One-sample T-test with unequal number of samples per voxel

Hello -

I’m plotting data from microelectrode recordings into anatomical space. Because patients can have variable recording sites, the spatial density of all plotted statistical is unequal. How can I test whether my plotted statistical values are observed above chance, while accounting for the variable number of recordings/samples at each voxel?

If it clarifies things, I have the data plotted both individually (1 statistical value - 1 recording site coordinate), and as combined datasets (1 containing the number of recordings at each voxel, 1 containing the summed statistical values at each voxel).

Thank you (and sorry if this is obvious),

and to further clarify, my ultimate question is whether statistical values are observed above chance in specific voxels within the entire recording site blob


To me that sounds like having missing data, which is not handled by 3dttest++.

However, 3dMEMA can handle missing data, so that might be a useful generalization? There is the “-missing_data 0” option there.


Hi Pete,

Actually, 3dttest++ can indeed handle this, but there has to be the same number of values per location (the datasets themselves cannot be sparse). For locations with missing data, specify the value as zero. Then use 3dttest++ -zskip to skip those zero values (so each location would have varying degrees of freedom, hence the automatic application of -toz).

As Paul noted, 3dMEMA is a possibility, but then you would need the first level t-stats, if you have them.

  • rick