This document is an Author Correction for a previously published research paper in Nature Communications. Because the provided text is a brief formal amendment rather than the full original study, this summary focuses on the technical statistical clarification provided by the authors.
TL;DR
The authors clarify the specific statistical methodology used to calculate effect sizes (Cohen’s d) in their study of adolescent brain asymmetry. By using fixed-effect estimates from Linear Mixed-Effects Models (LMMs) divided by residual standard deviation, they account for the hierarchical nature of their data while maintaining the validity of their original scientific conclusions.
What Problem Does This Paper Address?
In complex neuroimaging and behavioral studies, researchers often use Linear Mixed-Effects Models (LMMs) to handle "nested" or hierarchical data (e.g., multiple measurements from the same individual or participants grouped by study site).
The "classical" definition of Cohen’s d—a measure of effect size—is typically calculated using the difference between two group means divided by a pooled standard deviation. However, applying this classical formula directly to LMM outputs can be misleading because it doesn't account for the random effects (the "mixed" part of the model). The authors issued this correction to transparently define how they adapted this metric for their specific modeling framework to avoid potential misinterpretation by the community.
Methods
The authors specify that their calculation of Cohen’s d was performed as follows:
- Fixed-Effect Estimate: They took the coefficient (beta) from the LMM.
- Residual Standard Deviation: They divided that estimate by the residual standard deviation of the model.
This approach provides a semi-standardized effect size. While it differs from the textbook group-comparison method, it is a recognized and accepted approximation within the LMM framework. It allows the effect size to reflect the data's hierarchical structure (including random effects) more accurately than a simple mean-based calculation would.
Datasets
Not discussed in the correction. (The original title implies the use of adolescent neuroimaging and psychiatric data, but specific database names like ABCD or IMAGEN are not mentioned in this amendment).
Key Insights and Ideas
- Context-Dependent Metrics: The "standard" way of calculating statistics often needs to be adapted when moving from simple experimental designs to complex, multi-level modeling.
- Transparency in "Semi-Standardization": The authors emphasize that this method provides a "semi-standardized" effect size. This is a crucial distinction for meta-analysts who might try to compare these results with those from standard t-tests or ANOVAs.
- Robustness of Conclusions: Despite the technical nuance in how the effect size was calculated, the authors state that the underlying scientific findings regarding brain asymmetry and adolescent outcomes remain unchanged.
Key Results
- Methodological Validation: The authors cite four key papers (Feingold 2019, Brysbaert & Stevens 2018, etc.) to support the validity of their chosen effect size calculation within the LMM framework.
- Stability of Findings: The primary quantitative and qualitative conclusions of the original study are reaffirmed and remain unaffected by this clarification.
Limitations and Future Work
Not discussed in the correction. The document is strictly limited to clarifying the statistical reporting of the original work.
Summary updated