Publication - Research publication

Tackling child poverty delivery plan: forecasting child poverty in Scotland

Published: 29 Mar 2018

Child poverty projections for Scotland independently produced by Howard Reed at Landman Economics and Graham Start at Virtual Worlds Research.

82 page PDF

778.8 kB

82 page PDF

778.8 kB

Contents
Tackling child poverty delivery plan: forecasting child poverty in Scotland
Chapter 2. Methodology

82 page PDF

778.8 kB

Chapter 2. Methodology

2.1 Microsimulation of household net incomes

Our approach for forecasting child poverty over the years to 2030/31 relies on using a microsimulation model to estimate household net incomes under three different policy scenarios (as explained in Section 2.6 below). We use the IPPR/Resolution Foundation/Landman Economics tax-transfer model [11] (referred to hereafter in this report as the “tax-transfer model”, or the abbreviation TTM) to model the effects of reforms to the tax and social security system in Scotland and the rest of the UK. Broadly, the following parts of the system are modelled:

  • Income tax;
  • National Insurance Contributions ( NICs);
  • Council Tax and Council Tax-related benefits (including the Council Tax Reduction scheme in Scotland);
  • Means-tested and non-means tested benefits;
  • Tax credits;
  • Universal Credit, which is currently being rolled out to all families in the UK and replacing tax credits and most means-tested benefits.
    In addition to this, the analysis also models the introduction of the National Living Wage ( NLW) – an above-inflation increase in the minimum wage for employees aged 25 and over, introduced in 2015 and uprated every April since then, with the stated intention of increasing to 60 percent of median earnings by 2020.

The TTM is a microsimulation model which uses data from the UK Family Resources Survey ( FRS) (discussed further in Section 2.2 below). It calculates net incomes for households (and benefit units within households, and individuals within benefit units) given a set of tax-transfer parameters, and for a given tax year (e.g. 2017-18). The parameters are held in files in spreadsheet format; a set of parameters can describe the actual tax-transfer system in place at a given time, or a simulated system with one or more reforms implemented (for example, an increase in income tax rates).

The model takes account of all the reforms to the tax and social security system in Scotland and other countries of the UK that can be modelled using the information in the FRS data. The model also takes account of partial take-up of means-tested benefits, tax credits and Universal Credit. Not everybody who is eligible for particular means-tested social security payments based on their circumstances in the FRS data actually claims those payments, and it is important for the modelling to control for this where possible [12] . Appendix 2 contains more detail on these features of the TTM.

The model is fundamentally static in that it does not attempt to model the effect of reforms to taxes or transfer payments on people’s behaviour; the analyses in this report assume that behaviour is unchanged in response to policy changes. However, the algorithm used to reweight the FRS data for years between 2017/18 and 2030/31 (discussed in more detail in Section 2.4 below) does take forecasts of increased employment in Scotland (and the rest of the UK) into account.

The TTM has been used for analysis of the impact of reforms to tax and social security on child poverty rates in the UK before – for example in an analysis of the impact of increased employment on child poverty by 2020 for the Social Mobility and Child Poverty Commission (Reed and Portes 2014a), cumulative impact assessments of Budget and Spending Review policy decisions by the UK Government for the Equality and Human Rights Commission (Reed and Portes, 2014b and 2018) and an analysis of the impact of various policy choices on the welfare state for inequality and poverty in 2030 (Harrop and Reed, 2015). It is a model with a proven track record for realistic poverty forecasts.

2.2 Survey data

The Family Resources Survey ( FRS)

The FRS is an annual survey of around 20,000 households per year in the UK, collected on a tax-year basis [13] . The most recent release of FRS at the time of writing this report was 2015-16 [14] ; the 2015-16 dataset contains 2,704 households from Scotland, 13,840 households from England, 1,930 households from Northern Ireland and 848 households from Wales.

The FRS is generally acknowledged to be the best source of data on individual, family and household gross incomes and disposable incomes (incomes after payment of direct taxes and transfer payments) in the UK. For this reason, the FRS is used for the UK Government’s statistical publication on the income distribution Households Below Average Income ( HBAI) ( DWP 2017b) and also the Scottish Government’s own publication Poverty and Income Inequality in Scotland (Scottish Government, 2017d).

The FRS is also suitable for microsimulation modelling of changes in taxes and transfer payments in response to policy reforms as it contains individual, family and household attribute variables which establish eligibility to many elements of the tax and transfer payment system (e.g. age, single/couple and/or marital status, number of children in the family, housing tenure type, and so on). The FRS also contains information on housing costs and childcare arrangements and expenditure (but not expenditure on other goods and services).

