Improving Social Policy through Spatial Information:Application of Small Area Estimation and Spatial Microsimulation Methods in Geographical Targeting

Noriel Christopher C. Tiglao, Dr. Eng.National College of Public Administration and Governance, University of the Philippines

DDSS Conference, 7 July 2006Heeze, The Netherlands

2

Outline

IntroductionGeographical TargetingComputational Methods

Small area estimationSpatial microsimulation

Framework for Generation of Spatial InformationConcluding Remarks

3

Introduction

Social policy is the study of social welfare, and its relationship to politics and societyPrincipal areas include health administration, social security, education, employment services; also includes social problems, such as crime, disability, unemployment, mental health, and old ageMajor goal of social policy in developing countries is poverty alleviation

4

Role of Spatial Information in Social Policy

Strong influence on economic and technical developmentExtending beyond the scope of central government to include local government and civil societyCritical in processes of consultation and consensus building among policymaking groups

5

Social Policy Administration

Debate between universalism and selectivity (targeting)

Universal policies can reach everyone on the same terms and this has been the argument for public services such as roads or parks, including education and health services.Targeted programs are often considered as being more efficient, that is, it takes less money to realize the benefits.

6

Targeting of Social Programs

Universal programs are too expensive for most developing countries, and even many industrial countries find the rising welfare costs dauntingThe only viable option, therefore, is to use some form of targeting

Requires a careful choice of the targeting criteria, the observable indicators that will determine eligibility, and the programs that be fit the specific conditions of the country or locality

7

Targeting Methods

Targeting by activity – e.g. health care, educationTargeting by indicator - alternatives to income, that are expected to be correlated with poverty, are used to identify the poorTargeting by location, where area of residence becomes the criteria for identifying the target groupTargeting by self-selection or self-targeting, where programs are designed to be attractive only to the poor, e.g. work-for-food programs

8

Geographical Targeting

The optimum solution in welfare programs, from a theoretical point of view, is to identify the target population and design the most effective program for this groupIn most cases, however, it is not possible to identify the target population since this requires information that is not observable and thus difficult to verify

9

Geographical Targeting

In poverty alleviation programs, the target population is the group of households with incomes below a certain minimum level necessary to provide basic needs. Household income is often difficult to observe, however, and efforts to assess its value and thus identify the target group may involve prohibitive costs

10

Geographical Targeting

These costs consist not only of direct administrative expenses for collecting the necessary information on income, but also of indirect costs due to incentives that the program may give individuals either to modify their behavior or to falsify information on their income in order to qualify for the program’s benefits

Ex. Poverty alleviation programs such as income transfers or food subsidies to the poor, for example, may provide incentives to work less, cut earnings, or underreport income in order to qualify

11

Small Area Estimation

Small area estimation has received a lot of attention in recent years due to growing demand for reliable small area estimatorsTraditional area-specific direct estimators do not provide adequate precision because sample sizes in small areas are seldom large enoughSample surveys are used to provide estimates not only for the total population but also for a variety of subpopulations (domains)

12

Small Area Estimation

“Direct” estimators, based only on the domain-specific sample data, are typically used to estimate parameters for large domainsBut sample sizes in small domains, particularly small geographic areas, are rarely large enough to provide direct estimates for specific small domains

13

Types of Small Area Estimation Models

i ~ IID N(0, b2)

~ known positive constants

~ IID N(0, e2)

i ~ IID N(0, b2)

