Cognitive Load Theory: Failure?

The title of this post is meant to be a joke (not a troll).  The inventor of cognitive load theory (Sweller) and others labeled problem-based learning and other constructivist and inquiry-based instructional techniques a ‘failure’ in an oft-discussed 2006 paper I posted about earlier (no joke).

Recently the journal Instructional Science published some reflections by Ton de Jong on cognitive load theory itself, identifying some conceptual and methodological problems.  Roxana Moreno, who has an edited book on Cognitive Load Theory coming out next year, also published a nice summary of de Jong’s paper in the same journal.  So, below is a summary of the summary, plus a few extra things.

What is cognitive load theory?

Cognitive load theory is the idea, first published by Sweller in 1988, that instructional design should focus on not overloading a learner’s mental effort when designing instruction.  “Learning is hampered when working memory capacity is exceeded in a learning task” (de Jong, 2009).

The first version of cognitive load theory (1988-1998) had 2 elements, intrinsic and extraneous cognitive load.  The latter is “bad” and needs to be reduced because it hurts performance, and the former is out of our control – the intrinsic difficulty of the subject matter being learned:

1. Intrinsic cognitive load – This is driven by element interactivity of the material to be learned.  Memorizing something like independent commands has low/no element interactivity, whereas learning how to edit a photo in photoshop is something with high element interactivity (Paas, Renkl, Sweller, 2003).  We as instructional designers can’t control the intrinsic difficulty of the subject being learned.

2. Extraneous or ineffective cognitive load – This is the unnecessary information presented during instruction, or making learners do extra work that only delays or detracts from their learning.  Such as having to search for information rather than telling them where to find it.  Extraneous cognitive load is sometimes just a necessary constraint (like having to turn a page in a book rather than watching a video), but where it really hurts learning is when the element interactivity and instrinsic cognitive load are already high and pushing at one’s working memory limitations.  “As a consequence, instructional designs intended to reduce cognitive load are primarily effective when element interactivity is high” (Paas, Renkl, Sweller, 2003).

In 1998, Sweller published a paper that introduced a 3rd element, germane cognitive load.  Because of course if you take the first two types to the extreme, it suggests that we as instructional designers should do nothing but spoon feed content to learners, and make learners exert as little mental effort as possible.  Of course we know that that higher mental effort is sometimes necessary for learning, and not a bad thing at all – more like ‘hard fun’ and ‘serious play’.  An example is learning how to do math calculations by hand before using a calculator or computer to do them.  There are plenty of examples where higher cognitive load actually leads to much better learning, which would contradict the original cognitive load theory.

3. Germane cognitive load – This is the on-task mental ‘load’ or activity during learning.  This can and should be influenced by the instructional designer.  If one defines learning as schema acquisition and building, the ‘germane’ cognitive load is that which contributes to such schema acquisition, and the ‘extraneous’ cognitive load is that which does not.

Conceptual Problems with Cognitive Load Theory

1. Post-hoc explanation.  As soon as I first read about germane cognitive load (good) in 1998 vs. extraneous cognitive load (bad), cognitive load theory became unfalsifiable in my opinion.  You can justify any experimental result after the fact by labeling stuff that hurts performance as extraneous and the stuff that didn’t as germane.  Numerous contradictions of cognitive load theory’s predictions have been found, but with germane cognitive load, they can still be explained away.  de Jong does not use this term (unfalsifiable) but instead states that germane cognitive load is a post-hoc explanation with no theoretical basis: “there seems to be no grounds for asserting that processes that lead to (correct) schema acquisition will impose a higher cognitive load than learning processes that do not lead to (correct) schemas” (2009).

2. Can’t distinguish between germane and extraneous cognitive load.  Related to the above – one can’t objectively and before the fact tell whether something will be germane or not.  Sometimes something that induces extraneous load may also induce germane load and vice versa.  The type of load is highly dependent on learner characteristics and learning objectives (Moreno, 2009).

3. Lack of clarity about the cognitive load construct itself. Moreno (2009) describes a lack of clarity about terms such as cognitive load, mental load, and mental effort.  Mental load is a subjective rating or experience – it’s not ‘intrinsic’ to the material.

