Budget 2019 to 2020: feasibility of distributional analysis - study

3. The Model

The first step was to consider the range of existing models available to undertake this work. Table 1 outlines the models and their characteristics. It became apparent from this review that existing models did not provide what was needed for the purpose of a cumulative impact assessment of the budget. They were either not available to Scottish Government to use; did not provide sufficient coverage of taxes, benefits or services; or they were not without constraints about how we could use them or publish findings from them. Importantly, some of the models were not designed to undertake this type of cumulative distributional analysis – but had been developed to undertake 'what if' analysis. This led us to conclude that a bespoke model was necessary.

Table 1: Comparison of models

Model	Owner	Available to SG?	Main Data Source	Coverage
				Direct Tax	Benefits	Indirect Tax	Public Services
Tax and Welfare Model	Scottish Government	Yes	FRS	Yes	Yes	No	No
Tax Transfer Model[10]	Landman Economics	No	FRS	Yes	Yes	Yes	Yes
Intra-Gov Tax Benefit Model (IGOTM)[11]	HM Treasury	Yes	LCF	Yes	Yes	Yes	Yes – as add on
Policy Simulation Model (PSM)[12]	DWP	Partially	FRS	Yes	Yes	No	No
Euromod[13]	ISER, Univ. of Essex	Yes	FRS	Yes	Yes	No	No
TaxBen[14]	IFS	No	LCF	Yes	Yes	Yes	No
ONS microsimulation model[15]	ONS	No	LCF	Yes	Yes	Yes	Yes

To perform this work our preference was to have a single model. Hence we have created a model that comprises components to model the tax system, welfare benefits system and public service expenditure. It is a static model and does not take into account behavioural effects, such as how people might restructure their affairs in the event of a tax rise. More detailed information on each part of the model is contained in Annexes B, C and D. What follows is a brief technical overview of the model.

3.1 Approaches to modelling

Having decided upon our overarching principles - that it must be possible to allocate expenditure to individuals or households and that we are concerned only with taxes, benefits and public services which are paid or used within a given year – the next step was to decide which taxes, benefits and public service expenditure to include.

An overview of approaches taken in the modelling is presented below – and more detail is provided in the annexes.

Taxes^[16]

To model a given tax, we have to be able to assign it to either an individual or a household. Some taxes cannot be assigned to either because they are paid by businesses (e.g. corporation tax) or because there is insufficient data to model them (e.g. inheritance tax, capital gains tax).^[17] Other taxes may be paid by both households and businesses – e.g. around 70% of VAT is paid by households.^[18] Some taxes, such as Land and Buildings Transaction Tax (LBTT), are also challenging to model because they are paid infrequently. We decided to include only those taxes which can easily be assigned to individuals. This includes Income Tax, National Insurance (NI), VAT, Council Tax and all duties, including for example alcohol, tobacco and fuel duties.

In Scotland, non-savings non-dividend (NSND) income^[19] tax is modelled by the Scottish Fiscal Commission for budget purposes. To do this they use a model which is based on the Survey of Personal Incomes – the best source for modelling NSND income tax. It would be technically possible to constrain the NSND income tax output from our modelling to that forecast by the Scottish Fiscal Commission. However, in order to ensure the integrity of our modelling we took an approach in which our NSND income tax model shares the same assumptions as the Scottish Fiscal Commission, but the model is allowed to generate slightly different tax results, as would be usual with different models. The differences are very small and make no discernible difference to the analysis. See Annex B for further detail.

Welfare benefits

Welfare benefits may be paid to individuals or households – and therefore can be attributed, thus meeting our principles. There is considerable range in the amount of spend on benefits (from UK and Scottish governments) and the number of recipients of each benefit. The smaller benefits present some challenges for distributional analysis, for example the Scottish Welfare Fund or Best Start Grant. They may be difficult to model due to the relatively small number of recipients involved who may not appear in surveys. These smaller benefits are therefore unlikely to have discernible impacts in the distributional analysis outputs. However, they may have a significant impact on the people who receive them and they form part of the overall Scottish offer.

Public Services

Whilst the methodology to undertake distributional analysis of tax and welfare (cash) benefits is well established and has been used for some time – there is no standard methodology for carrying out a distributional analysis of public service provision. Studies that have been published have tended to be concerned with methodological issues^[20] or undertaken by external bodies^[21] – very few governments publish this type of analysis. There are several key areas that present particular difficulties, including:

• lack of information about who uses services and how this differs amongst different groups
• who benefits from use of the service
• lack of clarity about how different people/groups value the services they receive – in quantitative terms
• what impact changes in level of spending on services might have.

As a result of these challenges, modelling public services is the area of greatest uncertainty.

We – like others such as the HM Treasury – decided to exclude services that could not be attributed to individuals such as defence and environmental services. For most services we have adopted the usage principle – in that services are distributed amongst users in the analysis. However for health, following OECD^[22] we have adopted an insurance approach – so spend is spread across the entire population rather than by usage - as it is argued that the population as a whole benefits from the insurance and knowledge of free health provision, in particular in public health provisions. Detailed breakdowns of usage at individual level are also scarce – even if we wanted to use that approach. However, we have made adjustments for age, sex and deprivation as this reflects differences in usage at population level. This in part recognises that any increase or decrease in health spend would not be felt equally amongst different groups. Further details of approach are in Annex D.

Equality/protected characteristics

For some equality characteristics available data is scarce and even where information is available, small sample sizes often mean that reliable modelling cannot be carried out. Developments such as the recent boosts to the FRS Scottish sample has allowed more detailed breakdowns but information for some characteristics remains scarce.

