Reply to Chen et al.: Parametric methods for cluster inference perform worse for two-sided t-tests

Area of Science:

• Neuroimaging
• Statistical analysis

Background:

• One-sided t-tests are prevalent in neuroimaging research.
• Previous work focused on cluster false positive rates using one-sided tests.

Purpose of the Study:

• Extend previous findings on cluster false positive rates to two-sided t-tests.
• Evaluate the performance of statistical methods for two-sided hypothesis testing in neuroimaging.

Main Methods:

• Simulation studies to assess cluster false positive rates.
• Comparison of parametric and nonparametric statistical methods.
• Analysis extended from one-sided to two-sided t-tests.

Main Results:

• Parametric methods demonstrate poorer performance with two-sided t-tests.
• Nonparametric methods exhibit consistent performance for both one-sided and two-sided tests.
• False positive rates were evaluated for cluster-based analyses.

Conclusions:

• Two-sided t-tests should be the default statistical approach in neuroimaging unless strongly justified otherwise.
• Nonparametric methods offer a robust alternative for statistical testing in neuroimaging, particularly for two-sided hypotheses.