Analysis 6A — onset timing during stimulus omission

Macaque multi-area laminar electrophysiology · Exploratory analysis not publication-final

Overview

Which changes first during stimulus omission — the local field potential, or spiking?

Primary population-level result

A larger fraction of high-frequency LFP responses (beta, low gamma, high gamma) are temporally resolved than low-frequency responses (theta, alpha), and this replicates across sessions: median HIGH−LOW difference +7.27 percentage points, exact exhaustive sign-flip permutation p = 8.39e-05 (262,144 assignments enumerated, n = 21 sessions, 18 non-zero), Holm p = 1.7e-04. replicated across sessions

What the corpus contains

QuantityValueScope
Eligible single units4,130 SPK, all sessions
Omission responses detected637 a response exists
Omission latencies resolved187 estimators agree
Resolved units with a stimulus reference139 ΔT defined
Median resolved Tom 190 msIQR 120–250 ms
LFP area×frequency cells1,820 all sessions
LFP cells resolved114 6.3% of cells

The three session-level tests, and their three different answers

The session is the inferential unit. Units within a session and area×frequency cells within a session are not independent, so pooled counts below are labelled descriptive.

Testnmedianp (sign-flip) Standing
LFP high−low frequency, fraction resolved21 +7.27 pp8.39e-05 replicated
SPK increase-vs-decrease timing7 -60 ms0.875 not replicated
SPK omission−stimulus shift11 +20 ms0.0527 not significant

Three outcomes from the same test procedure. The third is insufficient session-level evidence, not an established null: at n = 11 sessions the smallest attainable exact p is bounded away from zero. The pooled unit-level tendency (104/139 = 74.8% of units with ΔT > 0) is descriptive and is weighted by unit count, not by session.

Held, deliberately

No common-axis LFP-vs-SPK figure. It is withheld pending a scientifically appropriate LFP representation: the LFP side must be interval- and censoring-based, not a cloud of points, because beta latencies are strongly left-censored (see LFP).

No area-latency hierarchy claim. Area and subject are partially confounded; no area was recorded in all subjects. Pooled area medians are descriptive only (see Coverage).

No LFP→SPK causal or directional claim. Nothing in 6A tests direction.