Go/No-Go Response Inhibition Paradigm

• The Go/No-Go paradigm is a task that measures the capacity to inhibit prepotent motor responses using controlled trial-based designs.
• It employs specific stimulus ratios and time constraints to extract precise behavioral metrics such as reaction times, commission, and omission errors.
• Computational models, ERP analysis, and information-theoretic methods reveal neural dynamics and executive control mechanisms underlying response inhibition.

The Go/No-Go response-inhibition paradigm is a foundational task in cognitive neuroscience, neuropsychology, and psychometrics for quantifying the capacity to inhibit prepotent motor responses. It operationalizes the construct of response inhibition via trial-based designs in which some stimuli (Go) require a rapid action, while others (No-Go) require the suppression of that action. The paradigm is widely used in behavioral experiments, electrophysiological studies, and model-based analyses to dissect executive control, neural dynamics, and cognitive processing at behavioral and neural levels (Miller et al., 2015, Lucas et al., 2021, Carlos et al., 22 Jul 2025, Pankaj et al., 2020).

1. Experimental Structure and Behavioral Metrics

Go/No-Go paradigms employ a variety of stimulus frameworks, but all share the response-contingent dichotomy: subjects must respond (typically by button press or lever release) to Go stimuli, while withholding responses to No-Go stimuli. Task parameters include the stimulus ratio (e.g., Go:No-Go = 7:1 or 3:1), randomized presentation order, explicit time limits ("timeouts" for response), and instructions emphasizing both speed and accuracy (Pankaj et al., 2020, Miller et al., 2015).

Behavioral measurements commonly extracted are:

• Reaction Time (RT): $RT_i = t_\mathrm{press,i} - t_\mathrm{onset,i}$ for Go (commission) trials;
• Commission Errors: inappropriate responses on No-Go trials;
• Omission Errors: failures to respond on Go trials (often corresponding to right-censored RTs);
• Overall Accuracy: $100\% -$(commission error rate $+$ omission error rate).

In typical data, mean correct Go RTs range from 300–400 ms, overall accuracy can exceed 87.5%, and omission/commission rates provide process-specific markers of inhibition performance (Pankaj et al., 2020).

2. Computational Models of Response Inhibition

Evidence accumulation models, particularly diffusion frameworks, dominate the quantitative analysis of Go/No-Go performance. Two principal models are used:

• Two-Boundary Wiener Diffusion Model: Models decision dynamics in 2AFC tasks as a Brownian motion process $x(t)$ drifting with mean rate $v$ from starting point $z$ toward absorbing boundaries at $x=0$ (error) and $x=a$ (correct), with non-decision time $T_\mathrm{er}$ added to the first-passage time. The hitting-time density for correct responses is given by:

$$f_W(t; v, a, T_\mathrm{er}, z) = \sum_{k=1}^\infty \frac{k\pi}{a} \exp\Bigl[-\frac{(v a - (2k-1)\pi z)^2}{2 (t - T_\mathrm{er})}\Bigr] \, \frac{1}{\sqrt{2\pi (t-T_\mathrm{er})^3}}.$$

• Shifted Wald (One-Boundary) Distribution: For tasks or conditions generating very high accuracy (few commission errors), the process reduces to a single boundary at $100\% -$0, yielding the shifted Wald (inverse Gaussian) first-passage time. With parameters $100\% -$1, $100\% -$2, and $100\% -$3:

$100\% -$4

Under high-accuracy conditions ($100\% -$5), the one-boundary and two-boundary models yield analytically equivalent RT distributions: $100\% -$6, $100\% -$7, $100\% -$8 (Miller et al., 2015).

3. Omission Errors and Response Censoring

Timeout-induced Go omissions are naturally modeled as right-censored RTs. For each trial $100\% -$9, the data point $+$0 indicates an observed button press ($+$1) at time $+$2 or a censoring (timeout; $+$3) at time $+$4. The censored shifted Wald likelihood is given by:

• Uncensored: $+$5 for $+$6,
• Censored: $+$7 for $+$8.

Maximum likelihood procedures, with explicit right-censoring, recover unbiased drift and boundary parameters even with substantial omission rates—whereas naïve exclusion or misclassification of these trials severely biases estimates (Miller et al., 2015).

Practical guidelines include:

• Use censored shifted Wald if omission (timeout) rate exceeds ~5%.
• Use competing-risks extended Wald models (separate for correct/error trials) if commission errors are substantial and error RTs are available.
• Diagnostic QQ plots of empirical vs. modelled response time quantiles are recommended for evaluating fit (Miller et al., 2015).

