Cognitive Load

In Kirs2006, Kirschner, Sweller and Clark wrote:

Although unguided or minimally guided instructional approaches are very popular and intuitively appealing…these approaches ignore both the structures that constitute human cognitive architecture and evidence from empirical studies over the past half-century that consistently indicate that minimally guided instruction is less effective and less efficient than instructional approaches that place a strong emphasis on guidance of the student learning process. The advantage of guidance begins to recede only when learners have sufficiently high prior knowledge to provide “internal” guidance.

Their paper set off a minor academic storm, because beneath the jargon the authors were claiming that allowing learners to ask their own questions, set their own goals, and find their own path through a subject, as they would when solving problems in real life, isn’t effective. This approach—called inquiry-based learning—is intuitively appealing, but the authors argued that it overloads learners by requiring them to master a domain’s factual content and its problem-solving strategies at the same time.

More specifically, cognitive load theory posited that people have to deal with three things when they’re learning:

Intrinsic load
is what people have to keep in mind in order to absorb new material. In a programming class, this might be understanding what a variable is, or understanding how assignment in a programming language is different from creating a reference to a cell in a spreadsheet. Intrinsic load can’t be reduced except by reducing the amount of content being taught.
Germane load
is the (desirable) mental effort required to link new information to old, which is one of the things that distinguishes learning from memorization. An example might be remembering that a loop variable is assigned a new value each time the loop executes.
Extraneous load
is everything else in the instructional material that distracts from learning, such as matching the highlight colors in the instructor’s examples to the different color scheme used by the learner’s own editor.

Cognitive load theory holds that people have to split a fixed amount of working memory between these three things. Our goal as instructors is to maximize the memory available to handle germane load, which means reducing the intrinsic load at each step and eliminating as much of the extraneous load as possible.

For example, searching for a solution strategy is an extra burden on top of actually applying that strategy. We can therefore accelerate learning by giving learners worked examples that break a solution procedure down into steps, each of which can be mastered on its own before being combined with other steps (which is a step in its own right).

One way to do this is to give learners a series of faded examples. The first example presents a nearly-complete use of the same problem-solving strategy just demonstrated, but with a small number of blanks for the learner to fill in. The next problem is of the same type, but has more blanks, and so on until the learner is asked to solve the entire problem. The material that isn’t blank is often referred to as scaffolding, since it serves the same purpose as the scaffolding set up temporarily at a building site.

Faded examples can be used in almost every kind of teaching, from sport and music to contract law. Someone teaching programming might use them by first explaining how to calculate the total length of a list of words:

# total_length(["red", "green", "blue"]) => 12
define total_length(list_of_words):
    total = 0
    for each word in list_of_words:
        total = total + word.length()
    return total

and then asking learners to fill in the blanks in this (which focuses their attention on control structures):

# word_lengths(["red", "green", "blue"]) => [3, 5, 4]
define word_lengths(list_of_words):
    list_of_lengths = []
    for each ____ in ____:
    return list_of_lengths

The next problem might be this (which focuses their attention on updating the final result):

# join_all(["red", "green", "blue"]) => "redgreenblue"
define join_all(list_of_words):
    joined_words = ____
    for each ____ in ____:
    return joined_words

Learners would finally be asked to write an entire function on their own:

# make_acronym(["red", "green", "blue"]) => "RGB"
define make_acronym(list_of_words):

Faded examples work because they introduce the problem-solving strategy piece by piece: at each step, learners have one new problem to tackle, which is less intimidating than a blank screen or a blank sheet of paper (s:classroom-practices). It also encourages learners to think about the similarities and differences between various approaches, which helps create the linkages in their mental models that help retrieval.

Efficiency vs. Extent

Seeing worked examples accelerates learning more than having students write lots of code themselves Skud2014. As we will see in s:pck, deconstructing code by tracing it or debugging it also increases the efficiency of learning Grif2016. However, this isn’t the same as saying that people learn more unless they see additional problems.

The key to constructing a good faded example is to think about the problem-solving strategy it is meant to teach. For example, the series of problems are all examples of the accumulator pattern, in which the results of processing items from a collection are repeatedly added to a single variable in some way to create the final result.

Critics have sometimes argued that any result can be justified after the fact by labelling things that hurt performance as extraneous load and things that don’t as intrinsic or germane. However, instruction based on cognitive load theory is undeniably effective. For example, Maso2016 redesigned a database course to remove split attention and redundancy effects, and provide worked examples and sub-goals. The new course reduced exam failure rate by 34% on an identical final exam and increased student satisfaction.

A decade after the publication of Kirs2006, a growing number of people believe that cognitive load theory and inquiry-based approaches are compatible if viewed in the right way. Kaly2015 argues that cognitive load theory is basically micro-management of learning within a broader context that considers things like motivation, while Kirs2018 extends cognitive load theory to include collaborative aspects of learning. As with Mark2018 (discussed in s:individual-strategies), researchers’ perspectives may differ, but the practical implementation of their theories often wind up being the same.

