Statistical inference

From FMRI Methods Wiki

Jump to: navigation, search

Inference on Statistic Image (thresholding)

NB: These points refer primarily to classical inference; an additional set of guidelines is necessary for Bayesian inference.


  • Type of search region considered, and the volume in voxels or CC.
    • If not whole brain, how region was found; method for constructing region should be independent of present statistic image.
    • If threshold used for inference and threshold used for visualization in figures is different, clearly state so and list each.
  • Explicitly state if inferences are corrected for multiple comparisons, and if so, what method and over what region.
    • If correction is limited to a small volume, the method for selecting the region should be stated explicitly.
    • If no formal multiple comparisons method is used, the inference must be explicitly labeled "uncorrected".
  • Voxel-wise significance? Corrected for Familywise Error (FWE) or False Discovery Rate (FDR).
    • If FWE found by random field theory (e.g. with SPM) list the smoothness in mm FWHM and the RESEL count.
    • If FWE found by simulation (e.g., AFNI AlphaSim), provide details of parameters for simulation
    • If not a standard method, specify the method for finding significance (e.g. "Custom in-lab software was used to construct statistic maps and thresholded at FDR<0.05 (Benjamini & Hochberg 1995)".
  • Cluster-wise significance?
    • State cluster-defining threshold (e.g. P=0.01)
    • State the corrected cluster significance level
      • E.g. "Statistic images were assessed for cluster-wise significance using a cluster-defining threshold of P=0.01; the 0.05 FWE-corrected critical cluster size was 103."
    • If significance determined with random field theory, then smoothness and RESEL count must be supplied.
  • Correction for multiple planned comparisons within each voxel?
  • False Negative Discussion
    • Any discussion of failure to reject the null hypothesis (e.g., lack of activation in a particular region) should be accompanied by an estimate of the actually observed effect (allows reader to infer power to estimate an effect).


ROI analysis

  • How were ROI's defined
    • E.g., functional versus anatomical localizer?
  • How was signal extracted within ROI?
    • E.g., average parameter estimates, FIR deconvolution?
    • If percent signal change reported, how was scaling factor determined (e.g. height of block regressor or height of isolated event regresor)? Is change relative to voxel-mean, or whole-brain mean?
Personal tools