Dear Andy,
The contrast vector tells you how to combine the beta estimates.
- With [1 1] you test whether 1* beta 1 + 1* beta 2 (= the sum) is significantly different from zero.
- With [0.5 0.5] you test whether 0.5 * beta 1 + 0.5 * beta 2 = (beta 1 + beta 2)/2 (= the average) is significantly different from zero.
Thus the contrast estimates obtained with [1 1] are twice as large.
Best
Helmut
|