Shin Juh Hyun
School of Nursing, State University of New York, 3435 Main Street, Kimball Tower 823, Buffalo, NY 14214-3079, USA.
Nurs Res. 2009 May-Jun;58(3):211-7. doi: 10.1097/NNR.0b013e318199b5ae.
The aims of this study were to describe how repeated-measures analysis of variance (ANOVA) and the hierarchical linear model (HLM) are used to evaluate intervention effect and to compare these methods, especially in relation to their requirements regarding assumptions, number of repeated measures, completeness of repeated measures, and equal intervals between measurements.
Alzheimer's Disease Assessment Scale (ADAS) data sets (101 residents in 14 nursing homes, five times) were analyzed to explain differences between repeated-measures ANOVA and the HLM.
More detailed information is available when HLM is used. For example, repeated-measures ANOVA showed that there is a statistically significant difference on overall mean ADAS scores between the married and nonmarried groups. The HLM analysis showed more detailed information; the ADAS score of the married group was higher by 6.4 than that of the nonmarried group on the adjusted average ADAS scores (during the whole data collection period). Repeated-measures ANOVA does not provide results on the within-subject changes with days. The HLM provides the specific conclusion that ADAS scores were increased by the one unit of "days" variable (0.017) when days were included.
Hierarchical linear model is a powerful statistical method that can be applied to longitudinal research to evaluate an intervention at multiple levels. The major differences between the repeated-measures ANOVA and the HLM can be summarized as follows: The HLM (a) has less strict assumptions, (b) has more flexible data requirements (dealing with the missing data), and (c) stresses individual change over group differences. More stringent assumptions should be satisfied in repeated-measures ANOVA than in the HLM. The HLM may resolve important statistical issues that have existed in repeated-measures ANOVA. The HLM has more flexible data requirements in that it (a) can be utilized when the measurement data collection points are unequal and (b) may be used when researchers do not have data for all follow-up points, whereas the repeated-measures ANOVA requires a fixed time series design (equal interval, equal number of time points).
本研究旨在描述如何使用重复测量方差分析（ANOVA）和分层线性模型（HLM）来评估干预效果，并比较这些方法，特别是在假设要求、重复测量次数、重复测量的完整性以及测量之间的等间隔方面。
分析阿尔茨海默病评估量表（ADAS）数据集（14家养老院的101名居民，共测量5次），以解释重复测量方差分析和HLM之间的差异。
使用HLM时可获得更详细的信息。例如，重复测量方差分析表明，已婚组和未婚组的总体平均ADAS得分存在统计学上的显著差异。HLM分析显示了更详细的信息；在调整后的平均ADAS得分（在整个数据收集期间）上，已婚组的ADAS得分比未婚组高6.4。重复测量方差分析没有提供随天数变化的个体内部变化结果。HLM提供了具体结论，即当纳入“天数”变量时，ADAS得分随“天数”变量增加一个单位（0.017）。
分层线性模型是一种强大的统计方法，可应用于纵向研究以在多个层面评估干预。重复测量方差分析和HLM之间的主要差异可总结如下：HLM（a）假设较宽松，（b）数据要求更灵活（处理缺失数据），以及（c）强调个体变化而非组间差异。重复测量方差分析比HLM需要满足更严格的假设。HLM可能解决重复测量方差分析中存在的重要统计问题。HLM的数据要求更灵活，因为它（a）可在测量数据收集点不等时使用，（b）当研究人员没有所有随访点的数据时也可使用，而重复测量方差分析需要固定时间序列设计（等间隔、等时间点数）。
Zotero 插件