Because of the relatively small sample size for the FRS in Scotland, our calculations of child poverty for this project use a pooled dataset of four years (2012/13, 2013/14, 2014/15 and 2015/16). This gives a total sample size of 13,337 Scottish households out of 92,501 UK households (excluding a small number of households dropped from the dataset which the UK Government uses for HBAI because of incomplete or poor quality data).

Understanding Society

In order to forecast persistent poverty, it is necessary to use a panel dataset where the same households and individuals are interviewed repeatedly (e.g. on an annual basis). FRS is a cross-sectional dataset and so cannot be used in isolation to produce persistent poverty forecasts. The best source of panel data on household incomes is Understanding Society ( USoc). This is a panel survey of around 25,000 households which has been running since 2009; 6 waves of data are now available, with the most recent interviews in 2015-16. The DWP and Scottish Government have both produced experimental statistics on persistent poverty using the USoc data ( DWP 2017c; Scottish Government 2017e).

The USoc data on household disposable incomes for consecutive waves are used to produce estimates of persistent poverty (being in poverty in at least 3 of the 4 waves 3,4,5 and 6) which are then combined with the FRS data using techniques described in Section 2.4 below, to produce estimates of persistent poverty for the FRS sample.

2.3 Reweighting and uprating the data

The pooled FRS dataset for the four years 2012-13 through to 2015-16 needs to be adjusted for each of the forecast years (2016-17 [15] through to 2030-31) so that, for each forecast year, the dataset better resembles forecasts of what the population and the distribution of earnings and other gross incomes in future years. This project uses two techniques [16] to adjust the FRS dataset so that it better resembles the population profile and earnings distribution in future years:

1. Re-weighting. Survey datasets such as the FRS contain weights that are inversely related to the probability of a given household being selected in a random sample; for example, if lone parent households are less likely than average to respond to the FRS questionnaire (or to be contacted by FRS interviewers in the first place), lone parent households will tend to have a higher-than-average weight in the survey. These weights ‘gross up’ to the totals of each type of household in the UK population. By varying the weights it is possible to transform the FRS data so that it more closely resembles forecasts of the demographic structure of the population in future years.

2. Uprating. If we expect earnings or other economic variables (such as housing costs) to rise or fall in real terms, the recorded values of these variables in the FRS data can be adjusted before carrying out the calculations in the TTM.

Appendix 1 gives a detailed account of the re-weighting and uprating methods used to adjust the FRS data to forecast the demographic structure and economic variables in Scotland (and the rest of the UK) over the forecast period.

2.4 Forecasting the child poverty measures

This section gives an overview of the methods required for forecasting each of the four child poverty measures specified in the Child Poverty (Scotland) Act 2017.

Measure 1: relative child poverty

The forecast for this measure in each target year (between 2016/17 and 2030/31) is calculated as follows:

i. The pooled FRS data for the years 2012/13 to 2015/16 are reweighted and uprated to the target year according to economic projections for employment and earnings growth, and growth of gross incomes from other sources (e.g. investment, property etc). More detail on the reweighting and uprating procedures is given in Appendix 1;

ii. The relevant tax-benefit parameters for the target year are applied (see Section 2.5 on ‘policy scenarios’ below);

iii. Median household AHC (equivalised) net incomes are calculated at a UK-wide level, taking simulated weights for the target year into account;

iv. The child poverty rate is calculated based on 60% of the AHC poverty line.

Measure 2: absolute child poverty

The method for Measure 2 is exactly the same as for Measure 1 except that the poverty line in step (iii) is the 2010-11 median income level, uprated for inflation [17] to the target year.

Measure 3: combined low income and material deprivation

The forecasting of poverty measure 3 requires additional assumptions compared to measures 1 and 2 because measure 3 depends in part on material deprivation and this cannot be forecast directly using the tax-transfer model in the way that disposable income can. Instead, we estimate a simulated deprivation score for the FRS subsample of households with children in the target year.

The following procedure is used to estimate a simulated material deprivation score for each household with children in the FRS:

1) Pooling data from the FRS for 2010-11 [18] through to 2015-16;

2) A regression of household deprivation score against variables including:

i. household demographics (e.g. number of children in various age groups, number of single and couple adults);

ii. region;

iii. housing tenure;

iv. household earned income;

v. household investment income;

vi. household income from transfer payments (e.g. benefits, tax credits);

vii. employment status.

3) Based on the regression coefficients (plus a randomly generated error term for each household) and the forecast values of the explanatory variables for the target year (some are time invariant such as household demographics, whereas others change between the base year and the target year, such as household incomes), a simulated deprivation score is predicted for each household in the target year.

