Tackling child poverty delivery plan: forecasting child poverty in Scotland

Appendix 2: The IPPR/Resolution Foundation/Landman Economics tax-transfer model

Overview

The calculations of the distributional effects of tax and transfer (benefit, tax credit and Universal Credit) policies is in this report were made using a tax-benefit microsimulation model (the tax-transfer model, or TTM) which was originally written by Landman Economics, and which is maintained jointly by Landman Economics, the Resolution Foundation and the Institute for Public Policy Research. This appendix gives a technical overview of the model.

Data and Outputs

The tax-benefit model uses data from the Family Resources Survey ( FRS) to analyse the impact of direct taxes, benefits, tax credits and Universal Credit, and the Living Costs and Food Survey ( LCF) to analyse the impact of indirect taxes. The information in the FRS and LCF allows payments of direct taxes and receipts of benefits and tax credits to be modelled with a reasonable degree of precision for each family in the surveys using either the current tax/benefit system which is in place at the moment, or an alternative system of the users’ choice. For example, the user can look at what the impact of an increase in the income tax personal allowance would be. Using a ‘base’ system (this is often the actual current tax and benefit system, although the model can use any system as the base) and one or more ‘reform’ systems, the model can produce several types of outputs, for example:

• Aggregate costings of each system (i.e. amount received in direct and indirect personal taxes, and amount paid out in benefits and tax credits);
• Distributional impacts of reform system compared with base system (e.g. change in incomes in cash terms and as a percentage of weekly income in the base system). The distributional effects can be broken down according to several different variables, as shown in the section "individual and household identifier variables" below.
• Proportions of exchequer savings/costs due to a particular reform or set of reforms paid for by/going to particular family types
• Average impact of reforms on the household incomes of particular types of individuals, eg children, working age adults and pensioners;
• Winners and losers from a particular reform or set of reforms (grouped according to size of cash gain or size of percentage gain);
• Impact of reforms on overall inequality of disposable incomes (Gini coefficient);
• Impact of reforms on household and child poverty rates (using various definitions, e.g. proportion of children below 60% of median income);
• Changes in Marginal Deduction Rates ( MDRs), i.e. the net gain to people in employment from an extra pound of earned income (which, for many individuals, will depend on income tax and National Insurance Contribution rates as well as the taper rates on means-tested benefits and tax credits).

Note that only the FRS data was used for this report because the child poverty measures contained in the Child Poverty (Scotland) Act do not take into account the effect of indirect taxes on household incomes. Therefore, the remainder of this Appendix only discusses the FRS component of the model.

Reforms modelled

The TTM is able to model most, but not all, of the features of the tax and social security systems in Scotland and the rest of the UK. For the purposes of modelling, the tax and social security system can be classified into three categories as follows

1. Features modelled with high accuracy. These include the following:

• Income-based taxation, e.g. income tax, National Insurance Contributions;
• Most parts of the benefit, tax credit and Universal Credit systems.

2. Reforms modelled with lower accuracy. Some aspects of the tax and social security reforms are modellable but with lower accuracy because the relevant information necessary to model the reforms with high accuracy is not available in the FRS dataset. The main examples of these are as follows:

• Council Tax payments and Council Tax support payments can only be approximated because the FRS data do not contain local authority information. This is less of a problem in Scotland where the Scottish Government’s Council Tax Reduction Scheme operates on a nationwide basis, and can be modelled with higher accuracy.
• The Local Housing Allowance for Housing Benefit claimants can only be approximated, again because of the lack of local authority data in the FRS.
• Assessment and re-assessment for disability-related benefits (in particular Employment and Support Allowance, and the replacement of Disability Living Allowance with Personal Independence Payment) cannot be modelled with full accuracy because the FRS does not have enough detail on the type and severity of disabilities which affect each claimant.

3. Reforms which can't be modelled. Some aspects of the tax and welfare reforms cannot be included in the analysis because the FRS data doesn't contain enough information to model them at all. The main examples of these are:

• Changes to the rules on income thresholds for repayment of tax credits when family income increases from one year to the next; these can't be modelled because the FRS doesn't contain information on the previous year's incomes for each household.
• Sanctions for JSA and ESA claimants as well as Universal Credit; the FRS doesn't contain information on whether claimants are being sanctioned or not.

