You just say "I want a FWE of 0.05" and it figures the rest out for you. *randomise has one MAJOR MAJOR perk: you can use Threshold-Free Cluster Enhancement (TFCE) which allows you to not have to chose a cluster forming threshold.FSL offers the great randomise*, SPM offers SnPM, and there is a tool called BROCCOLI by Eklund which promises GPU-fast permutation testing. For group-level full brain analysis: Use permutation testing. For ROI-correlations: instead of Pearson’s correlation, use Spearman’s rank correlation or Wilcoxon rank correaltion.The main takeaways from the presentation and papers below is the following: use non-parametric methods. In this presentation I cover two situations in which assumption infringement might cause misleading (or entirely erroneous) conclusions, suggesting that it might be better to apply non-parametric methods such as Spearman or Wilcox Skipped Correlations (for correlations) or permutation testing (for group level inference). BOLD time-series are known not to meet the several assumptions of parametric testing (see this paper for an overview), particularly with respect to homoschedasticity (i.e., the assumptions that the variances are equal across - for instance - your subjects) and homogeneity of the sample (i.e., outliers). Assessing whether your data meets the assumptions of the model you use to analyze it is fundamental to ensure validity of the analysis.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |