1. 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 2006 Heeze, The Netherlands

2. 2 Outline Introduction Geographical Targeting Computational Methods Small area estimation Spatial microsimulation Framework for Generation of Spatial Information Concluding Remarks

3. 3 Introduction Social policy is the study of social welfare, and its relationship to politics and society Principal areas include health administration, social security, education, employment services; also includes social problems, such as crime, disability, unemployment, mental health, and old age Major goal of social policy in developing countries is poverty alleviation

4. 4 Role of Spatial Information in Social Policy Strong influence on economic and technical development Extending beyond the scope of central government to include local government and civil society Critical in processes of consultation and consensus building among policymaking groups

5. 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. 6 Targeting of Social Programs Universal programs are too expensive for most developing countries, and even many industrial countries find the rising welfare costs daunting The 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. 7 Targeting Methods Targeting by activity – e.g. health care, education Targeting by indicator - alternatives to income, that are expected to be correlated with poverty, are used to identify the poor Targeting by location, where area of residence becomes the criteria for identifying the target group Targeting by self-selection or self-targeting, where programs are designed to be attractive only to the poor, e.g. work-for-food programs

8. 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 group In 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. 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. 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. 11 Small Area Estimation Small area estimation has received a lot of attention in recent years due to growing demand for reliable small area estimators Traditional area-specific direct estimators do not provide adequate precision because sample sizes in small areas are seldom large enough Sample surveys are used to provide estimates not only for the total population but also for a variety of subpopulations (domains)

12. 12 Small Area Estimation “Direct” estimators, based only on the domain- specific sample data, are typically used to estimate parameters for large domains But sample sizes in small domains, particularly small geographic areas, are rarely large enough to provide direct estimates for specific small domains

13. 13 Types of Small Area Estimation Models  i ~ IID N(0,  b 2 ) ~ known positive constants ~ IID N(0,  e 2 )  i ~ IID N(0,  b 2 ) Unit-level Model (Battese et al., 1988) Area-level Model (Fay and Herriot, 1979)

14. 14 Building footprint and land use data in GIS Small Area Estimation of Mean Household Incomes in Manila City y ij =x′ ij β + u ij Nested Error Linear Regression (EBLUP) y ij : mean household income for traffic zone j in city i υ i : i-th city effect e ij : randoms effect associated with zone j in city i covariate used is average dwelling unit size u ij = υ i + e ij

15. 15 CityNo. of zones EBLUPSurvey regression Direct estimates Manila540.0390.0350.031 Pasay110.0690.0310.029 Makati180.0600.1620.242 Mandaluyong80.0700.1850.304 San Juan40.0890.3900.560 Quezon City570.0380.0340.042 Caloocan170.0610.0260.036 Valenzuela90.0720.0340.028 Malabon70.0770.0480.097 Navotas50.0820.0320.560 Marikina80.0740.0460.050 Pasig110.0690.0540.051 Parañaque150.0630.1300.122 Muntinlupa70.0770.2150.141 Las Piñas80.0740.0830.056 [1][1] Zoning unit refers to Traffic Analysis Zones (TAZ). Standard error of estimates

16. 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 firms Models lifecycle by the use of conditional probabilities One major objective in spatial microsimulation is the estimation of microdata

17. 17 Spatial Microsimulation (cont.) Spatial microsimulation is increasingly applied in the quantitative analysis of economic and social policy problems (Clarke, 1996) Tax benefit incidence Income Housing Water consumption Transportation Health

18. 18 every hh) Steps1st2ndLast Head of household (hh) 1. Age, sex, and marital status (M) of hh 2. Probability of hh of give age, sex, and M being an owner-occupier 3. Random number (computer generated) 4. Tenure assigned to hh on basis of random sampling 5. Next hh(keep repeating until a tenure type has been allocated to Age: 27 Sex: male M: married 0.7 0.542 owner-occupied Age: 32 Sex: male M: married 0.7 0.823 rented Age: 87 Sex: female M: divorced 0.54 0.794 rented Source: Clarke (1996) Example of spatial microsimulation process

19. 19 HOUSEHOLD Variables Province-ID District-ID Barangay-ID Household-ID Household size Age of hh head Sex of hh head Marital status of hh head Education 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 type Roof type Wall type State of repair Year built (Household income) (Housing status) (Housing value) Methods GetEconomicActivityofHead GetOccupationofHead GetEmploymentSectorofHead GetEmploymentStatusofHead GetHouseholdIncome GetHousingStatus GetHousingValue... MEMBER Variables Province-ID District-ID Barangay-ID Household-ID Member-ID Relation to hh head Age Sex Marital status Education (Occupation) (Employment sector) (Income) Methods GetEconomictActivity GetOccupation GetEmploymentSector GetIncome... Baseline Characteristics Unobserved Characteristics Computational Objects/ Models Object representation of household microdata Target of spatial microsimulation

20. 20 HOUSEHOLD Variables Province-ID District-ID Barangay-ID Household-ID Household size Age of hh head Sex of hh head Marital status of hh head Education 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 type Roof type Wall type State of repair Year built (Household income) (Housing status) (Housing value) Methods GetEconomicActivityofHead GetOccupationofHead GetEmploymentSectorofHead GetEmploymentStatusofHead GetHouseholdIncome GetHousingStatus GetHousingValue... Spatial attribute of household microdata

21. 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. 22 Available data sets

23. 23 Different zoning systems City Traffic Zone Barangay Households

24. 24 Initialize base households using 1990 CPH data (age, sex, marital status, and education of household head, household size) Assign occupation of household head based on Monte Carlo sampling Assign employment sector of household head based on Monte Carlo sampling Compute occupation probabilities from Occupation Choice Model Compute employment probabilities from Employment Choice Model Estimate household income based on characteristics of household head Compute employment status probabilities and assign employment status by Monte Carlo sampling Compute bias-adjusted household income function based on employment status Compute economic activity rate of household head Estimate permanent Income of household Compute housing tenure status probabilities and assign housing tenure status by Monte Carlo sampling Compute bias-adjusted housing value function based on tenure status Estimate housing tenure and housing value Spatial microsimulation system for estimating household characteristics

25. 25 Simulated mean household incomes Low High Middle Low High

26. 26 Ground truths Smokey Mountain Port Area, Tondo Pandacan Punta

27. 27 Simulated housing tenure

28. 28 Simulated informal employment

29. 29 Inequality measures

30. 30 Framework for Generation of Spatial Information

31. 31 Framework for Generation of Spatial Information Validation of spatial microsimulation output

32. 32 Concluding Remarks Small Area Estimation and Spatial Microsimulation methods can overcome data problems in ‘data-poor’environments Resulting microdata enables analyst to make full use of existing but disparate data sets and produce reliable and spatially-disaggregate information Further research work should be pursued in developing the methods as practical tools for improving social policy

33. 33 Thank you!