Skip to contentScholarGate
LibraryBookshelfDeskReview StudioAssistant
Sign in
On this page
IntuitionHow it worksWhen to use itStrengths & limitationsCommon pitfallsApplicationsFrequently asked🔒 Read the full methodSourcesRelated methods
Cite this pageSpotted an issue on this page? Report or suggest a fix →
Home›Causal inference›Transfer Entropy
Machine learningInformation-theoretic causality

Transfer Entropy

Also known as: Schreiber Information Transfer, Directed Information Flow, Conditional Mutual Information (directed), Transfer Entropisi

Transfer Entropy (TE) is a non-parametric, information-theoretic measure of directed statistical dependence between two time series, introduced by Thomas Schreiber in 2000. Grounded in Shannon entropy, it quantifies how much information the past of one process Y reduces uncertainty about the next state of another process X, beyond what X's own past already provides. Unlike linear correlation or Granger causality, TE captures nonlinear interactions and requires no model assumptions about the underlying dynamics.

ScholarGate
  1. Machine learning
  2. v1
  3. 1 Sources
  4. PUBLISHED
Cite this page →
Tools & resources
Download slides
Learn & explore

Read the full method

Members only

Sign in with a free account to read this section.

Sign in

Method map

The neighbourhood of related methods — select a node to explore.

Transfer Entropy
Convergent Cross MappingGranger CausalitySample EntropyHiemstra-Jones Causality

When to use it

Transfer Entropy is appropriate when you suspect nonlinear directional coupling between two time series and cannot assume a parametric model. It is well-suited to neuroscience (neural spike trains), finance (market contagion), climatology, and complex systems research. Assumptions: the series should be stationary or made stationary, and the sample size must be large enough for reliable density estimation (typically N > 500 per embedding dimension). Limitations include sensitivity to binning choices, high computational cost for long embeddings, and inflated estimates under shared common drivers. Granger causality is a faster linear alternative; convergent cross-mapping may outperform TE on weakly coupled deterministic systems.

Strengths & limitations

Strengths
  • Model-free and nonlinear: detects directional dependencies of any functional form without assuming linearity or Gaussianity.
  • Directional and asymmetric: T_{Y→X} and T_{X→Y} can differ, revealing the net direction of information flow.
  • Rooted in rigorous information theory: directly interpretable as reduction in uncertainty measured in bits or nats.
  • Widely applicable across disciplines, from neuroscience and finance to climate science and engineering.
Limitations
  • Sensitive to the choice of embedding dimensions k and l and to probability estimation method (binning vs. kernel vs. kNN).
  • Requires relatively large sample sizes; performance degrades noticeably for N < 300–500 observations.
  • Does not distinguish direct causal influence from indirect paths or common-cause confounding without conditioning on third variables.
  • Computationally expensive for high-dimensional embeddings or multivariate extensions involving many variables.

Frequently asked

How does Transfer Entropy differ from Granger causality?

Granger causality tests whether past values of Y improve a linear predictive model of X. Transfer Entropy makes no linearity assumption: it measures the same directional predictive gain using entropy rather than variance reduction. For Gaussian linear processes both measures are equivalent, but TE detects nonlinear coupling that Granger causality misses. The cost is higher data demand and greater sensitivity to estimation choices.

What embedding dimensions k and l should I choose?

There is no universal answer. Common practice is to select k and l using the false nearest-neighbour method, Cao's criterion, or Bayesian model selection on the estimated TE. Starting with k = l = 1 and increasing until TE stabilises is a pragmatic first step. Unnecessarily large embeddings increase the curse of dimensionality and estimation variance.

Can Transfer Entropy establish true causality?

No. Like all observational measures, TE detects predictive information flow, not interventional causality in the Pearl or Rubin sense. A common latent driver, measurement noise, or non-stationarity can all produce positive TE without any physical causal link. Complementary approaches such as conditional TE, time-delayed mutual information, or controlled experiments are needed to strengthen causal claims.

Sources

  1. Schreiber, T. (2000). Measuring information transfer. Physical Review Letters, 85(2), 461–464. DOI: 10.1103/PhysRevLett.85.461 ↗

How to cite this page

ScholarGate. (2026, June 2). Transfer Entropy. ScholarGate. https://scholargate.app/en/causal-inference/transfer-entropy

Related methods

Convergent Cross MappingGranger CausalitySample Entropy

Which method?

Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.

  • Convergent Cross MappingCausal inference↔ compare
  • Granger CausalityEconometrics↔ compare
  • Sample EntropyComplex Systems↔ compare
Compare side by side →

Referenced by

Convergent Cross MappingHiemstra-Jones Causality

Similar methods

Nonlinear Granger CausalityConvergent Cross MappingGranger Causality TestGranger CausalitySample EntropyNonlinear Toda-Yamamoto CausalityRecurrence Quantification AnalysisBayesian Granger Causality

Related reference concepts

Causal InferenceSensitivity AnalysisPermutation TestsDirected Acyclic GraphHidden Markov ModelsDimension Reduction

Spotted an issue on this page? Report or suggest a fix →

ScholarGate — Transfer Entropy (Transfer Entropy). Retrieved 2026-07-21 from https://scholargate.app/en/causal-inference/transfer-entropy · Dataset: https://doi.org/10.5281/zenodo.20539026
Quick facts
Originator
Thomas Schreiber
Year
2000
Type
Non-parametric information-theoretic measure
Subfamily
Information-theoretic causality
Framework
Shannon entropy / Kullback-Leibler divergence
Data Requirement
Stationary or near-stationary time series
Related methods
Convergent Cross MappingGranger CausalitySample Entropy
ScholarGate

A content-first reference library for research methods — what each one is, how it works, and where it comes from.

Open data (CC-BY)

Explore

  • Library
  • Search the library…
  • Browse by field
  • Fields
  • Journey
  • Compare
  • Which method?

Reference

  • Subjects
  • Atlas
  • Glossary
  • Methodology
  • Philosophy

Your tools

  • Bookshelf
  • Desk
  • Chat

Company

  • About
  • Pricing
  • Contact
  • Suggest a method

Entries are compiled from published sources for reference. Verifying the accuracy and suitability of any information for your own use remains your responsibility.

© 2026 ScholarGate · A research-method reference library
  • Privacy
  • Cookies
  • Terms
  • Delete account