Multilevel analysis in CSCL Research

CSCL researchers are often interested in the processes that unfold between learners in online learning environments and the outcomes that stem from these interactions. However, studying collaborative learning processes is not an easy task. Researchers have to make quite a few methodological decisions such as how to study the collaborative process itself (e.g., develop a coding scheme or a questionnaire), on the appropriate unit of analysis (e.g., the individual or the group), and which statistical technique to use (e.g., descriptive statistics, analysis of variance, correlation analysis). Recently, several researchers have turned to multilevel analysis (MLA) to answer their research questions (e.g., Cress, 2008; De Wever, Van Keer, Schellens, & Valcke, 2007; Dewiyanti, Brand-Gruwel, Jochems, & Broers, 2007; Schellens, Van Keer, & Valcke, 2005; Strijbos, Martens, Jochems, & Broers, 2004; Stylianou-Georgiou, Papanastasiou, & Puntambekar, chapter #). However, CSCL studies that apply MLA analysis still remain relatively scarce. Instead, many CSCL researchers continue to use ‘traditional’ statistical techniques (e.g., analysis of variance, regression analysis), although these techniques may not be appropriate for what is being studied. An important aim of this chapter is therefore to explain why MLA is often necessary to correctly answer the questions CSCL researchers address. Furthermore, we wish to highlight the consequences of failing to use MLA when this is called for, using data from our own studies.