newcastle upon Tyne, United Kingdom

   "This ORCID iD identifies the author of this article:" 0000-0002-0937-6894

3. Howard Reed

   1. Department of Social Work, Education and Community Wellbeing, Northumbria University, Newcastle upon Tyne, United Kingdom
   2. Landman Economics, Colchester, United Kingdom

   "This ORCID iD identifies the author of this article:" 0000-0003-4577-1178

4. Daniel Nettle

   1. Department of Social Work, Education and Community Wellbeing, Northumbria University, Newcastle upon Tyne, United Kingdom
   2. Institut Jean Nicod, École normale supérieure, Paris, France

   "This ORCID iD identifies the author of this article:" 0000-0001-9089-2599

5. Matthew T Johnson

   Department of Social Work, Education and Community Wellbeing, Northumbria University, Newcastle upon Tyne, United Kingdom

   For correspondence

   matthew7.johnson@northumbria.ac.uk

   "This ORCID iD identifies the author of this article:" 0000-0002-9987-7050