SWITCH: A Tax-Benefit Model for Ireland Linked to Survey and Register Data

Changes to the income tax and social welfare system have the potential to impact upon the incomes of most, if not all of the population. Policy changes are often considered in terms of their effects on a number of “hypothetical families”. For example, in Ireland, analysis of the annual Budget often looks at specific types of people (e.g., a single person on the minimum wage, a lone parent with two children, one and two-earner couples on the average wage etc.). This approach has strong limitations as families within this category differ in terms of income, housing tenure, number of children and other characteristics that affect their tax-benefit position. More fundamentally, analysis of hypothetical families - no matter how well chosen - simply cannot give an overall picture of the impact of a policy change on incomes and work incentives at the population level. The use of tax-benefit microsimulation models, which are based on large-scale, nationally representative samples of households or rich administrative data, can provide a more representative picture of the impact of policy changes.

Microsimulation modelling has been used in the measurement of poverty and inequality for decades. This framework allows the simulation of income distributions for a given population, income structure and tax-benefit system. These income distributions are often also recoverable from the survey or administrative datasets underlying microsimulation models. However, linking these data to a microsimulation model allows the computation of counterfactual or “what-if” scenarios which is not possible using the underlying data alone. The ability to create counterfactual income distributions is critical to policy-makers and researchers who wish to estimate, for example, how the distribution of income will change due to population ageing (Dolls et al., 2019), if unemployment increases (Dolls et al., 2012) or if the parameters of the tax-benefit system change (Paulus and Tasseva, 2017). Well known microsimulation models include EUROMOD (Sutherland and Figari, 2013) for Europe; TRIM (Giannarelli et al., 2007) for North America and TaxBEN (Immervoll and Barber, 2006) for OECD countries.

The European Commission develops and maintains the harmonised European microsimulation model, EUROMOD. The first Irish policy system to be included in the EUROMOD model dated from 1998. Since then, the Irish component of the EUROMOD model has been used mainly for academic work in single country studies (Bargain and Doorley, 2011; McQuade et al., 2017) or comparative European work (Paulus et al., 2020; Sologon et al., 2021; Doorley et al., 2021). While an invaluable tool, there are some drawbacks to the EUROMOD model from a policy-maker perspective and these mainly stem from the limitations of the underlying data.

Broadly speaking, three types of data source can be linked to a microsimulation model. Survey data, such as the EU Statistics on Income and Living Conditions (EU-SILC), tend to be most straightforward to access and are widely used. Alternatively, administrative or register data can be linked to either EUROMOD or to a national model but these tend to be subject to strict access rules and do not always contain rich demographic information. A third option combines survey data with some administrative data to allow a more accurate simulation of income, benefit eligibility and tax liabilities.

The data underlying the Irish component of the EUROMOD model is EU-SILC in which the data reference period is the year prior to the data collection period. This results in a mismatch between the income reference period and policy rules/employment information. This coupled with the fact that the data underlying the EUROMOD model is not reweighted/nowcasted to match detailed employment or income control totals for the year of collection or, indeed, the current year, means that the simulated distributions of income, benefit receipt and tax revenue can be difficult to validate against external benchmarks. Additionally, EU-SILC contains aggregated rather than detailed information on benefit receipt. This makes the modelling of contributory and non-contributory benefits difficult as, without information on which particular benefit an individual is in receipt of, the researcher cannot determine this independently as social security contribution history is not available in EU-SILC. It is for these reasons that the Economic and Social Research Institute (ESRI) develops and maintains a separate microsimulation model of the Irish tax and benefit system. The model is used internally at the ESRI and is also provided to civil servants and the staff of the Parliamentary Budget Office (PBO).¹

The ESRI’s microsimulation model, SWITCH (Simulating Welfare, Income Tax Childcare and Health policies) is based on data drawn from the Survey on Income and Living Conditions (SILC) which is carried out by the Irish national statistical agency – the Central Statistics Office (CSO). This dataset falls into the third category described above – it is primarily a survey dataset but is linked to administrative information on current income and welfare. Using this dataset, rather than EU-SILC allows us to use a “current” concept for income and to model contributory and non-contributory benefits more precisely. Due to the more extensive set of variables provided in the national version of SILC, we can also model non-cash benefits, which is not possible with EUROMOD. This allows us to create a model that is more precise and easily validated against external benchmarks. Further advantages of the model include nowcasting of the underlying data and detailed automated results tailored to policy-makers.

The SWITCH model has been in operation for three decades and both policy makers and researchers have published extensively using the model (see, for example, Doorley et al., 2021; Boyle, 2018; Department of Social Protection, 2021, 2022; Bargain et al., 2017; Keane, 2015; Callan et al., 2006; 2014a; Callan et al., 2014b; Baicker et al., 2013; Callan and Sutherland, 1997; Bargain and Callan, 2010). A recent European Commission report indicates that policy makers in Ireland perform systematically more Distributional Impact Analysis than most other member states around Budget time and the SWITCH model underpins much of this analysis (Fiorio, 2021).

