2. Interactive Online Learning in Public Universities
Randomized Study of Interactive Online Statistics Course
Testing the Benefits of MOOCs
Rebecca Griffiths
Program Director for Online Learning, Ithaka S+R
Feb 7, 2012
3. Motivations for Online Learning Program
Budget crises and pressure to increase graduation rates,
particularly at public universities
Potential of sophisticated, interactive online systems, especially
when used in hybrid mode, to
• Improve student outcomes
• Reduce disparities in outcomes
• Reduce costs without sacrificing learning
Need for solid evidence about effectiveness or cost-saving
potential of such systems
4. Randomized Study of Interactive Online Statistics
Course
• Study conducted by William G. Bowen, Matthew Chingos, Tom
Nygren, Kelly Lack
• Completed spring 2012
• Test of introductory statistics course developed at Carnegie
Mellon University
Primarily text-based with cognitive tutors
• Hybrid vs. traditional face-to-face class sections
5. Summary of Research Design
Registration and recruitment (not necessarily in that order)
• Students register for introductory statistics course
• Students recruited to participate in study
Random assignment of participants
• Traditional section of statistics course
• Hybrid section of statistics course
Baseline measures (beginning of semester)
• Student background survey
• Score on standardized test of statistical reasoning (CAOS test)
Outcome measures (end of semester)
• Completion rate
• Pass rate
• Score on common final exam questions
• Score on CAOS test (second administration of test)
• Student satisfaction survey
6. Fall 2011 Study Sizes
Traditional Hybrid Total
Institution A 45 52 97
Institution B 112 117 229
Institution C 45 47 92
Institution D 7 9 16
Institution E
Department 1 16 15 31
Department 2 24 26 50
Institution F 43 47 90
Total 292 313 605
7. Demographics of Study Participants
Adjusted
Traditional Hybrid Signif.?
Diff.
Proportion of students who are black 14% 14% 0%
Proportion of students who are Hispanic 20% 14% -5%
Proportion of females 54% 61% 7% +
Proportion of students whose parents’ 49% 50% -2% +
income is <$50,000 a year
Proportion of students who have at least 49% 47% 2%
one parent with a college degree
Proportion of full-time students 90% 90% 0%
Mean cumulative college GPA 2.63 2.63 -0.01
Notes: Adjusted differences (average within-institution differences) control for institutional dummy variables.
“Signif.?” indicates whether the result is statistically significant from zero at ** p<0.01, * p<0.05, + p< 0.10.
8. How Participants Compare with Non-Participants
Non-
Adjusted
Participants participants Signif.?
Diff.
Proportion of students who are black 14% 13% 0%
Proportion of students who are Hispanic 17% 10% 3% *
Proportion of females 58% 56% 1%
Proportion of full-time students 90% 86% 5% **
Mean age 21.9 21.6 -0.3
Mean cumulative college GPA 2.63 2.24 0.12 *
Proportion of students who passed course 78% 81% -5% *
Notes: Adjusted differences control for institutional dummy variables. “Signif.?” indicates whether the result is
statistically significant from zero at ** p<0.01, * p<0.05, + p< 0.10.
9. Completion Rates and Pass Rates
Completion and Pass Rates (Percentages)
100
90 88%
84% 81%
80 78%
70
60
50 Traditional
Hybrid
40
30
20
10
0
Completion Rate Pass Rate
(n=605) (n=605)
Results depicted control for institution effects and were not significant at p<0.10.
10. Performance on End-of-Semester Assessments
Post-Course CAOS Scores and Scores on Common Final Exam Questions
(Percentage of Questions Answered Correctly)
100
90
80
70
57% 59%
60
50 47% 48%
Traditional
40 Hybrid
30
20
10
0
Post-Course CAOS Score Common Final Exam Questions
(n=458) (n=431)
Results depicted control for institution effects and were not significant at p<0.10.
11. Results by Subgroup
Subgroup Pass Post- Final
Rate CAOS Exam
Black/Hispanic 0.02 0.00 -0.00
N=188 N=143 N=131
White/Asian 0.05 0.01 0.03
N=406 N=308 N=292
Male 0.04 -0.00 -0.00
N=257 N=194 N=173
Female 0.05 0.01 0.04
N=348 N=264 N=258
First-generation college students 0.01 -0.00 0.02
N=316 N=231 N=258
Students who have at least one parent with a college degree 0.07 0.01 0.03
N=289 N=227 N=216
Pre-CAOS test low 0.02 0.01 -0.03
N=266 N=215 N=196
Pre-CAOS test high -0.02 0.00 0.06+
N=265 N=234 N=222
Notes: Significant at +p<0.10. Results depicted control for institution effects.