• Unit-level Model (Battese et al., 1988)

i i i iy x z

i i i ijy x e

• Area-level Model (Fay and Herriot, 1979)

iz

ije

14

Building footprint and land use data in GIS

Small Area Estimation of Mean Household Incomes in Manila City

yij =x′ij β + uij

• Nested Error Linear Regression (EBLUP)

yij : mean household income for traffic zone j in city iυi : i-th city effecteij : randoms effect associated with zone j in city i

covariate used is average dwelling unit size

uij = υi + eij

15

City No. of zones

EBLUP Survey regression

Direct estimates

Manila 54 0.039 0.035 0.031

Pasay 11 0.069 0.031 0.029

Makati 18 0.060 0.162 0.242

Mandaluyong 8 0.070 0.185 0.304

San Juan 4 0.089 0.390 0.560

Quezon City 57 0.038 0.034 0.042

Caloocan 17 0.061 0.026 0.036

Valenzuela 9 0.072 0.034 0.028

Malabon 7 0.077 0.048 0.097

Navotas 5 0.082 0.032 0.560

Marikina 8 0.074 0.046 0.050

Pasig 11 0.069 0.054 0.051

Parañaque 15 0.063 0.130 0.122

Muntinlupa 7 0.077 0.215 0.141

Las Piñas 8 0.074 0.083 0.056

Standard error of estimates

16

Spatial Microsimulation

Developed by Guy Orcutt in 1957; ‘A new kind of socio-economic system’Directly concerned with microunits such as persons, households, or firmsModels lifecycle by the use of conditional probabilitiesOne major objective in spatial microsimulation is the estimation of microdata

17

Spatial Microsimulation (cont.)

Spatial microsimulation is increasingly applied in the quantitative analysis of economic and social policy problems (Clarke, 1996)

Tax benefit incidenceIncomeHousingWater consumptionTransportationHealth

18

every hh)

Steps 1st 2nd Last

Head of household (hh)

1. Age, sex, andmarital status (M)of hh

2. Probability of hhof give age, sex,and M being anowner-occupier

3. Random number(computer generated)

4. Tenure assignedto hh on basis of random sampling

5. Next hh (keeprepeating until a tenure type hasbeen allocated to

Age: 27Sex: maleM: married

0.7

0.542

owner-occupied

Age: 32Sex: maleM: married

0.7

0.823

rented

Age: 87Sex: femaleM: divorced

0.54

0.794

rented

Source: Clarke (1996)

Example of spatial microsimulation process

19

HOUSEHOLD

VariablesProvince-IDDistrict-IDBarangay-IDHousehold-IDHousehold sizeAge of hh headSex of hh headMarital status of hh headEducation of hh head(Economic activity of hh head)(Occupation of hh head)(Employment sector of hh head)(Employment status of hh head)Members [Vector]Building typeRoof typeWall typeState of repairYear built(Household income)(Housing status)(Housing value)MethodsGetEconomicActivityofHeadGetOccupationofHeadGetEmploymentSectorofHeadGetEmploymentStatusofHeadGetHouseholdIncomeGetHousingStatusGetHousingValue...

MEMBER

VariablesProvince-IDDistrict-IDBarangay-IDHousehold-IDMember-IDRelation to hh headAgeSexMarital statusEducation(Occupation)(Employment sector)(Income)MethodsGetEconomictActivityGetOccupationGetEmploymentSectorGetIncome...

BaselineCharacteristics

UnobservedCharacteristics

ComputationalObjects/ Models

Object representation of household microdata

Target of spatial microsimulation

20

HOUSEHOLD

VariablesProvince-IDDistrict-IDBarangay-IDHousehold-IDHousehold sizeAge of hh headSex of hh headMarital status of hh headEducation of hh head(Economic activity of hh head)(Occupation of hh head)(Employment sector of hh head)(Employment status of hh head)Members [Vector]Building typeRoof typeWall typeState of repairYear built(Household income)(Housing status)(Housing value)MethodsGetEconomicActivityofHeadGetOccupationofHeadGetEmploymentSectorofHeadGetEmploymentStatusofHeadGetHouseholdIncomeGetHousingStatusGetHousingValue...

