Preweaning growth curves in Brown Swiss and Pirenaica calves with emphasis on individual variability 1 D. Villalba* , †, I. Casasu ´ s†, A. Sanz†, J. Estany*, and R. Revilla† *Departament de Produccio ´ Animal, Universitat de Lleida, Rovira Roure 177, 25198 Lleida, Spain and †Unidad de Tecnologı ´a en Produccio ´n Animal, Servicio de Investigacio ´n Agroalimentaria, Diputacio ´n General de Arago ´n, Zaragoza, Spain ABSTRACT: A quadratic polynomial model with random regression coefficients was used to describe pre- weaning growth curves of two beef cattle breeds widely used in the Spanish Pyrenees, according to genotype and season of birth effects. In addition, parameters of individual variability that can be used in a stochastic model were obtained. Data recorded indoors from birth to weaning of 217 Brown Swiss calves (3,509 observa- tions) born either in spring or autumn (BS-S, BS-A) and 101 spring-born Pirenaica calves (PI-S, 967 obser- vations) were analyzed. A quadratic model accurately fitted the preweaning weights (R 2 = .99). Use of random regression coefficients improved the weaning weight adjustment; the residual variance of the model with intercept and linear random coefficients (9.61 kg 2 ) was smaller than that of the model without them (130.03 kg 2 ). Brown Swiss-S and PI-S calves had similar birth weight (40.9 ± .96 vs 39.4 ± .73 kg), but BS-S calves Key Words: Beef Cows, Breeds, Calving Season, Genetic Variation, Preweaning Period 2000 American Society of Animal Science. All rights reserved. J. Anim. Sci. 2000. 78:1132–1140 Introduction Management in beef cattle production generally im- plies weaning at a fixed date so that calf age at weaning varies widely. Thus, an age-adjusted weaning weight is calculated, usually by means of a linear estimation of calf growth from birth to weaning (BIF, 1986). Al- though a linear adjustment is used (Gregory et al., 1978; Bolton et al., 1987; Boggess et al., 1991), the nonlinear- ity of calf growth can lead to bias in age-adjusted wean- ing weights (Brinks et al., 1962; Woodward et al., 1989). Frequently, only one or two weight records are available 1 We wish to thank the farm staff working at La Garcipoliera re- search station. Research funded by the projects CE DG VI-8001 CT 90.0002, CE DG VI-1124, INTERREG II and INIA 94-72 and 98-44. Students in receipt of grants from INIA and the Basque government. Received April 5, 1999. Accepted September 14, 1999. 1132 achieved significantly higher weaning weights at 150 d of age (175.2 ± 2.45 vs 158.4 ± 3.17 kg). Preweaning growth patterns were different for each season of birth, but there were no differences in weaning weight at 150 d of age (172.9 ± 2.01 BS-A vs 175.2 ± 2.45 BS- S). Standardization of weaning weights using a linear approximation could lead to biases, especially when comparing animals from the two calving seasons. The estimate of variances of random parameters should be done within breed and season of birth in order to take into account heteroscedasticity. The variances for BS- A, BS-S, and PI-S were 39.9, 57.6, and 32.2 kg 2 for the intercept, respectively, and .0159, .0141, and .0205 kg 2 for the linear coefficient. Covariance between the inter- cept and the linear coefficient (.34 kg 2 ) was only statisti- cally significant in the case of BS-S. The individual variance of weight at 150 d was 424.7 kg 2 and 526.7 kg 2 for BS-S and PI-S, respectively, almost 65% of the observed variance of weaning weight. per animal (birth and weaning weight) (Woodward et al., 1989; Rossi et al., 1992; MacNeil and Snelling, 1996), and growth patterns are estimated using a com- mon covariate for the entire population. The use of mixed models for the analysis of longitudi- nal data should provide a more accurate characteriza- tion of growth patterns because this methodology allows some parameters to be fixed and others to vary with animal, through random effects. Mixed models are a compromise between population models that do not take into account within-animal correlation and ani- mal-specific models that could be overparametrized and are inadequate when data are unbalanced. Also, the separation of variation within individuals from varia- tion between individuals for each of the parameters of the curve (Andersen and Pedersen, 1996; Littell et al., 1996) could be useful for stochastic models of growth. Simulations can be carried out using animals with dif- ferent growth patterns instead of an average animal with the average parameters of the population (Werth et al., 1991; Davis et al., 1994).