12. End-of-Semester Student Survey Responses
Mean Course Rating and Amount Learned at End of Semester
4.0
About the
3.5
vertical axis:
Rating / Amount Learned
3.0 0 = Rated course
much worse than
2.5 typical lecture-
2.10 2.10 based course /
1.85 + 1.89 + Learned much
2.0 Traditional less
Hybrid
1.5 4 = Rated course
much better than
typical lecture-
1.0 based course /
Learned much
0.5 more
0.0
Overall Rating Amount Learned
(n=435) (n=438)
Significant at +p<0.10. Results depicted control for institution effects.
13. Takeaways of Empirical Study
Students in the hybrid sections had roughly similar learning outcomes to
students in traditional-format sections.
Our finding of no significant differences is precisely estimated
We also calculated results separately for each institution, and for
subgroups of students, defined in terms of characteristics like
race/ethnicity, gender, parental education, primary language spoken, and
GPA.
• Results broken down by institution did not reveal any noteworthy patterns.
• We did not find any evidence that the hybrid-format effect varied by any
subgroup characteristics.
Worries that use of online courses may hurt basic student learning
outcomes do not appear to be well-founded.
14. Instructor Experience
25 Average Years of College-Level Teaching Experience
20
15
10
5
0
Hybrid Instructors Face-to-face Instructors
15. Instructor Interaction with Students
Average Classroom time Attendance Rates
Spent Lecturing
100
70
60 80
50
60
40
30 40
20
10 20
0 0
Hybrid Face-to-face Hybrid Face-to-face
Instructors Instructors Instructors Instructors
16. Instructors’ Assessment of the Hybrid Course
Long term impact on time spent:
• If hybrid course used regularly, 5 out of 10 instructors said much less or somewhat
less time would be spent in the long run; only 1 said somewhat more time
Evaluation of online course:
• Mixed reviews; most found it acceptable, but all mentioned a few areas of
mismatch.
• Many instructors believed their students had negative views of the online course.
CMU course is a good prototype, but there is room for
improvement
17. Next steps
• Encourage the development of more high quality, interactive,
customizable online learning systems and content
• More evidence
• Further exploration of the potential for cost savings
18. Along came 2012 Class of MOOCs – Massively
Open Online Courses
• Massive – some have attracted over 100,000 registrations
• Open – freely accessible to anyone with internet connection
• Online Courses –
• Led by an instructor at an institution
• Have a beginning and an end
• Have lectures, in-video quizzes, assignments, quizzes and tests
• Heavy reliance on peer collaboration, even for grading
• Offer certificates, exploring options for accreditation (e.g. testing
centers, ACE)
Not quite what institutions need, but is there a way to bridge
the gap?
19. Testing the Benefits of MOOCs
Partnership with the University System of Maryland to test the
hypothesis that MOOCs can be used to improve student
outcomes and/or reduce costs within a public university system.
Research plan::
» 5-7 controlled side-by-side tests
» 5-10 case studies
Why not randomized? Things are moving too fast!
20. What We Aim to Learn
• Can MOOCs be used to improve student outcomes?
• What implementation challenges arise, and how can these
be overcome?
• What models of adoption are there? What are the potential
benefits and challenges of each?
• What can we learn about cost savings?
21. Other Things We Might Learn
• Will MOOCs be adopted like multimedia textbooks?
• How can one tell a good MOOC from a bad MOOC?
• What are the key differences between MOOC platforms?
• Which features of MOOCs work well in a campus environment?
Which do not?
• What conditions are conducive to success?
22. How Do / Will MOOCs “Make Information Pay”?
• Student – instructor – developer feedback loops enable constant
improvement of courses
• More data and better analytics needed to model student profiles,
behavior, experience, knowledge, etc.
• Ownership of these data will be a key issue
24. Participating Institutions
City University of New York
• Baruch College
• Borough of Manhattan Community College*
• City College
State University of New York
• University of Albany
• SUNY-Institute of Technology
• Nassau Community College*
University of Maryland
• Baltimore County
• Towson University
Montgomery County College, Maryland*
* Data from these institutions were analyzed separately and are not included in this presentation.
Cautionary note: We cannot assume that the findings presented today for 4-year public institutions
necessarily hold for community colleges, nor can we compare outcomes at community colleges with
outcomes at 4-year institutions.