Estimaciones de pobreza para áreas pequeñas en Costa Rica: una aplicación de los estimadores de contracción de James–Stein

Abstract

This paper evaluates two methods in order to estimate the poverty at small areas, both were used to correct the direct estimators from the periodical data sources with the help of the only available parametrical data source (Census). The first method is so-called regression-method, of the type Pit = β0 + β1Pio , where Pit are the poverty values at the periodic source data and Pi0 are the corresponding parametric values (Census); and the second one is called shrinking-method, based on an approach of the James-Stein shrinking estimators:

with variance   consisting to “shrink” the periodical data source estimators ˜θi towards the parametrical values θi, when the variance of the estimator at the small area ψi is relatively big. Otherwise, to shrink towards the periodical data source estimators ˜θi when the parametrical variance ˆσ2 is relatively big. Suggested shrinking-method estimators had smaller average quadratic errors, than those from regression-method, and produces smaller confidence interval than both regressionmethod and direct estimations.