4. Electrophysiological and Neural Correlates

Event-Related Potential (ERP) studies of the Go/No-Go paradigm reliably differentiate inhibition processes through specific components:

• N200 (N2): Negative peak (150–250 ms), larger for No-Go stimuli, localized primarily to dorsal anterior cingulate cortex (BA 24/32), signaling conflict monitoring.
• P300 (P3): Positive peak (250–350 ms), larger and earlier for Go stimuli, with sources in bilateral inferior frontal gyrus (BA 44/45) and superior parietal lobule (BA 7), indicative of inhibitory control implementation.

Topographical analyses show modulation across frontal, parietal, and central clusters. Larger N2 amplitudes in No-Go trials are correlated (e.g., $+$9, $x(t)$0) with fewer commission errors (Pankaj et al., 2020).

sLORETA source imaging enables identification of distributed frontoparietal networks underlying action selection and inhibition. The ERP time-locked approach complements behavioral diffusion modeling by providing temporal and regional specificity for subprocesses of motor inhibition.

5. Information-Theoretic Analyses of Neural Time Series

Recent approaches utilize symbolic dynamics and information theoretic measures (permutation entropy, statistical complexity) to dissect Go/No-Go neural dynamics at finer resolution.

• Bandt–Pompe Symbolization: For a time series $x(t)$1 and embedding dimension $x(t)$2, construct overlapping $x(t)$3-dimensional vectors; each is mapped to its ordinal pattern (permutation), yielding an empirical symbol probability distribution $x(t)$4.
• Permutation Entropy: $x(t)$5 quantifies sequence unpredictability.
• Statistical Complexity: Martin–Plastino–Rosso complexity $x(t)$6, where $x(t)$7 is disequilibrium (Jensen–Shannon divergence from uniformity).
• Asymmetry Index: $x(t)$8 (for $x(t)$9) for quantifying trial-type differences.

Multi-scale embedding (varying delay $v$0) identifies optimal time scales for differentiation, typically 15–30 ms. In both human EEG and monkey LFP data, Go and No-Go trials diverge in the entropy–complexity plane only after the discrimination cue, robustly across central (Cz) and occipital (O1/O2) electrodes and cortical areas (Carlos et al., 22 Jul 2025, Lucas et al., 2021). Differences peak ~150–200 ms after trial-unique cues.

In some cases, complexity indices distinguish conditions earlier or more robustly than classical ERPs, and can be computed in 100 ms windows suitable for near–real-time monitoring of response inhibition (Carlos et al., 22 Jul 2025).

6. Data Analysis, Model Fitting, and Practical Recommendations

The choice of analytic model should match task characteristics:

• If Go accuracy $v$1 (rare errors): Uncensored shifted Wald or Wiener diffusion with fixed boundary parameter suffices.
• If significant timeouts/omissions: Censored shifted Wald is required; treat omission as right-censored RT.
• If both high commission and omission error rates: Use competing-risks censored shifted Walds for correct and error RTs, or fit two-boundary Wiener model if all RTs are recorded and the model is identifiable (Miller et al., 2015).
• For neural data: Information-theoretic analyses (Bandt–Pompe, complexity–entropy indices, asymmetry indices) reveal trial-type differences at the single-channel and group level, especially in temporo-central and occipital regions (Carlos et al., 22 Jul 2025, Lucas et al., 2021).

Empirical guidelines include stabilizing moment estimates with at least 150–200 Go RTs, modeling non-decision time variability if substantial, and validating model fits with QQ plots and survival curves (Miller et al., 2015).

7. Significance and Interpretation in Executive Control

The Go/No-Go paradigm, combined with evidence accumulation modeling and modern neural quantification (ERP, complexity indices), provides a rigorous operationalization of response inhibition. Behavioral and neural signatures localize the inhibition process to distributed frontoparietal networks, with distinct temporal dynamics measurable at both macro- and micro-temporal scales. Statistically principled treatment of omissions as censored data is essential for unbiased mechanistic inference. Symbolic information-theoretic analyses further augment classical ERP methodology by pinpointing not only the when and where but also the statistical structure underlying inhibition-specific neural dynamics.

The paradigm is thus indispensable for probing central executive inhibitory mechanisms in both typical and pathological populations, enabling dissociation of subprocesses in executive function, and supporting the development of real-time cognitive monitoring tools (Miller et al., 2015, Carlos et al., 22 Jul 2025, Lucas et al., 2021, Pankaj et al., 2020).