Provincial level estimation of the proportion of vitamin A deficient children aged 6 months to 5 years in the Philippines

This study aims to present estimates of the proportion of Vitamin A deficient children aged six months to five years in the Philippines at the provincial level. Two estimation procedures, namely; direct and indirect, were used and compared in this study. Direct estimation technique used data set from the 6th National Nutrition Survey (NNS), specifically, the level of plasma retinol which is used to determine Vitamin A deficiency. Indirect estimation technique employed model-based techniques, specifically regression-synthetic, empirical best linear unbiased prediction (EBLUP), and Poisson regression based techniques. Two data sets,namely: 2000 Census of Population and Housing (CPH) and 2002 Field Health Service Information System (FHSIS) were used as sources of possible auxiliary variables. Direct estimation generated provincial proportions which range from 0.040 (Davao Oriental) to 0.941 (Mt. Province), with mean square errors ranging from 0.0003 to 0.0422, and coefficients of variation where 9.59% of the computed coefficients are less than 10%. Int he regression-synthetic estimation technique, a weighted least squares model with an adjusted-R square of 27.18% was obtained with the following predictors: provincial proportion of deliveries attended by a doctor; households with access to safe water supply; barangays with barangay hall; persons who are high school undergraduates (completed with at least one high school level but did not graduate); and housing units under construction. The predicted proportions using the model range from 0.1359 (Sorsogon) to 0.8077 (Sulu), with mean square errors ranging from 0.0007 to 0.0168, and coefficients of variation where 34% of the computed coefficients are less than 10%. The empirical best linear unbiased prediction (EBLUP) technique is a weighted combination of direct and regression-synthetic estimators. The estimated proportions obtained using this technique range from 0.0467 (Davao Oriental) to 0.9146 (Mt. Province), with mean square errors ranging from 0.0003 to 0.163 and coefficients of variation where 11.27% of the computed coefficients are less than 10%. On the other hand, the Poisson regression was fitted using robust standard errors which resulted to a predicting model with Pseudo-R square of 55.57%. The predictors of the predicting model are: provincial total number of separated or divorced persons; households headed by a person who is a college graduate or an academic degree holder, and untrained 'hilot' in the deliveries of babies. The estimated proportions range from 0.0952 (Iloilo) to 0.988 (Quirino), with variances ranging from 0.0001 to 0.0262 and coefficients of variation of which 77.7% of the coefficients are less than 10%. Among the four estimation techniques, Poisson regression generated the 'best' set of estimates since this technique provided the most number of precise, accurate and reliable estimates.