Matching income and consumption: HBS - SILC matching in the UK & EU

Richard Tonkin

Measuring Poverty – Concepts, Challenges and Recommendations 17th April 2015, Belgrade

Income based poverty estimates

• Strengths:• Disposable income good proxy for material living standards –

amount of money households have available for spending/saving

• Able to analyse income by component – e.g. wages, property income, benefits, etc.

• Direct policy levers – government able to influence through tax/benefit changes

• Able to QA against other sources (e.g. earnings, administrative data)

• Weaknesses• Households with variable income (inc. Self-employed) may

engage in consumption smoothing• Evidence to suggest data quality may be lower for low income

households

Expenditure based poverty estimates

• Strengths:• Better proxy of living standards: Consumption of goods/services

more closely related to satisfaction of household’s needs• Consumption smoothing: Consumer decisions based on long-

term income expectations – consumption expenditure fluctuates less

• Arguably better data quality: Income under-reported for households with low levels of resources – reporting of expenditure relatively accurate

• E.g. Meyer & Sullivan, 2011; Brewer & O’Dea 2012

• Weaknesses:• Expenditure not the same as consumption• Collection of data often more expensive – therefore smaller

samples/more irregular data• Indirect policy levers

Income & expenditure poverty

• Considering multiple measures together may give us best insights for policy intervention

• Income poor but not expenditure or MD:• May be able to consumption smooth to maintain living standards

due to (expected) temporary low income

• Expenditure poor but not income or MD:• Possible uncertainty over future income levels / lack of assets• “zero hours” contracts

• Relative income poverty main measure used in EU• Aim of this work therefore to:

• Estimate measures of expenditure poverty for range of European countries – Belgium, Germany, Spain, Austria, Finland & UK

• Compare income, expenditure and material based measures of poverty across these countries

Statistical Matching: overview

• No single data source provides joint information on income, expenditure and material deprivation

• Solution - statistical matching of HBS expenditure variables onto EU-SILC

Recipient dataset (EU-SILC)

Donor dataset (HBS)

Matched dataset

X

XX

Y

YZ

Z

Statistical Matching: Reconciliation of data sources

• D’Orazio (2006): Consider variety of factors including:• Population• Definition of units• Reference period• Definition and classification of variables• Treatment of missing data

• Potential HBS/EU-SILC ‘common’ variables cover:• Household size and structure• Household and HRP characteristics• Income bands

Matching variables need to have similar distributions across the two datasets:• this was assessed using the Hellinger Distance with 5% cut-off• variables recoded to harmonise distributions, e.g. activity status

Statistical Matching: Choosing the matching variables

DV

_AC

TS

TA

T

DV

_AC

TS

TA

T2

DV

_AC

TS

TA

T3

DV

_AG

E2

DV

_AG

E3

DV

_CA

R

DV

_CO

NU

NI

DV

_DW

ELL

DV

_HH

SIZ

E

DV

_HH

TY

PE

DV

_LA

BO

UR

DV

_MA

RS

TA

DV

_MA

XE

DU

DV

_MA

XE

DU

2

DV

_OC

C

DV

_PC

DV

_PH

ON

E

DV

_RE

GIO

N

DV

_RO

OM

S

DV

_SE

X

DV

_TE

NU

RE

DV

_TV

S

DV

_UR

BA

N

DV

_WA

SH

INC

_BA

ND

INC

_BA

ND

2

INC

_BA

ND

3

0

5

10

15

20

25

30

35

40

45

50

Belgium Germany Spain Austria Finland UK

He

llin

ge

r d

ista

nc

e (

%)

• Need matching variables to be predictors of both mat dep and expenditure

• Examined using multiple linear and logistic regression models

Final matching variables:

VARIABLE  BELGIUM GERMANY SPAIN AUSTRIA FINLAND UK

AGE ● ● ● ● ● ●

REGION ● ●

LEVEL OF URBANISATION ●

HOUSEHOLD SIZE ● ● ● ● ●

HOUSEHOLD TYPE ● ●

TENURE STATUS ●

MARITAL STATUS ● ● ● ● ●

CONSENSUAL UNION ●

TYPE OF LABOUR CONTRACT ●

HIGHEST EDUCATIONAL ATTAINMENT ● ● ●

ACTIVITY STATUS ● ● ● ● ●

OCCUPATION ●

CAR OWNERSHIP ●

PC OWNERSHIP ●

INCOME BAND ● ● ● ● ● ●

Statistical Matching: Choosing the matching variables

Statistical Matching: methods

• Number of approaches to statistical matching• 3 methods investigated:

• Hotdeck (non-parametric)• Parametric • Mixed methods

• Various diagnostics were used to assess which method performed best

Results of statistical matching – mean expenditure by expenditure decile

• All three methods relatively effective in replicating mean expenditure by expenditure decile, particularly for Germany

• Some underestimation at the upper end of the distribution for Germany & throughout for UK

