Speed and accuracy instructions affect two aspects of skill learning differently

The study investigates how emphasizing speed versus accuracy during practice influences two distinct aspects of procedural skill learning: probability-based learning (short-range statistical regularities) and serial-order-based learning (global sequence order). Prior work provides mixed evidence on whether speed or accuracy instructions benefit procedural learning, possibly because procedural learning is multifaceted and supported by partially distinct mechanisms. The authors hypothesize that probability-based learning, which is rapid and typically incidental, would be more sensitive to instructional emphasis and potentially show improved expression under speed instructions, whereas serial-order-based learning, which develops more slowly and can be more intentional, might be less affected. The purpose is to test whether instructions differentially affect (1) probability- vs serial-order-based learning and (2) the expression of knowledge during learning versus the acquired knowledge when tested under neutral (equal speed and accuracy) instructions.

Prior studies in sports and motor learning suggest novices may benefit from accuracy-focused practice, whereas experts may perform better under speed constraints. Some reports found that accuracy instructions can impair the expression of implicit skill acquisition relative to speed instruction, and that speed instructions can enhance representations of repeating sequences, while accuracy may support better stimulus–response mapping. Conversely, earlier work by the present authors using a non-cued probabilistic sequence task found no effect of speed versus accuracy instructions on acquired procedural knowledge. Procedural learning encompasses distinct but parallel processes: probability-based learning (rapid, incidental acquisition of transitional probabilities <1) and serial-order-based learning (acquisition of deterministic transitional probabilities =1), which have different neural substrates and developmental profiles. These distinctions may account for conflicting results when the two aspects are not separated behaviorally. The current study leverages a cued version of the ASRT task to dissociate these components after relatively brief practice.

Design: Experimental, between-subjects manipulation of instructions (Speed vs Accuracy) using the cued Alternating Serial Reaction Time (ASRT) task to separately measure probability-based and serial-order-based learning. Two phases: Different Instruction Phase (practice under Speed or Accuracy emphasis) followed by a Same Instruction Phase (equal emphasis on speed and accuracy) after a brief rest. Participants: 56 healthy young adults recruited; 8 excluded for instruction noncompliance (6 due to sequence-report accuracy <30%, 2 due to general speed/accuracy criteria), leaving N=48 (43 females). Groups: Accuracy (N=26), Speed (N=22). Age 18–34 years (M=21.21±2.81), education M=14.10±2.01 years, counting span M=3.67±0.98. Handedness: 35 right-handed, 3 left-handed, 10 ambidextrous. No significant group differences in demographics or cognitive measures. Ethical approval obtained; informed consent provided. Task and stimuli: Cued ASRT. Four horizontally arranged circles mapped to keys Z, C, B, M. Targets were dog (pattern/cued) or penguin (random) icons appearing in one of four positions. After correct response, 120 ms response-to-stimulus interval. Sequence structure alternated predetermined (pattern) and random elements, yielding an 8-element alternating sequence (e.g., 2 r 4 r 3 r 1 r). Pattern and random elements were visually cued (dog vs penguin). Participants were informed that dog trials follow a repeating sequence and were encouraged to discover it; penguin trials were random. One of six unique 4-element pattern permutations assigned per participant for the whole task. Triplet structure and learning measures: Due to alternation, certain 3-item runs (triplets) occur with different probabilities. Trial types: (1) pattern high-probability (pattern element last of high-probability triplet), (2) random high-probability, (3) random low-probability. Probability-based learning index: RT difference random high-probability minus random low-probability (faster for high indicates learning). Serial-order-based learning index: RT difference random high-probability minus pattern high-probability (faster for pattern indicates order learning). To account for large RT differences induced by instructions, learning scores were standardized by dividing by the median RT of the corresponding epoch. Procedure: Practice Phase: 3 ASRT blocks with random stimuli to familiarize responses. Different Instruction Phase: 20 ASRT blocks (~1–1.5 min/block) under instruction to be either as accurate as possible (Accuracy Group) or as fast as possible (Speed Group). After each ASRT block, participants completed a post-block sequence report task, typing the order of pattern (dog) elements using up to 12 keypresses. Rest: 10 minutes. Same Instruction Phase: 5 ASRT blocks under instruction to be equally fast and accurate; post-block sequence reports continued. Data preprocessing and analysis: Blocks collapsed into epochs of five blocks each (Epochs 1–4: Different Instruction Phase; Epoch 5: Same Instruction Phase). Trials with incorrect responses, triplet trills (e.g., 1–2–1) and immediate repetitions (e.g., 1–1–1) were excluded due to pre-existing tendencies. Median RTs computed per trial type per epoch; accuracy ceiling effects expected under accuracy instruction, so primary analyses used RT-based measures. Standardized learning scores computed by dividing learning indices by epoch median RT. Statistical analyses: Mixed-design ANOVAs (within-subject factor Epoch; between-subject factor Group) assessed RT, accuracy, and standardized learning scores during Different Instruction Phase; Mann–Whitney U tests compared groups in Same Instruction Phase. Bayesian ANOVAs/Mann–Whitney with model averaging reported BF_exclusion. Software: JASP 0.16. Power analysis (G*Power 3.1.9.7) indicated N=30–38 sufficient for expected effects (ηp²≈0.12); final N=48 adequate. Explicit knowledge measure: Post-block sequence report accuracy (% correct keypresses relative to the repeating 4-item pattern), averaged within epochs. Same factors and statistical tests as for ASRT learning indices.

