One confirmatory endpoint for postural sway

The decision: keep one confirmatory endpoint

Standing sway was recorded for 28 participants across three visual contexts and four Mozart conditions. A design that large can support many summaries. The reanalysis names one confirmatory endpoint: log anterior–posterior root-mean-square center-of-pressure displacement.

The source recordings are the public PhysioNet force-plate files. This portfolio does not replay those files. The lab shows the 3×4 design and the statistics the paper reported.

The consequence: permutations stay inside the participant

Each person contributes repeated trials, so a test that shuffles trials as if they were independent people would overstate the evidence. The confirmatory test uses 99,999 permutations that keep each participant’s block intact and move music labels only inside that block.

The diagram of four blocks is a picture of that restriction. It is not one of the 99,999 permutations, and it is not a new p-value.

What survived, and what did not

Music condition affected displacement: F(3, 301) = 5.52, permutation p = 0.00003. The 0.1 Hz amplitude-modulated condition divided by no music was 1.129, with a 95% confidence interval from 1.038 to 1.230. The visual-by-music interaction was unsupported, which is why that ratio is not split by vision.

Multiplicity-controlled nonlinear measures were null, as was a post-result spectral family. A delayed-feedback surrogate failed its predictive and recovery criteria. Participant-held-out classifiers did not exceed chance.

• Read the ratio with its interval, not as a vision-specific effect.
• Keep the null families visible beside the one supported contrast.
• Treat classifier and surrogate failures as boundaries, not missing footnotes.

The boundary of the small check

The downloadable Python check confirms that the transcribed ratio sits inside the reported interval, that the interval excludes 1, and that the family outcomes match the abstract. A passing check is transcription arithmetic.

It does not reanalyze 334 trials, rerun the permutation distribution, or establish an effect beyond the source experiment. The paper calls the work a robustness reanalysis, not an independent discovery.