Huggard David J
Department of Zoology, University of British Columbia, 6270 University Blvd, V6T-IZ4, Vancouver, B.C., Canada.
Oecologia. 1994 Dec;100(4):470-474. doi: 10.1007/BF00317869.
I assessed the effects of the sampling error of input variables on the energy-maximizing diets of 14 grassland herbivores that Belovsky (1986) predicted using a linear programming model of optimal foraging. Monte Carlo simulations showed that the error reported in the estimates of the variables generated wide confidence intervals on predicted diets of the species. Given this imprecision in the predictions, the predicted diets that Belovsky reported were unexpectedly similar to the observed diets. The high correlation between predicted and observed diets reported by Belovsky was only attained in 0.01% of the simulation runs. Simulations assuming a variety of relationships between the sampling error in the different variables did not alter this conclusion. Incorporating the sampling error in even a single variable causes wide variability in the predicted diets. This analysis suggests that the high levels of accuracy reported for the linear programming approach will be difficult to repeat.
我评估了输入变量的抽样误差对14种草食性动物能量最大化饮食的影响，这些动物的饮食是贝洛夫斯基（1986年）使用最优觅食的线性规划模型预测的。蒙特卡洛模拟表明，变量估计中报告的误差在物种预测饮食上产生了广泛的置信区间。鉴于预测存在这种不精确性，贝洛夫斯基报告的预测饮食与观察到的饮食出人意料地相似。贝洛夫斯基报告的预测饮食与观察到的饮食之间的高相关性仅在0.01%的模拟运行中出现。假设不同变量的抽样误差之间存在各种关系的模拟并没有改变这一结论。即使在单个变量中纳入抽样误差也会导致预测饮食的广泛变异性。该分析表明，线性规划方法所报告的高精度水平将难以重现。