Baseline/practice: No pre-existing group differences in median RT (U=294, p=0.88, BF10=0.03) or accuracy (U=297.5, p=0.82, BF10=0.04). Instruction effects on overall performance (Different Instruction Phase): - RT: Decreased over epochs irrespective of trial type (Epoch main effect: F(1.65,75.71)=47.23, p<0.001, ηp²=0.51). Speed Group faster than Accuracy Group (M=348±13.22 ms vs 464±13.22 ms; Group main effect: F(1,47)=38.13, p<0.001, ηp²=0.45). No Epoch×Group interaction. - Accuracy: Decreased over time (Epoch main effect: F(1.23,56.57)=9.12, p=0.002, η²=0.17). Accuracy Group more accurate than Speed Group (M=98.6%±2.2 vs 83.3%±2.2; Group main effect: F(1,53)=24.18, p<0.001, η²=0.35). Accuracy decline across epochs evident mainly in Speed Group (Epoch×Group: F(1.23,56.57)=6.16, p=0.01, η²=0.12). Same Instruction Phase (equal emphasis): Speed Group remained slightly faster (M=331±7.68 ms vs 355±7.13 ms; U=396.5, p=0.02) and less accurate (M=92.3%±1.6 vs 96.3%±0.4; U=430.5, p=0.003). Probability-based learning: - Different Instruction Phase (standardized scores): Speed Group showed greater learning than Accuracy Group (M=0.08±0.01 vs 0.04±0.01; Group main effect: F(1,46)=7.64, p=0.008, ηp²=0.14). No Epoch effect or Epoch×Group interaction. - Same Instruction Phase: No group difference (U=221, p=0.18; BF01=2.20). Serial-order-based learning: - Different Instruction Phase (standardized scores): No group difference (M Accuracy=0.03±0.01, M Speed=0.02±0.01; Group: F(1,46)=0.12, p=0.74). No Epoch or interaction effects. - Same Instruction Phase: No group difference (U=295, p=0.86). Explicit sequence knowledge (post-block reports): Improved over time (Epoch: F(2.22,102.09)=17.19, p<0.001, ηp²=0.27). No overall group difference (F(1,46)=1.44, p=0.24); trend-level Epoch×Group interaction (F=2.94, p=0.05), with Accuracy Group higher in Epoch 1 (86.4%±2.3) than Speed Group (74.0%±2.4), converging by Epoch 4 (Accuracy 96.9%±2.6; Speed 94.5%±2.6). Same Instruction Phase: groups similar (U=309, p=0.53).

The findings demonstrate that instructional emphasis on speed versus accuracy differentially affects the expression of distinct procedural learning components. Speed instructions enhanced the expression of probability-based knowledge during learning, while serial-order-based learning was unaffected by instruction. Crucially, when tested under neutral instructions, the acquired knowledge was equivalent across groups for both learning types, indicating that instructions influenced momentary performance rather than the underlying competence. This dissociation underscores the need to distinguish performance from competence in studies of learning and when interpreting behavioral-neural relationships. The near-ceiling accuracy under accuracy instructions yet preserved learning indicates that response errors are not necessary for acquiring probability-based or serial-order-based procedural knowledge. From a process perspective, probability-based learning, likely more implicit and rapid, appears more susceptible to contextual performance factors (e.g., speed emphasis), whereas the more explicit or globally structured serial-order component is robust to such manipulations. The trend toward better early explicit sequence reporting under accuracy instructions suggests that accuracy emphasis might transiently facilitate explicit sequence awareness, potentially due to slower pacing aiding cue following. Alternative explanations (e.g., divided attention differences between cued and non-cued tasks) are unlikely given similar probability-based learning across versions and comparable sequence report performance. Overall, the results advise caution in drawing conclusions about learning mechanisms from single-context performance measures and motivate designs that assess both expression and underlying competence.

Speed and accuracy instructions selectively modulate the expression of probability-based regularities during learning without altering the ultimately acquired knowledge, while serial-order-based learning remains resistant to such instructional manipulation. The work clarifies that competence and performance can diverge in procedural learning, emphasizing the importance of testing knowledge under common instruction to assess acquisition. Practically, speed instructions can transiently boost performance on probabilistic regularities, whereas accuracy instructions may slightly aid early explicit sequence reporting. Future research should: (1) further delineate how explicitness/implicitness mediates instruction effects; (2) investigate the role of prediction errors and their behavioral correlates under different instructions; (3) examine potential moderating factors such as gender or individual differences; and (4) include control conditions to parse cueing and pacing influences.

The study lacks a control group without cueing, raising the possibility that cueing pattern elements might differentially affect responses; however, observed differences were confined to probability-based learning where only random (non-cued) elements were compared. The sample had an unequal gender distribution (predominantly female), limiting the ability to assess gender moderation. Additionally, although standard exclusions (trills/repetitions) were applied and ceiling effects in accuracy were addressed by relying on RT-based standardized scores, residual differences in overall RT and accuracy persisted into the Same Instruction Phase, potentially introducing small performance confounds.