Individual Differences in Working Memory Capacity: Measurement, Training, and the Filtering Account

Working memory capacity (WMC) refers to how well people can hold and work with information in mind while ignoring distractions. This essay reviews how researchers measure WMC, what distinguishes people with low versus high capacity, and whether training can improve working memory. Traditional lab tasks like complex span measures show good reliability, and newer online tools like the Mental Counters task allow for remote testing. A major contribution to understanding individual differences comes from Vogel and colleagues, who demonstrated using electrophysiological measures that high-capacity individuals excel at filtering irrelevant information from entering working memory, whereas low-capacity individuals inefficiently encode distractors alongside targets (Vogel, McCollough, & Machizawa, 2005). This filtering efficiency account has reshaped how researchers conceptualize WMC differences. Research on working memory training consistently finds that people get better at the specific tasks they practice. Some studies suggest that low-capacity individuals show larger improvements during training. However, transfer to other cognitive abilities—like intelligence or problem-solving—is limited and small in magnitude. Recent well-designed studies using varied training approaches have failed to show broader benefits, and individual differences do not appear to predict who benefits most. These findings suggest that while working memory is plastic, training improvements reflect learning how to do specific tasks rather than expanding general capacity. Theoretical and practical implications are discussed.

Introduction

Working memory is the cognitive system that lets you hold information in mind while working with it—like remembering a phone number while looking for your keys, or keeping track of the characters in a novel while following the plot. Since Baddeley and Hitch’s (1974) influential model, researchers have learned that this limited-capacity system is crucial for many aspects of everyday thinking.

Why should we care about working memory capacity (WMC)? Because individual differences in WMC predict a wide range of outcomes. People with higher WMC tend to score higher on intelligence tests (Conway, Kane, & Engle, 2003; Engle, Tuholski, Laughlin, & Conway, 1999; Kane et al., 2004), comprehend what they read better (Daneman & Merikle, 1996), solve problems more effectively (Wiley & Jarosz, 2012), and reason more accurately (Kyllonen & Christal, 1990). WMC also relates to real-world issues like depression (Joormann & Gotlib, 2008), how people perform under stereotype threat (Schmader & Johns, 2003), and even eyewitness accuracy (Jaschinski & Wentura, 2002). Some researchers have suggested that declining WMC might serve as an early warning sign for Alzheimer’s disease (Rosen, Bergeson, Putnam, Harwell, & Sunderland, 2002).

Given how much WMC predicts, two questions have driven research for decades. First, how can we measure WMC reliably and validly, especially now that researchers often need to test people online rather than in the lab? Second, can we improve WMC through training, and if so, do those improvements spread to other mental abilities? The second question matters because if we could boost WMC, we might help students learn better, workers perform more effectively, or older adults maintain their cognitive abilities longer.

This essay examines what we currently know about individual differences in working memory capacity, with particular attention to Vogel and colleagues’ influential work on attentional filtering. I’ll cover how researchers measure WMC, what makes low- and high-capacity people different according to the filtering account, and what the evidence says about whether working memory training works.

What Is Working Memory Capacity?

Working memory capacity isn’t just about how much you can remember. Most researchers now think WMC reflects how well you can control your attention—keeping relevant information active in mind while blocking out distractions (Engle, 2018; Barrett, Tugade, & Engle, 2004). This attention-based view distinguishes WMC from simple short-term memory, which you can measure by asking someone to repeat back a list of numbers with no other demands.

This distinction matters for understanding individual differences. People with high WMC are better at focusing on what matters and ignoring what doesn’t. They resist interference better (Kane & Engle, 2000, 2003), show less negative priming (Conway, Tuholski, Shisler, & Engle, 1999), and report fewer mind-wandering episodes during demanding tasks (Kane et al., 2007). People with low WMC, on the other hand, are more vulnerable to interference. They struggle more with proactive interference (Bunting, 2006; Lustig, May, & Hasher, 2001), fall for memory illusions more easily (Watson, Bunting, Poole, & Conway, 2005), and have more trouble filtering out irrelevant information (Conway & Engle, 1994).

These differences show up across many kinds of tasks. High WMC individuals outperform low WMC individuals on dichotic listening tasks where you have to pay attention to one ear while ignoring the other (Conway, Cowan, & Bunting, 2001). They’re less likely to miss changes in visual scenes (Colflesh & Wiley, 2013). They’re better at anti-saccade tasks where you have to look away from a flashing light instead of toward it (Kane, Bleckley, Conway, & Engle, 2001). All of this points to the same conclusion: WMC reflects domain-general attention control abilities, not just how much you can store in memory.

