多层线性建模在CSCL研究中的应用

The Application of Multilevel Linear Modeling in CSCL Research

CSCL研究中常需要处理小组变量和学习者个体变量两种数据,而个体嵌套在小组中,形成两层结构数据。传统的方差分析或线性回归模型仅能针对单层数据,处理多层数据时,易出现标准误差偏移,影响分析的可信度。多层线性建模尽管受CSCL领域样本数的限制,在组层次可能产生偏移量,但能处理稀疏数据,能比较、评估不同层次变异对总变异的贡献度,确定不同层次变量对因变量的影响程度,反映因变量测量随时间变化的发展轨迹,是CSCL领域比较合适的研究方法。

CSCL researches always deal with group variables and individual variables, whereas individuals are nested in groups, forming two level data. Traditional methods such as Analysis of variance （ANOVA） or the linear regression model only adapt to one level data. When they are applied into two level data, the standard variance bias would emerge out, which decreases the validity of analysis. Although Multilevel Linear Modeling （MLM） is likely to make a little bias in group level variance because of the small sample size of CSCL research, it can handle sparse data, evaluate the contribution of various level variables to the total variance, make clear the effect of various level variables on dependent variables, and reflect the growth trajectory of dependent variables along with the time change. Therefore, MLM is an appropriate research method in CSCL.