The performance of multilevel growth curve models under an autoregressive moving average process
Multilevel growth curve actually nests to hierarchical linear model (HLM) and conventional HLM analysis doesn’t consider the problem of time series. However, longitudinal data usually involves time series. In the present study, the authors conducted a two level linear growth model when ARMA process is present in the data and then the data are modeled as VC, AR, ARMA, and UN to examine the performance on parameter estimates and fix indexes. It was found that performance of fit index were not inconsistent and less accurate for conditions of small sample sizes and short series lengths. Moreover, type one error rates would be increased for some conditions.
Comments, Questions and Future Study
1) According to the paper, using false models to fit the data would cause biased estimates. What the guidelines that we can inspect the fit between date and candidate model if we have longitudinal data at hand?
2) Performance on growth can be extracted from analysis of longitudinal data. Would it be possible that different developmental stages may cause different relationship between times of the same time series for ability assessment?