The first of the four Evaluation lenses from Lesson 2 asks whether a design matches how people actually learn. This lesson is the evidence you answer that with. Seven findings, each with the researchers who established it, each with the design decision it changes. They are not schools of thought to choose between; each is a body of results that has held up across many studies, and a design that ignores them is not a matter of taste.
By the end you should be able to explain intrinsic, extraneous and germane cognitive load and spot extraneous load on a screen (4.1); explain spacing, retrieval practice and interleaving and say what each predicts about a design choice (4.2); apply Mayer's multimedia principles to judge a slide or screen (4.3); explain worked examples and the expertise reversal effect and say when a worked example helps and when it hurts (4.4); and define near and far transfer and name a design move that supports each (4.5).
In this lesson 7 sections
This lesson is the reading. The evidence is set out here with its sources named. Every researcher named below has a primary publication you can find; years are deliberately left out and should be checked against the source, not a summary, before you cite anything from this lesson in a portfolio or a design rationale.
4.1 Cognitive load
Working memory is small and short. John Sweller's cognitive load theory starts from that limit and asks where the load in a learning task comes from. Three sources:
- Intrinsic load is the difficulty inherent in the material for this learner: how many elements must be held in mind at once and how they interact. A dose calculation has more intrinsic load than a drug name. You cannot remove intrinsic load; you can sequence it (simple to complex) and chunk it.
- Extraneous load is load the design adds that has nothing to do with learning: decorative images, a narrator reading text that is also on screen, a diagram on one page and its explanation on another, a menu that has to be understood before the content can be reached. This is the load a designer is responsible for, and most first drafts are full of it.
- Germane load is the effort that goes into building the schema: connecting the new material to what is known, organizing it, practicing retrieval. It is the load you want. Reducing extraneous load frees capacity for germane load.
What it changes. When you review a screen, ask what the learner has to process that is not the thing being learned. A stock photo of a smiling team on a slide about payroll policy is extraneous load. Text on screen read aloud word for word is extraneous load (the redundancy principle, below). Twelve bullet points in one view is intrinsic load presented all at once, which becomes a design problem. The fix is nearly always removal or sequencing, not more explanation.
Check yourself: A compliance screen has a paragraph of policy text, a narrator reading it aloud, a background photograph of an office, and an animated icon in the corner. Name the extraneous load.
Three of the four elements. The narration duplicating on-screen text (redundancy), the photograph (decorative, unrelated) and the animation (attention pulled away from the text) all add processing that is not learning. The policy text is the intrinsic load and stays.
4.2 Spacing, retrieval and interleaving
Three findings from memory research, all robust, all routinely ignored by course designs that present material once and test it at the end.
Spacing. Practice distributed over time produces better long-term retention than the same amount of practice massed together. Nicholas Cepeda and colleagues synthesized a large body of studies on the effect; Robert Bjork's framing of "desirable difficulties" explains why it feels worse and works better: the forgetting between sessions makes each retrieval harder and therefore more effective. Design consequence: a topic met once in week two and tested in week ten will largely be gone. Revisit it, briefly, at intervals. This course is built on that finding; it is why the Unit B and Unit D assessments include cumulative items.
Retrieval practice. Pulling information out of memory strengthens it more than putting it in again. Henry Roediger and Jeffrey Karpicke's studies showed that learners who were tested on material remembered more later than learners who spent the same time rereading it, even though the rereaders felt more confident. Design consequence: a low-stakes quiz is a learning event, not just a measurement. The checkpoints in this course exist for this reason, and the "check yourself" prompts are the same thing at smaller scale. A course with no retrieval until the final is leaving most of its learning on the table.
Interleaving. Mixing problem types in practice (A, B, C, A, C, B) produces better discrimination and transfer than blocking them (A, A, A, B, B, B), because the learner has to decide which approach applies rather than repeating the one they were just shown. Doug Rohrer's work on mathematics practice is the clearest demonstration. Design consequence: if the job requires telling situations apart (which policy applies, which model fits, which error this is), practice must present them mixed. Blocked practice feels smoother and teaches less.
Check yourself: A manager wants all the practice questions for a module at the end of the module "so learners are not interrupted." What do you tell her, and on what evidence?
That the interruption is the learning. Retrieval practice (Roediger and Karpicke) strengthens memory more than rereading, and spacing it through the module rather than massing it at the end improves retention (Cepeda; Bjork). Propose short checks after each section and a mixed set at the end, and offer to show her the difference in the next cohort's results.
