This study had the objective of comparing weighted difasics logistic models applied to the study of Hereford females growth curves with three different error structures: independent errors (IE), first-order auto-regressive (AR (1)) and second-order auto-regressive (AR (2)) to weight-age data of 55 females of the Hereford breed, raised in the Bagé region, RS, Brazil, evaluated from birth to 675 days old. The weight and %AR options of model procedure, available in the software Statistical Analysis System (SAS), was used to fit data. The comparison among the models was carried out through the biological interpretation basis of the parameters and in the adjustment of quality measures (adjusted determination coefficient, Durbin-Watson test, residual standard desviation, number of iterations), beyond the Akaike information criteria (AIC) and the F test for model comparison. The models fitted to mean data indicated that the difasic logistic with AR(2) structure was the most efficient to describe the herd growth curve. In the individual fit, none of the models was accepted because they didn't produce consistent estimates.
non-linear models; auto-regressive errors; heterogenety of variances
