Evaluation of alternative nonlinear mixed effects models of swine growth

Professional Animal Scientist, Sep 2002 by Schinckel, A P, Craig, B A

Abstract

Alternative versions ofa common three-parameter nonlinear growth function were evaluated on two groups of gilts. Gilts were randomly assigned to be reared under all-in, all-out (AIAO; n = 96) or continuous flow (CF) management (n = 96). The addition ora single random effect, in which the mature BW of each pig varies, provided a substantially better fit and smaller parameter standard errors. This model predicts a constant coefficient of variation between pigs. The addition ofa second random effect further improved the likelihood statistics and reduced the residual standard deviation, although the impact was much smaller. The inclusion of the second random effect accounts for different patterns of growth between pigs. Variations in the growth patterns allow greater flexibility to describe the underlying variance/covariance structure of the serial live BW. For all mixed effects models, the mean and approximate variation in age required for pigs to reach a specific BW can be predicted. After 104 d of age, the growth of CF gilts declined more rapidly, and the standard deviation in days required to reach specific target BW (TBW) (110, 120, or

130 kg) increased more rapidly than for the AIAO gilts. These models are also easily adaptable to stochastic modeling. (Key Words: Mixed Effects Model, Nonlinear Growth Functions, Random Effects, Pig Growth.)

Introduction

Swine growth models are used to identify alternative strategies to improve the efficiency of swine production and to estimate daily nutrient requirements for pigs of various ages and genetic groups (16). For an effective application of these models, the growth potential of pigs must be accurately characterized. Several nonlinear growth equations have been used to fit BW as a function of age (3, 4, 10, 12). These equations take the form f(t:Theta) e^sub i,t^ where f is a function (with parameters Theta) describing the mean BW at age t, and e^sub i,t^ represents the residual deviation of the BW of pig i from this mean. These residuals are commonly assumed to be independent normal random variables with mean zero and constant variance. When fitting serial growth data, these assumptions can be troublesome because of competitive interactions and serial correlations (5). Heavier pigs at birth and weaning usually have a competitive advantage and remain heavier throughout their stay in the group (8). This results in increasing variance with age and correlated observations over time.

Implications

Animal growth models have been developed with the goal of optimizing production systems. These models require a parameterization of animal growth and the between and within pig sources of variation. Nonlinear mixed effects models allow a more accurate and precise estimation of animal growth functions than the traditional fixed effects models. Nonlinear mixed effects models provide parameters needed for stochastic modeling, which is needed to evaluate the economic impact of management changes that reduce the amount of variation.

Literature Cited

1. Boland, M. A., K. A. Foster, P. V. Preckel, and A. P. Schinckel. 1996. Analyzing pork carcass evaluation technologies in a swine bioeconomic model. J. Prod. Agric. 9(1):45.

2. Boland, M. A., P. V. Preckel, and A. P. Schinckel. 1993. Optimal slaughter weights under alternative price systems. J. Agric. Appl. Econ. 25:148.

3. Bridges, T. C., U. W. Turner, E. M. Smith, T. S. Stahly, and 0. J. Loewer. 1986. A mathematical procedure for estimating animal growth and body composition. Trans. ASAE 29:1342.

4. Brown, J. E., H. A. Fitzhugh, Jr., and T. C. Cartwright. 1976. A comparison of nonlinear models for describing weight-age relationships in cattle. J. Anim Sci. 42:810.

5. Craig, B. A., and A. P. Schinckel. 2001. Nonlinear mixed effects model for swine growth. Prof. Anim. Sci. 17:256.

6. Efron, B. 1982. The jacknife, the bootstrap and other resampline plans. In CBMS-NSF Conference Series in Applied Mathematics. Society Industrial and Applied Mathematics, Philadelphia, PA.

7. King, R. H. 1999. A Review-Nutritional constraints to pig performance and pig variability. In Manipulating Pig Production VII. P. D. Cranwell (Ed.). p 245. Aust. Pig Sci. Assoc., Werribee, Victoria, Australia.

8. Le Dividich, J. 1999. A Review-Neonatal and weaner pig: Management to reduce variation. In Manipulating Pig Production VII. P. D. Cranwell (Ed.). p 135. Aust. Pig Sci. Assoc., Werribee, Victoria, Australia.

9. Lindsey, J. K. 1996. Parametric Statistical Inference. Clarendon Press, Oxford.

10. Lopez, S., J. France, W. J. J. Gerrits, M. S. Dhanoa, D. J. Humphries, and J. Dijkstra. 2000. A generalized Michaelis-Menten equation for the analysis of growth. J. Anim. Sci. 78:1816.

11. Neter, J., M. H. Kutner, C. J. Nachtsheim, W. Wasserman. 1996. Applied Linear Statistical Models. (4th Ed). Richard D. Irwin, Inc., Chicago, IL.

12. Parks, J. R. 1982. A Theory of Feeding and Growth of Animals. Springer-Verlag, Berlin, Germany.

13. Patrick, G. F., C. A. Hart, and C. Overend. 1993. Marketing concerns in all-in/all-out production. In 1993 Swine Day Reports. p 53. Purdue University, West Lafayette, IN.