We tested movement initiation time and absolute angle error in 32 adults from OpenNeuro ds004021 version 1.0.1. The data cover two task days in the same people. We used raw error values and fixed blocks of 16 trial positions. The group mean gives equal weight to each person. The mean within-person slopes were -0.107 on day 1 and -1.177 on day 2, in degrees per doubling of time. The 95 percent person bootstrap intervals were [-1.685, 1.398] and [-4.346, 1.856]. Both included zero. The planned primary statistic was the day 1 within slope minus the between-person slope. It was -4.991 [-9.415, -0.037]. Several planned checks gave difference intervals that included zero. The result thus depends on the analysis rules. A larger bootstrap check and a check with the source paper's time limits were added after the primary results. Time was not assigned at random, and target direction is absent. The study does not establish a cause, a learning strategy, or absence of a relation. The package preserves code, raw sources, hashes, the local plan, all results, and instructions. Independent regression checks passed. A fresh source download and full replay matched all source files and all 17 result files.
Filters: arxiv: nlin.AO (Adaptation and Self-Organizing Systems) ×
Nakamuta et al. report that in developing rat cortical cultures on high-density electrode arrays, the PCA variance spectrum becomes a power law, PCs become spatially local, and neuronal avalanches appear. They explain this with heavy-tailed synaptic coupling, but they did not control for firing rate or for the number of active electrodes. We used their public data (DANDI 002020, 7 cultures, 113 of 129 daily sessions) and their methods, and added controls. The published changes reproduced. The spectral exponent increased from 0.54-0.98 to 1.02-1.33, and the participation ratio fell in 7 of 7 cultures. The PC locality slope became steeper in 7 of 7, and the branching ratio increased from about 0.6 to about 0.9 in 7 of 7. Firing rate and electrode count had no consistent day trend. The day trends remained with 200 matched electrodes and after partial control for rate and electrode count. We reduced a mature session to the rate and electrode count of each other day: this gave no day trend in the exponent, decay rate, participation ratio or locality. A surrogate with the same rates and burst timing but no pair structure gave flatter spectra (exponent 0.70 vs 1.11) and no PC locality. Thus pair structure is necessary. A model in which coupling depends only on inter-electrode distance gave the real PC locality (slope -0.35 vs -0.29, p = 0.59) and the fall in dimensionality. It gave a lower exponent (0.82 vs 1.11) and about one third of its developmental increase. When we added per-electrode coupling gains, the model gave about four fifths of the increase. But the gains became less heavy-tailed with maturation (Gini fell in 7 of 7 cultures). Sessions with heavier gain tails had flatter spectra beyond day. Spectrum shape and avalanche exponent did not co-vary beyond their shared day trend. Shot-noise correction showed that activity becomes lower-dimensional, not higher-dimensional. Limitations: unsorted channel-level detections, unknown days in vitro, 7 cultures, and functional (not anatomical) coupling proxies. Measured connectivity is necessary to test the heavy-tailed-synapse mechanism.