Estimating income, expenditure and time-use within small areas Ben Anderson, Paola De Agostini & Tony Lawson ESRC/BSPS Microsimulation Seminar Series University of Leeds July 2nd 2009

A story in five parts… What’s the problem? And what can we do about it? Income Small area income inequality Expenditure: Small area spending Time-use: A spatialised demand system model Current activities Census projection Agent-Based Microsimulation

What’s the problem?

And what can we do about it?

Income

Small area income inequality

Expenditure:

Small area spending

Time-use:

A spatialised demand system model

Current activities

Census projection

Agent-Based Microsimulation

What are ‘ small areas ’? Base map Parliamentary Constituencies Local/Unitary Authorities Wards Lower Layer Super Output Areas (LSOAs) Output Areas (OAs) [Postcodes]

Base map

Parliamentary Constituencies

Local/Unitary Authorities

Wards

Lower Layer Super Output Areas (LSOAs)

Output Areas (OAs)

[Postcodes]

Why small area data? ‘ Applied’ reasons Public & private resource allocation Public & private service delivery planning ‘ Research’ reasons ‘ small area’ level effects and trends Deprivation indices and correlates Environmental/ecological correlates Local area inequalities and social justice (Runciman) Intervention analysis Sampling frameworks Spatially embedded patterns of consumption … ?

‘ Applied’ reasons

Public & private resource allocation

Public & private service delivery planning

‘ Research’ reasons

‘ small area’ level effects and trends

Deprivation indices and correlates

Environmental/ecological correlates

Local area inequalities and social justice (Runciman)

Intervention analysis

Sampling frameworks

Spatially embedded patterns of consumption

… ?

BUT! Small area data on people are rare Census - down to OA level Administrative - postcode but restricted access Private/commercial - postcode but restricted access And the data captured are restricted Broad but shallow Rarely much on consumption or time-use And of sometimes uncertain quality

Small area data on people are rare

Census - down to OA level

Administrative - postcode but restricted access

Private/commercial - postcode but restricted access

And the data captured are restricted

Broad but shallow

Rarely much on consumption or time-use

And of sometimes uncertain quality

The requirement… Synthetic Census(es) Small area estimates Sociologically interesting phenomena Validated where possible And also What if? (tomorrow) What if? (in 5 years time) What if? (by 2021)

Synthetic Census(es)

Small area estimates

Sociologically interesting phenomena

Validated where possible

And also

What if? (tomorrow)

What if? (in 5 years time)

What if? (by 2021)

How to achieve it? Survey data cases Area 1 in Region X Area 2 in Region X If region = X Census ‘constraint’ tables Weights e.g. Iterative proportional fitting, combinatorial optimization

Survey data cases

Identifying constraints Two criteria: Exist in Census and Survey Good predictors of X in survey We need to know Which ones make the most difference? Which ones co-vary? What is the smallest effective set? Take microdata (survey) Stepwise regression of all possible constraints Which ones are best indicators?

Two criteria:

Exist in Census and Survey

Good predictors of X in survey

We need to know

Which ones make the most difference?

Which ones co-vary?

What is the smallest effective set?

Take microdata (survey)

Stepwise regression of all possible constraints

Which ones are best indicators?

Example: Income deprivation FRS 2004-6 (pooled) p(income < 60% UK median) Net income Equivalised Before housing costs Stepwise regression to select constraints

FRS 2004-6 (pooled)

p(income < 60% UK median)

Net income

Equivalised

Before housing costs

Stepwise regression to select constraints

How to achieve it? Survey data cases Area 1 in Region X Area 2 in Region X If region = X Census ‘constraint’ tables Weights e.g. Iterative proportional fitting, combinatorial optimization

Survey data cases

The result…

A story in five parts… What’s the problem? And what can we do about it? Income Small area income inequality Expenditure: Small area spending Time-use: A spatialised demand system model Current activities Agent-Based Microsimulation Census projection

What’s the problem?

And what can we do about it?

Income

Small area income inequality

Expenditure:

Small area spending

Time-use:

A spatialised demand system model

Current activities

Agent-Based Microsimulation

Census projection

Income Deprivation For each (L)SOA: % households below 60% UK median income % individuals in households below 60% UK median income % children … % working age adults % pensioners Why? Potential use in future IMD (CLG funded) Analyses of deprivation patterns (CRC, NISRA, WAG funded) Data Income: Family Resources Survey (FRS) 2004-6 Space: Census 2001 (LSOAs) and LFS 2004-6 (LA/UAs) Net equivalised income BHC/AHC

For each (L)SOA:

% households below 60% UK median income

% individuals in households below 60% UK median income

% children …

% working age adults

% pensioners

Why?

