Too anxious to control: the relation between math anxiety and inhibitory control processes

The study investigates how math anxiety, a specific form of anxiety characterized by tension and apprehension in math-related situations, influences inhibitory control, a key executive function under the central executive proposed in Attentional Control Theory (ACT). ACT posits that anxiety biases attentional control, diverting resources to threat-related information and increasing susceptibility to distraction by irrelevant internal or external stimuli, thereby reducing performance. Prior work suggests anxiety may impair inhibition, but inhibition comprises distinct modes of control: reactive (engaged after interference is detected) and proactive (anticipatory maintenance to prevent interference). The Dual Mechanisms of Control (DMC) theory predicts that anxiety, through reduced working memory availability, should impair proactive control; alternatively, heightened distractibility in anxiety could more strongly disrupt reactive control. The study aims to disentangle whether math anxiety relates more to deficits in reactive or proactive inhibitory control.

Math anxiety has been linked to reduced working memory and attentional biases, and to general (domain-general) inhibition difficulties even in math-unrelated tasks. Adult studies using Stroop-like tasks report worse inhibition for high math-anxious individuals, while child studies suggest not all inhibition components are equally affected. DMC posits proactive control depends on sustained goal maintenance (resource demanding), whereas reactive control is transient and more susceptible to distraction. Findings in general anxiety show mixed evidence: some report impaired proactive control, others show impairments in both reactive and proactive control. ERP work in math anxiety (e.g., numeric Stroop) indicated larger interference effects and patterns interpreted as reactive recruitment in high math-anxious individuals, though those tasks favored reactive strategies and may index reactive rather than proactive control. Hence, a paradigm that orthogonally manipulates reactive versus proactive control is needed to determine which component is most affected in math anxiety.

Participants: Ninety-eight university students were recruited; after exclusions for dyscalculia (1), high flanker errors (>20%; 3), and extreme median RT (>2.5 SD; 2), 92 participants (22 males; mean age 19.0 ± 1.3 years) remained. Ethical approval was obtained (KU Leuven G-2017 10 951) and informed consent collected. Apparatus: Stimuli displayed on 17-inch monitors (60 Hz; 1280×1024) at ~75 cm; E-Prime 2.0 controlled presentation and response recording. Inhibition task: Arrow flanker task with a central target arrow flanked by two arrows on each side. Congruent (<<<<<, >>>>>) and incongruent (<<><<, >><>>) stimuli (4° × 1°) appeared white on black. Responses: 'a' for left, 'p' for right. Trial structure: fixation 500 ms, blank 500 ms, then flanker stimulus until response; no feedback. Two 160-trial blocks manipulated proportion congruency: Mostly Congruent (MC; 80% congruent/20% incongruent) to induce reactive control; Mostly Incongruent (MI; 20% congruent/80% incongruent) to induce proactive control. Trial order randomized; block order counterbalanced; 30 s pause between blocks. Practice: 16 trials (50% congruent/incongruent) with accuracy feedback. Math anxiety: Dutch Abbreviated Math Anxiety Scale (AMAS; 9 items; 1–5 Likert; total 9–45); internal consistency α = 0.89. Mathematics achievement covariates: Arithmetic fluency (TTR; 200 basic facts; 30 s per column; total correct; α = 0.90). Complex arithmetic (CDR-5 subtests: Procedural calculation, Mathematical reasoning, Word problems with/without distraction; 20 items, 15 min; total correct 0–20; α = 0.75). Procedure: Participants first completed the flanker task individually, then group-administered paper tests; order of AMAS vs tests counterbalanced to assess order effects. Statistical analysis: RTs (ms) analyzed via linear mixed models (lme4 in R, ML estimation) with fixed factors Congruency (congruent vs incongruent) and Block (MC vs MI), covariates standardized AMAS (math anxiety), TTR, CDR, and stepwise random effects selection. Best-fitting random structure included by-subject intercepts and slopes for Congruency, Block, and their interaction (Model 5; lowest AIC; significant likelihood ratio tests). Satterthwaite’s adjustments provided t-statistics and df. Post-hoc tests used estimated marginal means with Tukey corrections. Error rates analyzed with generalized linear mixed models (logistic link), analogous fixed/random structures; Wald z reported. Data trimming: First trial of each block excluded; incorrect trials and RTs >1500 ms excluded from RT analyses; identical trimming (sans incorrect) for error models.

