[Paper Review] The lure of causal statements: Rampant mis-inference of causality in estimated connectivity

David M. A. Mehler, Konrad P. Körding|arXiv (Cornell University)|Dec 8, 2018

Functional Brain Connectivity StudiesNeuroscience18 citations

TL;DR

The paper argues that statistical methods like functional connectivity and Granger causality frequently mislead researchers into inferring causality from correlations in undersampled brain data, due to unobserved common inputs and a semantic redefinition of causal terms. It demonstrates that such methods cannot reliably reveal neural mechanisms because they ignore high-dimensional brain activity and the infinite set of unobserved confounders.

ABSTRACT

As neuroscientists we want to understand how causal interactions or mechanisms within the brain give rise to perception, cognition, and behavior. It is typical to estimate interaction effects from measured activity using statistical techniques such as functional connectivity, Granger Causality,or information flow, whose outcomes are often falsely treated as revealing mechanistic insight. Since these statistical techniques fit models to low-dimensional measurements from brains, they ignore the fact that brain activity is high-dimensional. Here we focus on the obvious confound of common inputs: the countless unobserved variables likely have more influence than the few observed ones. Any given observed correlation can be explained by an infinite set of causal models that take into account the unobserved variables. Therefore, correlations within massively undersampled measurements tell us little about mechanisms. We argue that these mis-inferences of causality from correlation are augmented by an implicit redefinition of words that suggest mechanisms, such as connectivity, causality, and flow.

Motivation & Objective

• To expose the widespread misinference of causality in neuroimaging studies using correlation-based statistical methods.
• To highlight the confounding influence of unobserved common inputs in low-dimensional brain measurements.
• To critique the implicit redefinition of terms like 'connectivity,' 'causality,' and 'flow' as if they imply mechanistic insight.
• To argue that high-dimensional neural dynamics are systematically ignored by current statistical models, undermining mechanistic claims.

Proposed method

• Analyzing the structural limitations of statistical techniques such as functional connectivity, Granger Causality, and information flow.
• Demonstrating that any observed correlation in undersampled data can be explained by an infinite number of causal models involving unobserved variables.
• Focusing on the confound of common inputs as a primary source of spurious causal inference in neural data.
• Using theoretical reasoning to show that low-dimensional measurements cannot capture the true high-dimensional nature of brain activity.
• Critiquing the semantic shift in terminology—e.g., 'causality' and 'flow'—that gives false legitimacy to statistical associations.

Experimental results

Research questions

• RQ1 Why do statistical methods like Granger Causality lead to incorrect inferences about neural mechanisms in brain data?
• RQ2 How do unobserved common inputs distort the interpretation of observed correlations in neural connectivity studies?
• RQ3 To what extent can observed correlations in undersampled neural data be explained by alternative causal models with hidden variables?
• RQ4 How does the redefinition of terms like 'connectivity' and 'causality' mislead researchers into believing they have identified mechanisms?

Key findings

• Observed correlations in neural data can be explained by an infinite set of causal models involving unobserved variables, making causal inference from correlation alone fundamentally unreliable.
• Statistical techniques such as functional connectivity and Granger Causality are prone to misrepresenting correlation as causality due to ignoring high-dimensional neural dynamics.
• The presence of countless unobserved common inputs in brain activity creates a pervasive confound that invalidates causal claims based on observed low-dimensional measurements.
• The use of terms like 'causality' and 'flow' in statistical outputs creates a false impression of mechanistic insight, even when no such mechanism is actually identified.
• The paper concludes that current methods cannot reliably reveal neural mechanisms because they fail to account for the high-dimensionality and unobserved variables inherent in brain activity.