Statmaps

Statistical Maps

By running a univariate test such as a t-test or ANOVA on every sample in a pattern, one can create a statistical map to determine which time samples/electrodes/frequencies show a given effect.

Creation

Statistical maps are created using pattern_statmap. It is flexible, and can use any function that follows one of the following forms:

[a,b,c,...] = f_stat(x, ...)

or

[a,b,c,...] = f_stat(x, group, ...)

If the function uses regressors. group is a cell array where each cell contains one factor, which may be represented as a numeric array or a cell array of strings. Each unique value in each factor represents a different group. The regressors are generated from the events structure associated with the pattern; the reg_defs input to pattern_statmap defines how the events should be divided up. reg_defs is a cell array, where each cell gives the definition for one factor. See make_event_index for allowed definition types.

The test will be run on each sample in the pattern, and saved to a new stat object. By default, the variables in the new file will be named p, statistic, and res. The statistics function may return either a scalar for each output, or vectors if it is testing multiple effects (e.g. an N-way ANOVA).

Note that in order to plot the statistical map with the toolbox plotting functions, significance values must be saved in a variable named "p."

Plotting

Plotting functions in the toolbox can plot statistics generated from pattern_statmap. First, you must indicate the stat_name you indicated when running pattern_statmap; this specifies which statistics object should be used. The significance map "p" will be loaded. There is also an optional input stat_index which allows you to choose which p-value to plot if the test includes multiple effects. stat_index may be an integer, or a string indicating the name of an effect if there is a cell array of strings called "names" saved in the stat file.

Finally, there are two params which determine whether a given sample should be plotted as significant: alpha and correctm. alpha sets the criterion for a p-value to be significant (default is 0.05). correctm specifies options for running correction for multiple comparisons, and can run Bonferroni or false discovery rate (FDR) correction (default is to plot uncorrected significance).