Simulation of cow-calf production with and without realistic levels of variability.

The Colorado beef cattle production model, a whole-herd, individual-animal, life-cycle simulation model, was used to determine if level of simulated variability affects simulation results. Beyond variability created by deterministic equations describing known biological relationships and direct input, the Colorado beef cattle production model can produce additional variation in a number of traits through its capacity to generate multinormal deviates for each animal. Runs simulating cow-calf production under ample and sparse levels of nutrition were performed with less than realistic and realistic levels of variability for mature weight, milk production, gestation length, maintenance requirements, appetite, and combinations thereof. Under poor nutrition, simulation with less than realistic variability altered means for pregnancy rate, postpartum interval, milk production, weaning weight, and mature weight by up to 14%, 8.6 d, 0.8 kg/d, 9.6 kg, and 19.1 kg, respectively, in addition to changing the output for numerous other variables. The level of simulated variability affected the means of output variables through 2 mechanisms: (1) change in potentials due to differential culling, which can only take place if potentials are allowed to vary; and (2) further interaction with the model's nonlinear, deterministic equations. Our findings indicate that by not simulating realistic levels of variability, models with nonlinear functions may yield misleading results.