Memory lesson 10 min read

Why a small store matters

Working memory would be a curiosity if it only told you how many digits someone can repeat. It earns its place in assessments because the ability to hold information while doing something else turns out to be closely tied to reasoning, reading and learning. This lesson sets out what the evidence shows, how strong it is, and where honest limits apply.

The most studied relationship is between working memory capacity and fluid intelligence, the ability to solve new problems that you cannot answer from knowledge. Two careful reviews, published side by side in 2005, disagreed about how close the two are, and the disagreement itself is instructive.

0.48Average correlation between working memory and general intelligence across 86 samples
0.72Median correlation between working memory and fluid reasoning factors across 14 latent-variable data sets (over 3,100 adults)
about 50%Share of variance the two constructs have in common, on the latent-variable estimate

Ackerman and colleagues pooled many studies that used individual tasks and found a solid but clearly imperfect relationship. Kane and colleagues replied that single tasks carry a lot of measurement noise; when you combine several tasks into a latent factor, which strips out that noise, the correlation rises to around .72. Both teams agreed on the important point: working memory and reasoning are strongly related, but they are not the same thing.

WorkingmemoryFluidreasoningsharedabout halfcontrolled attention: holding a goal against interference
What a correlation of about .72 looks like as shared variance: two abilities overlapping by roughly half. Strong enough to matter, far from identical.

Why it is the interference that counts

The link to reasoning is not about storage size. In Engle and colleagues' 1999 study, the factor built from tasks that only asked people to repeat lists (short-term memory) showed no strong connection to fluid intelligence, while the factor built from tasks that required holding information while doing something else (working memory) did. A 2004 follow-up with 236 participants found the same pattern across verbal and spatial material: working memory tasks loaded on one general factor that strongly predicted reasoning, while simple storage tasks were more tied to their own content.

The researchers' explanation is that working memory tasks and reasoning problems share one demand: keeping a goal and its pieces active while something else competes for attention. A matrix puzzle, a data-interpretation set and a complex span task all punish you for losing track of what you were doing when a distraction or a sub-step arrives.

This is the reason assessment designers prefer tasks with interference built in, such as GridRecall's symmetry judgements between storage steps, or Flashback's near-miss lures. Those features push the task toward the part of working memory that predicts the most.

Learning and school results

The evidence for reading and early schooling is also consistent.

How strongly working memory relates to other outcomes (correlations)
  • Working memory and fluid reasoning, latent factors0.72
  • Working memory and general intelligence, single tasks0.48
  • Working memory and reading, 197 studies0.29
  • Complex span and n-back tasks with each other0.2
  • Reading. A 2018 meta-analysis of 197 studies found a moderate correlation of .29 between working memory and reading. Before about the fourth year of school, all kinds of working memory related to reading similarly; after that, verbal working memory mattered most.
  • Early predictors. In a longitudinal study, Alloway and Alloway found that children's working memory at five years old was the best predictor of their literacy and numeracy six years later, with IQ accounting for a smaller unique share. Working memory was not simply a stand-in for IQ.

The practical meaning for an adult candidate is modest but real. Tasks that load working memory heavily, like multi-step quantitative problems, dense reading and data sets with several linked numbers, are where people with more capacity, or better strategies for saving capacity, pull ahead.

Job performance: the honest version

Here the evidence is indirect, and it is worth being exact about why.

For decades the headline figure in personnel selection has been Schmidt and Hunter's 1998 estimate that general mental ability correlates about .51 with job performance, making it one of the best single predictors. In 2022 Sackett and colleagues argued that earlier meta-analyses had over-corrected for range restriction, and that most validity estimates, including cognitive ability, should come down by about .10 to .20. Cognitive tests remain useful predictors, but less dominant than the older figure suggests; in their revised ranking, structured interviews came top.

What does not exist, as far as we can find, is a large meta-analysis of working memory tests on their own predicting job performance. The case employers and vendors make runs through the link above: working memory overlaps heavily with fluid reasoning, and reasoning ability predicts job performance. Aon, for example, describes its gamified working memory assessment as focusing on attention control instead of learned knowledge, partly to reduce group differences that knowledge-heavy tests can show.

Why it collapses under pressure

Working memory is also the resource that stress and fatigue attack first, which matters on test day.

Pressure. Beilock and Carr studied maths performance under pressure. Pressure hurt only the people with high working memory capacity, and only on problems with the highest working memory demands. Their explanation: worries about the outcome occupy the very capacity those people normally rely on.

Sleep. A meta-analysis of 70 articles on short-term total sleep deprivation found a medium-sized drop in working memory accuracy (Hedges' g of about −0.56), larger than the effect on reasoning accuracy, which was small and not statistically significant.

CalmUnder pressureratebasesteptotalall four pieces heldratebaseworryclockstep and total droppedSame capacity, different free space
Pressure as a lodger: if worry takes one or two of your roughly four slots, a problem that needed all four now fails, even though your capacity has not changed.

Neither finding means you should panic about one bad night. It does mean that a memory task taken tired or anxious understates you more than, say, a vocabulary test would, and that familiarity with the task format (which lowers the worry load) is a legitimate way to perform nearer your real level. The how to get better lesson turns this into a plan.

Where the evidence runs out

Being precise about limits will stop you drawing the wrong conclusions from a score.

  • One task, one day, is noisy. Single-task correlations are lower than latent ones precisely because any one measurement includes error. Your trend across sessions tells you more than any single run.
  • Correlation is not destiny. A correlation of .72 between factors still leaves plenty of people with strong reasoning and ordinary span, and the reverse.
  • Group results, not individual forecasts. The academic findings describe averages across thousands of learners. They do not say a given person will struggle.
  • Capacity is hard to raise. The training literature, covered in lesson four, finds that practising memory tasks improves those tasks and close relatives, with little reliable spread to reasoning or school outcomes. What improves reliably is your use of the capacity you have.

Reading your own memory score

Put together, the evidence suggests three ways to read a memory result, here or in a real assessment.

  • Compare facets, not just the total. A strong span score with a weak updating score points to a specific habit to fix (see the Flashback lesson), not a general weakness.
  • Look at interference. If your recall holds up only when you neglect GridRecall's figures, the gap between those trials is the part of working memory that matters most for reasoning, and it is trainable as a strategy even if capacity is not.
  • Discount bad days. A session taken tired, rushed or anxious understates you. Your trend over several sessions is the honest number, which is why our scores need several trials before they appear and weight recent evidence more heavily.

Key points

  • Working memory and fluid reasoning are strongly related (about .72 between latent factors, about .48 for single tasks) but distinct.
  • The predictive part is holding information against interference, which is why good memory tasks include distractions, lures or a second job.
  • Working memory relates moderately to reading (.29 across 197 studies) and early working memory predicted later school attainment better than IQ in one longitudinal study.
  • The link to job performance is indirect, through general cognitive ability, whose validity estimates have been revised downward.
  • Pressure and lost sleep eat into working memory directly, so test conditions matter.
Sources
  1. Ackerman, P. L., Beier, M. E., and 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
  2. Kane, M. J., Hambrick, D. Z., and Conway, A. R. A. (2005). Working memory capacity and fluid intelligence are strongly related constructs: comment on Ackerman, Beier, and Boyle (2005). Psychological Bulletin, 131(1), 66–71. doi:10.1037/0033-2909.131.1.66
  3. Engle, R. W., Tuholski, S. W., Laughlin, J. E., and Conway, A. R. A. (1999). Working memory, short-term memory, and general fluid intelligence: a latent-variable approach. Journal of Experimental Psychology: General, 128(3), 309–331. doi:10.1037/0096-3445.128.3.309
  4. Kane, M. J., Hambrick, D. Z., Tuholski, S. W., Wilhelm, O., Payne, T. W., and Engle, R. W. (2004). The generality of working memory capacity: a latent-variable approach to verbal and visuospatial memory span and reasoning. Journal of Experimental Psychology: General, 133(2), 189–217. doi:10.1037/0096-3445.133.2.189
  5. Peng, P., Barnes, M., Wang, C., et al. (2018). A meta-analysis on the relation between reading and working memory. Psychological Bulletin, 144(1), 48–76. doi:10.1037/bul0000124
  6. Alloway, T. P., and Alloway, R. G. (2010). Investigating the predictive roles of working memory and IQ in academic attainment. Journal of Experimental Child Psychology, 106(1), 20–29. doi:10.1016/j.jecp.2009.11.003
  7. Schmidt, F. L., and Hunter, J. E. (1998). The validity and utility of selection methods in personnel psychology: practical and theoretical implications of 85 years of research findings. Psychological Bulletin, 124(2), 262–274. doi:10.1037/0033-2909.124.2.262
  8. Sackett, P. R., Zhang, C., Berry, C. M., and Lievens, F. (2022). Revisiting meta-analytic estimates of validity in personnel selection: addressing systematic overcorrection for restriction of range. Journal of Applied Psychology, 107(11), 2040–2068. doi:10.1037/apl0000994
  9. Martin, N., and LaPort, K. (2017). Traditional cognitive testing got you down? Get your G.A.M.E. on with Aon. Aon. aon.com
  10. Beilock, S. L., and Carr, T. H. (2005). When high-powered people fail: working memory and choking under pressure in math. Psychological Science, 16(2), 101–105. doi:10.1111/j.0956-7976.2005.00789.x
  11. Lim, J., and Dinges, D. F. (2010). A meta-analysis of the impact of short-term sleep deprivation on cognitive variables. Psychological Bulletin, 136(3), 375–389. doi:10.1037/a0018883
  12. Redick, T. S., and Lindsey, D. R. (2013). Complex span and n-back measures of working memory: a meta-analysis. Psychonomic Bulletin and Review, 20(6), 1102–1113. doi:10.3758/s13423-013-0453-9
  13. Schmiedek, F., Lövdén, M., and Lindenberger, U. (2014). A task is a task is a task: putting complex span, n-back, and other working memory indicators in psychometric context. Frontiers in Psychology, 5, 1475. doi:10.3389/fpsyg.2014.01475