Math and verbal fluency across adulthood provide insights into aging and individual differences Authors: Yarden Gliksman, Itai Schwarz, Sharon Naparstek Published: 2026 ABSTRACT AbstractLife expectancy is increasing, leading to a growing population of older adults. Although cognitive functions decline with age, mathematical functioning in later adulthood has received relatively limited research attention. The present study examined age-related changes in mathematical fluency across all four basic arithmetic operations (addition, subtraction, multiplication, division) to determine whether aging affects these operations uniformly or whether specific operations show greater vulnerability. We also examined how individual differences contribute to math fluency in older adults, and whether mathematical and verbal fluency show similar patterns of age-related decline. Fifty-two participants (young and older adults) completed math fluency tasks using both an oral test and the Ben-Gurion University Math Fluency test (BGU-MF), a computerized paradigm measuring accuracy and reaction times. Verbal fluency was assessed using semantic and phonological verbal fluency tasks. Results showed that older adults demonstrated lower math fluency in both formats, but operations were not equally affected: division and subtraction showed the largest age differences, while addition was relatively preserved. Among older adults, education emerged as the strongest predictor of performance, while time since retirement negatively predicted accuracy and number of exercises solved. Verbal fluency also declined with age, though to a lesser extent than math fluency, pointing to a broader age-related reduction in fluency abilities alongside a heightened vulnerability of arithmetic processing. Together, these findings suggest that while fluency relies on partially shared domain-general mechanisms, mathematical fluency—particularly in more demanding operations—shows disproportionate age-related sensitivity, with consistent effects observed across computerized and oral math formats.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-41329-1.Keywords: Cognitive aging, Arithmetic operations, Computerized math fluency, Oral math fluency, Verbal sematic and phonological fluency, Individual differencesSubject terms: Psychology, Health care Introduction Cognitive aging is a complex process that affects various cognitive abilities, with significant variability depending on the type of task or information being processed1–3. Among these abilities, mathematical cognition represents a particularly important yet understudied domain. Mathematical skills are essential for daily functioning, from managing finances to making healthcare decisions, and their preservation or decline with age has significant practical implications for older adults’ independence and quality of life4–6. Despite extensive research on age-related changes in memory1 and language7–10, comparatively less attention has been paid to mathematical cognition in aging11. Mathematical fluency—the ability to rapidly and accurately perform basic arithmetic operations12–14 has been widely used to assess arithmetic competence across the lifespan, mainly among children and young adults13–16. In standard math fluency paradigms, participants are asked to solve basic arithmetic problems as quickly as possible, and fluency is quantified by the number of items solved within a limited time. However, math fluency is not a unitary construct. The four basic arithmetic operations (addition, subtraction, multiplication, and division) differ substantially in their underlying cognitive demands, making them differentially sensitive to age-related changes, as described below. Addition is considered a foundational arithmetic ability, with children as young as four years old able to perform simple addition (e.g., adding one to a given amount)17. It is typically the first operation formally taught18 and shows the least susceptibility to individual differences such as math anxiety19,20. Recent theoretical work suggests that addition of small numbers relies on automated counting procedures rather than retrieval from long-term memory21,22, which may account for its relative preservation with age. Indeed, older adults have been shown to use retrieval strategies for addition problems more frequently than younger adults23, suggesting that with extensive practice, addition solutions can be accessed efficiently from memory. Studies show that older adults maintain high accuracy in addition, though with slower response times compared to younger adults21,24. Because addition is typically mastered early and reinforced throughout schooling, it may be more resilient to age-related decline. This pattern aligns with developmental principles suggesting that earlier-acquired cognitive abilities are often better preserved in aging than skills learned later25–27). Multiplication, in contrast, is thought to rely primarily on retrieval of arithmetic facts from verbal long-term memory19,28. While older adults generally maintain high accuracy in multiplication, particularly for well-practiced problems from the multiplication table, they exhibit slower retrieval speeds29. Research suggests that older adults may be less sensitive to proactive interference in multiplication30, indicating that well-practiced operations maintain some resistance to decline. Arithmetic fact retrieval can be preserved or even enhanced in older adults21,29, though retrieval speed may decline31. This suggests multiplication may be more vulnerable to age-related slowing than addition despite both relying on retrieval processes29,32). Subtraction places greater demands on procedural and strategic processes. While simple subtraction problems may rely on retrieval, more complex problems often require computational strategies such as counting down or decomposition (e.g., 32 − 9), which depend heavily on working memory and executive control33,34. Research indicates that age impacts subtraction fluency differently depending on the complexity and strategy used: older adults are faster when using retrieval-based processes but slower when employing strategies requiring working memory33,34. This pattern suggests that age-related declines in working memory and executive functions disproportionately affect operations that cannot be solved through simple fact retrieval. Division remains the least studied operation in the aging literature, despite being considered one of the four basic operations. Division is unique in its heavy reliance on both computational procedures and executive functions, as it often requires multi-step problem-solving, estimation, and integration of multiplication knowledge. It was found that division performance depends critically on multiplication retrieval, suggesting potential vulnerabilities due to its procedural complexity and working memory demands35. Given its complexity and limited automatization compared to other operations, division may be particularly sensitive to age-related cognitive decline, yet empirical evidence examining division in older adults remains scarce. The distinct cognitive profiles of these operations suggest that aging may not affect mathematical fluency uniformly. Rather than a generalized decline, age-related changes may be operation-specific, with