Vogel and Colleagues’ Filtering Account of Individual Differences

The CDA as a Window into Working Memory

A major breakthrough in understanding individual differences in WMC came from Vogel and colleagues’ innovative use of event-related potentials (ERP). They developed a neurophysiological measure called the contralateral delay activity (CDA), a sustained negative wave over posterior scalp regions contralateral to the visual field where memory items are presented (Vogel & Machizawa, 2004). This activity persists throughout the memory retention period and increases systematically with the number of items being maintained, reaching asymptote at each individual’s capacity limit (McCollough, Machizawa, & Vogel, 2007).

The CDA provides a moment-by-moment window into the contents of visual working memory. Unlike behavioral measures that only reveal the final output of memory processes, the CDA allows researchers to track exactly what information individuals are maintaining during the retention interval. This methodological innovation made it possible to test hypotheses about why people differ in their apparent memory capacity (Vogel & Machizawa, 2004).

The Filtering Paradigm

In their landmark 2005 Nature paper, Vogel, McCollough, and Machizawa used the CDA to investigate whether individual differences in WMC reflect differences in storage capacity or differences in the ability to control what enters memory (Vogel et al., 2005). They designed a filtering task in which participants viewed arrays containing both relevant items (e.g., red rectangles) and irrelevant distractors (e.g., blue rectangles). Participants were instructed to remember only the relevant items for a later memory test.

The logic was straightforward: If the CDA reflects only the number of items being actively maintained, then its amplitude should indicate whether individuals are successfully excluding the distractors. For someone who filters efficiently, the CDA when viewing two relevant items plus two distractors should resemble the CDA for two relevant items alone. For someone who fails to filter, the CDA should resemble that for four relevant items, indicating that distractors are consuming valuable capacity (Vogel et al., 2005).

The results were striking. High-capacity individuals showed efficient filtering: their CDA amplitude for the distractor condition matched the amplitude for two-item arrays, not four-item arrays. Low-capacity individuals, however, showed a CDA amplitude for the distractor condition that was indistinguishable from the four-item condition (Vogel et al., 2005). As Vogel and colleagues summarized, “People differed systematically, and dramatically, in their ability to keep irrelevant items out of awareness” (Vogel et al., 2005, p. 502). This finding challenged the simple concept that memory capacity is just about storage space—it’s about the “bouncer” controlling what gets in.

Implications of the Filtering Account

The filtering account has several important implications. First, it suggests that low-capacity individuals aren’t necessarily storing less information overall—in some circumstances, they may actually store more information than high-capacity individuals, just not the right information (Vogel et al., 2005). As Vogel and colleagues noted, “low capacity individuals may actually store more information in memory than high capacity individuals. Indeed, this ancillary allocation of memory capacity to irrelevant objects may be a primary source of putative differences in overall storage capacity” (Vogel et al., 2005, p. 503).

Second, the filtering account connects WMC to broader attentional control processes. Subsequent work by Fukuda and Vogel (2011) examined whether low-capacity individuals are more susceptible to attentional capture by distractors or whether they simply take longer to recover from capture. Using psychophysical and electrophysiological methods, they found that high- and low-capacity individuals showed equivalent attentional capture effects in the initial moments following a distractor, but low-capacity individuals took much longer to recover. This suggests that the poor attentional control associated with low capacity reflects slow disengagement from distractors rather than enhanced capture per se (Fukuda & Vogel, 2011).

Third, the filtering framework has been extended to other domains and populations. Research has indicated that the filtering deficit generalizes beyond color-based filtering to location-based filtering and sequential memory updating (Vogel et al., 2005). Studies with older adults suggest that age-related declines in working memory may partly reflect filtering deficits (Lustig et al., 2001). Individual differences in cognitive style, such as field independence, have also been linked to filtering efficiency, with field-independent individuals showing better ability to exclude distractors than field-dependent individuals.

More recent work has connected filtering ability to real-world outcomes like demand avoidance. Nador, Minnery, Sherwood, Harel, and Juvina (2017) found that individuals with higher WMC and better cognitive filtering tended to select less demanding task alternatives, presumably because efficient filtering made them more sensitive to variations in cognitive demand. Inefficient filterers, who process more irrelevant information, were less sensitive to demand manipulations (Nador et al., 2017).