In 2020, the ESRI undertook a project to redesign the model and the underlying software to keep pace with international developments and provide more model flexibility. The updated SWITCH model, while retaining its advantages over the Irish component of EUROMOD, is now based on the EUROMOD platform, facilitating cross-country comparison, and uses the most recent pre-pandemic data available, SILC 2019. This paper describes the development and operation of the SWITCH model. The EUROMOD platform and the process of creating an input database for the model is discussed. Ensuring the model runs on a representative sample of the Irish population is essential to model accuracy and we discuss adjustments made to the underlying data to improve representativeness, such as reweighting and uprating. The simulation process is discussed along with the capabilities of the model in terms of results produced. We also present a model validation and show how the reweighting process helps improve the accuracy of the model.

EUROMOD is the tax-benefit microsimulation model for the European Union. It is maintained by the Joint Research Centre (JRC) of the European Commission, in collaboration with Eurostat and national teams from the EU countries. The Taxation, Welfare and Pensions team at the ESRI operate as the national team for the Irish component of EUROMOD. The model is open-source and consists of three main components – the software, the coded policy rules and the input data. The model consists of a set of xml files which contain the coded policy rules. The input data of the model is mainly based on the EU-SILC User Database (UDB).^2, The EU SILC survey is coordinated by EUROSTAT with CSO Ireland carrying out the survey in Ireland and supplying the data to EUROSTAT. The survey focusses on the income and living conditions of households in Ireland with a focus on poverty, deprivation and social exclusion. While the model itself is open-source, access to the input data (based on the EU SILC UDB) must first be authorised by EUROSTAT. The model has a flexible user interface that allows users to adjust tax-benefit parameters.

The model is also flexible in terms of input data – the data simply needs to be created in a harmonised manner to run with the EUROMOD platform. In some countries researchers have used alternative datasets to create an alternative input dataset for use with the EUROMOD model – for example, the use of administrative data in Greece.³

Simplified online tax-benefit calculators which use the EUROMOD platform exist for some countries such as the SORESI model for Austria or the EUROMOD-JRC Interface which allows users to simulate non-complex tax and benefit reforms without prior knowledge of the EUROMOD model and without the need to apply for data access.^4,

While the EUROMOD model covers Ireland, the SWITCH model, which is the focus of this paper, uses the EUROMOD platform along with a more detailed dataset to analyse tax-benefit changes. The key differences between the EUROMOD and SWITCH model are discussed throughout the paper and briefly summarised in Box 1 of the Appendix.

Ideally tax-benefit models would be based on administrative data capturing the entire population but this is not always possible. Administrative data, which is often subject to more restrictive access rules, may provide more accurate and timely information on income. Some drawbacks to administrative data include the limited demographic information they contain and the fact that observations are sometimes limited to taxpayers. This sometimes means an under-representation of the lower end of the income distribution (who may not pay tax) and difficulty in simulating the effect of policies on non-taxpayers. The use of administrative data may be further complicated by national statistical agency’s rules on data usage and reporting. Sometimes, the data must be used on the Virtual Desktop Interface of a statistical agency with a statistician charged with checking that statistical disclosure rules are not violated before results are transferred to the user’s own PC. The TAXIPP model used at the Institut des Politiques Publiques in France is an example of a national microsimulation model based on administrative data (Ben Jelloul et al., 2017).⁵

In Ireland, while administrative data on market incomes and social welfare income is available from the Revenue Commissioners and the Department of Social Protection (DSP),⁶ respectively, two main issues prevent the usage of such administrative data for microsimulation modelling in Ireland. Simulating tax and welfare entitlements depends on factors such as the number and ages of all household members. Those not in receipt of income (such as children, inactive adults) do not appear in these administrative records and Ireland does not have a population register as exists in many other countries. Benefit entitlements often also depend on other factors. For example, couples with children whose incomes are below the relevant income limit for the in-work income support, the Working Family Payment, must work at least 38 hours per fortnight combined and hours of work information is not available in income tax data. Simulating entitlements to additional benefits such as Medical Cards (the Irish public health insurance scheme) and to subsidies under the National Childcare Scheme is also only possible based on information not recorded in administrative data, such as hours of childcare used, housing costs paid etc.

For these reasons, the SWITCH model uses data from the CSO Ireland’s Survey of Income and Living Conditions (SILC). SILC is part of an EU-wide programme and surveys a sample of Irish households. The survey covers a wide range of issues, with a focus on income and living conditions, and is the official source of poverty indicators. The survey is carried out annually (since 2004) and normally covers around 4,000 private households/10,000 individuals each year. The most recent pre-COVID19 data year available, 2019, is the current dataset used in SWITCH. The survey captures information on all household members, and their relationships along with their types and levels of incomes, labour force status and work hours. This information is essential to accurately model income tax liabilities and entitlements to benefits. The majority of income information is captured from administrative sources such as Revenue and the Department of Social Protection which improves the accuracy of the data. For more information on the SILC data, see CSO (2020).

While the EU-SILC UDB data (provided to EUROSTAT by the CSO Ireland) is based on the same survey, the CSO Ireland provide a more detailed version of the SILC data that is only available under strict access conditions – the Researcher Microdata File (RMF). This contains more detailed information than the UDB version – for example in the UDB some benefits are aggregated together such as contributory and non-contributory pensions. This is an important difference between SWITCH and EUROMOD as information on the type of benefit received in reality is sometimes used in identifying eligibility to a particular payment (see Section 4.1 for more detail on this). Additional variables are also provided in the RMF compared to the UDB that improve the accuracy of the SWITCH model compared to EUROMOD – for example month of birth is provided in the RMF but not the UDB – this information can be used to identify a child’s entitlement to publicly funded pre-school, the Early Childhood Care and Education (ECCE) scheme, eligibility for which depends on the precise age of the child in months.

