MultipleComparisonCorrection¶
Correct p-values for multiple comparisons (multiple dependent variables).
This should typically be performed on mass-univariate statistics, e.g., the univar_stats output of the MANCOVA node. Each factor (i.e. explanatory-variable) level will be corrected independently, which are assumed to be indexed by a feature axis. Optionally each space axis entry (channel) may be corrected independently, as well. All other axes (indexing dependent variables) will be treated as flattened for the purposes of the correction. More Info... Version 1.1.0
Ports/Properties¶
data¶
Packet with stats chunk to process.
method¶
Method used for testing and adjustment of pvalues. Available methods to control for the family-wise error rate (FWER) are: - bonferroni : The highly conservative Bonferroni correction; best if the tested variables tend to be independent, and only suitable for at most a modest number of variables (least powerful in a statistical sense). - sidak : The Sidak correction; slightly less conservative / more powerful than Bonferroni (not conservative in the presence of negative dependencies between variables) - holm : A "step-down" (sequential testing) variant of Bonferroni, uniformly more powerful than the former, and can thus handle more variables. - holm-sidak : Step-down variant of the Sidak formula; uniformly more powerful and can handle more variables than Sidak, but inherits the caveat about negative dependencies. - simes-hochberg : A "step-up" procedure that is applicable for variables that are not negatively dependent, with different assumptions and caveats compared to holm and holm-sidak. - hommel : A "closed testing" method also based on Simes tests but with somewhat different assumptions and caveats. Not applicable in the presence of negatively-correlated variables. Slow for large arrays. Available methods to control for false discovery rate (FDR) are: - fdr_bh : The classic Benjamini/Hochberg correction for FDR; this method is not applicable if variables can be negatively dependent. - fdr_by : The Benjamini/Yekutieli correction for FDR; this method does handle any kind of dependency, including negative dependency. - fdr_tsbh : A two-stage variant of the Benjamini-Hochberg method that depends on the given alpha level (with higher statistical power than fdr_bh). Not applicable if variables can be negatively dependent. - fdr_tsbky : The Benjamini-Krieger-Yekutieli correction for FDR, also a two-stage variant, but one that can handle any kind of dependency structure (including negative). Corrected p-values depend o the alpha level, which needs to be supplied.
space_is_independent¶
Each item in the space axis (i.e . each channel) will be corrected independently of other items in the space axis. Since the correction will not be representative of the total number of comparisons made, this should be noted alongside the results, if used. If set to False, all comparisons present in the data will be accounted for (more conservative).
alpha¶
Desired alpha level for two-stage procedures. This value is only required for the two-stage FDR procedures and represents the false discovery rate that shall be maintained (i.e., the same alpha that shall be used in a subsequent significance test of the corrected p-values).
set_breakpoint¶
Set a breakpoint on this node. If this is enabled, your debugger (if one is attached) will trigger a breakpoint.
metadata¶
User-definable meta-data associated with the node. Usually reserved for technical purposes.