greater preservation of operations relying on early acquired, automated procedures (addition) and greater vulnerability in operations requiring working memory, executive control, or complex retrieval (subtraction, division, and possibly multiplication). Examining all four operations within a unified paradigm is therefore essential to determine the nature of age-related changes in mathematical fluency. Assessment methods and task form at considerations An additional consideration in measuring mathematical fluency concerns the format of assessment. Traditional assessments of numerical fluency, such as the Woodcock-Johnson Battery15,36, rely on paper-and-pencil formats. While widely used, these tools are limited by potential biases in administration and scoring, as well as their inability to provide objective measures like reaction times—which are particularly relevant for understanding age-related changes in processing speed37. Computerized tasks can address these limitations by providing precise timing measurements and standardized presentation13,39,40. Computerized formats may however, introduce confounds for older adults, due to unfamiliarity with keyboards and reduced motor speed, which could artificially inflate age-related differences in timed tasks. Studies examining mathematical cognition in aging have noted that task format can influence performance patterns, with production tasks (generating answers) typically showing greater age effects than verification tasks (judging correctness) due to differences in retrieval and response demands29,33. Comparing computerized and oral formats can help disentangle genuine arithmetic difficulties from format-specific confounds, providing a clearer picture of age-related changes in mathematical fluency itself. Individual differences in mathematical fluency among older adults As with many cognitive abilities, mathematical fluency is influenced by individual differences, with education and executive functions identified as key factors. Higher educational attainment is one of the most consistent predictors of preserved cognitive functioning in later life, with large-scale evidence showing that more years of education are associated with better numerical performance and reasoning efficiency in aging39,40. Education enables individuals to compensate for cognitive decline through more efficient processing strategies39,41. Executive functions (EFs) are higher-order cognitive processes that support goal-directed behavior, typically peaking in young adulthood before declining with age42,43. These functions are critical for arithmetic fluency, particularly when rapid retrieval, inhibition of irrelevant responses, or flexible switching between strategies is required44. Classic work in mathematical cognition demonstrates that older adults show slower and more variable performance when solving problems that rely on executive control, such as strategy selection or multi-step computation, relative to tasks relying on simple fact retrieval29,33. Consequently, age-related reductions in EF may contribute to declines in math fluency, and individual differences in EF preservation likely explain why some older adults perform substantially better than others, despite similar levels of schooling. Recent work has also suggested that time since retirement might play a role in cognitive aging45, though its impact remains debated. Some studies suggest that retirement accelerates cognitive decline, supporting the “use it or lose it” hypothesis46,47, while others report more modest effects48,49. Research specifically examining math fluency post-retirement is limited, but evidence indicates that numeracy may decline following retirement, with potential consequences for financial decision-making and everyday arithmetic performance6,45. Understanding how these individual difference factors—education, executive functions, and time since retirement—contribute to variability in mathematical fluency can inform targeted interventions and identify older adults at risk for arithmetic difficulties. Age-related sensitivity: mathematical versus verbal fluency Verbal fluency tasks, which require generating words under specific constraints, provide an important comparison for understanding the breadth of age-related changes in fluency abilities. Fluency is often viewed as an umbrella construct referring to the rapid and efficient generation of information, yet the specific cognitive demands involved depend strongly on task structure. Thus, while both mathematical and verbal fluency require rapid, efficient retrieval and generation of information under time pressure14,50, they differ in the nature of the responses they elicit and the processes they engage. Mathematical fluency typically involves producing a single correct response within a highly constrained context, whereas verbal fluency relies on generative search processes with multiple acceptable responses. Verbal fluency tasks differ in the cognitive functions involved depending on task demands. When instructed to produce as many words as possible, phonemic fluency (generating words beginning with specific letters) relies primarily on access to the phonological lexicon and strategic search processes, whereas semantic fluency (generating words within categories) depends on the organization of semantic memory and the coordination between conceptual and lexical representations7–10. Research shows that both phonemic and semantic fluency decline with age, though semantic fluency typically shows greater age-related decline, likely due to the deterioration of connections between conceptual and phonological networks10. These findings demonstrate that even within verbal abilities, different task requirements can lead to different patterns of age-related change. Comparing age effects across mathematical and verbal fluency can illuminate whether age-related differences in arithmetic reflect broader mechanisms affecting multiple cognitive abilities or vulnerabilities specific to numerical processing. Previous research has documented age-related slowing in processing speed51,52 and changes in retrieval efficiency53, but different cognitive abilities may be differentially sensitive to these changes. If mathematical fluency shows comparable age effects to verbal fluency, this would suggest shared underlying mechanisms such as general processing speed or executive control. However, if mathematical fluency shows disproportionate decline, this would point to factors more specific to maintaining arithmetic skills, such as reduced engagement with numerical operations in daily life or particular vulnerabilities in numerical processing systems. By assessing both mathematical and verbal fluency in the same individuals, we can determine whether age-related changes reflect domain-general mechanisms affecting multiple abilities or domain-specific patterns tied to the nature of numerical cognition54,55. The current study The current study investigated age-related differences in mathematical fluency using a comprehensive, multi-method approach. The first goal was examining performance across all four basic operations (addition, subtraction, multiplication, and division), to determine whether aging affects all operations similarly or whether operation-specific cognitive demands lead to