A Procedure for the Generation of Small Area Estimates of Philippine Poverty Incidence

Year: 2021       Vol.: 70       No.: 1      

Authors: Nelda A. Nacion and Arturo Y. Pacificador

Abstract:

The main purpose of this study is to propose an alternative procedure in the generation of small area estimates of poverty incidence using imputation-like procedures coupled with a calibration of estimates to ensure coherence in the regional estimates. Specifically, this study applied Deterministic Regression Approach, Stochastic Imputation-like procedure similar to Stochastic Regression, and applied the calibration techniques to ensure that the small area estimates conform to the known regional estimates. The difference of this methodology as compared to the ELL is that the error terms used for predictions is based on the empirical distribution of such residuals and thereby a protection against misspecification of the error model. At the same time, the procedure is simpler with available computing resources. In addition, the proposed methodology only utilized data from the census short form which is a 100% percent sample. Thus, eliminating another source of variation as compared to using the census long-form which is collected from a sample of households. This study used the Family Income and Expenditure Survey (FIES) of 2009 and the Census of Population and Housing (CPH form 2) 2010 to come up with reliable estimates of poverty incidence by municipal level. Since the CPH is conducted in the Philippines every 10 years, the CPH 2010 is the latest data that was used. The researcher was able to produce small area estimates of poverty in the Philippines at municipal level by combining survey data with auxiliary data derived from census. The study fitted different models for each region. Considering the results of this paper, the following conclusions were derived: The Stochastic Regression Imputation (SRI) is better to use as compared to Deterministic Regression Imputation (DRI) in attaching income to CPH. The SRI was able to preserve the distribution of 82% of the total number of regions. The DRI was able to preserve only 10.22% of the validation sets. Since the error in fitting the DRI in CPH does not follow a well-known distribution (such as the Normal distribution), the non-parametric way of estimating error was used to generate the errors attached in SRI. The technique is called Kernel Density Estimation (KDE) or the histogram method, which was found to be effective in using the SR. Using the calibration technique achieved municipal estimates that conforms to the regional estimates. The estimates of the poor households in CPH reflects the bottom 30% of the wealth index.

Keywords:

Download this article: