Xavier et al., An important secondary factor is the improved intersubject alignment provided by “MSMAll” alignment (based on areal features) over folding-based alignment (MSMSulc or fsaverage). Having better alignment of functionally defined areas increases effect sizes (and z-stats as well), making the type of analysis Matt recommends more robust whatever the number of subjects in a particular analysis.
These and a variety of other recommendations for "HCP-style” analysis are discussed more fully in Glasser et al., (Nature Neuroscience, 2016; https://www.ncbi.nlm.nih.gov/pubmed/27571196). David > On Jan 26, 2017, at 4:33 PM, Glasser, Matthew <[email protected]> wrote: > > Standard error scales with sample size, standard deviation does not. Things > like Z, t, and p all also scale with sample size and are measures of > statistical significance via various transformations. Thus for a large group > of subjects, Z and t will be very high and p will be very low. Z, t and p > are thus all not biologically interpretable, as their values also depend on > the amount and quality of the data. In the limit with infinite amounts of > data, the entire brain will be significant for any task, but wether a region > is statistically significant tells us little about its importance > functionally. Measures like appropriately scaled GLM regression betas, %BOLD > change, or Cohen’s d are biologically interpretable measures of effect size > because their values should not change as sample size and data amount go up > (rather the uncertainty on their estimates goes down). Regions with a large > effect size in a task are likely important to that task (and will probably > also meet criteria for statistical significance given a reasonable amount of > data). > > A common problem in neuroimaging studies is showing thresholded statistical > significance maps rather than effect size maps (ideally unthresholded with an > indication of which portions meet tests of statistical significance), and in > general focusing on statistically significant blobs rather than the effect > size in identifiable brain areas (which should often show stepwise changes in > activity at their borders). > > Peace, > > Matt. > > From: <[email protected] > <mailto:[email protected]>> on behalf of Xavier Guell > Paradis <[email protected] <mailto:[email protected]>> > Date: Thursday, January 26, 2017 at 3:46 PM > To: "[email protected] <mailto:[email protected]>" > <[email protected] <mailto:[email protected]>> > Subject: [HCP-Users] Very large z values for task contrasts in > S900_ALLTASKS_level3_zstat file: what does this mean in terms of statistical > significance? > > Dear HCP team, > I have seen that the zstat values for tasks contrasts are very large in the > HCP_S900_787_tfMRI_ALLTASKS_level3_zstat1_hp200_s2_MSMAll.dscalar.nii file, > to the point that one can observe areas of activation in task contrasts by > setting very high z value thresholds (e.g., a z threshold of +14). > I think (please correct me if I'm wrong) that the z values of the S900 file > are very large because the group is very large, therefore the standard > deviation is very small (given that there will be less variability in a group > if one takes a very large group of people rather than a small group of > people), and if the standard deviation is very small then even small > differences from the mean will lead to very large z values. > > I was wondering what implication does this have in terms of statistical > significance. A z value of 14 or larger would correspond to an extremely > small p value, i.e. it would be extremely unlikely to observe by chance a > measure which is 14 times the standard deviation away from the mean. Would it > therefore be correct to assume that the areas that we can observe in the S900 > tfMRI_ALLTASKS task contrasts with a very high zstat threshold (e.g., 14) are > statistically significant, without having to worry about multiple comparisons > or family structure? > > Thank you very much, > Xavier. > _______________________________________________ > HCP-Users mailing list > [email protected] <mailto:[email protected]> > http://lists.humanconnectome.org/mailman/listinfo/hcp-users > <http://lists.humanconnectome.org/mailman/listinfo/hcp-users> > > The materials in this message are private and may contain Protected > Healthcare Information or other information of a sensitive nature. If you are > not the intended recipient, be advised that any unauthorized use, disclosure, > copying or the taking of any action in reliance on the contents of this > information is strictly prohibited. If you have received this email in error, > please immediately notify the sender via telephone or return mail. > _______________________________________________ > HCP-Users mailing list > [email protected] <mailto:[email protected]> > http://lists.humanconnectome.org/mailman/listinfo/hcp-users > <http://lists.humanconnectome.org/mailman/listinfo/hcp-users> _______________________________________________ HCP-Users mailing list [email protected] http://lists.humanconnectome.org/mailman/listinfo/hcp-users
