Working memory and intelligence
One 1990 paper asked whether reasoning ability is "little more than" working-memory capacity. The question has not gone away, and the answer explains a lot about why matrix items feel the way they do.
On this page
What working memory is
Working memory is not "short-term memory" in the everyday sense. It is not a bucket that holds seven items. It is the capacity to hold information available while doing something to it, in the face of distraction.5
The distinction shows up in how it is measured. A simple span task asks you to repeat a list of digits - that is storage. A complex span task interleaves the storage with an unrelated processing demand: read a sentence, judge whether it makes sense, remember a letter, read another sentence, and so on, then recall the letters in order.
Complex span correlates far better with reasoning than simple span does. The processing requirement is what makes the difference, which is the first clue that the relevant construct is not storage but something closer to controlled attention - the ability to keep goal-relevant information active while resisting interference.3
The strong claim
In 1990 a study of several hundred participants used structural equation modelling to compare latent factors for reasoning ability and working-memory capacity. The correlation between them was so high that the authors titled the paper with a provocation: reasoning ability is (little more than) working-memory capacity?!1
That finding reframed a lot of research. If the two are nearly the same at the latent level, then reasoning tests might be measuring a fairly basic capacity constraint rather than anything mysterious - and the door opens to a mechanistic account of individual differences in intelligence.
Note the phrase "at the latent level". Latent-variable models strip out measurement error, so correlations between latent factors run substantially higher than correlations between the raw test scores you would actually observe. Two constructs can be near-identical as latent variables while any two individual tests of them correlate far more modestly.
How close are they really
The strong claim did not survive intact. A meta-analysis pooling 86 samples found the average correlation between working-memory measures and reasoning was around 0.5 - substantial, clearly one of the strongest relationships in differential psychology, but well short of identity.4
At r = 0.5, roughly a quarter of the variance is shared and three quarters is not. The current consensus position is that working-memory capacity is a strong correlate and plausible partial cause of reasoning ability, but that the two are separable constructs.2
This is also how the hierarchical model of abilities treats them. In the Cattell-Horn-Carroll framework, short-term and working memory is its own broad ability sitting alongside fluid reasoning, both loading on general ability but neither reducible to the other.8
Why matrix items load on it
If you have ever felt a matrix item slip away from you - you had the pattern, then lost it while checking the options - you have felt the working-memory component directly.
The computational analysis of matrix solving found that difficulty was driven mainly by the number of rules operating simultaneously, and that the bottleneck was the need to hold partial conclusions in mind while deriving further ones.6
That is a working-memory demand almost by definition. It also explains a pattern in this test:
- A one-rule matrix is easy for nearly everyone. The rule is visible and there is nothing to hold.
- A three-rule matrix is hard, not because any single rule is subtle, but because you must keep two conclusions alive while working out the third.
- People who fail hard matrix items often report finding a rule and stopping - which is what happens when capacity runs out before the search does.
Number and letter series work the same way: holding a candidate rule while testing it against later terms is a storage-plus-processing demand. This is part of why the two domains correlate as strongly as they do.
Why training it does not help
The tight link between working memory and reasoning produced an obvious hypothesis: train working memory, raise reasoning. It is the intellectual foundation of the entire brain-training industry.
It has not worked. A meta-analysis of 87 studies found that working-memory training reliably improves the trained task, produces some near transfer to very similar tasks, and yields effects close to zero on reasoning, verbal ability, arithmetic and attention once active control groups are used.7
Why the correlation exists but the intervention fails is genuinely instructive. A correlation between capacity and reasoning is consistent with several stories:
- capacity constrains reasoning (training should work);
- both depend on a third factor such as neural efficiency (training the task changes neither);
- the span task and the reasoning task both simply require sustained controlled attention, so they correlate without either causing the other.
The training failures are evidence against the first story. What improves with practice appears to be task-specific strategy rather than general capacity - you get better at n-back, and n-back is not reasoning. The full evidence is in can you improve your IQ.
Common questions
Is working memory the same as IQ?
No. The measured correlation averages about 0.5 across studies - one of the strongest relationships in the field, but that still leaves three quarters of the variance unshared. They are treated as separate broad abilities in the standard hierarchical model.
Is working memory the same as short-term memory?
No. Short-term memory is storage - repeating back a list of digits. Working memory is holding information available while actively doing something with it, under interference. Complex span tasks that combine storage with processing correlate with reasoning much better than simple span tasks do.
Does a big working memory make matrix puzzles easier?
It helps considerably on hard ones. Difficulty in matrix items is driven mainly by how many rules run at once, and the bottleneck is holding partial conclusions in mind while deriving further ones. On one-rule items it matters very little.
If they are so closely related, why does working-memory training not raise IQ?
Because a correlation does not identify the causal direction or rule out a common third cause. The training evidence suggests what improves is task-specific strategy rather than general capacity - people get better at the trained task and the gain does not spread.
References
- Kyllonen, P. C., & Christal, R. E. (1990). Reasoning ability is (little more than) working-memory capacity?! Intelligence, 14(4), 389-433.
- Conway, A. R. A., Kane, M. J., & Engle, R. W. (2003). Working memory capacity and its relation to general intelligence. Trends in Cognitive Sciences, 7(12), 547-552. doi:10.1016/j.tics.2003.10.005
- Engle, R. W. (2002). Working memory capacity as executive attention. Current Directions in Psychological Science, 11(1), 19-23.
- Ackerman, P. L., Beier, M. E., & Boyle, M. O. (2005). Working memory and intelligence: The same or different constructs? Psychological Bulletin, 131(1), 30-60. doi:10.1037/0033-2909.131.1.30
- Baddeley, A. (2000). The episodic buffer: A new component of working memory? Trends in Cognitive Sciences, 4(11), 417-423.
- Carpenter, P. A., Just, M. A., & Shell, P. (1990). What one intelligence test measures: A theoretical account of the processing in the Raven Progressive Matrices Test. Psychological Review, 97(3), 404-431.
- Melby-Lervag, M., Redick, T. S., & Hulme, C. (2016). Working memory training does not improve performance on measures of intelligence or other measures of far transfer. Perspectives on Psychological Science, 11(4), 512-534. doi:10.1177/1745691616635612
- Carroll, J. B. (1993). Human Cognitive Abilities: A Survey of Factor-Analytic Studies. Cambridge University Press.
Related reading
Find out where you land
Thirty questions, about 30 minutes, and a score with the uncertainty attached.
Start the test30 questions · about 30 minutes · no sign-up