1 2 3 4 5 6 7 8 9 10 0

10 000

20 000

30 000

40 000

50 000

60 000

70 000

80 000Germany

Expenditure decile

1 2 3 4 5 6 7 8 9 100

10000

20000

30000

40000

50000

60000

70000UK

Expenditure decile

Results of statistical matching – mean expenditure by income band

• All three methods effective at replicating distribution of expenditure by income band

• Expected expenditure ‘tick’ present but for Germany and UK it is underestimated in matched data

Under 5,000

5,000 - 9,999

10,000 -

14,999

15,000 -

19,999

20,000 -

29,999

30,000 -

39,999

40,000 -

49,999

50,000 or more

0

10 000

20 000

30 000

40 000

50 000

60 000Germany

HBS

Mixed

Hotdeck

Parametric

Income band (€)

Under 5,000

5,000 - 9,999

10,000 - 14,999

15,000 - 19,999

20,000 - 29,999

30,000 - 39,999

40,000 - 49,999

50,000 or more

0

10000

20000

30000

40000

50000UK

Income band (€)

Results of statistical matching – mean expenditure by age group

• All three methods relatively effective in replicating mean expenditure by age group 16-25 26-35 36-45 46-55 56-65 66-75 76+

0

5000

10000

15000

20000

25000

30000

35000UK

Age group

16-30 31-40 41-50 51-60 61-70 70+ 0

5 000

10 000

15 000

20 000

25 000

30 000

35 000Germany

HBS Mixed Hotdeck ParametricAge group

Income and expenditure poverty analysis

• No matching method consistently better than others• Mixed methods taken forward - broadly more effective

overall

• Expenditure poverty < 60% equivalised national median

0 5 10 15 20 250

5

10

15

20

25

UKDEBE

ES

ATFI

a) Income poverty

HBS

EU

-SIL

C

0 5 10 15 20 250

5

10

15

20

25

UK

DE BE

ESAT

FI

b) Expenditure poverty

HBS

EU

-SIL

C

Comparison of poverty measures:population breakdown by poverty status

• Degree of overlap high relative to the proportion experiencing at least one form of poverty in Germany

• Relatively low levels of overlap between measures in UK

• Greater overlap of material deprivation with income poverty than expenditure poverty

Germany

Expenditure poor

12.4%

3.8%

Expenditure poor Income poor

Deprived

4.3% 4.3%

4.5%

2.2%3.7%

0.7%

Deprived11.1%

Income poor

14.4%

UKExpenditure poor

Income poor

Deprived

6.7%10.5%

6.5%

4.7%

2.2%

2.5%2.4%

Expenditure poor

19.8%

Deprived13.5%

Income poor

16.0%

Material deprivation:overlap with other measures of poverty

Percentage of materially deprived individuals experiencing other forms of poverty, 2010

• In Finland over half are materially deprived only •Of those that are materially deprived:• 24% (Finland) to 39%

(Germany) are also expenditure poor

• 34% (UK) to 53% (Germany) are also income poor

Belgium Germany Spain Austria Finland UK0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

Materially deprived only

Income poor and materially deprived

Expenditure poor and materially deprived

Income and expenditure poor and materially deprived

Material deprivation by poverty status

• Material deprivation has a stronger relationship with income poverty than with expenditure poverty, particularly in Austria

• The relationship with expenditure poverty is stronger in Belgium, Germany and Spain than Austria, Finland and UK

BELGIUM GERMANY SPAIN AUSTRIA FINLAND UK0

5

10

15

20

25

30

35

40

45

Income-poor Non income-poorExpenditure-poor Non expenditure-poor

% m

ate

ria

lly d

ep

riv

ed

Low work intensity by poverty status

• Defined as living in a household where adults work < 20% of their full capacity

• Similar pattern is seen when considering low work intensity as with deprivation – but generally stronger relationship

BELGIUM GERMANY SPAIN AUSTRIA FINLAND UK0

5

10

15

20

25

30

35

40

Income-poor Non income-poorExpenditure-poor Non expenditure-poor

% li

vin

g in

ho

use

ho

ld w

ith

low

wo

rk in

-te

nsi

ty

Conclusions

• Statistical matching of EU-SILC and HBS encouraging:• Appears matching broadly effective across all 3 countries• No clear ‘winner’, though mixed and hotdeck approaches appear

marginally more effective overall• Statistical matching could be improved if variables measuring

comparable concepts were fully harmonised across ESS surveys

• Supports aim of EU-SILC legal basis TF in considering inclusion of ‘hooks’ to facilitate statistical matching with HBS and HFCS

• Clear evidence of relationship between both income & expenditure & other living standards measures• Limited overlap highlights importance of each measure in

identifying vulnerable groups

Further information

• Eurostat Working Paper: • Webber & Tonkin (2013):

Statistical Matching of EU-SILC and the Household Budget Survey

• Book chapter : • Serafino & Tonkin (forthcoming) Comparing poverty estimates

using income, expenditure & material deprivation, • In Marlier & Atkinson (Eds.)• Update/extension of Eurostat Working Paper also planned

• Contact:• richard.tonkin@ons.gsi.gov.uk• @richt2 on Twitter