Validation of models with constant bias: an applied approach

Revista MVZ Córdoba

Arjona-Suarez, Enrique⁴  Medina-Peralta, Salvador¹  Mendoza-Martinez, German²  Vargas-Villami, Luis¹  Jorge, Navarro A¹  Avendaño R, Leonel³  Colorado M, Luis¹ 

[1] Universidad Autónoma de Yucatán, Mérida, México;Universidad Autónoma Metropolitana, México, México;Universidad Autónoma de Baja California, Baja California, México;Colegio de Postgraduados, Montecillo, México

关键词: Statistics;    error;    confidence interval;    models (Source: MeSH);   

DOI  :  

学科分类：兽医学

来源: Universidad de Cordoba * Facultad de Medicina Veterinaria y Zootecnia

【 摘 要 】

Objective. This paper presents extensions to the statistical validation method based on the procedure of Freese when a model shows constant bias (CB) in its predictions and illustrate the method with data from a new mechanistic model that predict weight gain in cattle. Materials and methods. The extensions were the hypothesis tests and maximum anticipated error for the alternative approach, and the confidence interval for a quantile of the distribution of errors. Results. The model evaluated showed CB, once the CB is removed and with a confidence level of 95%, the magnitude of the error does not exceed 0.575 kg. Therefore, the validated model can be used to predict the daily weight gain of cattle, although it will require an adjustment in its structure based on the presence of CB to increase the accuracy of its forecasts. Conclusions. The confidence interval for the 1-α quantile of the distribution of errors after correcting the constant bias, allows determining the top limit for the magnitude of the error of prediction and use it to evaluate the evolution of the model in the forecasting of the system. The confidence interval approach to validate a model is more informative than the hypothesis tests for the same purpose.

【 授权许可】

Unknown