Metric Selection Effects in Consciousness Measurement: Lessons from Sleep EEG Analysis

A reanalysis of sleep-stage transitions challenges claims that consciousness changes follow two distinct patterns—gradual versus discrete. Using electroencephalography data from 12 healthy adults, the authors tested 622 transitions between sleep stages. When they selected analysis metrics tailored to each transition (e.g., alpha power for wake-to-sleep), the transition from light sleep (N1) to deeper sleep (N2) appeared uniquely steep, consistent with a discrete threshold. However, when they applied the same metrics uniformly across all transitions, no transition showed meaningful changes; the apparent steepness vanished. The authors conclude that the earlier finding resulted from circular reasoning—choosing metrics that already matched known neurophysiological changes—and serves as a methodological caution for consciousness research.

Study at a glance

Abstract

Abstract Background Integrated Information Theory (IIT) predicts gradual consciousness changes, while Global Workspace Theory (GWT) emphasizes discrete threshold events. We tested whether sleep stage transitions exhibit these distinct patterns using quantitative EEG analysis. Methods We analyzed 622 sleep stage transitions from 12 healthy adults (Wake→N1, N1→N2, N2→N3, REM→Wake) using the Consciousness Gradient Index (CGI = √(φ × ρ) × 10). We employed two approaches: (1) an adaptive method selecting transition-specific metrics (alpha power, spindle density, spectral entropy), and (2) control analyses applying uniform metrics (spectral entropy, Lempel-Ziv complexity) across all transitions. Results The adaptive approach showed N1→N2 as the steepest transition (slope = −2.161, d = 2.814, p < 0.001). However, control analyses revealed this pattern to be metric-dependent. With spectral entropy applied uniformly, all transitions showed near-zero slopes (0.001-0.004), with N1→N2 ranking 3rd in magnitude. With Lempel-Ziv complexity, N1→N2 showed the smallest magnitude (0.001). Neither control supported N1→N2 as uniquely steep. Conclusions The apparent “dual architecture” resulted from selecting metrics based on known neurophysiological changes at each transition, creating circular reasoning. Control analyses using unbiased metrics showed no meaningful consciousness changes during any transition. This study demonstrates the importance of validation with uniformly-applied metrics and serves as a methodological warning about metric selection effects in consciousness research. The adaptive CGI framework successfully detected known neurophysiological changes but did not reveal fundamental differences in consciousness transition mechanisms.