differential patterns of preservation and decline. Math fluency was assessed using a computerized task (BGU-MF), which recorded both accuracy and reaction times for single-digit arithmetic problems. An oral version of the same test was included to avoid confounding effects of keyboard familiarity or motor speed. The second goal was to examine how individual differences among older adults, contribute to variability in math fluency. Specifically, we measure education (i.e., “how many years of education”), retirement (i.e., “how many years passed since retirement” as a continuous measure), and EFs using a designated self-report questionnaire (the Behavioral Rating Inventory of Executive Functions; BRIEF56. The third goal of this research was to ask whether age-related differences in math fluency reflect a domain-general decline or a domain-specific vulnerability in arithmetic. To address this goal participants performed an additional test of fluency abilities, verbal fluency test. Verbal fluency was assessed using phonological and semantic word-generation tasks. Our questions and hypotheses were: Do young and older adults differ in their performance in math fluency? To address this question, young and older adults completed the computerized BGU-MF (Ben-Gurion University Math Fluency) test13,14,19. Following previous studies, we hypothesized that math fluency would be affected by age. Specifically, we predicted that: (a) RTs would be affected to a larger extent compared to accuracy; (b) Performance would differ by operation, with addition being more preserved due to its reliance on automated counting procedures21,22 and early acquisition26,27, while operations requiring greater working memory demands and retrieval from long-term memory (subtraction33,34; multiplication29,31; and division35 would show greater age-related decline; and (c) The difference between young and older adults would be smaller in the oral math fluency format, due to age-related differences in computer-use. Do young and older adults differ in their performance in math fluency? To address this question, young and older adults completed the computerized BGU-MF (Ben-Gurion University Math Fluency) test13,14,19. Following previous studies, we hypothesized that math fluency would be affected by age. Specifically, we predicted that: (a) RTs would be affected to a larger extent compared to accuracy; (b) Performance would differ by operation, with addition being more preserved due to its reliance on automated counting procedures21,22 and early acquisition26,27, while operations requiring greater working memory demands and retrieval from long-term memory (subtraction33,34; multiplication29,31; and division35 would show greater age-related decline; and (c) The difference between young and older adults would be smaller in the oral math fluency format, due to age-related differences in computer-use. How would individual differences affect math fluency among older adults? We hypothesized that more years of education, less time since retirement, and more preserved EF would be positively correlated with performance in the math fluency test. How would individual differences affect math fluency among older adults? We hypothesized that more years of education, less time since retirement, and more preserved EF would be positively correlated with performance in the math fluency test. Are math and verbal fluency similarly sensitive to age-related differences? We hypothesized that performance in verbal and math fluency tests would be positively associated50, suggesting fluency as an umbrella term across domains. However, we expected math fluency to show stronger age-related differences than verbal fluency53,57, suggesting domain-specific vulnerabilities in arithmetic rather than purely domain-general aging effects. Are math and verbal fluency similarly sensitive to age-related differences? We hypothesized that performance in verbal and math fluency tests would be positively associated50, suggesting fluency as an umbrella term across domains. However, we expected math fluency to show stronger age-related differences than verbal fluency53,57, suggesting domain-specific vulnerabilities in arithmetic rather than purely domain-general aging effects. Results Computerized math fluency Number of exercises Young adults solved more exercises than older adults (M = 105.5, SD = 22.2 vs. M = 45.3, SD = 11.3) t(46)= -12.3, p < .001, Cohen’s d=-3.422. The effect of operation was examined by a 4 (operation) X 2 (group) analysis of variance (ANOVA) with accuracy rates and RTs as dependent variables. Accuracy rates analysis Due to ceiling effects in accuracy rates (overall M = 95.8%), accuracy data were analysed using a binomial generalized linear model (GLM) with logit link function rather than ANOVA, as recommended for proportion data with restricted range58. The model included group and operation as fixed factors, with the number of correct and incorrect responses as the dependent variable weighted by total trials per participant. The binomial GLM revealed a significant main effect of group, χ²(1)= 21.90, p < .001, with young adults (M = 97.7%, SE = 0.3%) more accurate than older adults (M = 94.8%, SE = 0.6%) overall. There was no significant main effect of operation, χ²(3)= 4.73, p = .193. A significant group × operation interaction was found, χ²(3)= 9.08, p = .028. Planned comparisons revealed that older adults showed significantly lower accuracy than young adults in multiplication (92.7% vs. 97.9%, z = -3.76, p < .001), subtraction (93.9% vs. 97.9%, z = -3.15, p = .002), and division (95.9% vs. 98.6%, z = -2.56, p = .010), but not in addition (96.7% vs. 96.4%, z = 0.22, p = .830). Among older adults, multiplication showed the lowest accuracy, though the difference from other operations did not reach statistical significance after correction for multiple comparisons (multiplication vs. addition: z = 2.12, p = .102). These findings indicate that while older adults maintain high accuracy levels overall, age-related deficits are most pronounced in multiplication, followed by subtraction and division, with addition being relatively preserved. See Fig. 1A. (A) Accuracy rates as a function of group and operation. (B) RTs as a function of group and operation. * p < .05; ** p < .01; *** p < .001; n.s. – non significant; s – seconds. RTs analysis There was a main effect of group, as young adults were faster (M = 1.3 s, SD = 0.3) than older adults (M = 2.6 s, SD = 0.5), F(1,46) = 99.2, p < .001, = 0.7. Interaction between group and operation was significant, F(3,150) = 5.6, p = .001, = 0.1. Planned comparisons revealed that young adults were faster than old adults in all types of operations: Addition: (M = 1.2 s, SD = 0.3 vs. M = 2.2 s, SD = 0.6) t(46)= 7.1, p < .001; Subtraction: (M = 1.5 s, SD = 0.5 vs. M = 2.7 s, SD = 0.8) t(46)= 8.8, p < .001; Multiplication: (M = 1.3 s, SD = 0.2 vs. M = 2.4 s, SD = 0.5) t(46)= 8.1, p < .001; Division (M = 1.4 s, SD = 0.3 vs. M = 2.9 s, SD = 0.6) t(46)= 10.4, p < .001. The difference in RTs between groups was larger in division compared to addition, t(46)= 6.68, p = .01. Moreover, a significant difference between groups between operations of addition and multiplication was found, t(46)= 4.1, p = .04. Only older adults presented significantly shorter RTs in