Registered Report: Replication and Extension of Nozaradan, Peretz, Missal and Mouraux (2011)

This registered report, comprising 13 independent replications, failed to reproduce the original findings of Nozaradan et al. (2011) regarding neural signatures of conscious beat imagery, suggesting that the previously reported effects may not be reliable and highlighting the necessity of larger sample sizes for detecting such phenomena.

The Gist

The Big Question: Is Your Brain "Dancing" to the Beat?

Imagine you are listening to a metronome. Tick. Tick. Tick. It's a steady, boring sound. But, if you close your eyes and imagine a drumbeat, you might start hearing a pattern: Tick-TICK-tick-TICK (a binary beat) or Tick-TICK-tick-TICK-tick-TICK (a ternary beat).

For a long time, scientists believed that when you do this mental trick, your brain actually changes its electrical activity to match the pattern you are imagining. They thought your brain was "dancing" along with your imagination, even though the sound itself never changed.

This idea was based on a famous study from 2011. But, in science, big claims need big proof. So, a massive team of 13 different laboratories joined forces to test this idea again, this time with a much larger group of people and a very strict, pre-planned rulebook.

The Experiment: The "Mental Gym"

Think of this study as a giant, synchronized mental gym session.

1. The Stimulus (The Treadmill): Everyone listened to the exact same sound: a pure tone that ticked at a steady speed (2.4 times per second). It was like a treadmill set to a constant speed.
2. The Task (The Dance Moves):
   • Group A (Control): Just listen. Don't imagine anything special.
   • Group B (Binary): Imagine a "march" pattern (strong-weak, strong-weak).
   • Group C (Ternary): Imagine a "waltz" pattern (strong-weak-weak, strong-weak-weak).
3. The Test (The Check-in): At the end of the sound, a little "probe" tone would pop up. Participants had to guess: "Did that tone land on a strong beat or a weak beat?" This checked if they were actually paying attention and imagining the pattern correctly.
4. The Measurement (The Heart Monitor): While they did this, scientists used EEG caps (like swim caps with sensors) to read the electrical waves of their brains. They were looking for a specific "signature" in the brain waves that matched the imagined beat.

The Results: The Great Disappointment (and the Truth)

The original 2011 study said: "Wow! When people imagined the beat, their brain waves jumped up huge!" It was a loud, clear signal.

This new study, with 152 people (compared to the original 8), said: "We looked very carefully, and... we barely saw anything."

Here is the breakdown of what they found:

• The Signal was Tiny: The original study saw a brain signal that was about 4 to 5 times larger than what this new study found. In the new study, the "imagined beat" signal was so small it was almost lost in the background noise.
• The Confidence Interval: In science, we use a "confidence interval" like a safety net. If the net catches the number "zero," it means the effect might not exist at all. In this study, the safety nets for all the main results caught zero. This means the evidence that the brain changes based on imagination is very weak.
• The Real Hero: Interestingly, the only thing that did predict how well people did on the test was the brain's reaction to the actual sound (the ticking metronome), not the imagined beat. It seems the brain was just reacting to the physical noise, not the mental dance.

Why Did the Original Study Get It Wrong?

You might wonder, "If the effect is so small, why did the first study see it so clearly?"

Think of the original study like a lucky shot in a dark room. They had a very small group of people (only 8), and many of them were professional musicians. Musicians are like expert dancers; their brains are super-tuned to rhythm. The original study likely caught a "perfect storm" where a few highly skilled people showed a strong reaction, making it look like a universal human trait.

This new study was like turning on all the lights and inviting 152 people, including non-musicians. When you look at the whole crowd, the "magic" disappears. The effect is either non-existent or so incredibly tiny that you need a massive sample size to see it.

The Takeaway: A Lesson for Science

This paper is a classic example of scientific self-correction.

1. Replication is Key: Just because a study says something is true once doesn't mean it's true forever. We need to try it again, with more people, to be sure.
2. Beware of Small Samples: When you test only a few people, especially if they are experts, you might see a "ghost" effect that isn't really there for everyone.
3. The Method Might Need a Tune-Up: The technique used here (Frequency Tagging) is like a very sensitive radio. It can pick up the "station" of the physical sound perfectly. But, it might not be sensitive enough to pick up the "whisper" of the imagination. Scientists now need to figure out if this radio is the right tool for the job, or if they need a different one.

In short: The idea that our brains physically "sync up" to a beat we are only imagining is much less certain than we thought. It might happen, but if it does, it's a very faint whisper, not the loud shout the original study suggested. Science is working hard to figure out exactly how our brains hear music, and this study is a crucial step in getting the story right.

Technical Summary

1. Problem Statement

Cognitive neuroscience has long struggled to disentangle stimulus-driven (exogenous) neural processing from conscious perceptual (endogenous) processing. In the domain of music cognition, a pivotal 2011 study by Nozaradan et al. claimed to demonstrate that listeners could voluntarily impose a specific beat structure (binary or ternary) onto an ambiguous, isochronous auditory stimulus. Using frequency tagging (Steady-State Evoked Potentials, SSEPs), the original study reported significantly larger neural responses at the "imagined" beat frequencies (0.8 Hz for ternary, 1.2 Hz for binary) compared to control conditions, suggesting a neural correlate of top-down beat perception.

However, this finding has significant implications for theories of auditory entrainment and beat perception. If the effect is robust, it validates frequency tagging as a tool for studying conscious perception; if not, it suggests that previous findings may have been confounded by stimulus properties or small sample sizes (the original study had $N=8$). The core problem addressed by this Registered Report (RR) is whether the original effect is replicable across a larger, multi-lab sample and whether the neural activity at beat-related frequencies truly predicts conscious behavioral performance.

