Power-law dynamics in neuronal and behavioral data introduce spurious correlations

Area of Science:

• Cognitive Neuroscience
• Neuroimaging Analysis

Background:

• Relating behavioral and neuroimaging data is key to understanding brain function.
• Current methods often ignore temporal dependencies in single-trial data, potentially causing errors.

Purpose of the Study:

• To address data-analytic pitfalls arising from ignoring long-range temporal dependencies in neural and behavioral measures.
• To quantify the influence of temporal dependencies on observed correlations.

Main Methods:

• Simulations were used to quantify the impact of temporal dependencies on correlations.
• Analysis of real electroencephalography (EEG) data from a reaction time task was performed.
• A surrogate-data procedure was developed to control for power-law dynamics.

Main Results:

• Ignoring temporal dependencies can introduce spurious correlations between neural and behavioral measures.
• Reaction times could be predicted from EEG data of another subject due to shared power-law dynamics.
• Correlations between EEG data from different subjects were also affected by temporal dependencies.

Conclusions:

• Standard correlation methods are insufficient when data exhibit long-range temporal dependencies.
• A novel surrogate-data method can accurately assess correlations while accounting for power-law dynamics.
• Proper analysis is crucial for reliable conclusions about brain activity and behavior relationships.