addition compared to multiplication, t(46)= 2.9, p = .03, while young adults did not, t(46)= 1.1, p = .6. The effect of group in RTs on subtraction and division was not significant. A main effect of operation was found, F(3,150) = 28.1, p < .001, = 0.4. Planned comparisons revealed that addition was significantly faster (M = 1.7 s, SD = 0.7) than subtraction (M = 2.1 s, SD = 0.9), multiplication (M = 1.8 s, SD = 0.7) and division (M = 2.1 s, SD = 0.9), t(46)=-7.1, p < .001; t(46)=-2.8, p = .01; t(46)=-7, p < .001; respectively. Subtraction was significantly faster than multiplication but had no difference with division, t(46)= 4.3, p < .001; t < 1, respectively. Multiplication was significantly faster than division t=-5.2, p < .001. See Fig. 1B. Because older adults were overall slower across conditions, we ran a post-hoc ANCOVA to examine whether group effects persisted after adjusting for general slowing. Following previous studies59,60, who controlled age-related slowing using baseline RT in single-pure trials as a covariate, we included RT in the simplest condition (addition RT) as a baseline speed estimate. The analysis included the three remaining operations (subtraction, multiplication, and division) in a 3 (operation) × 2 (group) repeated measures ANCOVA. The covariate (addition RTs) was significant, F(1,45)= 136.46, p < .001, indicating that baseline processing speed significantly predicted performance across operations. Critically, even after controlling for baseline RT, a significant main effect of group remained, F(1,45)= 18.05, p < .001, =0.27, demonstrating that age-related differences in mathematical fluency persist beyond general processing speed differences. The interaction between group and operation remained significant, F(2,98) = 11.55, p < .001, =0.19. A main effect of operation was also found, F(2,98) = 11.28, p < .001. To provide converging evidence that the group differences reflect mathematical abilities rather than switching demands or greater reliance on fast, rule-based processing61–63, we conducted two additional post-hoc analyses examining switching costs and performance on rule-based items. To examine switching-cost each trial was coded as a repeat (same operation as the previous trial) or switch (different operation) trial, analyzing only sequences of two consecutive correct responses. RTs for the second trial in each sequence were submitted to a 2 (group: young vs. older adults) × 2 (condition: repeat vs. switch) mixed ANOVA. Results showed a significant main effect of group, F(1,50)= 151.29, p < .001, = 0.72, confirming that older adults were slower overall. However, there was no significant main effect of switching, F(1,50)= 0.83, p = .367, and no group × condition interaction, F(1,50)= 1.54, p = .220. Switch costs (difference between switch and repeat trials) were minimal and non-significant in both older adults (M = 0.11 s, t(23)= 1.05, p = .304) and young adults (M = -0.02 s, t(27)= -0.59, p = .563). These findings suggest that the observed age-related slowing in our math fluency task cannot be attributed to impaired cognitive flexibility or switching costs, and instead likely reflects genuine differences in mathematical processing. To examine rule-based processing we repeated the group × operation ANOVA including only non–rule-based trials. Results closely mirrored the full analysis, with significant main effects of group, F(1,50)= 94.92, p < .001, = 0.556, operation, F(3,162) = 24.40, p < .001, = 0.115, and a significant interaction, F(3,162) = 4.87, p = .003, = 0.025. Planned comparisons confirmed that older adults were slower than young adults across all operations (all p < .001). Large effect sizes (all d > 1.9) indicate that age-related RT differences in mathematical fluency remain robust when restricting the analysis to standard, non–rule-based exercises. Math fluency formats: computerized vs. oral Young and old adults solved exercises correctly in the oral math fluency format, (M = 75.4, SD = 35.1; M = 46, SD = 8.6, respectively. The effect of format (computerized vs. oral) was examined by a 2 (format) X 2 (group) ANOVA with accuracy rates and number of correctly solved exercises as dependent variables. Accuracy rates analysis There was a main effect of format, as accuracy rates were higher for the oral format (M = 97.8%, SD = 4) than the computerized format (M = 95.9%, SD = 3.6), F(1,46)= 9.4, p = .003, = 0.2. There was a main effect of group, as accuracy rates were higher for young adults (M = 98.3%, SD = 1.2) than old adults (M = 95.4%, SD = 3.4), F(1,46)= 16.2, p < .001, = 0.2. No significant interaction was found, F < 1. Number of exercises There was a correlation in performance between oral and computerized math fluency formats, r=.81, p<.001, ensuring that both tasks examine similar abilities. There was a main effect of format, more exercises were solved for the computerized format (M = 75.4, SD = 35.1) than the oral format (M = 46, SD = 8.6), F(1,46)= 202, p < .001, = 0.8. There was a main effect of group, as young adults (M = 78.7, SD = 13.2) solved more exercises than old adults (M = 42.6, SD = 7.9), F(1,46)= 142.1, p < .001, = 0.7. Format and group significantly interacted, F(1,46)= 136.6, p < .001, = 0.7. Planned comparisons revealed that young adults solved more exercises than old adults in the computerized version (M = 105.5, SD = 22.2 vs. M = 45.3, SD = 11.3) t=-16.4, p < .001; and in the oral format (M = 51.9, SD = 5.9 vs. M = 40, SD = 6.6), t=-3.2, p=.005. The difference between groups between formats was higher in the computerized format, t(46)= 11.9, p<.001, see Fig. 2. Interaction between fluency type and group. *** p < .0001; ** p < .001; * p < .001 n.s. – non-significant. To address concerns that rule-based items may influence performance between formats, we examined their proportion in both the oral and computerized BGU-MF tests. Rule-based items represented 19.1% of the computerized test and 28.3% of the oral test, with no significant difference between formats, χ²(1)= 2.65, p = .103, indicating similar representation across the two versions. We further examined difference between groups in the oral math format analyzing only non-rule correct items. Analysis included 51 participants (25 older adults, 26 young adults; one older participants’ oral data was unavailable). Young adults correctly solved significantly more non-rule items (M = 41.04, SD = 5.20) than older adults (M = 30.57, SD = 5.22), t(45)= -7.03, p < .001, d = 2.01. Results suggest that age-related math fluency deficits persist in the oral test format cannot be attributed to rule-based items. Individual differences among old adults To examine the role of individual differences on math fluency in aging, we conducted correlations between factors, and hierarchical regression analyses separately for the number of exercises, accuracy rates, and RTs. Each dependent variable in the BGU-MF was modeled using a set of predictors, including age, time since retirement, education, and the scaled global score (Global Executive Composite, GEC) from the BRIEF questionnaire. We evaluated the incremental contribution of these predictors across models and assessed multicollinearity using Variance Inflation Factor (VIF). Correlations Analysis revealed a significant negative correlation between accuracy rate and