It was used 10,238 weekly test day records from 388 first lactations of Saanen goats with the objective of comparing random regression models (RRM). Firstly, it was compared five models, whose exponential terms of Wilmink function assumed the following values: -0.0350; -0.0500; -0.0565; -0.0680 and -0.1000 (W0350, W0500, W0565, W0680 and W1000, respectively) by considering homogeneity of residual variance over the lactation period. The value -0.0500 was kept in the model W0500 whereas models W0350, W0565, W0680 and W1000 used values -0.0350; -0.0565; -0.0680 and -0.1000, respectively, replacing the value -0.0500, proposed by the original model used in dairy cattle. After choosing the best model according to ln L, homogeneity and heterogeneity for residual variance: homogeneity, two classes, three classes, four classes, five classes and six classes along the lactation were evaluated by using AIC, BIC and ln L criteria. According to criterion used, W0350 presents the best fit among the evaluated models. With regard to residual variance, the use of six classes over lactation is indicated by AIC, BIC, ln L and likelihood ratio test. Heritability estimates over lactation, for the best model, ranges from 0.07 (2nd lactation week) to 0.25 (20th lactation week).
milk; milk yield on test day; modelling; residual variance; Wilmink function
