Thanks, that reference cleared a lot up. But this seems like a case of making the intercept implicit by using homogeneous coordinates, rather than eliminating the intercept altogether. This is the approach that SPM always takes, and it doesn't seem to cause trouble when a model is just the interaction of several variables (quantitative or factorial), corresponding to an R model formula of (IIUC) `y ~ factor1*factor2`. I haven't used SPM enough to know how it handles `y ~ factor1 + factor2` though.