Another important distinction between SWITCH and EUROMOD is the income reference period. The input data prepared for SWITCH focusses on current income, most of which is drawn from administrative sources.⁷ The focus of EUROMOD and, more generally, of SILC and EU-SILC data is annual income, where the income reference period is the previous year. Poverty rates calculated by CSO Ireland and Eurostat are based on annual employment and benefit income. While this makes sense in examining changes in income levels and calculating poverty rates, tax liabilities and welfare entitlements depend on current income and labour market status. For example, an individual who was unemployed at some point during the previous year and received an unemployment benefit at that time might be employed at the date of interview, with no current entitlement to unemployment benefits. Likewise an unemployed individual eligible for Jobseeker’s Allowance, the means tested unemployment benefit, will be means tested based on their current income as opposed to their annual income.

The SILC questionnaire asks about receipt of social welfare payments over the course of the previous year. However, the CSO Ireland also links the survey data to administrative information from the DSP which indicates what (if any) benefits are currently received by the respondent. The last (usually weekly) amount is also provided. This allows us to identify a person’s current benefit entitlement which may be used in identifying an individual’s eligibility for a particular benefit scheme in the model. For example, knowing if an individual is currently unemployed does not provide sufficient information to determine the type of unemployment benefit they may be eligible for, as the SILC data does not contain social security (PRSI) contribution histories which determine which kind of Jobseeker’s support is granted.⁸

2.3. COVID employment shock

While incomes in the SWITCH model are uprated to current levels, the SWITCH model usually assumes a ‘constant’ population i.e., the demographic and economic characteristics of the population (such as the age and employment profile) are assumed to remain constant at the year the SILC data was collected. The model does, however, allow for the implementation of changed economic conditions. Given the substantial impact of the COVID 19 pandemic on employment rates, the 2019 data must be adjusted in order to be representative of the 2021 population in terms of unemployment rates. O’Donoghue et al. (2020) does this via a sophisticated nowcasting approach using the NUI Galway microsimulation model. Almeida et al. (2021) employ a reweighting method linked to macroeconomic scenarios in the EUROMOD model. In the SWITCH model, we opt for an ‘unemployment shock’ whereby a certain proportion of workers in an industry are assumed to have either lost their job or to have been placed on the wage subsidy scheme put in place to retain jobs – the Temporary COVID-19 Wage Subsidy Scheme (TWSS). The number of individuals to either lose their job or receive the TWSS (or its successor, the EWSS) is calibrated from publicly available CSO Ireland data on the number of people in receipt of the Pandemic Unemployment Payment (PUP),¹² the additional unemployment benefit introduced in response to the pandemic, and from Revenue data on the number of recipients of the EWSS.¹³ Both account for the industry breakdown of recipients of either scheme. Extra weight is given to workers under 25 in the employment shock simulations as administrative data shows that this group was hardest hit by pandemic related unemployment. The proportion of workers simulated in each age and gender category to be receiving the PUP and TWSS is compared to CSO Ireland statistics in August 2020 in Figure 1. Overall, we can see that the simulation of those in receipt of the PUP and wage subsidies compares well with official estimates along gender and age grounds.

Figure 1

Download asset Open asset

These calibrations can be changed by users of the model to reflect PUP and TWSS numbers at the time of analysis or estimated future numbers. Likewise, the unemployment shock can be omitted so that unemployment rates remain at 2019 levels if users wish to simulate the forecast income distribution when unemployment is back to pre-pandemic levels.

The SWITCH model, along with the reweighting process described in Section 3.4, is compared to external benchmarks where possible. Doing so allows us to examine in detail how well the model represents the Irish population – a particular focus is on ensuring income levels and benefit recipient numbers compare well to external estimates as these characteristics will be key in determining tax and benefit liabilities and entitlements. This allows users of the model to produce credible results. Where possible, the external benchmarks used are taken from administrative data to ensure that the model results compare favourably to statistics representing the entire (non-survey based) population.

Comparing the total number of tax units in the SILC data when applying the original SILC weights to administrative data from Revenue we see that the total number of taxpayer units is considerably lower in SILC (see Table 4) – Revenue Commissioner administrative information shows 2.2 million taxpayer units in 2019 for annual incomes above €10,000 while SILC only shows 1.6 million.²⁶ This is likely to be due to the fact that, in addition to tax-paying benefit recipients, administrative statistics count all tax-units that worked over the calendar year, even for just a day. SILC, on the other hand, is in effect a snapshot of income and employment status in a given month (the interview month) and thus seasonal or temporal employment is not captured.

