Joint ONS/DWP research on developing adjustments using administrative data which improve the measurement of top incomes. Presentation given at 2019 Royal Statistical Society Conference.

1. Measuring household
disposable income:
Developing adjustments using
tax data which improve the
measurement of top incomes
Office for National Statistics & Department for Work & Pensions
Martin Shine, Peter Matejic, Richard Tonkin & Dominic Webber
5 September 2019
Royal Statistical Society
Conference 2019
1

2. Please note that all results and figures in this presentation
are provisional results from experimentally applying a new
top income adjustment and are NOT official government
statistics.
2

3. The need for a top income adjustment
Background
3

4. UK Household income statistics
Two main sources:
1) Households Below Average Income (HBAI) produced
by Department for Work and Pensions based on the
Family Resources Survey.
2) Effects of Taxes and Benefits (ETB) produces by Office
for National Statistics based on the Living Costs and
Food Survey.
4

5. Issues measuring incomes of top
earners:
• Well reported issue that survey data under-reports and
has under-coverage of individuals with top incomes
• This has consequent impacts that the income of highest
income households and overall inequality are both
higher than reported
5

6. Source: Office for National Statistics
0.8
1.0
1.2
1.4
1.6
1.8
2.0
90 90.5 91 91.5 92 92.5 93 93.5 94 94.5 95 95.5 96 96.5 97 97.5 98 98.5 99 99.5
Quantile
2013/14 2014/15 2015/16
Ratio of
gross
income
in tax
data to
survey
data
Those with highest incomes under-report their income in surveys
6

7. Current status of top income
adjustments
• DWP already have a top income adjustment in place for
HBAI
• ONS in process of introducing top income adjustment
• Both make use of HMRC’s Survey of Personal Incomes
(SPI), a stratified sample of tax records, hence top
income adjustment known as “SPI adjustment”
7

8. Analysis on ONS data
Proposed new top
income adjustment
8

9. Developing new adjustment:
Criteria:
1) methodologically sound, based on academic standards
2) transparent and understandable by users
3) adjustment is made on underlying microdata rather
than aggregates
Builds on methods implemented by DWP and later
adapted by Burkhauser et al. (2018)
9

10. How it works – quantile method
1) Rank ETB and SPI data by gross income
2) Decide a threshold, and size of quantile groups above
threshold, e.g. 96% threshold, 0.5% quantile groups
4) Results in dataset which represents original data, but
incomes at top are adjusted to represent the top data
3) Impute the
mean average
for each SPI
quantile group
onto
individuals in
equivalent
survey quantile
groups
Effects of taxes and benefits data
Survey of personal income data
10th decile9th decile
87th 88th 89th 90th 91st81st 82nd 83rd 84th 85th 86th 97th 98th 99th 100th92nd 93rd 94th 95th 96th
9th decile 10th decile
87th 88th 89th 90th 91st81st 82nd 83rd 84th 85th 86th 97th 98th 99th 100th92nd 93rd 94th 95th 96th
Effects of taxes and benefits data (adjusted)
9th decile 10th decile
80th 81st 82nd 83rd 84th 85th 86th 87th 94th 95th 96th 97th 98th 99th88th 89th 90th 91st 92nd 93rd87th 88th 89th 90th 91st81st 82nd 83rd 84th 85th 86th 97th 98th 99th 100th92nd 93rd 94th 95th 96th
10

11. How it works – reweighting method part 1
1) Rank ETB and SPI data by gross income
2) Decide a threshold, and size of quantile groups above
threshold, e.g. 98% threshold, 1% quantile groups
3)Calculate
income
boundaries for
quantile
groups in SPI
4) Create bands
in ETB using
these
boundaries
Effects of taxes and benefits data
Survey of personal income data
10th decile9th decile
86th 87th 88th 89th 90th80th 81st 82nd 83rd 84th 85th 96th 97th 98th 99th91st 92nd 93rd 94th 95th
9th decile 10th decile
80th 81st 82nd 83rd 84th 85th 86th 87th 94th 95th 96th 97th 98th 99th88th 89th 90th 91st 92nd 93rd
1 2
Effects of taxes and benefits data
Survey of personal income data
10th decile9th decile
86th 87th 88th 89th 90th80th 81st 82nd 83rd 84th 85th 96th 97th 98th 99th91st 92nd 93rd 94th 95th
9th decile 10th decile
80th 81st 82nd 83rd 84th 85th 86th 87th 94th 95th 96th 97th 98th 99th88th 89th 90th 91st 92nd 93rd87th 88th 89th 90th 91st81st 82nd 83rd 84th 85th 86th 97th 98th 99th 100th92nd 93rd 94th 95th 96th
87th 88th 89th 90th 91st81st 82nd 83rd 84th 85th 86th 97th 98th 99th 100th92nd 93rd 94th 95th 96th
11