Ongoing Questions About the CDA

While the CDA has proven valuable, researchers continue to refine understanding of what it reflects. Some evidence suggests that the CDA may be sensitive not just to the number of objects but also to their spatial distribution. Wang, Most, and Hoffman (2010) found that CDA amplitude was highest for two items presented at different locations, while amplitude for two items at the same location was equivalent to that for a single item. This suggests that at least some components of the CDA reflect the number of locations requiring attention rather than the number of objects per se (Wang et al., 2010). Such findings highlight the importance of continued methodological refinement in understanding the neural bases of individual differences.

Measuring Working Memory Capacity

Traditional Lab Measures

The way researchers measure WMC has evolved since Daneman and Carpenter (1980) introduced the reading span task. Complex span tasks—which mix memory items with distracting processing demands—have become the standard. The operation span task (Unsworth, Heitz, Schrock, & Engle, 2005) has you solve math problems while trying to remember letters for later recall. The reading span task involves reading sentences while remembering their last words. The symmetry span task requires judging whether patterns are symmetrical while remembering sequences of locations in a grid.

These tasks have good psychometric properties. A recent comprehensive study by Robison, Miller, and Unsworth (2026) looked at 24 cognitive measures administered twice over two weeks. Complex span tasks showed adequate to high reliability (typically above .70) and good test-retest reliability. Using statistical techniques to extract what multiple tasks have in common, the researchers confirmed that working memory measures form coherent factors distinct from processing speed, primary memory, and fluid intelligence. Importantly, the measurement structure stayed stable across testing sessions—the tasks were measuring the same things each time (Robison et al., 2026).

Researchers often use latent variable approaches because no single task perfectly measures WMC. Any individual task reflects task-specific skills along with true capacity. By giving multiple tasks and extracting their shared variance through factor analysis, researchers get cleaner estimates of WMC (Conway et al., 2005). These latent estimates correlate more strongly with important outcomes than any single task does, because the task-specific noise gets removed.

New Online Assessment Tools

The COVID-19 pandemic pushed researchers to develop reliable online assessment tools. Draheim, Mashburn, Martin, and Engle (2024) recently validated an online version of the Mental Counters task, originally developed by Larson, Merritt, and Williams (1988). In this task, you track three independent counters that randomly go up or down, then report their final values.

The Online Mental Counters task has several advantages for remote testing. First, it’s non-verbal, so it avoids cultural and linguistic biases that can affect verbal tasks like reading span when administered to non-native speakers (Farmer, Fine, Misyak, & Christiansen, 2017; Olsthoorn, Andringa, & Hulstijn, 2014). Second, its fast-paced, novel format makes cheating difficult—unlike with verbal stimuli, participants can’t easily write down what they need to remember (Hicks, Foster, & Engle, 2016). Third, it only takes about ten minutes to administer, making it practical to include in larger test batteries.

Draheim et al. (2024) found that Online Mental Counters correlates moderately with established measures like Picture Span (r ≈ .25) and Paper Folding (a measure of fluid intelligence). These correlations match what previous lab studies have found (Colom & Shih, 2004; Mackintosh & Bennett, 2003), supporting the task’s validity. Tools like this make it possible to do large-scale individual differences research while maintaining good measurement standards.

Working Memory Training: What the Evidence Shows

The Promise of Cognitive Enhancement

Since WMC predicts so many important outcomes, it’s natural to wonder whether we can increase it through training. If WMC reflects a fundamental constraint on thinking, then interventions that boost it might produce broad cognitive benefits—improving intelligence, academic achievement, and everyday functioning. This possibility has generated tremendous research interest and commercial activity. Computerized training programs like Cogmed are now widely used in schools and clinics.

Researchers debate exactly how training might work. Early accounts suggested that repeatedly using working memory systems strengthens underlying neural circuits, increasing capacity in a general way (Klingberg, Forssberg, & Westerberg, 2002). Alternative views emphasize skill learning: rather than expanding capacity, training helps people develop task-specific strategies that improve performance on trained and similar tasks (Gathercole, Dunning, Holmes, & Norris, 2019; Laine, Fellman, Waris, & Nyman, 2018). These different accounts make different predictions about transfer—whether training benefits spread to untrained tasks.

Do Low- and High-Capacity People Benefit Differently?

An important question is whether training works better for people who start with lower or higher WMC. Low-capacity individuals might have more room to improve, or they might lack the attention control needed to benefit from training. Evidence from different populations gives mixed answers.