2. Methodology

This study is a multi-lab Registered Replication involving 13 independent laboratories and a total of 152 participants (after exclusions).

• Protocol: The study strictly adhered to the original Nozaradan et al. (2011) protocol, with specific extensions to improve rigor.
  • Stimulus: A 33-second amplitude-modulated pure tone (333.3 Hz carrier, 2.4 Hz modulation) with pseudo-periodic irregularities.
  • Conditions: Three conditions were tested:
    1. Control: Detect a brief sound interruption.
    2. Binary Imagery: Imagine a binary beat (every 2nd event).
    3. Ternary Imagery: Imagine a ternary beat (every 3rd event).
  • Extension: Unlike the original, this study added a behavioral probe task on every trial. After the stimulus, a probe tone was presented at either a "beat" or "off-beat" position. Participants indicated if the probe was on the beat, providing a direct measure of conscious perception.
  • Covariates: Detailed data on music and dance training years were collected to test for moderation effects.
• Data Collection & Exclusion:
  • Labs were required to pre-register their implementation plans.
  • Exclusions: 59 of 212 initial participants were excluded for failing imagery training, excessive EEG artifacts (>50% of data), or experimenter error.
  • EEG Processing: Data were processed using a standardized pipeline (MATLAB/EEGLAB/FieldTrip) involving high-pass filtering, ICA for artifact removal, and Fourier transformation to calculate SNR-corrected amplitudes at target frequencies (0.8 Hz, 1.2 Hz, 1.6 Hz, and 2.4 Hz).
• Analysis Plan:
  • Primary: Pre-registered random-effects meta-analyses of raw mean differences in SNR-corrected amplitudes between conditions.
  • Secondary: Logistic regression (GEE) to predict behavioral accuracy from neural amplitudes.
  • Exploratory: Analyses using median differences (due to sphericity violations) and moderation by musical/dance expertise.

3. Key Contributions

• Scale and Rigor: This is one of the largest multi-lab replications in auditory neuroscience to date, addressing the "small sample size" limitation of the original study ($N=8$) with a combined $N=152$.
• Behavioral-Neural Link: It is the first study to directly correlate trial-by-trial beat imagery neural responses with a concurrent behavioral measure of beat perception, testing the assumption that larger SSEPs equal better perception.
• Expertise Moderation: It systematically tests whether musical or dance training moderates the beat-imagery effect, a factor hypothesized to be crucial but untested in the original work.
• Open Science: The study utilized a Registered Report format, with protocols, analysis scripts, and raw data publicly available on OSF and Harvard Dataverse, ensuring transparency and reducing researcher degrees of freedom.

4. Results

The results indicate a failure to replicate the main findings of the original study with the same magnitude or statistical certainty.

• Effect Size Discrepancy:
  • Original Study: Reported large effect sizes ($\eta_p^2 = .62–.76$) with raw amplitude differences of 0.12–0.20 µV.
  • Current Study: Meta-analytic estimates showed significantly smaller raw mean differences (0.03–0.04 µV).
  • Significance: All four pre-registered meta-analytic effect sizes (Binary vs. Control, Binary vs. Ternary, Ternary vs. Control, Ternary vs. Binary) had 95% confidence intervals overlapping with zero ($p$-values ranging from .052 to .132).
• Replication Rate: Only 1 out of 13 labs showed a significant condition effect for binary imagery, and 2 out of 13 for ternary imagery. No lab replicated more than one effect.
• Behavioral Prediction (Crucial Finding):
  • Contrary to the hypothesis, neural activity at the imagined beat frequencies (0.8 Hz / 1.2 Hz) did NOT predict behavioral accuracy.
  • Instead, the SNR-corrected amplitude at the stimulus frequency (2.4 Hz) was the only significant predictor of trial accuracy. Higher attention/encoding of the physical stimulus predicted better performance, regardless of the imagery condition.
• Moderators: Musical and dance training years did not significantly moderate the effect sizes.
• Exploratory Analyses: When using medians (to address non-normal data distributions in some labs), the meta-analytic effects became statistically significant ($p < .001$), but the effect sizes remained very small (0.02–0.03 µV), suggesting that even if a true effect exists, it is minute and requires massive sample sizes to detect reliably.

5. Significance and Implications

• Questioning the Utility of Frequency Tagging for Beat Perception: The findings suggest that frequency-tagged neural responses at beat-related frequencies may not be a robust marker of conscious beat perception or top-down imagery. The lack of correlation between beat-frequency SSEPs and behavioral performance challenges the interpretation that these signals reflect "entrainment" to a perceived meter.
• Stimulus vs. Perception: The fact that the stimulus frequency (2.4 Hz) predicted behavior, while beat frequencies did not, implies that the measured SSEPs may reflect general stimulus encoding or attentional resources rather than specific metrical inference.
• Reproducibility Crisis: The study highlights the "winner's curse" in the original literature, where small samples with high effect sizes (likely inflated) led to a field of research that may be built on fragile foundations.
• Future Directions: The authors argue that while stimulus-driven frequency encoding is reliable, using it to infer conscious perception requires much larger samples and potentially alternative methods (e.g., temporal response functions, phase-locking values, or other frequency bands like $\beta$ and $\gamma$). The study calls for caution in reverse-infering beat perception from frequency-tagged EEG data without robust behavioral validation.

In summary, this Registered Report provides strong evidence that the large neural effects of beat imagery reported in 2011 are not robustly replicable in a larger, diverse sample, and that the relationship between beat-related neural oscillations and conscious perception is far more complex and weaker than previously assumed.