The income distribution of tax units also differs substantially. If left unchanged, these factors would have serious implications for the accuracy of the modelling – for example, in estimating the exchequer impact of income tax liabilities. The reweighting carried out addresses this issue by ensuring that the total number of taxpayer units increases substantially (hitting the target of 2.2 million on incomes above €10,000) and that the income distribution of these taxpayer units matches those of Revenue. We do not control for numbers in the lowest income band (€0 – €10,000). Doing so would present difficulties as the weighting process is arrived at based on employment and self-employment income. Those with zero employment/self-employment income are unlikely to show up in Revenue numbers (as they have no income, therefore no tax liability). This lowest income band may also contain people receiving taxable benefits (i.e., non-employment income) income. In theory most benefits are taxable, but it is unlikely they actually incur a tax liability as an individual’s tax credits will usually cover any liability. Those for whom benefits are their sole income source may therefore also not show up in Revenue numbers. It is usually only those individuals who have an alternative income source (for example rental income or occupational pensions) that will incur a tax liability on benefits received. These issues mean that the numbers in the low-income band as recorded by Revenue will differ from the actual number of low-income individuals in the population, therefore an appropriate, comparable control total does not exist for this income group. This has implications for the simulation of some means-tested benefits, as discussed later in this section.

We also compare how the original SILC data and SWITCH model compares to administrative data on benefit receipt, all for the year 2019. In the reweighting procedure we control for benefit recipients in the input data (i.e., based on the benefit receipt data provided by the DSP in the SILC data) so we provide a comparison of both the input data and the simulated output data to administrative information on benefit receipt.²⁷ This is an additional validation of the policy simulation in the model as, if the benefit recipient numbers in the output data were radically different to the input data, it might suggest an error in the simulation of benefit entitlements in SWITCH. There can be valid reasons, however, why an individual is simulated as receiving a benefit or entitlement but does not report actually receiving it – for example they may be eligible but not take up their entitlement.

We find that, in most schemes, the existing CSO Ireland weight already performs rather well in terms of benefit recipient numbers (Table 5, column (2)). Excluding Maternity Benefit, the original SILC weight captures 90% - 122% of the administrative numbers reported on each scheme (column (1)).²⁸ Regarding Maternity Benefit, some recipients do not receive the payment directly but opt to have it paid directly to their employer, who may top them up to their full pay. These individuals will not be captured in the DSP data linked to SILC. The numbers receiving Jobseeker’s Allowance and Jobseeker’s Benefit appear high in the SILC data using the original 2019 weight when compared to administrative numbers, reflecting the higher number of unemployed people in the SILC data when using the original 2019 weight. Recipient numbers simulated by SWITCH using the original 2019 weight (column (3)) are close to the numbers shown in the SILC (input) data which reassures us that the simulation processes in the model are accurately modelling benefit entitlements.

To examine the impact of the reweighting process, we compare the benefit recipient numbers simulated by SWITCH with the original and the revised weights to the official statistics. By design, as the reweighting method ensures that the administrative numbers used as targets/control totals are met exactly, our reweighting process results in the weighted number of recipients in the input data (column (4)) matching the administrative numbers. Examining the ratios to administrative statistics using the SWITCH output data (column (5)) again provides strong reassurance that the model is simulating benefit entitlements correctly as the ratio of recipients in the output data using the revised weight remain close to 100% for most schemes. Simulated Maternity Benefit numbers are higher than those actually reported in the input data and may reflect the fact that some people opt to have the benefit paid directly to their employer, therefore SWITCH simulates higher numbers of eligible individuals. Simulated Working Family Payment (WFP) recipient numbers are 15% higher than administrative estimates. SWITCH incorporates a take-up adjustment for WFP based on an estimated take-up rate of 33% (see Callan and Keane, 2008). The gap suggests that take-up may have changed since these estimates. Simulated numbers on Supplementary Welfare Allowance (SWA) are too low to report but this reflects the fact that some SWA recipients are individuals with little or no income who are awaiting the processing of a claim for a different social welfare benefit and may receive SWA as an emergency measure in the meantime. The SWITCH model assumes people receive benefits as soon as they are entitled to them, therefore processing delays are likely to explain this gap.

Remaining gaps tend to be small and more frequent in means tested (non-contributory) benefits such as the State Pension (non-contributory), Jobseeker’s Allowance and Disability Allowance. This may reflect improvements in some recipient’s financial situation as the SWITCH model assesses the person based on their current income/means – some recipients may have qualified for one of these payments at the point of application but may have experienced an improvement in their financial circumstances so that they are no longer eligible for the payment based on their current income levels. Alternatively, in the case of Jobseekers’ Allowance, for example, SWITCH may simulate the person in receipt as eligible for a different payment, such as the One-Parent Family payment (if they appear from the data to be a lone parent) or the Working Families payment (if they are only part-time unemployed).²⁹ Lastly, as the administrative figures we use in the reweighting process do not contain those who pay no tax, we do not reweight the lowest band of the income distribution (€0-10,000). This is likely to result in an underestimation of the number of tax-units in this income band as SILC estimates a lower population of tax units in general. This has repercussions for the simulation of means-tested benefits in general, with estimates of such benefits likely to come in below administrative figures for receipt and amount.

Finally, we also compare benefit recipient numbers as simulated by the Irish component of EUROMOD to administrative statistics. Much larger discrepancies are found here – in fact only two schemes fall within 10% of administrative numbers. While the reweighting process used in SWITCH explains part of these discrepancies, the differences are much larger than when using SWITCH estimates and the original CSO Ireland weight. Therefore, these differences are also driven by input data and modelling issues. The use of the more detailed RMF data in the SWITCH model, as opposed to the UDB data in EUROMOD, results in significantly more precise eligibility simulation in SWITCH compared to EUROMOD. For example, the aggregation of benefits in the UDB makes it more difficult to identify precisely eligibility for benefits that use receipt in the input data to identify eligibility (i.e. part-simulated benefits). In addition, the UDB records the total amount received in benefits in the last year but not whether the benefit is currently being received. The DSP register data indicating current receipt of benefits (at a dis-aggregated level) present in the RMF thus allows SWITCH to identify exactly what benefit a person is currently receiving. The resulting part-simulated benefit recipient numbers are significantly more precise. The focus on current incomes in SWITCH, as opposed to annual incomes in EUROMOD, will also play a role in better identifying current entitlements to means-tested benefits.