It is important to note that the TTM is able to model the following Scotland-specific tax and benefit policies with high accuracy:

• Additional funding from the Scottish Government to mitigate the impact of the “bedroom tax”.
• Recent increases to multipliers for the four top Council Tax bands (E,F,G and H).
• The Council Tax Reduction Scheme for low-income households.
• The increase in Carers’ Allowance to the level of JSA from 2018/19 onwards.
• The introduction of the Best Start Grant, replacing the Sure Start Maternity Grant.
• The changes to income tax rates in Scotland from 2018/19 onwards.

Modelling partial take-up

The take-up algorithm

Previous forecasts of child poverty using the TTM (for example Reed and Portes, 2014) assumed full take-up of means-tested benefits, tax credits and UC. In 2017 a new partial take-up algorithm was developed for the TTM. For a range of means-tested benefits (Housing Benefit, Income Support, income-based Employment and Support Allowance, income-based Jobseeker’s Allowance and Pension Credit) and for tax credits, the algorithm operates as follows:

First, actual benefit or tax credit receipt is compared with modelled receipt of the benefit or tax credit.

Second, the benefit unit is assigned to a quadrant based on the decision matrix in Table A1.1 below, and action is taken (or not taken) based on the assignment.

Table A1.1 Decision matrix for partial take-up algorithm: actual receipt vs modelled receipt

Benefit unit status:	Modelled as receiving benefit/tax credit	Not modelled as receiving benefit/tax credit
Actually receiving benefit/tax credit	Award benefit	Don’t award benefit
Not actually receiving benefit/tax credit	Award benefit based on take-up algorithm	Don’t award benefit

The next course of action for each benefit unit depends on which box of the decision matrix the benefit unit is assigned to, based on a comparison of actual and modelled receipt. Four options are possible:

1. If the benefit unit is actually receiving the benefit (or tax credit), and is also modelled as receiving the benefit in the TTM, the benefit is paid.

2. If the benefit unit is not receiving the benefit, and is modelled as not receiving the benefit, the benefit is not paid.

3. If the benefit unit is actually receiving the benefit, but is modelled as not receiving the benefit, the benefit is not paid.

4. If the benefit unit is not actually receiving the benefit, but is modelled as receiving the benefit, the partial take-up algorithm is applied.

All the remaining explanation in this section relates to option 4) – benefit units who are modelled as receiving a benefit (or tax credit) but do not actually receive that benefit or tax credit.

The partial take-up algorithm for each benefit works as follows:

For benefit units who are modelled as receiving a benefit or tax credit, a take-up regression is estimated. The regression is a probit regression with the dependent variable being actual take-up of the benefit or tax credit in question and the regressor variables being as follows:

• Ethnicity
• Disability (core group, wider group)
• Family demographic status (couple with children, couple without children, lone parent, single person with no children)
• Region
• Employment
• Housing tenure type (social tenant, private tenant, owner-occupier ^[31]

The predictions from this regression (plus a random error term for each benefit unit) are used to create a ranking (from 0 to 100) which is used to calibrate take-up of each benefit and tax credit in the FRS so that the grossed-up percentage of benefit units claiming each benefit in the model matches published DWP and HMRC statistics.

Table A1.2 compares estimated take-up rates from the pooled FRS data in the tax-transfer model – calculated as number of benefit units actually taking up each benefit, divided by number of benefit units modelled as receiving each benefit – with published take-up statistics from DWP (2017c) and HMRC (2017)– calculated in the same way, but using administrative data combined with FRS-based modelling. The table shows that estimates from the TTM for take-up proportions of each featured benefit and tax credit are below DWP and HMRC’s published statistics. This means that the estimated take-up rate in the FRS data needs to be adjusted upwards in the TTM so that estimated take-up matches published take-up rates. For example, our ‘raw’ estimate of take-up in the TTM is 42%; this needs to be adjusted upwards by 20 percentage points to match DWP’s Pension Credit take-up statistics.

Table A1.2. Comparison of estimated take-up rates for FRS data in tax-transfer model with published take-up statistics from DWP and HMRC, by caseload

Benefit/tax credit	TTM estimate (%)	DWP or HMRC estimate (%)	Difference, DWP/ HMRC minus TTM (ppts)
Pension Credit	42	62	20
JSA	46	50	4
IS/ESA	65	82	17
Working Tax Credit	48	68	20
Child Tax Credit	75	87	-12

Source: comparison of TTM estimates with DWP (2017c) and HMRC (2017)

Using the prediction ranking from the take-up regressions (as explained above), it is possible to adjust the simulated take-up rate for each benefit or tax credit in the TTM to match any percentage total between zero and 100%. The parameter files provide the flexibility to do this separately for each of the benefits and tax credits in Table A1.2. In the simulations presented in this report we assume that the take-up rates for each benefit and tax credit match DWP and HMRC’s latest published statistics.