No suitable data is available on maternity and pregnancy, gender reassignment, or sexual orientation.^[23] In addition to the Family Resources Survey, other potential sources such as the Scottish Household and Scottish Health surveys were considered. While these surveys are useful for sense checking and providing contextual information, they did not provide data that could be used in the modelling – because of small sample sizes. These characteristics are therefore excluded from the analysis.

In keeping with other equality analysis produced by Scottish Government and with the approach taken by EHRC in their distributional analysis^[24] - we have produced some analysis at the level of the household. We have also produced analyses for individuals.

Income sharing within households

Sharing of income within households has been a problematic issue for analysis of this type, particularly for gender analysis. However, recent data from the Scottish Social Attitudes Survey shows that the most common way people living with a partner organise their own income is to put all of it, or nearly all of it, into a joint pool and the majority of those living with a partner also organise their income in the same way their partner does.^[25] The survey also reported that women were much more likely (60%) than men (8%) to have child benefit paid into their own account. However, detailed information about withdrawal from/use of pooled income is not available.

On the basis of an equal joint pool, we assume that taxes paid at the household level (e.g. council tax and indirect taxes such as VAT) are divided equally between the adults in the household. Similarly, household benefits – such as Council Tax Reduction and Housing Benefit – are assumed to benefit all adults equally in the household. We know that this is not the case in all instances, but the absence of data curtails our ability to make more accurate allocations within the household.

All other income from earnings or benefits is assumed to be retained at an individual level. For example, if a person receives all the child benefit for a household, this is retained by that person. Further sensitivity testing around income sharing is recommended if the model is to be utilised in future years.

3.2 The model - brief technical overview

• The model uses six years of the Department for Work and Pension's Family Resources Survey (FRS), from 2012/13 to 2017/18. The reason for choosing this survey was that it is the main source of household income data for the UK. Whilst distributional analysis work done by HM Treasury uses the Living Cost and Food Survey (LCFS), the small sample size in this survey for Scotland meant that it was unsuitable for our analysis. Multiple years of the FRS were combined to reduce volatility in the estimates of the Scottish Rate of Income Tax.
• We correct for benefit under-reporting in the dataset. It is reported that a fifth of benefit spending is missing from the best source of household income data^[26] - the Households Below Average Earnings (HBAI) dataset based on the FRS.
• The six-year combined FRS dataset is re-weighted to the age-sex profile for Scotland contained within the Mid-Year 2019 population estimates. The FRS is a survey of private households. Therefore, by design, it will not include students living in halls of residence, prisoners or care home residents. By re-weighting the data, we are making the assumption that young people missing from the FRS have similar incomes to people in the survey of the same age and sex. The same is true for care home residents missing from the survey.
• As we have combined multiple years of the FRS, we have uprated the incomes in each survey year to bring them to 2019/20 prices. Using the same method as the Office for National Statistics,^[27] we uprate employment and self- employment with average earnings, private pensions are increased using the retail price index (RPI) and incomes from bank and building society interest are increased using deposit rates (RDEP).

Students

• In the construction of income deciles, we have assigned higher education students to the household net income decile of their parents using the typical population level socio-economic background of their parents.^[28] Using these adjusted income deciles for students, higher education funding is distributed across the income deciles. Without such an adjustment, higher education funding would fall only within the lowest deciles as higher education students have low incomes.Whilst higher education students have been moved for modelling purposes to a different household net income decile, their individual income is not changed in the model. There is no adjustment required for further education students because in the vast majority of cases they continue to live with their parents.

3.3 Overview of what's in the Model

Applying our principles and approach to modelling, the taxes, benefits and public services included in the model are summarised in Table 2. More detailed information on how each component has been modelled is provided in annexes B, C and D.

Table 2: Model - Summary of What's Included, Excluded and Coverage

Taxes	Benefits	Public Services
What's Included
Direct Taxes Non Savings Non Dividend Income Tax Employee NI Contributions Council Tax Indirect Taxes VAT Insurance Premium Tax Fuel Duty Alcohol Duty Tobacco Duty Land and Buildings Transaction Tax Air Passenger Duty Vehicle excise duty Total customs duties Total intermediate taxes[29] Taxes on final goods and services Betting taxes Television licence Camelot national lottery fund	State Pension Pension Credit Winter Fuel Payments Attendance Allowance Jobseeker's Allowance Employment and Support Allowance Income Support Working Tax Credit Child Tax Credit Child Benefit Disability Living Allowance Personal Independence Payment Housing Benefit Universal Credit Council Tax Reduction Statutory Maternity Pay Carer's Allowance Industrial Injuries Disablement Benefit Discretionary Housing Payments Over 75 TV Licences Bereavement Benefit / Widow's Benefit Scottish Welfare Fund Maternity Allowance Severe Disablement Allowance Cold Weather Payments Funeral Expenses Payments Sure Start Maternity Grant Incapacity Benefit Scottish Child payment when this comes on stream in 2021	Health Schools Social Work – includes social care such as care homes Universities and Colleges Apprenticeships Early learning & childcare Transport - Rail, bus and concessionary fares
What's Excluded (Main Exclusions only)
VAT paid by businesses Scottish Water charges Employer National Insurance All other taxes		services that are not frontline services where spend is not in the given year services where benefit from service is not realised in same given year services that cannot be attributed to individuals such as defence, environment
Coverage
£33.247 billion	£19.273 billion	£24.514 billion
58% of total Non-North Sea Taxes paid in Scotland.	100% of welfare benefits in Scotland.	73% of resource funding and local government expenditure on public services.
Further Information
See Annex B	See Annex C	See Annex D