Working memory tests measure something narrower than the word suggests

Two meanings of one phrase

In ordinary use, working memory describes the everyday experience of holding a phone number or a set of instructions briefly in mind. In cognitive research it refers to a theoretical construct involving limited-capacity maintenance combined with active manipulation of information. The construct is inferred from performance on specific tasks rather than observed directly, which means the tasks define it operationally.

Different tasks intended to measure the same construct correlate with each other, and the correlations are weaker than the shared name implies. That gap between the construct and the measurements used to index it is where most of the confusion about cognitive training originates. It also explains why two people can describe themselves as having good or poor working memory while performing similarly on the laboratory tasks.

What the standard tasks require

One widely used task presents a stream of items and asks whether the current one matches the item a set number of steps back. Another asks a person to hold items while performing an unrelated processing task, then recall the held items in order. These tasks demand attention control, resistance to interference and strategy use in addition to any pure maintenance capacity.

Because performance depends on that mixture, an improvement in score does not identify which component actually changed. Distinguishing the components requires comparing performance across several task variants that differ deliberately rather than relying on any single measure. Research groups do exactly that, and consumer applications reporting a single score almost never have the task variety to support such a distinction.

Practice effects and the transfer problem

Repeated performance of any cognitive task improves the score, partly through genuine change and partly through learned task-specific strategy. Improvement on a trained task therefore establishes very little on its own, which is why controlled designs include untrained comparison tasks. Near transfer, meaning improvement on closely similar tasks, has been reported reasonably consistently across the training literature.

Far transfer to unrelated abilities such as reasoning has been much harder to demonstrate, and results have often shrunk under stricter designs. The pattern is consistent with people becoming better at the trained activity rather than acquiring a broader capacity.

Why the control condition decides the result

Participants who know they are in a training condition expect improvement, and expectation influences effort on subsequent testing. Studies comparing training against no activity at all cannot separate the training from the expectation it creates. Designs using an active control, where the comparison group performs an engaging but different task, produce systematically smaller effects.

That difference between design types accounts for a substantial share of the disagreement in this literature. It is a general lesson about behavioural research rather than a criticism specific to cognitive training products.

What the measurement supports

Working memory measures correlate with academic and occupational outcomes at population level, which is why they are studied at all. A correlation observed across many people does not establish that raising one person's score changes their outcomes. Capacity also varies within a person by time of day, by sleep, by stress and by whether the task is interesting, which limits single measurements.

Marketing that presents a test score as a fixed personal attribute is overstating the stability of the measurement considerably. Concerns about memory or attention that interfere with daily life warrant assessment by a clinician rather than a training application.