A. Iparragirre, I. Barrio, J. Aramendi, I. Arostegui
In many cases, official statistics are based on the results of surveys based on complex designs. In this data, each sampled unit has assigned a sampling weight, which indicates the number of units that it represents in the population. In the context of dichotomous outcomes, logistic regression models are developed based on complex survey data, for which the discrimination ability is commonly measured by the area under the receiver operating characteristic (ROC) curve (AUC). We believe that if the AUC of the fitted model is estimated without considering the sampling weights, biased estimates could be obtained. Therefore, in this work we propose a new estimator for the AUC, which considers the sampling weights. We evaluate the behaviour of this estimator by means of a simulation study, in which we compare the weighted estimates and the unweighted ones to the true AUC of the models. The results suggest the use of this proposal to estimate the AUC when working with complex survey data.
Keywords: AUC, complex survey data, sampling weights
Scheduled
Posters IV
June 10, 2022 10:10 AM
Faculty Hall
