NotesFAQContact Us
Search Tips
Peer reviewed Peer reviewed
PDF on ERIC Download full text
ERIC Number: ED593206
Record Type: Non-Journal
Publication Date: 2018-Jul
Pages: 7
Abstractor: As Provided
Is the Doer Effect Robust across Multiple Data Sets?
Koedinger, Kenneth R.; Scheines, Richard; Schaldenbrand, Peter
International Educational Data Mining Society, Paper presented at the International Conference on Educational Data Mining (EDM) (11th, Raleigh, NC, Jul 16-20, 2018)
The "doer effect" is the assertion that the amount of interactive practice activity a student engages in is much more predictive of learning than the amount of passive reading or watching video the same student engages in. Although the evidence for a doer effect is now substantial, the evidence for a causal doer effect is not as well developed. To address this, we mined data for evidence of a causal doer effect across multiple domains. We examined data from two online courses in Psychology, one in Biology, one in Statistics, and two in Information Science, applying causal discovery algorithms in Tetrad to each. Assuming that factors driving a student's choices regarding how to spend their time in an online course are temporally prior to their performance on quizzes and exams, we found evidence of a causal relationship in every domain we studied. We did not find evidence that a unique causal model held in every domain we studied, but when we estimated the size of the causal relationships in the models we found in each domain, we did find evidence in every case that doing has a much stronger quantitative effect on learning than either reading or watching video. This work may be the first EDM effort to explore the generalizability of a causal claim about learning across multiple datasets from a variety of courses and contexts of use. It makes vivid the role of causal data mining algorithms in educational research. The evidence presented furthers the case for doer effect causality, but also recommends a need for richer data with more student background and learning process variables to better isolate causal directionality without assumptions about temporal order and unmeasured confounds. [For the full proceedings, see ED593090.]
International Educational Data Mining Society. e-mail:; Web site:
Publication Type: Speeches/Meeting Papers; Reports - Research
Education Level: Higher Education; Postsecondary Education
Audience: N/A
Language: English
Sponsor: National Science Foundation (NSF)
Authoring Institution: N/A
Identifiers - Location: Georgia; Maryland; Pennsylvania (Pittsburgh)
Grant or Contract Numbers: ACI1443068