Elsevier

NeuroImage

Volume 59, Issue 2, 16 January 2012, Pages 1228-1229
NeuroImage

Technical Note
The danger of systematic bias in group-level FMRI-lag-based causality estimation

https://doi.org/10.1016/j.neuroimage.2011.08.015Get rights and content

Abstract

Schippers, Renken and Keysers (NeuroImage, 2011) present a simulation of multi-subject lag-based causality estimation. We fully agree that single-subject evaluations (e.g., Smith et al., 2011) need to be revisited in the context of multi-subject studies, and Schippers' paper is a good example, including detailed multi-level simulation and cross-subject statistical modelling. The authors conclude that “the average chance to find a significant Granger causality effect when no actual influence is present in the data stays well below the p-level imposed on the second level statistics” and that “when the analyses reveal a significant directed influence, this direction was accurate in the vast majority of the cases”. Unfortunately, we believe that the general meaning that may be taken from these statements is not supported by the paper's results, as there may in reality be a systematic (group-average) difference in haemodynamic delay between two brain areas. While many statements in the paper (e.g., the final two sentences) do refer to this problem, we fear that the overriding message that many readers may take from the paper could cause misunderstanding.

Highlights

► Group-level FMRI simulations can be useful to test methods such as Granger causality. ► Simulation results need careful evaluation and interpretation. ► There is ample evidence of haemodynamic variability across regions and voxels. ► Lag-based FMRI causality analysis may be biassed by such variation. ► This confound should be considered when reporting lag-based results.

Section snippets

Likelihood of a true direction given a detected direction

When the main conclusions are drawn using Table 6 (in Schippers et al., 2011), it is effectively assumed that if the differential haemodynamic lag between two brain areas is unknown, then on average (across all subjects in a study), it is zero. This arises because the results are pooled across the full (negative and positive) range of potential differential haemodynamic lags. In reality, a given pair of brain regions will have a specific population-average differential haemodynamic lag, so in

Which null to use for causality p-values

We are also concerned that the claim of controlling false positives (first quotation in the Abstract) is in danger of being misunderstood. The false positive evaluations only considered the specific null hypothesis of complete independence between two areas. In practice, most applications of lag-based causality estimation pre-suppose that two timeseries are correlated.2

Conclusion

There is ample evidence of haemodynamic variability—between neighbouring voxels within a functional region, between different functional regions, over different cognitive conditions, and between different subjects (e.g., see discussions and references in Miezin et al., 2000 and Bandettini, 19993

Acknowledgments

We are very grateful to Christian Keysers and Alard Roebroeck for helpful discussions.

Cited by (0)

View full text