Cognitive Apprenticeship

An alternative model of learning and instruction that also uses scaffolding and fading is cognitive apprenticeship, which emphasizes the way in which a master passes on skills and insights to an apprentice. The master provides models of performance and outcomes, then coaches novices as they take their first steps by explaining what they’re doing and why Coll1991,Casp2007. The apprentice reflects on their own problem solving, e.g., by thinking aloud or critiquing their own work, and eventually explores problems of their own choosing.

This model tells us that instructors should present several examples when presenting a new idea so that learners can see what to generalize, and that we should vary the form of the problem to make it clear what are and aren’t superficial features. (For a long time, I believed that the variable holding the value a function was going to return had to be called result because my instructor always used that name in examples.) Problems should be presented in real-world contexts, and we should encourage self-explanation, since it helps learners organize and make sense of what they have just been taught. This is discussed in more detail in s:individual-strategies.

Parsons Problems

Another kind of exercise that can be explained in terms of cognitive load theory is called a Parsons Problem (named after one of their creators Pars2006). If you are teaching someone to speak a new language, you could ask them a question, and then give them the words they need to answer the question, but in jumbled order. Their task is to put the words in the right order to answer the question grammatically, which frees them from having to think simultaneously about what to say and how to say it.

Similarly, when teaching people to program, you can give them the lines of code they need to solve a problem, and ask them to put them in the right order. This allows them to concentrate on control flow and data dependencies, i.e., on what has to happen before what, without being distracted by variable naming or trying to remember what functions to call. Multiple studies have shown that Parsons Problems take less time for learners to do, but produce equivalent educational outcomes Eric2017.

Labelled Subgoals

Subgoal labelling means giving names to the steps in a step-by-step description of a problem-solving process. Marg2016,Morr2016 all found that students with labelled subgoals solved Parsons Problems better than students without, and the same benefit is seen in other problem domains Marg2012. Returning to the Python example used earlier, the subgoals in finding the total length of a list of words or constructing an acronym are:

  1. Create an empty value of the type to be returned.

  2. Get the value to be added to the result from the loop variable.

  3. Update the result with that value.

Labelling subgoals works because grouping related steps in a chunk (s:memory-pattern) and giving each chunk a name helps learners distinguish between generic information and information that is specific to the problem at hand, which reduces cognitive load. It also helps them build a mental model of that kind of problem, so that they can solve other problems of that kind, and gives them a natural opportunity for self-explanation (s:individual-strategies).

Split Attention

Research by Mayer and colleagues on the split-attention effect is closely related to cognitive load theory Maye2003. Linguistic and visual input are processed by different parts of the human brain, and linguistic and visual memories are stored separately as well. This means that correlating linguistic and visual streams of information takes cognitive effort: when someone reads something while hearing it spoken aloud, their brain can’t help but check that it’s getting the same information on both channels (a topic we’ll return to when discussing dual coding in s:individual-strategies).

Learning is therefore more effective when information is presented simultaneously in two different channels, but when that information is complementary rather than redundant. For example, people generally find it harder to learn from a video that has both narration and on-screen captions than from one that has either the narration or the captions but not both, because some of their attention has to be devoted to checking that the narration and the captions agree with each other. Two notable exceptions to this are people who do not yet speak the language well and people with hearing exercises or other special needs, both of whom may find that the extra effort is a net benefit.

This explains why it’s more effective to draw a diagram piece by piece while teaching rather than to present the whole thing at once. If parts of the diagram appear at the same time as things are being said, the two will be correlated in the learner’s memory. Pointing at part of the diagram later is then more likely to trigger recall of what was being said when that part was being drawn.

The split-attention effect does not mean that learners shouldn’t try to reconcile multiple incoming streams of information—after all, this is something they have to do in the real world Atki2000. Instead, it means that instruction shouldn’t require it while people are mastering unit skills; instead, using multiple sources of information simultaneously should be treated as a separate learning task.

Not All Graphics Are Created Equal

Sung2012 presents an elegant study that distinguishes seductive graphics (which are highly interesting but not directly relevant to the instructional goal), decorative graphics (which are neutral but not directly relevant to the instructional goal), and instructive graphics (directly relevant to the instructional goal). Students who received any kind of graphic gave significantly higher satisfaction ratings to material than those who didn’t get graphics, but only students who got instructive graphics actually performed better.

Similarly, Stam2013,Stam2014 found that having more information can actually lower performance. They showed children pictures, pictures and numbers, or just numbers for two tasks: fraction equivalence and fraction addition. For equivalence, having pictures or pictures and numbers outperformed having numbers only. For addition, however, having pictures outperformed pictures and numbers, which outperformed just having numbers.

