Multiple-Comparisons Correction in Qdec
see also this page for info on m-c correction using mri_glmfit
Muliple comparisons correction refers to the need to correct a significance level for the number of hypothesis tests performed. In brain mapping, the number of hypothesis tests is typically associated with the number of voxels or surface vertices and is therefore massive. Methods for multiple comparisons correction include:
With so many vertices in the significance maps produced in Qdec, it is likely that many vertices will appear signficant purely by random chance (ie, a false positive). This is known as the "Problem of Multiple Comparisons". Qdec implements two forms of correction: False Discovery Rate (FDR), and simulation (cluster analysis). In a cluster analysis, only clustered vertices are retained, the idea being that false positives will not appear next to each other.
False Discovery Rate
In qdec, once your analysis is complete, and the Design tab displays your hypothesis in the 'Scalars' menu, and you have selected one of these hypothesis for display, such that their significance values are painted on the average surface, then to use the FDR method of correcting for multiple comparisons, you simply press the Set Using FDR button found beneath the Scalars menu list. The rate used defaults to 0.05, and is adjustable via the entry box next to the FDR button (note: press ENTER after updating that value!). When the FDR button is pressed, the minimum threshold is set to that found from the FDR calculation. The mid-point is merely a small fraction above that minimum threshold, and the max is 2.25 above the min (2.25 is arbitrarily chosen). If no values meet the FDR criterion (when the FDR algorithm is run when the 'Set Using FDR' button is pressed), then a value slightly smaller than the smallest p is used as the minimum threshold.
Qdec supports correction for multiple comparisons by method of Monte Carlo simulation. To speed this simulation, some precomputed data is used, which limits the functionality to particular p-value thresholds, as found in the drop-down menus. In particular, thresholds of 1.3, 2, 2.3, 3, 3.3 and 4, corresponding to p-values of 0.05, 0.01, 0.005, 0.001, 0.0005 and 0.0001, which are common thresholds. The other drop-down menu allows selection of abs, pos or neg, which in case of pos and neg selection, limits values to positive or negative. The default, abs, considers the absolute value. Pressing the 'Run' button runs the simulation, lasting just few seconds, and then the cluster results are displayed graphically, and to the terminal. Interpretation of the cluster results is described here.
Note: The method used is based on: Smoothing and cluster thresholding for cortical surface-based group analysis of fMRI data. Hagler DJ Jr, Saygin AP, Sereno MI. NeuroImage (2006).