Avaliação da eficiência de planos amostrais em big data de registros administrativos

Abstract

New technologies of the XXI century provided a great increase in the production of

data, which caused the phenomenon known as Big Data. Increasingly complex and difficult- to-analyse databases are emerging that require new infrastructure and increased computational

processing power. A solution to this problem would be the use of probabilistic sampling techniques. The purpose of this study was to carry out a simulation study considering different sample designs through the software R and with the assistance of a server. Simple and complex sample designs were compared based on the standard error and coverage level in order to obtain estimates with the best properties from reduced sample sizes in order to allow data analysis also to the user of a desktop. In this study, data from the CadÚnico (Single Register for Social Programs of the Government), which has more than 20 million records and is distinguished as the platform for access to Bolsa Família, were considered. Simple Random Sampling has stood out as the fastest and most accurate method even in the smallest sample sizes.