Transcript ppt

CLT Conference Heerlen
Does Cognitive Load Theory account for
the beneficial effects of worked
examples in tutored problem solving?
Ron Salden, Ken Koedinger, Vincent Aleven, & Bruce McLaren
(Carnegie Mellon University, Pittsburgh, USA)
Worked examples and tutored problem solving
• Worked examples mostly investigated in
untutored problem solving environments
• Cognitive Tutor is Intelligent Tutoring System
 provides step-by-step guidance during complex
problem solving practice
Worked examples and tutored problem solving
• Cognitive Tutor provides a tougher control
condition of tutored problem solving
– It is this tutored part that in our view reduces
extraneous load
– And, sometimes, increases germane load
Cognitive Tutors
• Longstanding tradition in improving students’ learning
• Grounded in cognitive theory (ACT-R, Anderson & LeBière, 1998)
• Methods for reducing WM load
– Scaffolding or prompting of sub-goals
– Step by step feedback & hints (i.e., guided learning)
• Use cognitive model of student thinking
• Many full-year classroom evaluations show improved
math competence (Koedinger & Aleven, 2007)
Cognitive Tutors
Elements that reduce extraneous
cognitive load.
Standard Cognitive Tutor:
control condition
1a. Corrective feedback
7/24/2016
Pittsburgh Science of Learning Center
6
Standard Cognitive Tutor:
control condition
1b. Implicit positive feedback
7/24/2016
Pittsburgh Science of Learning Center
7
Standard Cognitive Tutor:
control condition
2. Stepwise hints:
last hint level is bottom-out
hint  problem fading into
example
Standard Cognitive Tutor:
control condition
3. Problem sub-goals
are given
Standard Cognitive Tutor:
control condition
4. Student’s self explanation
This feature is not so much about
reducing extraneous load but about
increasing germane load
7/24/2016
Pittsburgh Science of Learning Center
10
Studies
• Shih et al (Geometry)
• McLaren et al (Chemistry)
• Salden et al (Geometry)
Not addressed in this talk
• Anthony et al (Algebra) obtained similar results as the other
three studies
 Does Cognitive Load Theory explain beneficial
effects of examples in tutored problem solving?
Shih et al
• Study: re-analyzing prior study (Aleven & Koedinger,
2002)
– Logged response data on bottom-out hint usage
• One type of “Gaming the system” behavior
– Can be hint abuse due to students skipping
abstract hints to obtain the concrete answer
– Can also be helpful when bottom-out hints act as
worked examples
Shih et al
• Developed a model to distinguish between good
student use of bottom-out hints from bad
student use of bottom-out hints
• Two key elements of model are time spent on:
– Reflecting about prior step (after bottom-out hint)
– Thinking about next step (prior to next action)
• Subtraction method to isolate reflection (selfexplanation) time
– Use other data, when bottom-hints are not requested,
to estimate next step time
Shih et al results
• High correlation of time spent reflecting on
bottom-out hint with learning (pre-to-post gain)
• Spending time on hints is beneficial to learning
for all students
• Difference between students’ hint usage:
– Good usage = spending more time on bottom-out
hint
– Bad usage = spending less time on bottom-out hint
• Thus students who study bottom-out hint as
worked example obtain higher learning gains
McLaren et al
• Conducted three studies comparing
– Tutored Alone vs. Worked Examples + Tutored
• Examples are alternated with isomorphic
problems
Stoichiometry Tutor: control condition
Worked Example condition
• Students watch video of a worked example plus do
prompted self-explanations following the example:
McLaren et al results
• No differences on posttest performance
• BUT, students in Examples condition did learn
more efficiently, using 21% less time to finish
same problem set
Salden et al
• Conducted lab and classroom study
comparing:
– Tutored problem solving
– Fixed example fading
– Adaptive example fading
• Adaptive fading based on students’ selfexplanations of the example steps
– Students who self explain well receive fewer examples than
students who self explain poorly
Standard Cognitive Tutor:
control condition
7/24/2016
Pittsburgh Science of Learning Center
20
Example-enhanced Tutor:
experimental condition
Worked out value
step with calculation
shown by Tutor
Student still has
to self explain
the worked out
value step!
Salden et al results
• Lab study:
– Adaptive fading condition needed fewer examples
than fixed fading condition
– Adaptive fading > both fixed conditions on posttest
and delayed posttest
• Classroom study:
– Adaptive fading condition needed fewer examples on
several theorems than fixed fading condition
– Adaptive fading > problem solving on delayed posttest
Summary of results
• Shih et al: Students can effectively use bottom-out hints as
worked examples and achieve higher learning gains
• McLaren et al: Students working with examples can complete
learning phase needing 21% less time while obtaining the
same learning outcomes
• Salden et al: Students learning from adaptively faded examples
obtained higher immediate and delayed posttest performance
• Fourth study by Anthony et al (using Algebra Tutor): Students
who learned with examples attained better long term
retention
 Also measured mental effort: examples = tutored problem
solving
Does CLT explain these beneficial effects of
worked examples in tutored problem solving?
• Cognitive Tutor is a harder control condition than
untutored environments
– Students can effectively use bottom-out hints as worked examples
• The tutoring seems to reduce possible extraneous
cognitive load
– Anthony study even showed no difference in mental effort between
control and experimental condition
• Stepwise feedback & hints, self-explanation prompts
geared to increase germane cognitive load
Does CLT explain these beneficial effects of
worked examples in tutored problem solving?
Possible explanations
•
•
•
Without the information (guidance) provided by examples,
students waste time tackling new skills during problem
solving
–
McLaren study: examples lead to same learning gains but
needed 21% less time
–
Two Freiburg lab studies: examples lead to same learning gains
needing roughly 17.5% and 25% less time
Motivation
–
Goal of understanding v. performing (Shih et al)
–
Frustration after unsuccessful solution attempt
Where is the cognitive load?
Questions?