Resumo

MEDINA-PERALTA, S et al. Comparision of deviance measures for validation of models without bias, constant or proportional bias. Universidad y ciencia [online]. 2010, vol.26, n.3, pp.255-263. ISSN 0186-2979.

Diverse approaches and techniques have been documented in the literature for model validation in the prediction of systems. Two techniques involve the comparison of observed and predicted values by 1) graphical methods and 2) deviance measures. In the present study, comparisons of three measures of deviation, Modeling Efficiency (MEF), Coefficient of Model Determination (CD), and Coefficient of Error (C), is performed for models without bias (SS), and with constant (SC) and proportional bias (SP). The three measures indicate a good fit between observed and predicted values for a SS model, in agreement to what is observed using graphical methods. For a SC model, CD contradicts the interpretations given by the graphical methods. MEF, CD and C all contradict what it is seen in the graphical methods for a SP model. According to these results, it is preferable a combination of graphical methods and deviance measures for the purpose of model validation.

Palavras-chave : Model; validation; deviance measures; bias.

        · resumo em Espanhol     · texto em Espanhol     · Espanhol ( pdf )