In addition to examining benefit recipient numbers, we also compare how the SWITCH model performs in terms of annual benefits’ expenditure using the original and revised weight (Table 6). Doing so helps to highlight potential modelling errors, for example incorrect benefit payment rates in the model. It also allows us to examine if the model is representative not just of the total number of recipients of each benefit but also the distribution of people within benefits – for example if the model simulates a similar number of actual benefit recipient numbers but vastly over or underestimates the expenditure on a scheme it may suggest that the model simulates too many higher/lower rate recipients of, say, a means tested benefit. It is worth bearing in mind, however, that administrative statistics on the expenditure per scheme and SWITCH estimates may not be fully comparable due to a ‘stock’ versus ‘flow’ issue – the administrative recipient numbers show the ‘stock’ of recipients at a point in time (as at 31^st of December 2019) rather than the total number of people (or ‘flow’) who received the benefit during the year. SWITCH numbers also show the ‘stock’ of people eligible for a benefit at the point of interview.³⁰ While this may not match exactly the people eligible as at December 31st it is significantly more comparable than comparing to the flow of all recipients throughout the year, particularly for benefits that have shorter term receipt, such as Jobseeker’s or Maternity Benefit. Expenditure numbers, however, capture all expenditure on a benefit over the entire year while the SWITCH expenditure estimate is generally calculated by annualising (i.e. multiplying by 52) the weekly amount received by those unemployed and in receipt of one of these payments at the point of interview.³¹ Therefore, we would expect more divergence between SWITCH and administrative values for benefits that are received for a shorter period of time or that experience large month to month changes in recipient numbers.

Examining the estimated expenditure of the benefits as simulated by SWITCH when applying the revised weights (column (3)), we see that the total expenditure estimated by SWITCH using the revised weight compares favourably to administrative statistics (column (1)) for the most part - capturing between 90% to -106% of the administrative numbers on all but four schemes. The costs for Jobseeker’s Allowance, Jobseeker’s Benefit, Maternity Benefit and Working Family Payment are underestimated (capturing 57% to 80% of their administrative cost) – in fact the expenditure numbers using the original weight (column (2)) for these four schemes are actually closer to the administrative total. Despite this, it is still more likely the case that the SWITCH estimates using the revised weight are more accurate in capturing total expenditure at a point in time, given the fact that simulated recipient numbers on these four schemes compare well to administrative statistics, as shown in Table 5. Instead, seasonal variations in unemployment and the stock/flow issue described above help explain the discrepancies between SWITCH and administrative expenditure numbers on Jobseekers Allowance and Benefit as receipt of these benefits tends to be short-term so that considerable turnover on these schemes will occur. Maternity Benefit is only payable for 26 weeks; therefore it is not surprising that the administrative figure (capturing total expenditure on the scheme for all recipients during the year) is double that of SWITCH (which is calculated by multiplying the weekly simulated entitlement of those eligible for Maternity Benefit at the point of interview by the 26 week entitlement).

Finally, we look at how non-cash benefit recipients (Medical Card and GP-Visit cardholders as well as beneficiaries of ECCE and NCS) are captured in SWITCH with the original and the revised weights compared to administrative data. These schemes do not compare as well to administrative numbers as cash benefits, using either the original CSO Ireland weights or the revised weight as shown in Table 7 – in fact the comparison to administrative numbers worsens once the revised weights are used. Possible reasons for these discrepancies are discussed below.

Regarding Medical and GP-Visit Cards there is evidence of substantial non-take-up of these cards, particularly for the latter, with Keane et al. (2021) estimating a non-take-up rate of Medical Cards of 31% and 90% for means-tested GP-Visit cards (Keane, 2018). Take-up of public healthcare entitlements amongst the eligible population is seldom universal in countries without universal healthcare coverage such as Ireland and the U.S. (Baicker et al., 2013). This issue can help explain why SWITCH estimates may differ substantially from administrative figures as the numbers shown are based on the numbers eligible (therefore assumed to be availing of the entitlement). The SWITCH model does have the option to specify a take-up rate for these cards and therefore model random non-take-up. Non-take-up helps explain why SWITCH overestimates entitlement to GP-Visit Cards.

