Dear Marilu,
>We are still a bit struggling with this....following the example you
>gave below, what would we be testing if we are doing [1 1]? A main
>effect of all regressors? More explicitly, if we are looking at scores
>in naming living and non-living items, would the [1 1] contrast give us
>the areas that correlate with both or not? And what would we be looking
>at if we do 1 -1?
No, it would give the areas that have, on average, a positive correlation
with the two regressors. It is easier to think about the contrast of
parameter estimates in terms of regression coefficients, as opposed to
correlations. If you plot the data against one regressor, the parameter
estimate is the slope of the ensuing regression. The contrast [1 1]
tests for the average of the slopes if you plotted the data against both
regressors. The contrast [1 -1] is the difference in slopes. An F-contrast
matrix [1 0; 0 1] tests for either a significant slope in one, the other
or both slopes. A conjunction of [1 0] and [0 1] tests for positive slopes in
both. Is this clearer?
With very best wishes,
Karl
|