The lure of misleading causal statements in functional connectivity research (1812.03363v3)

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.

• The paper reveals that interpreting FC correlations as true causation is flawed due to undersampling and unobserved confounders.
• It uses Simpson’s paradox to demonstrate how confounding variables can reverse apparent causal relationships in neural data.
• The authors advocate for perturbation-based methods and simulations to achieve more accurate causal inference in neuroscience.

Insights on Misleading Causal Statements in Functional Connectivity

The paper "The lure of misleading causal statements in functional connectivity" by Mehler and Kording addresses the prevalent issue in neuroscience regarding the misinterpretation of functional connectivity (FC) results as indicative of true causal, mechanistic interactions in the brain. Recognizing this misunderstanding as the primary concern, the authors explore the methodological flaws and the semantic discrepancies contributing to this fallacy.

The central argument posited by the authors is the logical flaw prevalent in the interpretation of correlation-based outcomes of functional connectivity analyses as causal insights into brain mechanisms. Traditional approaches such as functional connectivity, Granger Causality, and dynamic causal modeling often overstretch their conclusions due to massive undersampling and unobserved confounding. The very foundation of these approaches depends on the causality inferred from correlations, which does not align with the nuanced, high-dimensional nature of neural data.

Challenges in Estimating True Causality

There are several confounding variables unaccounted for when employing FC methodologies. Two main criteria—back-door and front-door—are cited as prerequisites for establishing causality. The impossibility lies in measuring all causally relevant variables or blocking every possible confounding pathway, especially considering the complexity and interconnectedness of neuronal networks. The observed correlations within the data are therefore merely epiphenomena, incapable of distinguishing causal relationships due to this vast confounding.

The authors employ the Simpson's paradox as a demonstrative tool to highlight how confounders can arbitrarily affect FC estimations and even invert perceived causal directions. In extensive network systems where only a fraction of activities are recorded—e.g., fMRI data for large cortical areas—the possibility of genuine causal inference diminishes drastically, underscoring the inadequacy of storing low-dimensional projections.

Methodological and Interpretational Implications

Beyond the logical and statistical arguments, the paper argues there is a fundamental issue surrounding the language used in describing FC research. Terms such as "connectivity" and "causality" imply a predetermined mechanistic basis that the current methodological frameworks cannot substantiate. This semantics issue contributes heavily to the propagation of misconceived causal interpretations in the literature.

The misinterpretation problem is further exacerbated by the misalignment between the linguistic representation and the statistical reality of FC measures. Researchers often pursue studies with causal inferences drawn implicitly from models meant primarily for predictive purposes.

Future Directions and Practical Implications

Explicitly, the authors suggest a move towards rigorous perturbation-based methods that allow for more direct causal investigation. Techniques like optogenetic and invasive brain stimulation can bridge the gap by providing experimental dimensions to correlate connectivity measures with causal influence. However, even here, much care is needed to account for compensatory neural mechanisms that might mask causal inferences.

Simulations and computational models are recommended as vital complements to empirical studies, allowing for controlled experiments that identify potential pitfalls and validate assumptions. There exists optimism for applying FC-related algorithms more extensively to small-scale networks or complete systems where data acquisition is feasibly exhaustive.

In conclusion, while the quest for understanding brain causality remains compelling, this paper strongly advises refining the language and methodologies used in FC studies. A disciplined balance between statistical analysis and semantic clarity is necessary—both to enhance the reliability of conclusions drawn from FC studies and to pave the road for more meaningful neuroscientific discoveries.