三水平线性回归模型软件实现及实例应用

摘要

Objective Through the analysis of practical data with hierarchical structure, to give a guide for software im-plementation of three-level linear regression model for repeated measurements and non-repeated measurements in SPSS and SAS software. Methods By taking the data of a study of the effect of sustained nearwork on the ocular development of Juvenile Rhe-sus Monkeys as an example, this article illuminated the pre-condition and software implementation of three-level linear regression model. Results If the variances of random effects items of level 2 and level 3 were statistically significant, it’s necessary to ap-ply three-level linear regression. Conclusion Through judging whether there are hierarchies in the data based on professional knowledge and combining the result with the hypothesis test of high levels’ random items, we can determine if it’s necessary to consider high levels’ random effects. In SPSS, with MIXED procedure it can be easily implemented. For SAS, it can be imple-mented through PROC MIXED and note how to specify the high level units correctly. The results of SPSS and SAS are the same, but the adjustment methods of degree of freedom are a little difference.%目的：通过对具有层次结构的实例数据进行分析，给出重复测量和非重复测量数据三水平线性回归模型在 SPSS 和 SAS 中的实现方法。方法以持续近距离工作对幼年恒河猴眼球发育影响研究的实验数据为例，阐述三水平线性回归模型的应用条件及软件实现方法。结果如果三水平零模型中水平2和水平3的随机项方差有统计学意义，则说明有必要采用三水平线性回归分析方法。结论基于专业知识判断资料是否有层次结构并结合高水平随机项方差的假设检验，判断是否有必要考虑高水平随机效应。 SPSS MIXED 模块可实现三水平线性回归模型，相对容易操作；SAS中可以用 PROC MIXED 实现，需注意正确指定高水平单位。 SPSS 和 SAS 软件计算结果相同，但是对统计量的自由度调整方法略有差异。