Sampling procedures in scientific experiments have long been under scrutiny. The use of power testing and estimating replications has long been utilized as methods for designing experiments. In the same sense, simulation of reanalyzed data has become an approach for understanding on a more applied basis. In the current analysis, two recent trials were simulated 1000 times using a Monte Carlo design to represent performance and quality data of broilers reared to 52 and 56 days of age. Data consisted of male and female Ross 708 final live weights and then subsequently deboned into breast, tender, wing, and leg quarters to calculate yield. Additionally, broiler breasts were scored for woody breast and white striping. Pens contained 18, 21, 24, or 27 birds and all birds were processed at the time of sampling. Data for the current analysis were then analyzed as the maximum difference, as a percent, amongst all simulated pens and the true mean for all pens. Sample size was considered for each pen on the basis of increasing birds sampled per pen starting with 5 birds and concluding with 16 or 20 birds sampled per pen. Upon analysis, differences were observed between the performance characteristics and myopathy characteristics. Amongst live weight and carcass part yields, variance between the sample size of broilers processed was relatively minimized. Selecting any performance parameter expressed a maximum likelihood of 6.44% difference from the true mean when selecting 5 broilers for processing. However, myopathy differences did not follow a similar pattern. When assessing either woody breast or white striping, differences between sampled birds in the selected range could persist as high as 45% and 28%, respectively. This idea proves that myopathy characteristics are much harder to depict when sampling a smaller proportion of a pen rather than the culmination of a more robust sample size. In addition, a sample size of 5 to 20 birds consistently produces a tenfold difference when comparing performance and myopathy measures. Therefore, setting an acceptable threshold of possible error should be considered when sampling.
Maynard et al. (Sun,) studied this question.