Spatial attribute of household microdata

21

Spatial microsimulation of informal households in Manila City

• Manila City (54 traffic zones, 900 barangays, 1.59 million pop. in 1990, 308,874 households)

22

Available data sets

• Detailed household and housing characteristics

• No income/employment variable• Non-response on housing variables• All households in 1990 (1,567,665

households)

1990 Census of Population and Housing (CPH)

Barangay

Description/ CoverageData SetZone System

1996 MMUTIS Land Use GIS1997 Building Footprint Data

1996 Metro Manila Urban Transportation Integration Study (MMUTIS)

1997 Family Income and Expenditure Survey (FIES)

• Urban land use zoning map for entire Metro Manila

• Building footprints for most cities

GIS

• Selected household demographics• Member/ household income• 50,000 samples for Metro Manila

Traffic Zone

• Household demographics, some housing variables

• Detailed household incomes and expenditures

• 4,030 samples for Metro Manila

City

• Detailed household and housing characteristics

• No income/employment variable• Non-response on housing variables• All households in 1990 (1,567,665

households)

1990 Census of Population and Housing (CPH)

Barangay

Description/ CoverageData SetZone System

1996 MMUTIS Land Use GIS1997 Building Footprint Data

1996 Metro Manila Urban Transportation Integration Study (MMUTIS)

1997 Family Income and Expenditure Survey (FIES)

• Urban land use zoning map for entire Metro Manila

• Building footprints for most cities

GIS

• Selected household demographics• Member/ household income• 50,000 samples for Metro Manila

Traffic Zone

• Household demographics, some housing variables

• Detailed household incomes and expenditures

• 4,030 samples for Metro Manila

City

23

Different zoning systems

City Traffic ZoneBarangay

Households

24

Initialize base householdsusing 1990 CPH data

(age, sex, marital status,and education of household

head, household size)

Assign occupation ofhousehold head based

on Monte Carlo sampling

Assign employment sectorof household head basedon Monte Carlo sampling

Compute occupation probabilities from

Occupation Choice Model

Compute employmentprobabilities from

Employment Choice Model

Estimate household incomebased on characteristics

of household head

Compute employment status probabilities and

assign employment statusby Monte Carlo sampling

Compute bias-adjustedhousehold income function

based on employment status

Compute economic activityrate of household head

Estimate permanentIncome of household

Compute housing tenure status probabilities and

assign housing tenure statusby Monte Carlo sampling

Compute bias-adjustedhousing value functionbased on tenure status

Estimate housing tenureand housing value

Spatial microsimulationsystem for estimating household characteristics

25

Simulated mean household incomes

Low

Low

High

Middle

Low

High

26

Ground truths

Smokey Mountain

Port Area, Tondo

Pandacan

Punta

27

Simulated housing tenure

28

Simulated informal employment

29

Inequality measures

30

Estimation of mean household incomes

at the traffic zone level

Reliable estimates of mean household incomes

at traffic zone level

Estimation of household characteristics including

household income

VALIDATION

STATISTICAL SMALL AREA ESTIMATION

SPATIALMICROSIMULATION

Calculate mean household incomes at

traffic zone level

Survey data of household incomes

Estimation of mean household incomes

at the traffic zone level

Reliable estimates of mean household incomes

at traffic zone level

Estimation of household characteristics including

household income

VALIDATION

STATISTICAL SMALL AREA ESTIMATION

SPATIALMICROSIMULATION

Calculate mean household incomes at

traffic zone level

Survey data of household incomes

Framework for Generation of Spatial Information

31

Validation of Microsimulation ResultsEBLUP-DWELLING Benchmark

0

5000

10000

15000

20000

25000

30000

0 5000 10000 15000 20000 25000 30000

Estimated True Values

Sim

ula

ted

Va

lue

s

Framework for Generation of Spatial Information

Validation of spatial microsimulation output

32

Concluding Remarks

Small Area Estimation and Spatial Microsimulation methods can overcome data problems in ‘data-poor’environmentsResulting microdata enables analyst to make full use of existing but disparate data sets and produce reliable and spatially-disaggregate informationFurther research work should be pursued in developing the methods as practical tools for improving social policy

33

Thank you!