Análise de dados amostrais das Pesquisas AIBF I e II: efeito do plano amostral na estimação de médias

Abstract

In cross-sectional and longitudinal surveys, to ignore the sampling design, especially when it is complex, is still common practice in quantitative analysis, either by the absence of statistical software that allows for analysis based on the sampling design or by the unawareness from the data analysts of the importance of incorporating the sampling design to obtain estimates of the parameters of interest, like means; totals and proportions. A complex sampling design is defined by characteristics such as stratification; clustering; sample weights and unequal probabilities of selection, among other attributes that make it diverge from the simple random sampling presuppositions. When the complex sampling design is ignored, the obtained estimates may be biased. In large-scale longitudinal surveys, like the Pesquisa de Avaliação de Impacto do Programa Bolsa Família (AIBF), used as data source for this monograph, the selection of a simple random sampling when drawing the population elements is not economically feasible, which is the reason for the adoption of stratification and clustering techniques by the survey statisticians who are in charge of the sampling design. Thus, this monograph‟s aim is to compare alternative statistical techniques (ignoring or incorporating the sampling design) for the exploratory analysis of the impact of the Programa Bolsa Família on a parameter strictly associated with a condition to receive the benefit, that is the mean of the number of prenatal consultations, estimated for each group of households of the survey, and reinforce the importance of considering the sampling design. Finally, a measure of how large is the „cost‟ of ignoring the sample design characteristics is presented: the misspecification effect.