retirement, r=-.4, p=.05. Moreover, a significant correlation was found between the number of exercises to education r=.4, p = .03; and to GEC, r=-.4, p = .05. No other significant correlations were found. Number of exercises In the first regression model, age and retirement were included as predictors. This model accounted for 16.6% of the variance (R² = 0.166, adjusted R² = 0.093, p = .125), though neither age (β = 0.370, p = .088) nor retirement (β = -0.371, p = .087) reached statistical significance. Adding education in the second model significantly improved the model fit (R² = 0.330, adjusted R² = 0.238, p = .029). Education emerged as a significant positive predictor (β = 0.428, p = .030), while retirement became a significant negative predictor (β = -0.410, p = .043), indicating that individuals who had been retired for a longer period exhibited lower performance in math fluency. Age remained non-significant (β = 0.252, p = .214). In the final model, GEC was added to account for executive function contributions. However, this model did not significantly improve explanatory power (R² = 0.331, adjusted R² = 0.190, p = .091), and GEC was not a significant predictor (β = -0.100, p = .678). Education was not statistically significant in the final model (β = 0.358, p = .109), and time since retirement remained in the expected negative direction but was no longer significant (β = -0.377, p = .131). Multicollinearity checks confirmed that all predictors had acceptable VIF values (< 1.6), indicating no severe multicollinearity concerns. Accuracy rates For accuracy rates, the first model with age and retirement was marginally significant and explained 22.2% of the variance (R² = 0.222, adjusted R² = 0.154, p = .056). Retirement emerged as a significant negative predictor (β = -0.500, p = .020), while age was not significant (β = 0.307, p = .140). Adding education in the second model increased explanatory power (R² = 0.232, adjusted R² = 0.128, p = .114), but education itself was not a significant predictor (β = 0.108, p = .589), and retirement remained a significant negative predictor (β = -0.510, p = .021). Adding GEC in the final model did not significantly improve the model (R² = 0.226, adjusted R² = 0.097, p = .226), and none of the predictors were significant (all p > .05), although the coefficient for retirement remained in the same negative direction as in previous models (β = -0.536, p = .048). RTs The first regression model, including age and retirement, explained only 6.6% of the variance (R² = 0.066, adjusted R² = -0.015, p = .454), with neither predictor reaching significance (age: β = -0.281, p = .214; time since retirement: β = 0.114, p = .609). When education was added, the model slightly improved (R² = 0.222, adjusted R² = 0.115, p = .131), with education emerging as a significant negative predictor (β = -0.416, p = .048), suggesting that higher education levels were associated with faster RTs. The final model, incorporating GEC, did not improve explanatory power (R² = 0.220, adjusted R² = 0.056, p = .292), and none of the predictors remained significant (education: β = -0.332, p = .165; age: β = -0.132, p = .600; retirement: β = 0.102, p = .698; GEC: β = 0.151, p = .562). This suggests that EFs did not independently contribute to response time variance beyond education. Summary of regression models Across all three measures, education consistently emerged as the strongest predictor of math fluency, particularly for the number of exercises and RTs, where higher education was associated with greater accuracy and faster performance. Time since retirement negatively predicted the number of exercises and accuracy rates, suggesting that longer retirement duration may be linked to performance declines. However, age alone was not a significant predictor in any model, indicating that time since retirement may be a more sensitive indicator of changes in numerical abilities than chronological age. GEC did not significantly contribute to any of the models, suggesting that executive function, as measured by the BRIEF questionnaire, does not independently predict numerical abilities in this sample. Math and verbal fluency In the phonological fluency task, young adults generated more words than older adults across all three letters: /b/ (M = 13.44, SD = 4.03 vs. M = 9.88, SD = 3.79), /g/ (M = 13.04, SD = 3.99 vs. M = 11.42, SD = 4.05), and / š / (M = 13.64, SD = 4.29 vs. M = 12.85, SD = 4.94). In the semantic fluency task, young adults also outperformed older adults in animals (M = 25.40, SD = 7.04 vs. M = 18.35, SD = 4.34) and fruits and vegetables (M = 22.32, SD = 4.85 vs. M = 19.88, SD = 4.38), while performance was comparable in vehicles (M = 13.48, SD = 3.61 vs. M = 13.58, SD = 2.72). Performance was in the normal range for both age groups10. We calculated for each participant the total number of correct items in each fluency (phonological and semantical). In order to examine the association between different fluency tasks, correlations between the number of solved items in computerized math fluency and phonological fluency and semantic fluency were conducted. Significant correlations emerged between both math fluency (computerized, oral), and verbal fluency (phonological, semantic) tasks, as well as between the computerized math fluency task and the phonological and semantic fluency tasks. See Table 1. Table 1.Variable12341. Computerized Math–2. Oral Math0.808***–3. Verbal Phonological0.274*0.162–4. Verbal Semantic0.386**0.2540.660***–Values are Pearson correlation coefficients (two-tailed). *p < .05, **p < .01, ***p < .001. . Values are Pearson correlation coefficients (two-tailed). *p < .05, **p < .01, ***p < .001. Next, a 4 (fluency: computerized math vs. oral math vs. phonological vs. semantic) X 2 (group) ANOVA with number of correct items as dependent measures was conducted. A main effect of fluency was found, F(3,150) = 113.99, p< .001, = 0.695. Planned comparisons showed differences between all fluency types, as computerized math fluency (M = 75.4, SD = 35.07) was significantly higher than oral math (M = 45.98, SD = 8.63; t(46)= 13.58, p < .001, d = 1.88), semantic (M = 56.31, SD = 12.32; t(46)= 8.8, p < .001, d = 1.2), and phonological fluency (M = 37.4, SD = 11.56; t(46)= 17.5, p < .001, d = 2.4). Oral math exceeded both semantic (t(46)= 4.77, p < .001, d = 0.66) and phonological fluency (t(46)= 3.96, p < .001, d = 0.55), and semantic fluency was also higher than phonological (t(46)= 8.72, p < .001, d = 1.21). There was also a significant main effect of age group, F(1,46)= 103.06, p < .001, = 0.673, reflecting higher fluency scores overall in younger adults (M = 64.01, SD = 19.28) compared to older adults (M = 42.5, SD = 16.23). The interaction was significant, F(3,150) = 70.131, p < .001, = 0.584. Planned comparisons confirmed that younger adults outperformed older adults in math fluency conditions, in both formats, computerized math fluency, (M old = 45.27, SD old = 11.27; M young = 105.5, SD young = 22.22; t(46)= 17.62, p < .001, d = 5.63), and oral math, (M old = 40.04, SD old = 6.64; M young = 51.92, SD young = 5.87; t(46)= 3.47, p = .008, d = 0.35). However no differences were found in semantic fluency (M old = 51.81, SD old = 8.69; M young = 60.81, SD young = 13.85; t(46)= 2.63, p = .08, d = 0.03), or in phonological fluency, (M old = 34.15, SD old = 10.56; M young = 40.65, SD young =;11.8; t(46)= 3.2, p = .37, d = 0.22. See Fig. 2. Discussion The aim of the current study was to examine the effects of aging over mathematical fluency across all 4 operations, and how this ability is affected by individual difference. We further examined whether the effect of age is similar across domains by comparing the performance of the same individuals on the math and verbal fluency tasks. Let us summarize our main results: 1. Compared to young adults, old adults presented overall lower math fluency as exhibited by a lower number of exercises solved, lower accuracy rates and higher RTs. 2. Addition was the operation least affected by age. 3. Among old adults, years of education significantly predicted math fluency. 