12. 6) Reweight
ETB bands,
so that their
weights are
the same as
the SPI
quantiles
8) Results in dataset with top incomes adjusted and reweighted to match
top of SPI data
How it works – reweighting method part 2
7) Reweight rest of dataset to population totals
Survey of personal income data
9th decile 10th decile
80th 81st 82nd 83rd 84th 85th 86th 87th 94th 95th 96th 97th 98th 99th88th 89th 90th 91st 92nd 93rd87th 88th 89th 90th 91st81st 82nd 83rd 84th 85th 86th 97th 98th 99th 100th92nd 93rd 94th 95th 96th
Effects of taxes and benefits data
Survey of personal income data
10th decile9th decile
86th 87th 88th 89th 90th80th 81st 82nd 83rd 84th 85th 96th 97th 98th 99th91st 92nd 93rd 94th 95th
9th decile 10th decile
80th 81st 82nd 83rd 84th 85th 86th 87th 94th 95th 96th 97th 98th 99th88th 89th 90th 91st 92nd 93rd
1 2
Effects of taxes and benefits data
Survey of personal income data
10th decile9th decile
86th 87th 88th 89th 90th80th 81st 82nd 83rd 84th 85th 96th 97th 98th 99th91st 92nd 93rd 94th 95th
9th decile 10th decile
80th 81st 82nd 83rd 84th 85th 86th 87th 94th 95th 96th 97th 98th 99th88th 89th 90th 91st 92nd 93rd
1 2
Effects of taxes and benefits data
Survey of personal income data
10th decile9th decile
86th 87th 88th 89th 90th80th 81st 82nd 83rd 84th 85th 96th 97th 98th 99th91st 92nd 93rd 94th 95th
9th decile 10th decile
80th 81st 82nd 83rd 84th 85th 86th 87th 94th 95th 96th 97th 98th 99th88th 89th 90th 91st 92nd 93rd
1 2
5) Impute the mean average for each SPI quantile group onto individuals in
equivalent survey bands
12

13. Things to note:
• Both methods impute separate SPI incomes for retired
and non-retired individuals
• For most recent years (2016/17 and 2017/18) projected
SPI data is used as full outturn data is not yet available
• There is currently no SPI data available for 2008/09 –
projections are used instead
13

14. To be determined about adjustment:
• Threshold – How far down should data be adjusted?
95%-99%.
• Width – What size should the quantile groups be?
0.25%,0.5%,1%.
• Method – Quantile or reweighting?
14

15. The impact of including a top income adjustment on ONS’
data
Effect on ETB estimates
15

16. Quantile method, 0.5% width
Source: Office for National Statistics
31
32
33
34
35
36
37
Unadjusted 95% Threshold 96% Threshold
97% Threshold 98% Threshold 99% Threshold
Different thresholds – Gini of ETB disposable income
16

17. Quantile method, 97% threshold
Source: Office for National Statistics
31
32
33
34
35
36
37
Unadjusted 0.25% Quantile Group Size
0.5% Quantile Group Size 1% Quantile Group Size
Different widths – Gini of ETB disposable income
17

18. 97% threshold, 0.5% width
Source: Office for National Statistics
Different methods - Gini of ETB disposable income
31
32
33
34
35
36
37
Quantile Reweighting Unadjusted
18

19. Quantile method, 97% threshold, 0.5% width
Source: Office for National Statistics
65,000
70,000
75,000
80,000
85,000
90,000
95,000
100,000
105,000
110,000
Disposableincome(£pa)
Unadjusted 97% ThresholdAdjusted
Adjusting increases average income of the top decile
19

20. Source: Office for National Statistics
Adjusting allows new analysis – income share of top 1%
0
1
2
3
4
5
6
7
8
9
10
Shareoftotaldisposableincome(%)
20

21. Effect on DWP’s results
21

22. Different methods and SPI data options compared
Quantile method, 97% threshold, 0.5% width
Source: Department for Work and Pensions
31
32
33
34
35
36
37
Original method - projected SPI Original method - actual SPI
Quantile method - projected SPI Quantile method - actual SPI
22

23. Small increases in inequality
• Going from the current method to the quantile method
or using outturn rather than projected SPI both increase
the Gini by similar orders of magnitude
• Trends are less affected
23

24. Source: Department for Work and
Pensions & Office for National Statistics
Comparing HBAI and ETB
30.0
31.0
32.0
33.0
34.0
35.0
36.0
37.0
38.0
Original adjusted HBAI Original unadjusted ETB
ETB reweighting method ETB quantile method
24

25. Future outlook
25

26. Future work:
• Initial article released on the ONS website at the end of
February
• Intend to publish joint ONS/DWP ESCoE paper
• Top income adjustment to be included in ONS’ headline
income statistics for 2018/19 data
• Adjustment hoped to be applied to whole time series
26