The SWITCH estimate for Medical Cards is, however, lower than administrative figures. A number of reasons can help explain this. Some individuals hold a card who are not currently eligible based on their means. These can be categorised into three distinct groups. First, discretionary cards (i.e. cards awarded to those above the income limit) account for around 10% of administrative numbers (see Prior et al. (2019)) and these cannot be modelled by SWITCH as there is no identifiable criteria for a card on this basis.³² Second, card holders can retain their card for three years after returning to employment after a spell of long-term unemployment. SWITCH cannot model retained cards as longitudinal information on benefit receipt is not available in the SILC data. There is no administrative information relating to how many retained cards there are in the total figure. Third, cards are issued on a three-year basis – the obligation is on cardholders to report if their means have increased since the initial granting of the card but some may not do so and still hold the card despite an increase in income. Simulation of means for the purposes of the cards may also not be exact in the SWITCH model – while the underlying data in the model contains the vast majority of information needed to calculate an individual’s means for Medical and GP-Visit card purposes it does not contain information on travel to work costs which are allowed against income in the means calculation. Therefore, the model may overestimate means, and therefore underestimate the numbers entitled to the card. Robustness checks carried out which included in the model average travel to work costs for individuals did not, however, significantly impact the numbers modelled as eligible for the cards (see Keane et al., 2021).

ECCE beneficiaries are underrepresented with both weights, accounting for 60% and 64% of the administrative data when the original CSO Ireland and the revised weights are respectively applied whereas the beneficiaries of NCS are overrepresented - accounting for 247% and 264% of the administrative data beneficiaries respectively. ECCE eligibility is based on age (children must be between 2 years 8 months and 5 years, 6 months old and not in primary school) and the input data (applying the revised weight) compares favourably to administrative population statistics overall matching nearly exactly the total number of 2-5 year olds in the population. However, simulation of ECCE entitlement is also linked to reported childcare use with the subsidy only awarded to those who signal that their child in in a childcare setting for at least 15 hours a week (the pre-school entitlement). This reporting of childcare usage may be somewhat inaccurate as it is framed as “usual hours” per week of childcare. The ECCE scheme is only provided for 38 weeks of the year with a break over the summer months. Parents interviewed during the summer months may, therefore, report zero hours (and therefore the model will not simulate them as availing of the ECCE scheme) even if their child attends pre-school during the academic year.

Regarding the NCS scheme, a number of reasons may account for the difference in the number of beneficiaries. Firstly, the NCS was only introduced in 2019, with a number of legacy schemes operating in parallel for the first year. Entitlement to these legacy schemes is complicated and not possible to simulate as the SILC data does not contain all the eligibility information needed, therefore it is not possible to identify how many people may be eligible for the NCS but remain under the old legacy scheme. The available administrative data for the first year of the scheme rollout may, therefore, not be capturing fully the number of NCS beneficiaries as people take time to transition to the new scheme. Possible take-up issues might also arise from the fact that this was the first year that the scheme was operational and thus people eligible may not have been aware of its existence or that they were eligible for a subsidy. Another possible reason could be the way that participation in the scheme is captured in SWITCH. Administrative data captures the average number of beneficiaries throughout the year, accounting for potential lower participation in the scheme during summer. In SWITCH, we assume that the weekly hours of participation (based on the reported ‘usual’ weekly hours used of formal childcare at the date of interview) are the same throughout the year, which is likely to overestimate average participation.

While larger discrepancies occur between the non-cash benefits modelled in SWITCH compared to administrative statistics it is important to bear in mind the more limited impact of this than if large discrepancies occurred between SWITCH and administrative numbers on cash benefits. As these are non-cash benefits, no monetary value is generally placed on them, so these discrepancies have no impact on the estimated (cash) income of a household and therefore any model results estimating distributional impacts, poverty and inequality rates. The main advantage of simulating eligibility for these schemes is to allow policymakers and government users of the model to estimate the impact of changes to tax-benefit rules or to parameters of these schemes, such as the means test on the numbers eligible for them. It is also likely that comparisons to administrative numbers may improve over time – for example if the take-up rates of Medical or GP-Visit cards improves and as the NCS becomes more established.

The SWITCH model has been used by researchers and policy makers to simulate the effect of policy changes in Ireland for three decades. Recently, the model was updated by harmonising the modelling platform, modelling language and simulation engine with that of EUROMOD, the European Union’s microsimulation model.

While the model runs on the EUROMOD platform, SWITCH presents some advantages compared to the Irish component of EUROMOD. First, it is linked to more detailed register data which is reweighted to better represent the reference year population. This also allows the simulation of extra policies, such as non-cash benefits. Second, SWITCH is updated in a more timely manner than EUROMOD, in line with the Irish budgetary cycle. The more up to date nature of SWITCH is essential due to the fact that the model is provided to users across a variety of government departments and allows results from the model to feed into the annual Budget process, for example by allowing the analysis of the cost and distributional impact of tax-benefit changes to take place before the changes are actually implemented. The harmonisation of SWITCH with the European microsimulation model also presents an opportunity for researchers by facilitating cross-country research.

This paper provides information on how the input data for the SWITCH model is prepared and verified. It also details why, and how, reweighting of the input data is carried out and how the data is adjusted to be representative of the current population through the use of uprating. A description of how the impact of the COVID pandemic labour market shock is implemented is also provided.

The model is extensively validated against external administrative sources. The reweighting process helps ensure that the underlying data in the model represents to a much greater extent the labour force, income distribution and cash benefit recipiency population which greatly improves the accuracy of the model and provides users with additional faith in results from the model. While discrepancies occur between model estimates of entitlements to in-kind benefits the fact that such benefits do not carry a monetary value means that such discrepancies have no impact on the main results of the model as they are based on cash incomes. Some comparability issues with administrative statistics for these non-cash benefits have also been highlighted and help explain why such discrepancies exist.