4. Across age, math and verbal fluencies were correlated. 5.Compared to young adults, older adults presented lower verbal fluency for both semantic and phonological tasks, yet these differences were less pronounced compared to differences in math fluency. Age-related deterioration in math fluency The results of deterioration in performance in math fluency in older adults are in line with our hypothesis and replicated previous finding regarding addition21,24, subtraction33,53, and multiplication29. Our current findings extend these findings to division and suggest deterioration occurs in all four operations. Comparing math fluency performance within and across all four basic operations revealed more subtle age-related patterns. As expected, we found that addition was the least affected by age. Addition skills are considered a congenital mechanism64. Children at the age of 4 years-old are able to perform simple addition exercises (i.e., add one to a given amount65. Addition preservation aligns with developmental principles of cognitive aging: earlier-acquired abilities tend to be more resilient to age-related decline than later-acquired skills25–27. Addition is the first arithmetic operation children master and the first formally taught18, providing a foundation of lifelong practice. This early acquisition and extensive practice may explain why addition fluency is relatively preserved in older adults while later-learned operations like division show greater vulnerability27,66. Moreover, addition in older adults relies heavily on automated counting procedures21–23 rather than effortful retrieval from long-term memory, which may further contribute to its preservation. Interestingly, in the current study, older adults exhibited longer RTs in addition compared to younger adults; however, their RTs were shorter than those of sixth graders who completed the same task in our previous work13. Given that experience enhances arithmetic fact retrieval67, these RT patterns demonstrate that addition fluency continues developing from childhood through young adulthood, reaching peak performance before showing gradual decline in older adults. In old adults, the repeated exposure and experience accumulated over time enables to compensate for age-related slower processing speed and reduced attention68. Similar results regarding preserved semantic math knowledge in aging were recently reported69. Regarding division, the current study is the first to examine its performance among old adults. Results showed significantly lower accuracy rates and longer RTs compared to young adults. Whereas these differences were also present for subtraction and multiplication, they were largest for division. Given that division relies heavily on working memory and executive functions and processing speed, it is possible that age-related declines in these domains contribute to these differences. Interestingly, division was also the most affected operation among young adults with high levels of math anxiety19. Older adults exhibit general slowing37,53 and may also have less keyboard experience than younger adults. These differences are inherent to aging research and must be considered when interpreting RT data in our computerized task. In the present study, however, converging evidence suggest that the observed age effects cannot be attributed solely to general processing speed or motor familiarity. First, when using the addition operation as a baseline for processing speed, age-related differences remained robust even after statistically controlling for this factor. Second, the pattern of results was replicated in the oral math fluency task, in which responses were spoken rather than typed, eliminating keyboard demands altogether. The convergence of these findings strengthens the conclusion that the age effects observed here reflect genuine differences in mathematical fluency rather than artifacts of motor slowing. Nevertheless, to further sharpen the distinction between domain-general and numerical processes, future studies could incorporate a simple non-numerical speed task or assess computer-use experience as complementary measures. Additionally, in the computerized fluency task, there were no differences in the switching-cost between younger and older adults, suggesting preserved flexibility despite slower responses. This pattern is consistent with meta-analytic findings indicating that age effects are more pronounced in global switching (i.e., the overall cost of operating in a mixed-task context compared to single-task blocks) than in local switching (i.e., the trial-by-trial cost of switching versus repeating within a mixed block)3,70. The use of computerized versus oral tests in older adults Comparison of math fluency in both computerized and oral formats revealed that age differences were smaller in the oral format. One possible explanation for this difference is that older adults are less experienced in computer use which might negatively affect their performance71. Alternatively, it is possible that these differences originate from the fact that verbal abilities are relatively preserved in aging, while speed of motor responses declines72. As the two tests differ in response modes, these differences might account for the smaller age effect in the oral test. Using the computerized math fluency test enabled to reveal age-related differences in RTs per exercise and operation, rather than focusing only on differences in accuracy rates and number of correct exercises, as measured in other tests15,36. As age affected mostly RTs, measuring speed of response is highly important. As the use of computerized tests and measures is gaining popularity in health and mental health37,73, these findings have important implications on its application in the field of geriatrics. Specifically, researchers and clinicians in the field should consider that while computerized measures enable to reveal subtle age-related differences otherwise overlooked, they are also prone to effects of motor speed that is abrupted in old adults. Education and time since retirement affect math fluency The present study also examined the role of individual differences in fluency, focusing on age, education, executive functions, and time since retirement on math fluency in older adults. Across all measures, education emerged as the strongest predictor, with higher educational attainment linked to better performance in accuracy and response times. This aligns with cognitive reserve