Home » reflection » NAPLeS Webinar: Multilevel Analysis

NAPLeS Webinar: Multilevel Analysis

On 1st April, I participated a webinar in Naples, the theme was ‘multilevel analysis’ by Ulrike Cress. The ‘multilevel analysis’ (MLM) is a new knowledge for me, and I have never come into contact with it in the learning process. After reading the suggested article and joining the webinar, I understand what the ‘multilevel analysis’ is, and why it is an appropriate statistical approach in collaborative learning (CL) and computer-supported collaborative learning (CSCL) research. However, the hierarchical linear model (HLM), as the statistical approach for the data multilevel analysis, is complex for me, so I am not able to understand it deeply and even apply it.

Generally, the individuals’ interdependence and their learning process are always taken into account in CL and CSCL. Because it is admitted that individual is able to get benefit when working in the group, and collaboration and interaction can improve learning (Cress, 2008). Therefore, the research about CL and CSCL are necessary to be divided into two levels: individual level and group level. For the individual level, the variables involve individual learners’ prerequisites, knowledge construction, cognition, etc. For the group level, the variables involve the tools, learning environment, teachers’ instructions, etc. If the aim of research is to analyze CL or CSCL based on the group level, the multilevel structure of data needs to be used (Powerpoint by Cress, 2014).

The multilevel structure of data may cause stochastically non-independent problems. And the stochastic non-independence ‘can have three different causes: compositional effects, common fate, and reciprocal influences’ (Cress, 2008, pp.72). Compositional effects can happen when group members are already similar before the study begins, thus it is not possible to assign students to the groups randomly. Stochastic non-independence may also happen when group members share the common fate, which results in the increasing of similarity among group members during the experiment. This situation occurs in most CL and CSCL settings. The third cause of stochastic non-independence is reciprocal influence and it is often existed, especially in the small group. A single member is able to decide the whole interaction procedure within group members sometimes. For example, an active member enables to motivate the entire group members to discuss and communicate energetically; but an inactive member with destructive behaviors and speech enables to destroy the good atmosphere and positive discussion among other group members. The interactional and reciprocal influences among members within groups increase differences between members of various groups (Cress, 2008).

Through comparing those three effects in CL or CSCL research, we find out that it is possible to minimize the compositional effect by randomization of individuals to different groups, but common fate and reciprocal influence cannot be eliminated at all (Cress, 2008). That is because CL and CSCL are based on the ideas of creating non-independency, and the individuals are expected to interact and learn from each other. Thus, the aim of CL and CSCL should consider the effect of non-independency, and the multilevel structure of data will be worked as an intended effect (Powerpoint by Cress, 2014).

In Cress’s presentation, she does not list possible solutions for the analysis of multilevel structure of data (hierarchical data), such as working with fakes, groups as unit of analysis, slopes as outcomes, hierarchical linear analysis, but also their pros and cons and other useful messages. Through the learning in the webinar, it helps me to rethink about the research of CL and CSCL in the following learning process, e.g. what kind of level (individual level or group level) will be examined in the research? What analysis approach is appropriate to this level?, and so forth. If we are able to implement research in an accurate and advanced method, we will find out the key elements which improve learners’ learning process and enhance their learning outcome in CL and CSCL.

References:

Cress, Ulrike (2008). The need for considering multilevel analysis in CSCL research-An appeal for the use of more advanced statistical methods. Computer-Supported Collaborative Learning, 3, 69-84.

Powerpoint by Cress, Ulrike, 2014

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s