Type I and Type II error concerns in fMRI research: re-balancing the scale

Area of Science:

• Neuroimaging
• Statistical analysis
• Cognitive neuroscience

Background:

• fMRI statistical thresholding has become more conservative to reduce Type I errors (false positives).
• This shift aims to match traditional behavioral science standards but has unintended consequences.
• Conservative thresholds may increase Type II errors (false negatives) and bias research focus.

Purpose of the Study:

• To examine the negative consequences of increasingly conservative statistical thresholds in fMRI.
• To evaluate the impact on Type II errors, effect size bias, and research focus.
• To propose alternative thresholding strategies for a better balance of error rates.

Main Methods:

• Analysis of statistical thresholding trends in fMRI over the past decade.
• Power analyses to assess the impact of reduced P-values on Type II error rates.
• Simulations to evaluate combined intensity and cluster size thresholds (e.g., P < 0.005 with 10 voxel extent).

Main Results:

• Conservative P-values significantly increase Type II error rates in fMRI.
• A combined threshold of P < 0.005 with a 10 voxel extent balances Type I and Type II errors effectively.
• This joint threshold yields a false discovery rate (FDR) comparable to typical behavioral science research.

Conclusions:

• Overly stringent fMRI thresholds lead to missed true effects and biased research.
• Combined intensity and cluster size thresholds offer a more balanced approach to statistical inference.
• Prioritizing replication and meta-analysis over single-study significance is crucial for scientific truth.