4. Lack of additivity. The assumption of cognitive load theory is that intrinsic, extraneous, and germane cognitive load all add up, and cannot exceed our working memory resources if learning is to occur.  As in point 2 above, de Jong thinks that intrinsic cognitive load is different ontologically from the other two types, and “we cannot add apples and oranges” (Moreno, 2009), and Moreno also describes recent studies that refute the additivity hypothesis.

Methodological Problems with Cognitive Load Theory Research

1. No reliable, valid measure of cognitive load.  Most people have used a one-item scale of perceived mental effort to measure cognitive load.  A one item measure can’t even be analyzed for reliability and validity properly.

2. Poor external validity of lab-based studies.  Moreno doesn’t touch on something in the de Jong article – the fact that most cognitive load (and multimedia learning) studies are conducted in labs that “includes participants who have no specific interest in learning the domain involved and who are also given a very short study time” (de Jong, 2009), often only a few minutes.  Quite a number of findings from these studies have not held up as strongly when tested in classrooms or real-world scenarios, or have even reversed (such as the modality effect, but see this refutation and this other example of a reverse effect).

3. Ignores, or selectively ignores, other educational and cognitive research.  Cognitive load theorists vehemently argue for the basis for their model in cognitive research, and yet ignore quite a huge swath of it.  It accepts the information processing view of cognition (most popular in the 1980s) and Baddeley’s model of working memory from the 1970s.

Conclusion

So is cognitive load theory a failure or wrong?  Is that important?  Like I said, the question is a joke. From one perspective Newton’s laws are wrong and were superseded by Einstein’s theories, but of course Newton’s laws are still quite useful and correct enough for everyday scenarios.  The more important question is whether a theory is useful, or is there a better, more useful theory.

Moreno (2009) concludes that cognitive load theory is at an impasse, and dissatisfaction with it is growing.  She cites Labaree’s (1998) paper about the learning sciences having to “live with a lesser form of knowledge” than the hard sciences.  Lesser because it doesn’t build on existing research, it isn’t subject to direct, reliable and valid measurements across different studies, and is more subject to bias.

I believe an area of future interest should be in exploring how post-cognitive theories may provide more useful explanations for some of the phenomena uncovered by cognitive load theory research.  Kaptelinin and Nardi describe some post-cognitive theories in chapter 9 of their book Acting with Technology. Unfortunately, chapter 9 was taken down from the First Monday site for some reason.  But the four theories they compare include activity theory, phenomenology, distributed cognition, and actor network theory.  They (and I) lean heavily toward activity theory and (embodied) phenomenology, both of which have significant overlap.  This is part of a larger phenomenon of researchers exploring post-cognitive models in education, human-computer interaction, and numerous other areas.  As the very short cognitive psychology page states on Wikipedia: “The information processing approach to cognitive functioning is currently being questioned by new approaches in psychology, such as dynamical systems, and the embodiment perspective.”

Further Reading

• Wolfgang Schnotz also published an article identifying conceptual problems with cognitive load theory in 2007, I just didn’t have time to re-review it in depth here.  Some quotes from the abstract:  “Various generalizations of empirical findings become questionable because the theory allows different and contradicting possibilities to explain some empirical results” and: “reduction of cognitive load can sometimes impair learning rather than enhancing it.”  Another 2005 article by Schnotz described why reduction of cognitive load can have negative results on learning.
• Moreno also wrote that cognitive load theory might be at an impasse in another article in 2006.
• Given that cognitive load theorists have at times hung some of their work on top of schema theory, evolutionary theory, ACT-R and so forth, one might be interested in the work connecting those theories with embodied cognition and/or activity theory (Vygotsky).  The McVee et al. (2005) article Schema Theory Revisited made this connection in the context of reading and literacy instruction.  Interestingly, Paivio, inventor of dual-coding theory which many multimedia learning theorists cite, writes a strong rejoinder against this article and all schema theory.  McVee responded to that and other critiques.
• Some of the terms cognitive load and multimedia learning researchers use such as “active” and “interactive” can be a bit vague, as well as other terms such as “constructive” and “passive” learning.  Chi (2009) recently proposed more operational and precise definitions of these terms in an article titled Active-Constructive-Interactive: A Conceptual Framework for Differentiating Learning Activities.