Zinke, Zeintl, Eschen, Herzog, and Kliegel (2012) studied working memory training in very old adults (average age about 87), a group with substantially reduced cognitive function. Participants completed ten sessions training on five working memory tasks. The training group improved on four of the five tasks, with medium to large effects. Critically, these improvements came mainly from low-capacity individuals, who gained on all trained tasks. High-capacity participants showed more limited improvement. This suggests that people with lower starting capacity may benefit more from training, at least among older adults.

Research with intellectually disabled populations similarly indicates that low-capacity individuals can benefit from working memory interventions. Daniels, Wittkowski, and Riby (2015) conducted a meta-analysis of working memory training studies involving people with intellectual disabilities. Across ten studies with 28 comparisons, they found a small but significant overall effect. Notably, mixed training approaches that included both verbal and visuospatial components and emphasized strategy instruction produced medium effects, while pure visuospatial training (used in 60% of comparisons) produced non-significant effects near zero. This suggests that what you train and how you train it matters for low-capacity populations.

However, recent well-designed studies challenge simple conclusions about individual differences in training response. Soveri, Karlstedt, Wikgren, and Laine (2025) conducted a preregistered, double-blind randomized controlled trial examining visuospatial N-back training with different amounts of practice. While longer training produced bigger gains on the trained task itself—people could handle higher N-levels—the training group showed no near or far transfer effects compared to active controls. Importantly, individual differences in cognitive ability, need for cognition (enjoying effortful thinking), and beliefs about whether intelligence can grow did not predict transfer effects. Baseline WMC failed to predict who would benefit from training, despite the study having enough statistical power to detect such effects.

These conflicting findings may reflect an important distinction between training gains (getting better at the practiced tasks) and transfer effects (improvement on untrained tasks). Low-capacity individuals may show larger training gains, as Zinke et al. (2012) found, without showing broader transfer. Alternatively, the relationship between baseline capacity and training outcomes may depend on the population studied—older adults and intellectually disabled individuals may differ from the young adults typically examined in training studies.

Transfer Effects: Near and Far

The transfer question is central to both theory and practice. Near transfer means improvement on tasks closely related to those trained—like doing better on an untrained complex span task after practicing operation span. Far transfer means improvement on qualitatively different tasks—like doing better on fluid intelligence tests or measures of impulse control.

Meta-analyses provide consistent answers about transfer magnitude. Melby-Lervåg, Redick, and Hulme (2016), in a comprehensive meta-analysis, concluded that working memory training produces reliable near-transfer effects but little evidence of far transfer to fluid intelligence or academic skills. Sala and Gobet (2017) similarly found that transfer is mostly limited to tasks that share structural features with training. More recently, Ishitani and Koyasu (2024) critically re-assessed nine meta-analyses and reported that effects on fluid intelligence, executive function, and academic performance average below .20, with methodological issues like placebo effects and small samples complicating interpretation.

These meta-analytic conclusions align with recent large-scale randomized controlled trials. Ritakallio et al. (2022) developed a novel varied training protocol in which the task paradigm, stimulus type, and sequence predictability changed constantly across four weeks of training. This approach aimed to promote general strategy learning that would transfer broadly. Compared to traditional training (repetitive practice with a single adaptive task) and active controls, varied training failed to produce broader transfer. Traditional training showed strong task-specific near transfer, while varied training showed only limited near transfer on one measure. Neither group showed evidence of far transfer.

Soveri et al. (2025) similarly found no near or far transfer following N-back training, despite using a double-blind design that minimized expectancy effects. Notably, this null result held across training dosages ranging from five to twenty sessions. Even the highest dosage group, which showed substantial training gains, failed to transfer improvements to untrained tasks. These findings challenge the idea that insufficient training intensity explains previous null results.

Connecting Filtering to Training Outcomes

The filtering account of individual differences raises important questions for training research. If low-capacity individuals’ primary deficit is inefficient filtering of irrelevant information, then training that specifically targets filtering mechanisms might be more effective than general working memory practice. Some evidence supports this possibility. Schmicker et al. (2021) examined a single session of distractor inhibition training combined with transcranial direct current stimulation. They found complex interactions with baseline capacity: high-capacity individuals showed enhanced transfer following stimulation, while low-capacity individuals showed reduced transfer. The authors suggested that stimulation might have disrupted compensatory mechanisms that low-capacity individuals typically rely on.

However, most working memory training studies have not directly targeted filtering processes, and none have used CDA measures to track whether training changes filtering efficiency. This represents an important gap in the literature. Future research could combine the methodological rigor of recent training trials with the sophisticated electrophysiological measures developed by Vogel and colleagues to test whether successful training operates through improved filtering or through other mechanisms.

