好多同学手上有纵向数据,想看轨迹,看人群异质性,咨询做法,今天给大家写两个方法,一个叫潜增长模型Latent Class Growth Analyses (LCGA) ,一个叫增长混合模型Growth Mixture Modeling (GMM)。
这个异质性怎么看呢,就是基于人的不同发展的轨迹的出来的:
They can be used to identify latent subgroups, classes or clusters of individuals based on their common growth trajectories over time.
两个模型都可以看作是增长模型growth mode的拓展:
LCGA can roughly be seen as an extension of a fixed effect growth model, whereas GMM can be seen as an extension of a random effect growth model
LCGA和GMM的区别
这两种方法都是将传统增长模型与潜类别分析相结合的模型,既可以刻画增长趋势又可以考虑群体异质性的目的。
二者的区别主要在于类别组内的发展轨迹是否考虑增长曲线内部的个体扰动,潜增长模型可以看作是增长混合模型的特例,就是说增长混合模型不考虑随机效应的时候就可以认为是潜增长模型:
A special case of GMMs is latent class growth analysis (LCGA)[15],[16] which does not allow for departure from the average trajectory within each latent class。Thus, in contrast to mixed effects models where each subject's intercept and slope are drawn from a normal distribution or GCMMs where they are drawn from a mixture of normal distributions,LCGAs allow only for a limited set of discrete options。
实例操练
我现在手上有500个重复测量的数据集,100个观测,每个观测连续测量5次。数据大概长这样:
这个数据还有一个协变量covar,是一个取0和1的二分类数据,对于这么一个纵向数据,我可以先非常直观地把每个个体的轨迹画出