Design-based and model-based estimation of poverty measures at the provincial level in the Philippines

This study focused on estimating the Foster-Greer-Thorbecke (FGT) poverty measures (poverty incidence, poverty gap, severity of poverty) at the provincial level in the Philippines. Three estimation procedures (design-based, model-based, EBLUP) were used and estimates were compared based on a accuracy, precision and reliability as measured by bias, standard error, mean square error and coefficient of variation. Design-based estimation were employed using the data from 2000 Family Income and Expenditure Survey (FIES); model-based estimation using the data from 2000 Census of Population and Housing (CPH) and empirical best linear unbiased prediction (EBLUP) which is the weighted average of the first two mentioned estimation procedures. For provincial estimates for poverty incidence, the model-based estimation procedure produced reliable estimates in 71 provinces as compared to the design-based and EBLUP reliable estimates in 28 and 39 provinces only, respectively. The province of Sulu has the highest estimate for poverty incidence for the design-based and EBLUP methods while Lanao del Sur has the highest estimate from model-based estimation method. The model came up with a coefficient of determination (R square) of 63.55 percent and obtained the following predictors: average number of male persons per household; the provincial proportion of the total barangays which have street pattern, the provincial proportion of total barangays which have access to the national highway, and the provincial proportion of the total number of barangays with the surroundings as waste disposal system. On the other hand, the higher number of reliable provincial estimates for poverty gap was obtained from model-based estimation procedure again. It produced reliable estimates of more than 75 percent of the provinces in the Philippines. The provinces which have highest estimate for the 3 methods are Romblon for design-based, Lanao del Sur for model-based, and Maguindanao for EBLUP. The obtained model for provincial poverty gap have the following predictors: provincial average number of members aged between 7 and 14; the provincial proportion of the number of housing units with floor area at most 10 sq. m. and the provincial number of housing units of single type. These predictors explain 62.67 percent of the variations in provincial poverty gap. While for provincial estimates for severity of poverty, 56 provinces have reliable estimates from model-based estimation method and the provinces which have highest estimate for the 3 methods are Romblon for design-based, Sulu for model-based, and Maguindanao for EBLUP. The model obtained for provincial poverty severity index came up with three predictors at an R square of 51.08 percent and the predictors are the provincial proportion of the number of housing units with floor area at most 10 sq. m., the provincial proportion of the number households with members aged less than 7, and the provincial proportion of barangays with the surroundings as waste disposal system. Based on the three methods used in estimating the FGT poverty measures, the model-based estimator for FGT poverty measure at the provincial level was found to be the best. Majority of the estimates produced using model-based estimation method were reliable as measured by the coefficient of variation and were accurate and precise.