Discussion

The evidence on working memory training presents a puzzle. Training reliably improves performance on trained tasks, sometimes substantially. Low-capacity individuals, including older adults and intellectually disabled populations, often show particularly large training gains. Yet transfer beyond the trained tasks remains elusive, with meta-analyses consistently reporting small or null effects on fluid intelligence and other cognitive abilities.

Several explanations for this pattern deserve consideration. The skill learning account proposes that training enables acquisition of task-specific strategies rather than expanding underlying capacity (Gathercole et al., 2019). According to this view, people learn how to perform particular tasks more effectively—through chunking, better rehearsal strategies, or more efficient attention allocation—but this knowledge doesn’t generalize to structurally different tasks requiring different approaches. The limited transfer observed even in varied training protocols, which explicitly aimed to promote general strategy development (Ritakallio et al., 2022), suggests that strategy learning may be more task-bound than previously assumed.

Another explanation focuses on measurement issues. Robison et al. (2026) demonstrated that cognitive measures contain multiple sources of variance: stable trait-level variance, task-specific variance, state-specific variance, and measurement error. Training studies typically examine observed scores that mix all these sources together. If training mainly enhances task-specific skills, observed improvements may reflect increased task-specific variance rather than changes in the underlying trait of interest. Latent variable approaches that isolate common variance across multiple tasks might reveal training effects obscured in single-task analyses, though such designs require larger samples and more extensive testing than typical training studies use.

The absence of individual differences in training response, as reported by Soveri et al. (2025), carries important implications. If training effects don’t vary systematically with baseline capacity or motivational factors, this suggests that the mechanisms underlying improvement are fairly uniform across people. Alternatively, null moderation findings may reflect limited statistical power for detecting interactions, which typically require larger samples than main effects. However, the consistency of null findings across multiple potential moderators strengthens confidence in this conclusion.

Vogel and colleagues’ filtering account suggests another possibility: perhaps what matters for training outcomes is not global WMC but specific filtering ability. Individuals with poor filtering might benefit from training that specifically targets distractor exclusion, whereas those with good filtering might need different interventions. This hypothesis remains untested but could guide future research.

Methodological Considerations and Future Directions

Working memory training research has become increasingly methodologically sophisticated. Early studies often lacked active control groups, blinding, or preregistration, potentially inflating effect estimates through placebo effects and experimenter expectations. Contemporary research increasingly adheres to rigorous standards, including preregistration, double-blind designs, active control conditions, and comprehensive transfer batteries. The convergence of findings across such studies—consistent training gains without far transfer—suggests that earlier positive findings may have reflected methodological artifacts rather than genuine cognitive enhancement.

Future research should address several outstanding questions. First, what mechanisms underlie training-induced improvements? Process-oriented assessments like the CDA could track whether training changes filtering efficiency or merely improves task-specific strategies. Second, can training protocols be designed to specifically target filtering mechanisms? Vogel and colleagues’ paradigm provides a model for such interventions. Third, do training effects accumulate over multiple training periods, or do gains plateau quickly? Longitudinal studies with extended follow-up could address whether maintained practice produces qualitatively different outcomes.

Conclusion

Working memory capacity is a stable individual-differences variable that predicts diverse cognitive outcomes. Vogel and colleagues’ groundbreaking work using the CDA has fundamentally reshaped understanding of what underlies these individual differences. Rather than reflecting simple storage limits, WMC differences largely reflect variation in the ability to filter irrelevant information from entering working memory. High-capacity individuals excel at this filtering; low-capacity individuals inefficiently encode distractors alongside targets (Vogel et al., 2005).

Measurement approaches have evolved from laboratory-based complex span tasks to online instruments that enable remote assessment while maintaining psychometric quality. Working memory training reliably improves performance on trained tasks, and low-capacity individuals may show larger training gains. However, transfer to untrained tasks—particularly far transfer to fluid intelligence—remains limited and small in magnitude. Recent well-designed trials using varied training protocols have failed to demonstrate broader transfer, and individual differences do not consistently predict outcomes.

The filtering account suggests new directions for training research. Perhaps interventions that specifically target filtering mechanisms, combined with electrophysiological measures to track their effects, might produce different outcomes than the generic working memory training studied to date. For now, the evidence indicates that while working memory shows plasticity, training-induced improvements largely reflect task-specific skill acquisition rather than fundamental capacity expansion. The theoretical and practical implications of this conclusion continue to shape research agendas, with important consequences for educational and clinical interventions seeking to enhance cognitive function.