Hi Philippe,

Here's what I understand of conjunction analyses from a qualitative standpoint, from several readings of "Cognitive Conjunction: A new approach to brain activation experiments" Price and Friston, Neuroimage 5, 261-270 (1997).

These explications are qualitative, so please correct me if I misuse terms...

Orthogonal means "independent" from a mathematical standpoint. In the parlance of a conjunction analysis, I believe this means to use separate data sets in each group. So a conjunction of A vs. B and C vs. B would each use half of the available data from B, i.e., B1 and B2, so that none of the data sets overlap and the observations are independent in a mathematical sense.

The finding of a conjunction, according to Price and Friston, is the presence of a main effect in the absence of an interaction.

We have tried the following and it seems to work from a qualitative standpoint, although, honestly, the mathematics in the appendix of this article have so far eluded me.

To force your data into a conjunction analysis, you will need to use 3dRegAna if you don't have equal numbers of datasets in A, B1, B2, and C. Otherwise 3dANOVA2 might work.

A main effect of task would be represented by (A-B1)+(C-B2) and in a single 3dRegAna column one would assign A and C datasets +1, and B1 and B2 datasets -1. An interaction, represented by (A-B1)-(C-B2), would in a single 3dRegAna column assign A and B2 datasets +1, and B1 and C datasets -1.

Your "Conjunction Analysis" results would then be contained in the output for these two columns, with one caveat. The main effect shows you where activation is conjoint(?). The interaction shows where there are differences between the difference of each Experimental-Control task pair. The caveat is...If you see a main effect, it constitutes a conjunction if and only if the same voxels do not show an interaction. You could use 3dcalc to mask the main effect brik with the interaction brik.

Good luck and let us know what you figure out.