1. 1
   The impact of COVID-19 on households´ income in the EU

   1. V Almeida
   2. S Barrios
   3. M Christl
   4. S De Poli
   5. A Tumino
   6. W van der Wielen

   (2021)

   Journal of Economic Inequality 19:413–431.

   https://doi.org/10.1007/s10888-021-09485-8

   • PubMed
   • Google Scholar
2. 2
   The Oregon experiment--effects of Medicaid on clinical outcomes

   1. K Baicker
   2. SL Taubman
   3. HL Allen
   4. M Bernstein
   5. JH Gruber
   6. JP Newhouse
   7. EC Schneider
   8. BJ Wright
   9. AM Zaslavsky
   10. AN Finkelstein
   11. Oregon Health Study Group

   (2013)

   New England Journal of Medicine 368:1713–1722.

   https://doi.org/10.1056/NEJMsa1212321

   • Google Scholar
3. 3
   Analysing the effects of tax-benefit reforms on income distribution: a decomposition approach

   1. O Bargain
   2. T Callan

   (2010)

   The Journal of Economic Inequality 8:1–21.

   https://doi.org/10.1007/s10888-008-9101-4

   • Google Scholar
4. 4
   Changes in income distributions and the role of Tax-Benefit policy during the great recession: an international perspective

   1. O Bargain
   2. T Callan
   3. K Doorley
   4. C Keane

   (2017)

   Fiscal Studies 38:559–585.

   https://doi.org/10.1111/1475-5890.12113

   • Google Scholar
5. 5
   Research in Labor Economics (Research in Labor Economics)

   1. O Bargain
   2. K Doorley

   (2011)

   307–339, In-Work Transfers in Good Times and Bad: Simulations for Ireland, In:, Research in Labor Economics (Research in Labor Economics), Vol, 33, Emerald Group Publishing Limited, pp.

   • Google Scholar
6. 6
   Le modèle de microsimulation TAXIPP – version 1.0, guide Methodologique IPP, Septemberr 2019

   1. M Ben Jelloul
   2. A Bozio
   3. T Douenne
   4. B Fabre
   5. C Leroy

   (2017)

   1–52.

   • Google Scholar
7. 7
   Spending review 2018: analysis of replacement rates

   1. C Boyle

   (2018)
8. 8
   Micro-Simulation in Action

   1. T Callan
   2. K Coleman
   3. J Walsh

   (2006)

   Emerald Group Publishing Limited.

   • Google Scholar
9. 9
   Non-take-up of means-tested benefits: national report for Ireland

   1. T Callan
   2. C Keane

   (2008)
10. 10
    Ireland’s economic adjustment: from crisis to recovery

    1. T Callan
    2. C Keane
    3. M Savage
    4. JR Walsh
    5. B Colgan

    (2014a)

    Inter Economics 49:300–322.

    https://doi.org/10.1007/s10272-014-0514-y

    • Google Scholar
11. 11
    Crisis, response and distributional impact: the case of Ireland

    1. T Callan
    2. B Nolan
    3. C Keane
    4. M Savage
    5. JR Walsh

    (2014b)

    IZA Journal of European Labor Studies 3:9.

    https://doi.org/10.1186/2193-9012-3-9

    • Google Scholar
12. 12
    The impact of comparable policies in European countries: Microsimulation approaches

    1. T Callan
    2. H Sutherland

    (1997)

    European Economic Review 41:627–633.

    https://doi.org/10.1016/S0014-2921(97)00006-8

    • Google Scholar
13. 13
    Standard report on methods and quality for the 2019 survey on income and living conditions (EU-SILC)

    1. CSO

    (2020)
14. 14
    The unequal inequality impact of the COVID-19 pandemic (no.2021/06)

    1. ZM Darvas

    (2021)

    Bruegel Working Paper.

    • Google Scholar
15. 15
    TTax strategy group social protection budget package – budget 2022 issues

    1. Department of Social Protection

    (2021)

    1–34.

    • Google Scholar
16. 16
    Social impact assessment of the welfare and income tax measures in budget 2022

    1. Department of Social Protection

    (2022)
17. 17
    Calibration estimators in survey sampling

    1. J-C Deville
    2. C-E Särndal

    (1992)

    Journal of the American Statistical Association 87:376–382.

    https://doi.org/10.1080/01621459.1992.10475217

    • Google Scholar
18. 18
    Demographic change and the European income distribution

    1. M Dolls
    2. K Doorley
    3. A Paulus
    4. H Schneider
    5. E Sommer

    (2019)

    The Journal of Economic Inequality 17:337–357.

    https://doi.org/10.1007/s10888-019-09411-z

    • Google Scholar
19. 19
    Automatic stabilizers and economic crisis: US vs

    1. M Dolls
    2. C Fuest
    3. A Peichl

    (2012)

    Europe," Journal of Public Economics, Elsevier 96:279–294.

    • Google Scholar
20. 20
    What drove income inequality in EU crisis countries during the Great Recession, 42

    1. K Doorley
    2. T Callan
    3. M Savage

    (2021)

    Fiscal Studies.

    https://doi.org/10.1111/1475-5890.12250

    • Google Scholar
21. 21
    Role of Distributional impact assessment

    1. C Fiorio

    (2021)

    Presentation to Conference on Addressing Economic Inequalities organised by the European Commission.

