Habtemariam T, Cho Y
J Am Vet Med Assoc. 1983 Dec 15;183(12):1440-6.
In an attempt to develop a predictive model for poultry inspection at processing plants, a systems research design was developed to examine poultry population health dynamics from hatching to processing. Five cooperating Alabama poultry firms with 5 to 7 growers from each firm were identified to form the study unit. The farms were stratified to represent good, average, and poor producers. Via epidemiologic causal diagrams, variables with potential influence on poultry condemnation due to diseases were identified for hatchery, broiler, and processing subsystems. Field and/or laboratory data were generated for each study unit and for each variable. Using stepwise multiple regression and discriminant analysis, a predictor model with a multiple correlation coefficient (R) of 0.91 and multiple coefficient of determination (R2) of 0.82 was developed. A discriminant analysis model for classifying a flock into high or low condemnation group, using 0.1% to 2.0% as the demarcation line, was also developed. This latter model had overall correct classification probabilities ranging from 0.88 to 1.0. The 2 decision-making models were then computerized, using Beginner's All-purpose Symbolic Instruction Code (BASIC). On the day of processing, the grower provides the inspector with selected information, which is entered on a computer. Alternatives for scaled-down inspection or others are then systematically evaluated by the computer, and decision-making information is provided to the user.
为了开发一种用于加工厂家禽检验的预测模型，开展了一项系统研究设计，以研究从孵化到加工过程中的家禽群体健康动态。确定了阿拉巴马州五家合作的家禽公司，每家公司有5至7名养殖者，组成研究单元。这些农场被分层以代表优质、中等和劣质生产者。通过流行病学因果图，确定了孵化场、肉鸡和加工子系统中对因疾病导致的家禽判废有潜在影响的变量。为每个研究单元和每个变量生成了现场和/或实验室数据。使用逐步多元回归和判别分析，开发了一个预测模型，其多重相关系数（R）为0.91，多重决定系数（R2）为0.82。还开发了一个判别分析模型，用于将鸡群分为高判废组或低判废组，以0.1%至2.0%作为分界线。后一个模型的总体正确分类概率在0.88至1.0之间。然后使用初学者通用符号指令代码（BASIC）将这两个决策模型计算机化。在加工当天，养殖者向检验员提供选定的信息，这些信息被输入计算机。然后计算机系统地评估按比例缩减检验或其他检验的替代方案，并向用户提供决策信息。