Descriptives: AMAS mean 24.0 (SD 6.8; range 11–42); no significant gender difference in AMAS. TTR mean 97.2 (SD 16.5), CDR mean 10.4 (SD 3.2). Math anxiety did not significantly correlate with TTR or CDR. Higher CDR associated with faster RTs and fewer errors on flanker task; slower RTs correlated with lower error rates. RT analysis (Table 3): Main effects: Congruency, faster responses on congruent vs incongruent trials (approx. 464 vs 608 ms; Estimate for Congruency = -72.24 ms, p < 0.001 reflects coding; large congruency effect). Block: longer RTs in MC vs MI (approx. 545 vs 528 ms; Block Estimate = 8.83 ms, p < 0.001). CDR: each unit increase associated with ~21 ms faster RT (Estimate = -20.75 ms, p = 0.007). Significant Congruency × Block interaction: larger congruency effect in MC (≈211 ms) than MI (≈77 ms), confirming the proportion congruency effect. Block × CDR interaction: steeper RT decrease with higher CDR in MC (-28 ms/unit) than MI (-13 ms/unit). Congruency × math anxiety: congruency effect increased with higher math anxiety (Estimate = -7.70 ms, p = 0.018), driven by slower incongruent RTs. Block × math anxiety: greater RT increase with math anxiety in MC (≈12 ms/unit) than MI (≈2 ms/unit) (p = 0.031). Critically, Congruency × Block × math anxiety interaction significant (Estimate = -5.62 ms, p = 0.012). Follow-ups: In MC (reactive context) Congruency × math anxiety interaction significant (t(84.9)=2.72, p=0.008); congruency effect increased by ~27 ms per unit increase in math anxiety. In MI (proactive context) this interaction was absent (t(84.9)=0.85, p=0.40); Bayes factor BF10=0.1 supported the null. The effect of math anxiety on RTs was significant only for incongruent trials in MC (slope ~+26 ms per unit; t(84.9)=2.27, p=0.026) and was larger than the corresponding effects in congruent MC (-1 ms), congruent MI (0 ms), and incongruent MI (4 ms) with Tukey-corrected comparisons all p<0.05. Error analysis (Table 4): More errors on incongruent (4.2%) than congruent (0.4%) trials (p<0.001). More errors in MI (2.0%) than MC (0.8%) blocks (p=0.004). No significant interactions involving math anxiety on errors.

The study addressed whether math anxiety disrupts reactive or proactive components of inhibitory control. Results showed that higher math anxiety specifically impaired performance under reactive control demands: in the mostly congruent context, where interference is rare and cannot be anticipated, math anxiety predicted slower responses on incongruent trials. In contrast, proactive control appeared intact: in the mostly incongruent context, where interference is frequent and goals can be maintained, math anxiety did not relate to RTs. This pattern suggests that distractibility—elevated both by the reactive task context and by anxiety—may reduce efficient detection of interference necessary to trigger reactive control. The maintained goal state in proactive contexts may shield against distraction, preserving performance despite anxiety. The absence of a general performance decrement and the use of math-unrelated material support a domain-general inhibition issue linked to math anxiety rather than solely math-specific content. However, proactive deficits might emerge under higher working-memory demands than imposed by the present flanker task, as suggested by prior work using more demanding tasks (e.g., n-back). The unexpected lack of correlation between math anxiety and math achievement in this sample may reflect the simplicity of the achievement measures for college students, particularly the CDR’s suitability and working-memory load for this population.

By manipulating the context to elicit reactive versus proactive control within an arrow flanker task, the study demonstrates that math anxiety selectively hampers reactive inhibitory control, while proactive control remains unaffected under the present task demands. The findings advance theory by linking math anxiety to interference detection and control deployment when control must be invoked reactively. Future research should increase working-memory load or task complexity to test whether proactive control deficits emerge in math anxiety, include measures of other anxiety constructs to assess specificity, and examine transfer of math-anxiety effects to general cognitive processes more directly.

The study did not include measures of general or test anxiety, limiting conclusions about specificity to math anxiety. The flanker task likely imposed relatively low working-memory demands, which may have masked potential proactive control impairments. The task context was math-unrelated, possibly reducing general performance effects of math anxiety. The complex arithmetic test (CDR) may not have been optimally challenging for a university sample, potentially attenuating relations with math anxiety. The sample comprised young adult students, which may limit generalizability.