    • Google Scholar
22. 22
    Estimating the anti-poverty effects of changes in taxes and benefits with the TRIM3 microsimulation model

    1. L Giannarelli
    2. J Morton
    3. L Wheaton

    (2007)

    Washington, DC: Urban Institute.

    • Google Scholar
23. 23
    Supporting low income working families: Enabling resilience

    1. J Gray
    2. C Rooney

    (2018)

    s.l: MUSSI Working Paper Series.

    • Google Scholar
24. 24
    Can Parents Afford to Work? Childcare Costs, Tax-Benefit Policies and Work Incentives, IZA Discussion Papers, No.1932

    1. H Immervoll
    2. D Barber

    (2006)

    Bonn: Institute for the Study of Labor (IZA).

    • Google Scholar
25. 25
    Trends in income inequality, pro-poor income growth, and income mobility

    1. SP Jenkins
    2. P Van Kerm

    (2006)

    Oxford Economic Papers 58:531–548.

    https://doi.org/10.1093/oep/gpl014

    • Google Scholar
26. 26
    Assessing individual income growth

    1. SP Jenkins
    2. P Van Kerm

    (2016)

    Economica 83:679–703.

    https://doi.org/10.1111/ecca.12205

    • Google Scholar
27. 27
    Irish public finances through the financial crisis

    1. C Keane

    (2015)

    Fiscal Studies 36:475–497.

    https://doi.org/10.1111/j.1475-5890.2015.12077

    • Google Scholar
28. 28
    The Take-up of medical and GP-Visit cards in Ireland

    1. C Keane

    (2018)

    Powerpoint presentation, https://www.youtube.com/watch?v=NXR6q8WbYdw, Accessed, January 2022.

    • Google Scholar
29. 29
    Failure to take-up public healthcare entitlements: evidence from the medical card system in Ireland

    1. C Keane
    2. M Regan
    3. B Walsh

    (2021)

    Social Science & Medicine 281:114069.

    https://doi.org/10.1016/j.socscimed.2021.114069

    • Google Scholar
30. 30
    European Economy - Economic Briefs 028, Directorate General Economic and Financial Affairs (DG ECFIN), European Commission

    1. P McQuade
    2. S Riscado
    3. S Santacroce

    (2017)

    Personal Income Tax in Ireland: The Future of the Universal Social Charge, In:, European Economy - Economic Briefs 028, Directorate General Economic and Financial Affairs (DG ECFIN), European Commission.

    • Google Scholar
31. 31
    Modelling the Distributional impact of the COVID-19 crisis

    1. C O'Donoghue
    2. DM Sologon
    3. I Kyzyma
    4. J McHale

    (2020)

    Fiscal Studies 41:321–336.

    https://doi.org/10.1111/1475-5890.12231

    • Google Scholar
32. 32
    Sreweight: a Stata command to Reweight survey data to external Totals

    1. D Pacifico

    (2014)

    The Stata Journal: Promoting communications on statistics and Stata 14:4–21.

    https://doi.org/10.1177/1536867X1401400103

    • Google Scholar
33. 33
    Indexing out of poverty? fiscal drag and benefit erosion in Cross‐National perspective

    1. A Paulus
    2. H Sutherland
    3. I Tasseva

    (2020)

    Review of Income and Wealth 66:311–333.

    https://doi.org/10.1111/roiw.12413

    • Google Scholar
34. 34
    EUROMOD Working Papers EM9/17, EUROMOD at the Institute for Social and Economic Research

    1. A Paulus
    2. I Tasseva

    (2017)

    Decomposition of changes in the EU income distribution in 2007-2011, In:, EUROMOD Working Papers EM9/17, EUROMOD at the Institute for Social and Economic Research.

    • Google Scholar
35. 35
    SSpending review 2019 costing framework for the expansion of GP care

    1. S Prior
    2. N Duff
    3. R Scott

    (2019)

    1–36.

    • Google Scholar
36. 36
    Accounting for differences in income inequality across countries: tax-benefit policy, labour market structure, returns and demographics

    1. DM Sologon
    2. P Van Kerm
    3. J Li
    4. C O’Donoghue

    (2021)

    The Journal of Economic Inequality 19:13–43.

    https://doi.org/10.1007/s10888-020-09454-7

    • Google Scholar
37. 37
    EUROMOD: the European Union tax-benefit microsimulation model

    1. H Sutherland
    2. F Figari

    (2013)

    International Journal of Microsimulation pp. 1–24.

    • Google Scholar
38. 38
    Technical Paper on the Measurement of Household Joblessness in SILC and QNHS, 2004-2012: An Analysis of the Central Statistics Office (CSO) Survey of Income and Living Conditions (SILC) and the Quarterly National Household Survey (QNHS

    1. D Watson
    2. B Maître
    3. H Russell

    (2015)

    Technical Paper on the Measurement of Household Joblessness in SILC and QNHS, 2004-2012: An Analysis of the Central Statistics Office (CSO) Survey of Income and Living Conditions (SILC) and the Quarterly National Household Survey (QNHS, Social Inclusion Technical Paper no.6, Dublin, Department of Social Protection.

    • Google Scholar

Author details

1. Claire Keane

2. Karina Doorley

3. Theano Kakoulidou

   For correspondence

   theano.kakoulidou@esri.ie

4. Seamus O’Malley

   Economic and Social Research Institute, Dublin, Ireland