4) The proportion of households with a deprivation score of over 25 is calculated (based on the simulated sample weights for the target year) – this is the forecast of material deprivation in the FRS for the target year.

5) Forecast material deprivation is combined with a variant of the forecast of low income from Measure 1 above (using a higher poverty line, of 70% of median equivalised AHC income) to produce the forecast of combined low income and material deprivation which is Measure 3.

Note that this procedure can be repeated a larger number of times with differently randomly generated errors to build up a distribution of simulated forecasts. In Chapter 3 we use this technique to produce statistical confidence intervals for the estimates of Measure 3 over the forecast period.

Measure 4: persistent child poverty

The method for estimating persistent child poverty is similar to the method for forecasting material deprivation used for measure 3 above, but there is an additional complication because the persistent child poverty measure uses Understanding Society ( USoc) data rather than FRS.

Bearing this in mind, the forecast of persistent poverty proceeds as follows:

1. A pooled USoc dataset from waves 3 to 6 is used;

2. For households in waves 3 to 6, a (probit) regression of persistent child poverty is estimated (defined as child poverty in at least 3 of the last 4 years) against variables including:

i. household demographics (e.g. number of children in various age groups, number of single and couple adults);

ii. region;

iii. housing tenure;

iv. household earned income;

v. household investment income;

vi. household income from transfer payments (e.g. benefits, tax credits);

vii. employment status.

3. The probability of being in persistent poverty is predicted using the FRS sample for the target year, based on the regression coefficients from the USoc regression (plus a randomly generated error term for each household) and the forecast values of the explanatory variables for the target year

4. The proportion of households in persistent poverty in the FRS is calculated (based on the simulated sample weights for the target year). This is our estimate of measure 4.

As with measure 3 above, we perform a robustness analysis of the confidence intervals on the persistent child poverty results derived using this approach by replicating steps (iii) and (iv) a large number of times with different sets of random errors.

2.5 Policy scenarios

The four child poverty measures are estimated for three different policy scenarios. Each of these scenarios uses the actual tax and social security systems in place in Scotland and the rest of the UK for years up to and including the current tax year (2017/18). After 2017/18, the three scenarios make different assumptions about the future parameters of the tax and social security system, as follows:

Scenario (a): uprating only

Scenario (a) assumes that the 2017-18 tax and social security system is kept in place and simply uprated for future years using the default uprating assumptions (‘triple lock’ uprating for the state pension and Consumer Prices Index uprating for other benefits, tax credits, and tax and National Insurance thresholds). None of the further reforms or changes scheduled to take effect after 2017/18 – for example the roll-out of Universal Credit, and the continuation of the working-age social security uprating freeze for two more years – are implemented.

Scenario (b): Westminster reforms

Scenario (b) includes reforms announced by the UK Government which are scheduled to come into effect after 2017/18, but does not include reforms announced by the Scottish Government (the changes to income tax rates in Scotland scheduled for 2018/19, the increase in Carers’ Allowance to the level of Jobseekers’ Allowance, and the Best Start Grant).

Scenario (c): Westminster plus Scotland reforms

Scenario (c) includes the reforms announced by the UK Government plus the reforms announced by the Scottish Government detailed above.

By comparing between the three scenarios it is possible to establish the projected impact of future announced reforms from the UK Government, and the forthcoming changes to the Scottish tax and social security system, on the projected path of child poverty.

2.6 Robustness analysis

Variant scenarios

It is useful and instructive to re-calculate the child poverty measures using variations to many of the assumptions on the macroeconomy, demographic structure, and some of the key assumptions underlying the microsimulation modelling of household net incomes using the TTM. With this in mind, Chapter 3 of the report includes some analysis of variants to the child poverty estimates produced using the following variants to the modelling methodology:

  • Varying the population forecasts for the Scottish economy;
  • Varying the forecasts for employment growth in the Scottish economy;
    Zero real wage growth;
  • Wage growth lower than forecast (e.g. 0.5 percentage points per year lower);
  • Wage growth higher than forecast (e.g. 0.5 percentage points per year lower);
  • No reweighting (essentially stripping out all the reweighting assumptions after 2017-18);
  • Using OBR forecasts across the whole UK economy (including Scotland);
  • Using a longer transition period for the roll-out of Universal Credit;
  • Assuming 100% take-up of means-tested social security payments.

Comparison with IFS results

In October 2017 the Institute for Fiscal Studies published forecasts for child poverty going up to the 2022/23 tax year which included a subset of results for Scotland (Hood and Waters, 2017). We compare our results with the IFS results and attempt to explain any differences between the two sets of results.


Contact