Bodong Chen

Crisscross Landscapes

Notes: Agudo-Peregrina. (2014). Can we predict success from log data in VLEs



Citekey: @Agudo2014

Agudo-Peregrina, Á. F., Iglesias-Pradas, S., Conde-González, M. Á., & Hernández-García, Á. (2014). Can we predict success from log data in VLEs? Classification of interactions for learning analytics and their relation with performance in VLE-supported F2F and online learning. Computers in Human Behavior, 31, 542–550. doi:10.1016/j.chb.2013.05.031



The emergence of distance learning fostered a shift towards the decentralization of the learning process, with higher effort put on course structure and content creation (p. 542)

However, it is very difficult to identify what is the net contribution of each type of interaction to the learning process (p. 542)

Hidden assumptions here? What is learning? Running against learning as interaction metaphor. (p. 542)

Traditional in-class or face-to-face (F2F) learning has historically been a teacher-centric process (p. 542)

Not true. Modality not necessarily linked to ways of teaching. (p. 542)

Since in-class learning interactions are complex and informal by nature, it is difficult to systematically quantify them. (p. 543)

Not clear why f2f interactions in a classroom become ’informal’. (p. 543)

Learning data analysis, most popularly known as learning analytics, is an emerging and promising – the estimated time-to-adoption horizon is between 2 and 3years (Johnson, Adams, & Cummins, 2012) – discipline in educational research. (p. 543)

poor design
Learning analytics is not ‘learning data analysis’. (p. 543)

1.2. A proposal for system-independent classification of interactions in VLEs (p. 544)

1.2.1. Based on the agent (p. 544)

The first classification of interactions in learning processes to reach wide acceptance was proposed by Moore (1989), who identifies three different types of interactions associated to distance learning: (p. 544)

Student–student interactions (p. 544)

Student–teacher interactions: (p. 544)

Student-content interactions (p. 544)

R1. Is it possible to define a system-independent framework of interaction data characterization for learning analytics? R2. If the answer to the previous research question is affirmative, is there any relation between interaction data and academic performance? R3. If so, do results depend on course characteristics, such as instruction mode? (e.g. VLE-supported face-to-face learning versus online learning). (p. 544)

  1. Method (p. 545)

1.2.2. Based on the frequency of use (p. 545)

Thus, Malikowski et al. identified three different levels of use and a total of five categories: (p. 545)

Most used: this level groups interactions related to the transmission of content. The category includes delivery and access to learning resources, general announcements and information about course grades. (p. 545)

Moderately used: this level includes two different types of interactions: creating class discussions and evaluating students. (p. 545)

Rarely used: in this level we may find interactions related to the evaluation of courses and teachers – e.g. course/teaching quality or satisfaction surveys (p. 545)

1.2.3. Based on the participation mode (p. 545)

Rovai and Barnum (2003) differentiate between two types of interaction: active and passive. (p. 545)

2.3. Data analysis (p. 546)

Multiple linear regression was used to find the different relations between student interactions in the VLE and their academic performance. In the context of this research, the independent variables were the number of interactions of each type registered for each user on the VLE – i.e., the output from the Interactions tool – and the dependent variable – academic performance – was represented by the final course grade achieved by each student. (p. 546)

  1. Results (p. 546)

From Table 1, there is a significant relation between the different types of interactions and the student’s academic performance – all of them significant at p < 0,01 except for transmission of content, (p. 546)

context of VLE-supported F2F courses. Nevertheless, the results show that the three different classifications may be useful to predict student achievement in online courses. (p. 547)

Table 3 shows the final models for each of the classifications; criteria for exclusion in each step of the backwards stepwise regression was p < 0.1. According to the results, it follows that final students’ academic performance is determined, depending on the classification used, by: (1) the interactions they have in the VLE with their peers and – mainly – with the teachers; (2) the interactions related to student assessment; and (3) interactions involving active participation. Moreover, the values of VIF (Variance Inflation Factor) suggest that multicollinearity effects may be ruled out in this analysis. (p. 547)

In Tables 2 and 3, we have not been able to statistically establish any predictive model for final student learning outcomes in the (p. 547)

As for the final predictive model for online courses, the results from this study emphasize the importance of having teachers involved in the course (An, Shin, & Lim, 2009) and the promotion of active student participation as a lever to improve the learning process and its results. (p. 548)

However, teacher– student interaction is generally the least scalable type of interaction (p. 548)

Student–student interactions in online learning have been found to be the most important predictors of student success in prior studies (Macfadyen & Dawson, 2012). (p. 548)

Is exploration of interaction data without considering pedagogy ill-informed? (p. 549)