N-of-1无残留效应定量数据混合效应模型的模拟研究
发布时间：2018-12-17 04:50
【摘要】：目的通过对不同时间点间存在相关、无残留效应的N-of-1定量数据进行模拟研究,比较不同检验方法的统计性能。方法模拟参数设样本量为10(模型1),研究周期为3(模型2-4),不同时间点间相关系数为0.8(模型5-7),无残留效应,根据固定的效应差值产生多元正态分布数据,建立配对t检验、混合效应模型和差值的混合效应模型。使用效应差值估计值的Ⅰ类错误、检验功效、平均误差(ME),平均绝对误差(AE),均方误差(RMSE)评价各种模型。结果所有模型估计值的均数都非常接近效应差值,所有模型估计值的ME、AE、RMSE都较小。除了模型7,其他模型的Ⅰ类错误概率都约等于0.05。随着效应差值的增大,所有模型的检验功效都随之增大。当两组的效应差值较小时(1.0),模型5的检验功效最大,模型2至模型4的功效较小。当两组的效应差值较大时(≥1.0),所有模型的功效都小于0.010。结论混合效应模型比配对t检验更适合存在相关关系的N-of-1数据。混合效应模型的效果优于差值的混合效应模型,效果最优的模型是CS结构的混合效应模型。
[Abstract]:Aim to compare the statistical performance of different test methods by simulating quantitative N-of-1 data with correlation and no residual effect at different time points. Methods the sample size was 10 (model 1), the research period was 3 (model 2-4), the correlation coefficient between different time points was 0.8 (model 5-7), and there was no residual effect. According to the fixed effect difference, the multivariate normal distribution data are generated, and the paired t test, mixed effect model and mixed effect model of the difference are established. In this paper, we use the class I errors of the estimation of the effect difference to test the effectiveness, the average error (ME), mean absolute error (AE), mean square error (RMSE) to evaluate various models. Results the average values of all models are very close to the effect difference, and the ME,AE,RMSE of all models is smaller. With the exception of model 7, the class I error probability of other models is about 0.05. With the increase of the effect difference, the test effect of all models increases. When the effect difference between the two groups was small (1.0), the efficiency of model 5 was the greatest, and that of model 2 to model 4 was smaller. When the difference between the two groups was greater (鈮，
本文编号：2383681
[Abstract]:Aim to compare the statistical performance of different test methods by simulating quantitative N-of-1 data with correlation and no residual effect at different time points. Methods the sample size was 10 (model 1), the research period was 3 (model 2-4), the correlation coefficient between different time points was 0.8 (model 5-7), and there was no residual effect. According to the fixed effect difference, the multivariate normal distribution data are generated, and the paired t test, mixed effect model and mixed effect model of the difference are established. In this paper, we use the class I errors of the estimation of the effect difference to test the effectiveness, the average error (ME), mean absolute error (AE), mean square error (RMSE) to evaluate various models. Results the average values of all models are very close to the effect difference, and the ME,AE,RMSE of all models is smaller. With the exception of model 7, the class I error probability of other models is about 0.05. With the increase of the effect difference, the test effect of all models increases. When the effect difference between the two groups was small (1.0), the efficiency of model 5 was the greatest, and that of model 2 to model 4 was smaller. When the difference between the two groups was greater (鈮，
本文编号：2383681
本文链接：https://www.wllwen.com/yixuelunwen/jichuyixue/2383681.html
