The purpose of this thesis is to propose estimators for the coefficients of scalability associated with Non Parametric Item Response Theory (NIRT), namely: Hij, Hi and H, and their variance estimators, based on the approach of sampling finite populations. To investigate empirically the quality of these estimators are considered the reference populations that are formed by students attending the 9th year of elementary school, in public, in urban areas of the states of Roraima and Rio de Janeiro, who participated Prova Brasil 2007. The responses of students to a set of 10 dichotomized items that measure the economic status of their families were used in the construction of the coefficients of scalability of the Homogeneity Model. Repeated samples were selected from each reference population using two sampling plans: AC1S (cluster sampling single stage) and AC2-SAEB (with selecting schools and classes, stratification and draw units of the first stage with probability proportional to a measure of school size). The point estimate is based on the approach of the Model Overpopulation. Two techniques were treated to estimate the variance: Ultimate Cluster method and Delete - 1 Jackknife. The usual measures: relative bias, mean relative error, confidence intervals and effect of the sampling plan is used to assess the quality of the estimators in terms of the properties of bias and accuracy. The study notes that the estimators have good properties and, in addition, the estimator of the variance corrected by the correction factor for finite population is the most appropriate in terms of accuracy and bias. The complex sampling (AC2-SAEB) impacted the point estimate and variance of the estimators of the coefficients of scalability.