Guo I Y, Schneiderman E D, Kowalski C J, Willis S M
Department of Public Health Sciences, Baylor College of Dentistry, Dallas, TX 75246.
Int J Biomed Comput. 1994 Feb;35(1):39-46. doi: 10.1016/0020-7101(94)90047-7.
We consider the problem of growth prediction in the context of the two-stage (or random coefficients) one-sample polynomial growth curve model and provide a PC program, written in GAUSS386i, to perform the associated computations. The problem considered is that of estimating the value of the measurement under consideration for a 'new' individual at the Tth time point given measurements on that individual at T-1 previous points in time and the values of the measurement on N 'similar' individuals at all T time points. The times of measurement t1, t2, ..., tT need not be equally spaced, but we assume that each of the N individuals comprising the normative sample were measured at these times. The method and the program are illustrated using the data set previously considered (Schneiderman and Kowalski, Am J Phys Anthrop, 67 (1985) 323-333) consisting of mandibular ramus height measurements (in mm) for 12 male rhesus monkeys at T = 5 yearly intervals (coded 1, 2, 3, 4, and 5). Results are compared with those obtained under a less restrictive set of assumptions concerning the covariance matrix of the observations than is made in the context of the two-stage model. It is seen that the accuracies of prediction of the two methods, for this and other data sets, are quite close, suggesting that the less restrictive model may be preferred in many situations.
我们在两阶段（或随机系数）单样本多项式增长曲线模型的背景下考虑生长预测问题，并提供一个用GAUSS386i编写的PC程序来执行相关计算。所考虑的问题是，给定一个个体在之前T - 1个时间点的测量值以及N个“相似”个体在所有T个时间点的测量值，估计该个体在第T个时间点“新”个体的测量值。测量时间t1、t2、...、tT不必等距，但我们假设构成标准样本的N个个体中的每一个都在这些时间进行了测量。使用先前考虑的数据集（Schneiderman和Kowalski，《美国物理人类学杂志》，67 (1985) 323 - 333）对该方法和程序进行了说明，该数据集由12只雄性恒河猴在T = 5个年度间隔（编码为1、2、3、4和5）的下颌升支高度测量值（以毫米为单位）组成。将结果与在关于观测值协方差矩阵的一组限制较少的假设下获得的结果进行了比较，这些假设比两阶段模型背景下的假设限制更少。可以看出，对于这个和其他数据集，两种方法的预测准确性相当接近，这表明在许多情况下，限制较少的模型可能更受青睐。