Minimal Manuals

The most extreme use of cognitive load theory may be the “minimal manual” method introduced in Carr1987. Its starting point is a quote from a user: “I want to do something, not learn how to do everything.” Carroll and colleagues therefore redesigned training to present every idea as a single-page self-contained task: a title describing what the page was about, step-by-step instructions of how to do something really simple (like how to delete a blank line in a text editor), and then several notes how to recognize and debug common problems.

Carroll and colleagues found that rewriting training materials this way made them shorter overall, and that people using them learned faster. Later studies like Lazo1993 confirmed that this approach outperformed the traditional approach regardless of prior experience with computers.

Looking back, Carr2014 summarized this work by saying:

Our “minimalist” designs sought to leverage user initiative and prior knowledge, instead of controlling it through warnings and ordered steps. It emphasized that users typically bring much expertise and insight to this learning, for example, knowledge about the task domain, and that such knowledge could be a resource to instructional designers. Minimalism leveraged episodes of error recognition, diagnosis, and recovery, instead of attempting to merely forestall error. It framed troubleshooting and recovery as learning opportunities instead of as aberrations.

He goes on to say that at the time, instruction decomposed skills into sub-skills hierarchically and then drilled people on the sub-skills. However, this meant context was lost: the goals weren’t apparent until people had learned the pieces. Since people want to dive in and do real tasks, well-designed instruction should help them do that.


Create a Faded Example (pairs/30)

It’s very common for programs to count how many things fall into different categories: for example, how many times different colors appear in an image, or how many times different words appear in a paragraph of text.

  1. Create a short example (no more than 10 lines of code) that shows people how to do this, and then create a second example that solves a similar problem in a similar way, but has a couple of blanks for learners to fill in. How did you decide what to fade out? What would the next example in the series be?

  2. Define the audience for your examples. For example, are these beginners who only know some basics programming concepts? Or are these learners with some experience in programming but not in Python?

  3. Show your example to a partner, but do not tell them what level it is intended for. Once they have filled in the blanks, ask them what level they think it is for.

If there are people among the trainees who don’t program at all, try to place them in different groups, and have them play the part of learners for those groups. Alternatively, choose a different problem domain and develop a faded example for it.

Classifying Load (small groups/15)

Working in groups of 3–4, choose a short lesson that one of you has taught or taken recently, make a point-form list of the ideas, instructions, and explanations it contains, and then classify each as intrinsic, germane, or extraneous. (The exercise “Noticing Your Blind Spot” in s:memory-exercises will give you an idea of how detailed your point-form list should be.)

Create a Parsons Problem (pairs/20)

Write five or six lines of code that does something useful, jumble them, and ask your partner to put them in order. If you are using an indentation-based language like Python, do not indent any of the lines; if you are using a curly-brace language like Java, do not include any of the curly braces. Again, if your group includes people who aren’t programmers, try using a different problem domain, such as making guacamole.

Minimal Manuals (individual/20)

Write a one-page guide to doing something simple that your learners might encounter in one of your classes, such as centering text horizontally or printing a number with a certain number of digits after the decimal points. Try to list at least three or four incorrect behaviors or outcomes the learner might see, and include a one- or two-line explanation of why each happens and how to correct it (i.e., go from symptoms to cause to fix).

Cognitive Apprenticeship (pairs/15)

Pick a small coding problem (something you can do in two or three minutes) and think aloud as you work through it while your partner asks questions about what you’re doing and why. As you work, do not just comment on what you’re doing, but also on why you’re doing it, how you know it’s the right thing to do, and what alternatives you’ve considered but discarded. When you are done, swap roles with your partner and repeat the exercise.

Critiquing Graphics (individual/30)

Maye2009 presents six principles for designing good diagrams for teaching. As summarized in Mill2016a, they are:

visually highlight the most important points that you want students to retain so that they stand out from less-critical material.
Spatial contiguity:
if using captions or other text to accompany graphics, place them as close to the graphics as practical to offset the cost of shifting between the two. If using diagrams or animations, place captions right next to relative components instead of putting them in one big block of text.
Temporal contiguity:
present spoken narration and graphics as close in time as practical—presenting both at once is better than presenting them one after another.
when presenting a long sequence of material or when students are inexperienced with the subject, break up the presentation into shorter segments and let students control how quickly they advance from one part to the next.
if students don’t know the major concepts and terminology used in your presentation, set up a module just to teach those concepts and terms and make sure they complete that module beforehand.
students learn better from pictures plus audio narration than from pictures plus text, unless there are technical words or symbols, or the students are non-native speakers.

Choose a video of a lesson or talk online that uses slides or other static presentations, and rate its graphics as “poor”, “average”, or “good” according to these six criteria.