Review papers with raw data transparency

Concise appraisal

The authors present three well-controlled psychophysical experiments showing that a high contrast non-target reduces orientation sensitivity for a target presented 250 ms before or after it, consistent with contrast-dependent temporal normalization and with bidirectional temporal suppression across hundreds of milliseconds Primary paper on temporal normalization

Detailed review and critique

What the paper did (summary)

The authors ran three human psychophysics experiments that manipulated the contrast of two sequential Gabor stimuli separated by 250 ms to test whether a high contrast non-target suppresses perception of a target as predicted by temporal normalization. Experiment 1 (discrimination) showed reduced sensitivity when the non-target was high contrast; Experiment 2 replicated this effect for suprathreshold trials using visibility reports; Experiment 3 (continuous estimation plus probabilistic mixture modeling) showed reduced precision and increased swap errors when the non-target was high contrast. Together the results support bidirectional contrast-dependent suppression across ~250 ms, consistent with temporal normalization Primary paper on temporal normalization

Strengths

• Clear, falsifiable prediction from temporal normalization operationalized as contrast-dependent suppression across time; hypothesis testing across three complementary tasks (discrimination, suprathreshold visibility-filtered discrimination, continuous estimation with probabilistic modeling) strengthens inference Mixture modeling in Experiment 3
• Replication across experiments with different durations/contrasts and analysis choices (visibility filtering) reduces the chance that the effect is an artifact of one task design Experiment 2 replication
• Open data and code availability at OSF (vf2ac) supports reproducibility and reanalysis by others OSF data repository vf2ac
• Use of probabilistic mixture modeling (von Mises components for target and non-target plus uniform guess) in Exp 3 gives mechanistic insight into whether suppression reduces precision versus increases swaps/guesses Modeling details

Weaknesses, potential confounds, and blindspots

1. Sample sizes are modest per experiment (Exp1 n=8; Exp2 final n=12; Exp3 final n=20) and some participants overlapped across experiments including two authors in Exp1; this raises concerns about statistical power, generalizability across populations, and potential experimenter-participant effects Sample descriptions and overlap
2. Although visibility reports were used in Exp2 and Exp3 to restrict analyses to trials where stimuli were seen, subjective visibility can be biased and may not fully rule out attentional lapses or order confusions; alternative objective measures (e.g., peripheral probes, forced-choice temporal order judgments) could strengthen claims about preserved temporal visibility Visibility filtering approach
3. The observed backward suppression (T2 affecting T1 precision) could be partially explained by decision-level biases or failures of temporal order encoding (swapping), not solely early perceptual normalization; the modeling in Exp3 addresses this but cannot fully separate late decisional from early sensory suppression without concurrent neural measures Precision and swap findings
4. Tasks used only two items separated by 250 ms; it remains unknown whether temporal normalization generalizes to richer temporal streams, different feature domains (color, motion), or longer/shorter SOAs — authors acknowledge this as needed future work Authors note on future directions

How convincing is the temporal normalization interpretation?

The behavioral pattern — contrast-dependent decreases in sensitivity, reductions in precision, and increases in swap errors when a high-contrast non-target is present both before and after the target — maps onto the predictions of a divisive temporal normalization computation, and is consistent with a unified spatiotemporal normalization model (D-STAN) proposed in related theoretical work that can produce bidirectional suppression via overlapping temporal receptive fields D-STAN and bidirectional suppression

Confidence is therefore moderate-to-high that temporal normalization contributes to the observed effects, but complete mechanistic attribution would require neural measures (e.g., EEG/MEG or intracranial recordings) to show divisive suppression of sensory responses across the same timescale and contrast dependence.

Reproducibility and best practices

• Data and code are available on OSF (vf2ac) — this is excellent and makes direct reproduction straightforward if one follows the reported stimulus parameters and ANOVA/modeling pipelines OSF data repository vf2ac
• To maximize reproducibility authors provided explicit stimulus specs (Gabor parameters, contrasts, SOA, durations), hardware and eye tracking details, and analysis code references — good practice that supports high reproducibility scores.

Concrete suggestions to strengthen follow-up work

1. Acquire neurophysiological measures (EEG/MEG or intracranial/EEG+fMRI) with the same behavioral paradigm to test for contrast-dependent divisive suppression of early sensory responses across 200–500 ms windows and to distinguish sensory suppression from late decision effects.
2. Vary SOA systematically (e.g., 50, 100, 250, 500 ms) and include longer streams with more than two items to map the temporal window of normalization and its decay function.
3. Include objective temporal order judgments or brief temporal tags to reduce reliance on subjective visibility reports and to better separate swapping (order confusions) from sensory suppression.
4. Increase sample sizes and test more diverse populations to examine generality beyond young university samples and to improve statistical power for interaction tests.

Practical takeaways

Behavioral evidence now supports the idea that a normalization computation operates across time and can suppress perception both forward and backward over at least ~250 ms; this provides a plausible mechanistic link tying together adaptation, masking, and temporal crowding phenomena and suggests measurement opportunities for neurophysiology and computational modeling General discussion conclusion

Interactive resources and next actions

You can run an iterative bioinformatics/analysis research agent to reanalyze the public OSF data, fit alternative models (e.g., hierarchical Bayesian swap+precision models), and produce reproducible figures — click below to start.

Author review links