What Working Memory Research Actually Predicts for Students

Every few years a parent emails me about a brain-training app their child has been using, usually after paying a subscription for twelve months, always with a wistful hope that the grades will have shifted. The grades almost never have. The app almost always has some impressive-looking graphic of a brain lighting up. Somewhere in the marketing copy is a phrase like scientifically designed to boost working memory. The phrase is doing a lot of work, most of it deceptive.

To understand why, it helps to know what working memory actually is, and what cognitive psychology has and has not established about it.

The modern concept traces to Alan Baddeley and Graham Hitch, who in 1974 proposed that short-term memory was not a single bucket but a small system of coordinated components. Baddeley, working at the University of Cambridge and later at the University of York, spent the following four decades refining the model. In its mature form it includes a central executive that directs attention and coordinates cognition, a phonological loop that holds verbal information for a few seconds and refreshes it by silent rehearsal, a visuospatial sketchpad that holds images and spatial layouts, and an episodic buffer that binds information from these subsystems into integrated chunks. The whole system has limited capacity, and that capacity is stubbornly difficult to expand.

This last point is where most of the pop psychology goes wrong.

The idea that working memory capacity can be trained, and that training it will transfer to real-world performance like school grades or reasoning ability, was a boom industry in the early 2010s. Susanne Jaeggi and Martin Buschkuehl published work in 2008 suggesting that a task called dual n-back could raise fluid intelligence. Commercial products built on this idea sold widely. Then the replication studies arrived and were unkind. Thomas Redick, Zach Shipstead, and colleagues at Georgia Tech, working with Randall Engle, ran a carefully controlled study published in 2013 in the Journal of Experimental Psychology: General. They trained participants on working memory tasks for weeks and found exactly what critics had predicted: people got better at the trained task and showed no meaningful improvement on other cognitive tasks, no transfer to fluid intelligence, and no transfer to academic outcomes. Later meta-analyses, notably by Monica Melby-Lervag and Charles Hulme, reached the same conclusion. If you train someone to do n-back, they get better at n-back. That is essentially all.

So what does working memory research actually say about studying?

First, capacity is a bottleneck but not a fate. A typical adult can hold something like four to seven unrelated chunks of verbal information in mind at once. This is small. It is why a phone number read aloud is a challenge and why a textbook paragraph in an unfamiliar subject feels like walking uphill. Any study technique that exceeds this limit produces comprehension failure without the student quite noticing why. This is also the foundation of cognitive load theory, and it is the reason the general case for reducing extraneous cognitive load matters so much. You cannot train the bottleneck to be wider. You can work around it.

Second, chunking is the one genuine escape hatch. A chess master looking at a mid-game board holds more pieces in mind than a novice not because the master’s phonological loop is larger, but because the master sees patterns, each a single chunk made up of many underlying elements. Expertise in any domain is in part the accumulation of such chunks. This is why long-term memory matters more for studying than working memory capacity. The richer your schemas, the more each unit of working memory can hold.

Third, the phonological loop and visuospatial sketchpad are somewhat independent. This has direct implications for study design. If you combine a verbal explanation with a relevant diagram, you are loading two different subsystems and roughly doubling effective capacity. If you combine a verbal explanation with written text of the same explanation, you are loading the phonological loop twice, once through reading and once through inner speech, and you are interfering with yourself. Richard Mayer at UC Santa Barbara has built a career out of testing this and related predictions of the modality effect. His work suggests that well-designed multimedia outperforms text-heavy instruction, but only when the designer knows which channel each piece of information is competing for. This is also one of the reasons drawing to learn outperforms rereading: the act of drawing forces you to integrate verbal and visuospatial representations instead of relying on either alone.

Fourth, the central executive is taxed by interruption. Every time your phone buzzes and you glance over, the executive has to swap tasks, flushing some of what you were holding and reloading the study context from long-term memory. This is expensive, and the cost accumulates. The familiar advice to study in focused blocks is not a matter of discipline theater. It is a consequence of how the central executive actually works.

There is a related point about spacing and working memory that is sometimes misstated. Spaced repetition does not work because it trains working memory. It works because long-term memory consolidation is a time-dependent process and because retrieval, especially after partial forgetting, strengthens traces in ways restudy does not. Most of the ways students misuse spaced repetition come from treating the software as the point rather than the schedule of retrieval as the point. Working memory is the doorway; long-term memory is the house.

Fifth, anxiety and sleep loss chew into effective working memory capacity in ways that dwarf any plausible training effect. Sian Beilock’s work on math anxiety at the University of Chicago, and later Barnard, demonstrated that anxious test-takers perform worse on working-memory-heavy tasks because intrusive thoughts occupy the same resources the problem requires. The practical implication is that a tired or anxious student has less capacity than a rested and calm one, and no training regime is going to fix that gap as cleanly as a night of sleep.

None of this means brain-training apps are scams in the legal sense. They deliver what they literally promise, which is a lot of practice on specific tasks. They just do not deliver what parents hope they deliver, which is better performance in school. Working memory science, read carefully, points in a different direction entirely. It suggests that students will gain more from building rich long-term schemas, from arranging their study environments to reduce interference, from coordinating verbal and visual channels in their notes, and from sleeping properly than from any amount of n-back. The evidence has been pointing that way for a while. The apps, somehow, keep selling.

Photo via Unsplash.