theory39,41,74, suggesting that education supports efficient processing strategies that help sustain numerical abilities despite aging40,75. However, even in our highly educated sample, older adults showed declines, indicating that education does not fully compensate for age-related changes. These declines may be even more pronounced in populations with lower education levels53,76. Notably, we did not collect information on participants’ academic field or degree type, limiting our ability to separate the effects of domain-specific training (e.g., STEM) from total years of education. Although a few reports suggest possible spatial advantages among adults with STEM backgrounds77, evidence remains sparse and inconsistent, while large-scale analyses more consistently emphasize years of education as a key predictor of late-life cognitive performance40. Population-based findings even indicate no broad numeracy or problem-solving advantage for STEM degree holders over non-STEM peers78. Thus, while field may shape specific skills, our results and the broader literature suggest that education duration is a more robust driver of math-fluency outcomes in aging. Future studies should incorporate educational specialization to determine whether math-intensive backgrounds yield additional benefit beyond schooling years. Time since retirement negatively predicted math fluency performance, suggesting that longer retirement duration may contribute to declines in math fluency. These findings support the “use it or lose it” hypothesis46,48, which suggests that disengagement from cognitively demanding tasks can accelerate decline. While research on retirement and cognition has yielded mixed results49, our findings highlight numerical abilities as a potentially vulnerable domain in retired individuals. In contrast, self-reported executive function (GEC) didn’t predicted performance, suggesting that life-course factors such as education and retirement status may be more sensitive indicators of numerical ability than aging alone. The pattern of results highlights the importance of maintaining engagement with numeracy in later adulthood. Monitoring numerical skills and offering opportunities for adults to practice everyday arithmetic may support functional numeracy, particularly in activities such as financial and medical decision-making4. Tools such as the BGU-MF test could be valuable assessing numeracy in aging populations. Future research could investigate whether executive functions mediate the relationship between education and numerical performance and whether targeted numeracy-focused programs or activities might help support numerical skills in later life. What drives the age-related differences in math fluency? Processing speed slowing is well-established in aging51, but if this were purely a global process affecting all cognitive domains equally, it should impact mathematical and verbal abilities similarly, and within mathematics, affect all operations uniformly52. Our results are partly in-line with this explanation as old adults were slower in all tasks, but the difference was smaller in addition exercises, suggesting operation-specific vulnerabilities beyond general slowing. To directly address whether general processing speed fully accounts for these differences, future studies should incorporate a pure non-numerical speed measure as a baseline comparison. A second potential mechanism involves executive functions. Among older adults, all participants reported intact EF performance (T < 65). In line with previous work, EFs negatively correlated with math fluency79. However, EF’s predictive power diminished in regression models when considered alongside education and retirement status. This pattern suggests that EFs contributions to numerical accuracy may be weaker than those of life-course factors such as education and cognitive engagement. Notably, EFs were assessed via self-reports which may not capture the full range of executive abilities as sensitively as objective measures. Future studies should examine EF abilities using objective behavioral tasks measuring RTs and accuracy which may more sensitively detect EF-math fluency associations and reveal age-related changes not apparent in self-report data37,80. Vulnerability to fluency across domains Verbal fluency performance of both young and old adults was within normal range10. As expected, younger adults outperformed old adults in both semantic and phonological tasks, however this difference did not reach significance. In the math fluency task, differences between age groups were highly significant suggesting that decline in math performance in older adults was much greater than that of verbal performance. However, these age-related differences might be explained not only by the domain examined (i.e., linguistic versus numerical) but also by the task demands in each domain. Specifically, math fluency was examined employing a constrained task, as only one specific response is required for each exercise. Verbal fluency, in contrast, was examined using a less constrained task, as each exercise (category) required several responses. Previous work has suggested that age-related deficits in cognitive flexibility affect constrained tasks to a larger extent81. Thus, it is possible that older adults performed worse due to the nature of the task and not the domain examined. Future work could examine the combined effects of task demands and domain, for example by designing semantic fluency tasks with more constrained responses, to better match the structure of math fluency tasks. The correlation between math and verbal fluency contributes to the ongoing discussion on shared versus unique mechanisms in cognitive aging54,55. Although modest, the correlations in the current study suggest that different fluency tasks share some underlying cognitive mechanisms, consistent with the umbrella term for rapid and efficient information generation. On one hand, the observed correlation suggests a common fluency mechanism that extends beyond task-specific demands or type of information being processed, across fluency tasks. Similar findings in children further support this idea, demonstrating associations between verbal and mathematical fluency50. On the other hand, the more pronounced age-related decline in math fluency compared to verbal fluency highlights distinct mechanisms that may contribute uniquely to each domain53,57. Taken together, these findings suggest that math fluency involves both shared cognitive processes—such as executive control, processing speed and retrieval efficiency—as well as domain-specific mechanisms tied to numerical cognition82,83. From a developmental perspective, it is possible that the shared fluency mechanism plays a more prominent role in childhood and young adulthood, while aging introduces distinct patterns of decline in numerical abilities due to the increasing impact of processing speed and working memory limitations. Conclusion The aim of this study was to examine mathematical fluency in aging and the effect of age on mathematical versus verbal fluency. The findings demonstrate that aging affects these domains differently, with mathematical fluency showing a greater decline. However, even within mathematics, not all operations are equally impacted—addition appears to be