Potential use in future IMD (CLG funded)

Analyses of deprivation patterns (CRC, NISRA, WAG funded)

Data

Income: Family Resources Survey (FRS) 2004-6

Space: Census 2001 (LSOAs) and LFS 2004-6 (LA/UAs)

Net equivalised income BHC/AHC

Results: Households BHC AHC

BHC

Results: Pensioners BHC AHC

BHC

Validation Spearman correlation results at English LSOA level for household poverty rates and IMD 2007 income domain score Outliers where IMD income domain score low but % HHBMI high: areas with many student HRPs

Spearman correlation results at English LSOA level for household poverty rates and IMD 2007 income domain score

Outliers where IMD income domain score low but % HHBMI high:

areas with many student HRPs

We have an income distribution… Local income inequality So we can calculate a gini coefficient for each zone… Income inequality within small areas

We have an income distribution…

So we can calculate a gini coefficient for each zone…

Income inequality within small areas

Local income inequality - results Mean equivalised household income Equivalised household income gini

Mean equivalised household income

Equivalised household income gini

Local income inequality - gini FRS 2005 East of England, LSOA LSOA gini: Mean = .35 (UK = .36 [UN]) Rural zones Max = Central Cambridge (gini = .408, % HHBMI = 22%) Min = South East Ipswich (gini = .268, %HHBMI = 31.8%)

FRS 2005

East of England, LSOA

LSOA gini:

Mean = .35 (UK = .36 [UN])

Rural zones

Max = Central Cambridge

(gini = .408, % HHBMI = 22%)

Min = South East Ipswich

(gini = .268, %HHBMI = 31.8%)

A story in five parts… What’s the problem? And what can we do about it? Income Small area income inequality Expenditure: Small area spending Time-use: A spatialised demand system model Current activities Agent-Based Microsimulation Census projection

What’s the problem?

And what can we do about it?

Income

Small area income inequality

Expenditure:

Small area spending

Time-use:

A spatialised demand system model

Current activities

Agent-Based Microsimulation

Census projection

Expenditure Modelling Approach Census 2001 Expenditure & Food Surveys Demand system models Spatial microsimulation Why? Resource/consumption ‘poverty’ and inequality modelling ‘ Share of wallet’ & revenue modelling for BT

Approach

Census 2001

Expenditure & Food Surveys

Demand system models

Spatial microsimulation

Why?

Resource/consumption ‘poverty’ and inequality modelling

‘ Share of wallet’ & revenue modelling for BT

Water 2005/6 Simulated household weekly water expenditure (EFS 2005/6) EEDA, LSOA level Water gini

Simulated household weekly water expenditure (EFS 2005/6)

EEDA, LSOA level

Water gini

Water Poverty 2005/6 % Household who spend > 3% of income on water (EFS 2005/6) AHC = After Housing Costs income EEDA, LSOA level Probably needs equivalised income to highlight low income / high demand

% Household who spend > 3% of income on water (EFS 2005/6)

AHC = After Housing Costs income

EEDA, LSOA level

Probably needs equivalised income

to highlight low income / high demand

Telephony: 2005/6 Simulated household weekly telephone bill (landlines) (EFS 2005/6) EEDA, LSOA level Ward level comparison with BT billing data (EEDA, Ward level) Spearman rho = 0.7796, p < 0.001

Simulated household weekly telephone bill (landlines) (EFS 2005/6)

EEDA, LSOA level

Ward level comparison with BT billing data (EEDA, Ward level)

Spearman rho = 0.7796, p < 0.001

Broadband tax…Noddy’s view Simple ROI model £500 cost per household 20% increase in household level spend on internet subscription 50% of additional income to repay investment No spatial cross-subsidisation How long to pay back?

Simple ROI model

£500 cost per household

20% increase in household level spend on internet subscription

50% of additional income to repay investment

No spatial cross-subsidisation

How long to pay back?

A story in five parts… What’s the problem? And what can we do about it? Income Small area income inequality Expenditure: Small area spending Time-use: A spatialised demand system model Current activities Agent-Based Microsimulation Census projection

What’s the problem?

And what can we do about it?