4.3 Mayer's multimedia principles
Richard Mayer's research program on how people learn from words and pictures together produced a set of principles that are the closest thing the field has to a checklist for a screen. The ones you will use most:
| Principle | People learn better when | So do not |
|---|---|---|
| Coherence | Extraneous words, pictures and sounds are excluded | Add "interesting" material that is not the point |
| Signaling | Cues highlight the organization of the essential material | Leave the learner to find the structure |
| Redundancy | Graphics are paired with narration rather than with narration plus identical on-screen text | Read the slide aloud |
| Spatial contiguity | Words and the pictures they describe are placed near each other | Put the label in a legend across the page |
| Temporal contiguity | Corresponding words and pictures are presented at the same time | Explain the diagram, then show it |
| Segmenting | A complex lesson is presented in learner-paced segments | Autoplay a twelve-minute continuous animation |
| Modality | Graphics are explained by narration rather than by on-screen text | Pair a complex diagram with a dense caption |
| Personalization | Words are in conversational rather than formal style | Write like a policy manual when you are teaching |
Two cautions. These are principles, established on average across many learners, and they interact with prior knowledge: the modality effect, for instance, weakens for experts. And they are about learning, not about looking professional; a screen that violates several of them can be beautiful. In QA, the question is not "is this attractive" but "which principle does this break, and would a learner learn more if it did not."
Check yourself: A slide shows a process diagram with eight steps; each step's explanation is in a numbered list beside the diagram, and the narrator reads the list. Which principles are at risk?
Spatial contiguity (the explanations are beside the diagram rather than on it), redundancy (narration duplicating on-screen text), and possibly segmenting if all eight steps appear at once. A better version reveals steps one at a time with the label on the step and the narration carrying the explanation.
4.4 Worked examples and the expertise reversal effect
A worked example shows a complete solution with its steps, and asking a novice to study worked examples before attempting problems produces better learning than having them solve problems from the start (the worked-example effect, from Sweller's group). The reason is load: a novice solving a problem unaided spends working memory on search rather than on learning the pattern. Fading works well: a full worked example, then one with the last step missing, then one with two steps missing, then a full problem.
The expertise reversal effect, documented by Slava Kalyuga and colleagues, is the other half. Instructional support that helps novices becomes neutral and then harmful as expertise grows: the expert already has the schema, and the worked example is now extraneous load. The same design that is right for a first-week learner is wrong for a third-year one.
What it changes. "Who is the audience and how expert are they" is not a courtesy question; it decides whether guidance helps or hurts. Paul Kirschner, John Sweller and Richard Clark's argument against minimal guidance for novices rests on this evidence, and so does this course's decision to put a structured floor before the first project. It also explains why a single course for "everyone" so often satisfies nobody: the novice needs the worked example the expert resents.
4.5 Transfer
Transfer is the use of something learned in one situation in a different one, and it is the actual goal of nearly all workplace learning: nobody pays for what a learner can do inside the course. Near transfer is to situations closely resembling the learning situation (the same software, a slightly different form). Far transfer is to situations that look different on the surface but share deep structure (applying a negotiation principle learned in a sales role to a budget conversation). Far transfer is rarer and harder, and claims of it should be treated with suspicion until measured.
David Perkins and Gavriel Salomon described two families of design move. Hugging supports near transfer by making the practice resemble the target situation closely: realistic scenarios, real tools, real constraints. Bridging supports far transfer by having learners abstract the principle and deliberately connect it to other contexts: "where else would this apply, and what would change." Most designs hug and hope; the ones that also bridge are the ones whose learners use the material somewhere new.
Check yourself: A course teaches a five-step de-escalation model using three retail scenarios and tests it with a fourth retail scenario. What kind of transfer does it support, and what one addition would support the other kind?
Near transfer, by hugging: the practice resembles the target. For far transfer, add a bridging activity: have learners name a situation outside retail (a family argument, a tense meeting) and work through which steps hold and which change. Then test with a non-retail scenario.
How this lesson gets used
These seven findings are what you will reach for under the first Evaluation lens, in QA cycle one and every project after it. A design rationale that says "this uses spaced retrieval because the material is met once and used months later" is worth ten that say "this is engaging." In Lesson 11 you will run an AI-generated draft through exactly these findings, and you will find that fluent drafts violate several of them by default.
Check yourself: Back to the sign-off. The AI-built module had a narrator reading every slide's text aloud, stock photos on every slide, and one quiz at the end. Name the finding each one breaks, and the change you'd ask for.
Narration reading the on-screen text breaks the redundancy principle: keep the narration and cut the on-screen text to short labels. Stock photos on every slide break coherence and add extraneous load: remove any image that isn't the thing being learned. One quiz at the end ignores retrieval practice and spacing: add short checks after each section and a mixed set at the end. That's a sign-off note built on evidence, not taste. Compare it with what you said at the start of the lesson.
You wrote, before signing off:
You wrote, to the client:
Before the checkpoint
Eight questions on the five objectives. The load types, the three memory findings, at least five of Mayer's principles, the reversal effect and the two kinds of transfer should all be within reach.
Sources named in this lesson: John Sweller (cognitive load theory, the worked-example effect); Nicholas Cepeda and colleagues (spacing); Robert Bjork (desirable difficulties); Henry Roediger and Jeffrey Karpicke (retrieval practice); Doug Rohrer (interleaving); Richard Mayer (multimedia principles); Slava Kalyuga and colleagues (expertise reversal); Paul Kirschner, John Sweller and Richard Clark (guidance for novices); David Perkins and Gavriel Salomon (transfer, hugging and bridging). Verify each against the primary publication before citing.