more resilient to aging, while division is more vulnerable, particularly in response times. These findings highlight the complexity of cognitive aging and suggest that different numerical and linguistic abilities rely on distinct mechanisms that may be differentially affected over time. Future research should further investigate how life experiences, such as education and cognitive engagement, contribute to healthy cognitive aging and whether targeted interventions can mitigate age-related declines in fluency. Methods Participants Fifty-two subjects participated in the study. Young adults (N = 26, 14 females, aged 21–25 years old, M = 23.1, SD = 1; years of education: M = 12.5, SD = 0.1) were recruited from Ben-Gurion University of the Negev and participated in the study for course credit. Old adults (N = 26, 14 females, aged 66–74, M = 69.9, SD = 2.9; years of education: M = 14.9, SD = 0.5), free of any motor, cardiovascular, or neurological disease, were recruited via advertisements in the local community, and participated in the study for payment (approximately $10 per hour). We calculated the statistical power using G*Power version 3.1.984 to calculate a sample-size determination85, in ANOVA repeated measure, within-between factors. Following previous study86, we chose effect size specification as in Cohen (1988)87. With an alpha of 0.05, of 0.25, power of 0.8, 2 groups, and 8 number of measurements across different analyses, the recommended sample size was 46 participants. Participants were naive as to the purpose of the study. The study follows the guidelines of the protocol approved by the Ethics Committee of Ruppin Academic Center (Clinical Trial Number 2020–62 L/Lb). All methods were carried out in accordance with relevant guidelines and regulations. Informed consent was obtained from all participants prior to the experiment. The data presented in the manuscript is anonymous. Apparatus Math fluency Oral math fluency test The oral math fluency test includes math fact exercises similar to the ones presented in the BGU-MF. In each trial, the experimenter reads out loud one exercise and participants are instructed to answer the exercise orally. Participants are instructed to answer as accurately and quickly as possible. The test lasts 180 s. For each participant, the number of solved exercises and accuracy are recorded (see Supplemental Material). Computerized - BGU-MF The BGU-MF test is a computerized math fluency test14. It includes simple math exercises of the four operations: addition, subtraction, multiplication, and division. The addition and subtraction exercises use numbers 0–20 and answers range 0–20 (e.g., 7 + 6). Multiplication and division exercises are taken from the multiplication table up to 10 × 10 (e.g., 56 ÷ 7). Task included both non-rule and rule-based exercises (i.e., addition and subtraction with operand 0; multiplication with operand of 0 or 1, division with operand 1). The test lasts 180 s in one block of randomized trials, where all four types of math operations (addition, subtraction, multiplication, division) are intermixed throughout the block. The randomization creates repetitions (i.e., performing the same operation) and switches (i.e., performing different operations) in consecutive trials. The exercises are displayed one by one, requiring participants to type in their answers. The typed answer was presented on the screen. Participants are instructed to be as accurate and fast as possible. Depended-measures were the number of correct exercises solved (henceforth, number of solved exercises (Cronbach’s alpha 0.94), accuracy rates per operation (Cronbach’s alpha 0.41), and RTs per exercise (Cronbach’s alpha 0.82). RTs are calculated from the beginning of the trial until the key press of the first digit in the answer. The online version of the BGU-MF is available here: https://yardengliksman.github.io/BGU-MF. Instructions of using and editing the task are available at: https://github.com/yardenGliksman/BGU-MF. Verbal fluency test Verbal fluency was assessed using both phonemic and semantic fluency tasks. In the phonemic fluency condition, participants were asked to produce as many different words as possible within 60 s for each of three Hebrew letters: Bet (/b/), Gimel (/g/), and Shin (/š/). Repetitions, proper names, and morphological variants of the same root were not counted. In the semantic fluency condition, participants generated as many exemplars as possible within 60 s in each of three categories: animals, fruits and vegetables, and vehicles. Across all fluency conditions, the instruction was to produce as many unique, non-repeating words as possible within one minute for each letter or category. Each phonemic or semantic task was presented separately, with participants prompted before each trial with the specific letter or category. The dependent-measure of performance was the total number of correct items across all three categories in the phonological and semantic test separately. Following previous work10, the order of letters and categories within each task, and the order of the phonemic and semantic tasks, was constant across participants). Executive functions Participants completed the Behavior Rating Inventory of Executive Function – Adult Version (BRIEF-A56. BRIEF-A is a 75-question self-report survey evaluating EF in daily life. Participants rate the frequency of different problems on a 3-point Likert scale (never, sometimes, often). Responses are summed up to create the Global Executive Composite (GEC) score56. Procedure The study was conducted online via Zoom program in two sessions lasting approximately 20 min each. Participants were instructed to conduct the experiment in a quiet room. In the first session the computerized math fluency and verbal fluency tests were conducted with order of the tests (math first or verbal tests first) counterbalanced across participants. In the second session the oral math fluency and EF questionnaire were completed. The responses of the BGU-MF were recorded on the server All verbal fluency (oral math fluency and the verbal semantic and phonological fluency) responses were collected orally, recorded, and coded by the experimenter. All analyses were conducted with JASP. All p-values for planned comparisons analysis reported after Holm’s correction. Information regarding participants’ years of education, treated as a continuous variable, was collected during the second session. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Author contributions Conceptualization: YG & SN; Data curation: YG & IS; Formal analysis: C; Funding acquisition: YG & SN; Investigation: YG& SN; Methodology: YG & SN; Project administration: YG & IS; Resources: YG & SN; Software: YG, IS & SN; Supervision: YG; Validation: YG & SN; Visualization: YG; Writing—original draft: YG & SN; Writing—reviewand editing: YG & SN. Funding The authors report no funding for this research. Data availability The dataset is available here: [https://osf.io/d74e2/files/osfstorage/67dc153df7b7f1dc43f32a4f](https:/osf.io/d74e2/files/osfstorage/67dc153df7b7f1dc43f32a4f). Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement The dataset is available here: [https://osf.io/d74e2/files/osfstorage/67dc153df7b7f1dc43f32a4f](https:/osf.io/d74e2/files/osfstorage/67dc153df7b7f1dc43f32a4f).