VPAL Research, HarvardX work to Enable Adaptive Learning in New Online Course

February 2, 2017

HarvardX, a Harvard University strategic initiative, is running a massive open online course (MOOC) featuring adaptive learning and assessment algorithms, which tailor course material in response to student performance. 

Adaptive learning functionality–never before offered in a HarvardX course and featured in a few courses across the edX online learning platform–has been deployed in Super-Earths and Life, instructed by Harvard’s Phillips Professor of Astronomy Dimitar Sasselov. The effort aims to gain a preliminary assessment of the technological feasibility and implications of adaptive functionality to online course design.

“Adaptive learning programs are very good at speeding up information acquisition and lengthening retention, as well as individualizing learning to help learners see where they have difficulty,” said Peter K. Bol, Harvard’s Vice Provost for Advances in Learning (VPAL) and Charles H. Carswell Professor of East Asian Languages and Civilizations. 

“There is strong evidence, broadly, in the field that adaptive learning is one particularly exciting way to leverage the possibilities of a digital platform,” said Robert A. Lue, HarvardX Faculty Director and Professor of the Practice of Molecular and Cellular Biology. “Our hope is that adaptive functionality will over time become a feature in a significant percentage of HarvardX classes, and provide powerful new assignment types in the Harvard classroom.”

“The idea is to tailor the experience so that students are more likely to get what they need,” said Dustin Tingley, VPAL Research Faculty Director and Harvard Professor of Government. “The broader mission is to make sure that students are really benefitting from the online learning experience.”

VPAL researchers and HarvardX technologists collaborated with educational technology company TutorGen to implement an adaptive learning algorithm within a subset of Super-Earths and Life content. VPAL researchers leveraged the existing algorithm to build-out, in-house, an adaptive learning technology specifically designed for use within a HarvardX course in a variety of different ways.

Super-Earths and Life users were evenly divided into experimental and control groups.  On course homework pages, control group learners received a pre-determined, non-adaptive set of problems of varying difficulty (easy, medium, and hard). The experimental group received the same assessments, but their problems were served sequentially, one-by-one, in an order based upon prior performance. 

“The adaptive approach works well for Super-Earths because the class covers a very interesting topic, but also because it involves physics and biology–high level science,” said Sasselov. “Students on an advanced level are able to push further, and the adaptive course is the ultimate way to accomplish that.”

In an effort to examine the effect of adaptive technology on learner performance, engagement and completion in online courses, VPAL Research analyzed Super-Earths and Life course data between Oct.19, 2016 (course launch) and Jan. 4, 2017. Differences in learner efficiency and performance, on average, were noted in the adaptive experimental group in a number of key areas:

  • The experimental group outperformed learners in the control group by achieving a larger knowledge gain (19% higher) after completing adaptive assessments, compared with non-adaptive settings in the control gro
  • Experimental group students tended to move faster through the course materials, showing a lower net-time on-task (4.37 hours) than the control group (4.80 hours)
  • Experimental group students showed more persistence in course assessments, making more attempts on advanced problems (2.99 attempts/problem) than the control group (2.41 attempts/problem)
  • In addition to making more attempts per-problem, the experimental group attempted fewer problems overall (3.99) than the control group (5.42)

Despite the experimental group progressing more quickly through the course than the control group, no statistically significant differences in rates of course completion or certification were found between the groups. 

VPAL Research and HarvardX look forward to expanding adaptive learning functionality in future HarvardX courses and further exploring how the technology can improve online learning outcomes. 

More detailed analysis of adaptive learning and assessment algorithms deployed in Super-Earths is available on the blog section of the VPAL Research website.