Income

Small area income inequality

Expenditure:

Small area spending

Time-use:

A spatialised demand system model

Current activities

Agent-Based Microsimulation

Census projection

Time in Space: Work Simulated ‘work time’ ONS Time-Use Survey (2001) and Census 2001 East of England, LSOA Validation: (Spearman rho = 0.8404, p < 0.001) Strong correlation with Census 2001 ‘work time’

Simulated ‘work time’

ONS Time-Use Survey (2001) and Census 2001

East of England, LSOA

Validation:

(Spearman rho = 0.8404, p < 0.001)

Strong correlation with Census 2001 ‘work time’

Time in Space - Mass media Simulated total ‘mass media’ time (2001) - TV, radio, DVD, video ONS Time-Use Survey 2001 & Census 2001 East of England, LSOA

Simulated total ‘mass media’ time (2001) - TV, radio, DVD, video

ONS Time-Use Survey 2001 & Census 2001

East of England, LSOA

What if? Estimate base model time(v 1 ) = a(have_internet) + b[socio-demographics] … . time(v n ) = a(have_internet) + b[socio-demographics] demand system ( Deaton & Muellbaeur, Blundell…) Change a parameter e.g: make internet access = 100% Predict new values of time(v 1-n ) for each household Re-run spatial microsimulation

Estimate base model

time(v 1 ) = a(have_internet) + b[socio-demographics]

… .

time(v n ) = a(have_internet) + b[socio-demographics]

demand system ( Deaton & Muellbaeur, Blundell…)

Change a parameter

e.g: make internet access = 100%

Predict new values of time(v 1-n )

for each household

Re-run spatial microsimulation

An example - Internet & TV We model ‘what if 100% internet access’ on TV time in 2001? Internet access has no main effect But has interaction effects with constraints (demographics) Which is why TV time increases for some but not others Survey effects (microsimulation) Spatial effects (spatial microsimulation)

We model ‘what if 100% internet access’ on TV time in 2001?

Internet access has no main effect

But has interaction effects with constraints (demographics)

Which is why TV time increases for some but not others

But it’s not that simple The demand system eats time Only converges slowly, sometimes never ‘ Instant’ change is unlikely Need a forecasting time frame And ‘internet effects’ are marginal Certainly in 2000 data (but maybe not now?) Other changes have more effect on time-use => Work in Progress

The demand system eats time

Only converges slowly, sometimes never

‘ Instant’ change is unlikely

Need a forecasting time frame

And ‘internet effects’ are marginal

Certainly in 2000 data (but maybe not now?)

Other changes have more effect on time-use

=> Work in Progress

A story in five parts… What’s the problem? And what can we do about it? Income Small area income inequality Expenditure: Small area spending Time-use: A spatialised demand system model Current activities Agent-Based Microsimulation Census projection

What’s the problem?

And what can we do about it?

Income

Small area income inequality

Expenditure:

Small area spending

Time-use:

A spatialised demand system model

Current activities

Agent-Based Microsimulation

Census projection

What if? (by 2021) Projected census Fixing geography over time EDs (71,81,91) -> LSOAs (2001) Projecting census counts/distributions Projected survey population Dynamic agent-based microsimulation model Individuals within households Demographic change and consumption Combine results: 2006,2011,2016,2021

Projected census

Fixing geography over time

EDs (71,81,91) -> LSOAs (2001)

Projecting census counts/distributions

Projected survey population

Dynamic agent-based microsimulation model

Individuals within households

Demographic change and consumption

Combine results: 2006,2011,2016,2021

Tony Lawson’s PhD (ESRC/BT) Implementation Netlogo BHPS for transition probabilities Actuarial tables for mortality ONS and EFS for prices Validation BHPS 1991 seed, run to 2005 and compare The dynamic model…

Tony Lawson’s PhD (ESRC/BT)

Implementation

Netlogo

BHPS for transition probabilities

Actuarial tables for mortality

ONS and EFS for prices

Validation

BHPS 1991 seed, run to 2005 and compare

The dynamic model… 3 examples on web Model 1 – dynamic demographic microsimulation model Model 2 – simplified model 1 with food/fuel and housing expenditure model Model 3 – full model Ongoing work Work location, broadband and fuel consumption

3 examples on web

Model 1 – dynamic demographic microsimulation model

Model 2 – simplified model 1 with food/fuel and housing expenditure model

Model 3 – full model

Ongoing work

Work location, broadband and fuel consumption

Thank you benander@essex.ac.uk [email_address] [email_address] http://cresi.essex.ac.uk

benander@essex.ac.uk

[email_address]

[email_address]

http://cresi.essex.ac.uk