Take-up of Universal Credit

Universal Credit ( UC) presents an additional problem because there are, as yet, no official statistics from DWP on the take-up rate. However, it is generally assumed that the take-up rate for UC will be higher than the take-up rate for the benefits and tax credits it replaces, for one particular reason: there are currently many benefit units who are eligible for more than one of the benefits or tax credits which are being replaced by UC, but who do not claim the whole package of benefits. For example, there are benefit units who are eligible for tax credits and Housing Benefit but whom claim only one or the other. Because UC is a single payment replacing several different benefits, when a claim is processed, it is equivalent to a situation in which the benefit unit applied for all the ‘legacy’ benefits and tax credits, and this should result in a boost in take-up rates.

To estimate the extent to which UC might be expected to boost take-up rates, all other things being equal, we used the TTM to calculate the number of benefit units who claimed any of the benefits being replaced by UC (Income Support, income-based JSA, income-based ESA, Housing Benefit, Working Tax Credit and Child Tax Credit) as a proportion of the number of benefit units modelled as eligible to receive any of those benefits in the TTM. The calculation (adjusted for the gap between TTM estimates of take-up rates for the individual benefits and DWP/ HMRC estimates) was a UC take-up rate of 87%. This is a relatively high take-up rate compared to the DWP/ HMRC estimates for most of the individual benefits and tax credits. However, we adjust this assumption slightly downwards, by 5 percentage points, to take account of recent evidence from UC sanctions statistics that the sanction rate for claimants of UC is substantially higher than the average sanctions rate for the benefits and tax credits it replaces (Webster, 2017). Thus, 82% is our headline take-up rate assumption for Universal Credit in the reform scenario.

Simulating changes in child poverty rates

Modelling of the impact of reforms to direct taxes and transfer payments on the number of children in poverty proceeds in six stages as follows.

Firstly, the FRS data from the pooled 2012-13, 2013-14, 2014-15 and 2015-16 Households Below Average Income ( HBAI) dataset is analysed to identify households who are below 60% of equivalised median household disposable income on the relative and absolute After Housing Costs ( AHC) measure.

Next, the pooled FRS data for 2012-13 through to 2015-16 are run through the TTM using the relevant parameter files for the data year and for 2015-16 (the starting year for the simulations). For each household, we also calculate a ‘calibration factor’ equal to the difference between modelled income and actual HBAI income for each household in the data base year (2012/13, 2013/14, 2014/15 or 2015/16 depending on which year of the FRS data the household is from). This calibration factor is added back into the modelled income estimate for each year. The objective of this procedure is to ensure that the starting child poverty rates in 2015/16 match the published statistics from Poverty and Inequality in Scotland.

For each of our scenarios (a), (b) and (c) (explained in Section 2.5 in the main report), the pooled FRS data are run through the TTM for all tax years between 2016/17 and 2030/31 inclusive. The earnings and other gross incomes in the model are uprated to the relevant tax year in each case. This produces a total of (4 data years) x (3 scenarios) x (15 forecast years) = 180 model runs.

New simulated child poverty rates for each data year, each scenario and each forecast year are calculated based on modelled net incomes (after applying the calibration factor).

This procedure is used for all poverty measures 1 and 2 (relative AHC and absolute AHC). The only difference between the relative and absolute poverty calculation procedures is that for the relative poverty measures, the poverty line is recalculated based on the modelled distribution of incomes in the baseline and reform scenarios, whereas for the absolute poverty measures, the AHC poverty line in the 2010-11 tax year are used (uprated by the Consumer Price Index).

For poverty measure 3 (combined low income and material deprivation), relative AHC poverty using a higher threshold (70% of median equivalised household income) is calculated and this is then combined with regression-based prediction of material deprivation as outlined in Section 2.4 in the main report.

For poverty measure 4 (persistent poverty), household net income in each scenario (a), (b) and (c) is calculated using the TTM and this is combined with the results from the regression for the probability of persistent poverty in the USoc data to produce a predicted probability of being in persistent poverty for each household in the pooled Scottish FRS sample.