Determination of suitable sample size and number of simulations (resampling) for predicting dry matter intake of feedlot cattle

Abstract

The main purpose of this study is to determine the appropriate combinations of simulation numbers and resampling numbers that can be used in practice in animal husbandry. Close-out information, submitted by Iowa cattle producers to the Iowa State University Feedlot Performance and Cost Monitoring Program, was used to develop a suitable sample size and the number of simulations for predicting dry matter intake in feedlot cattle. Close-out information consisting of 3452 pens of cattle included information on start and end dates, cattle per pen, sex, housing type, days on feed, initial and sale weight, feed conversion (FC), the proportion of concentrate, average daily gain (ADG), percent death loss, feed cost and total cost per 45.35 kg gain, breakeven sale price, non-feed variable cost, non-feed fixed cost, and corn price. Dry matter intake (DMI) was not provided but was calculated as DMI = ADG x FC.  The average number of cattle per pen was 147 thus totaling the number of cattle fed to 507444. Since the number of cattle fed was enormous, it was assumed that this could represent the population. In order to determine a suitable sample size and the number of simulations, different sample sizes (3, 5, 10, 15, 20, 30, 50, 70, and 100) and different numbers of simulations (1000, 3000, 5000, and 10000) combinations were run. Samples were chosen as sample with replacement among 3452 pens.  This sampling procedure was carried out 1000, 3000, 5000, and 10 000 times.  Results showed that sample size of 15 pens and above gave better results than smaller sample sizes. As the sample size increased above 15, results became more dependable but considering the time and money constraints it is advised to have a sample size of 15. The optimum number of simulations was found as between 3000 and 5000 by looking at the distribution shape, standard error of the mean, and the similarity of DMI to actual DMI.  When the simulation was run for 10000, the distribution shape, standard error of the mean, and the similarity of DMI to actual DMI were nearly perfect. However, considering the time constraint and the advanced computers needed to run a large number of simulations it is advised to have simulations between 3000 and 5000. In conclusion, using the results in practice in animal husbandry can provide time, labor, and economic contributions to the breeders.