Hi Robie,

So though there are conceptual 30s blocks, some
individual stimulus types last 1, 2, or 8 seconds,
is that correct? What are the typical durations
in the dmBLOCK cases?

Also, do you ever want to contrast the ex or inc
betas with the but, bug or ins betas?


The latex-inc contrast in the second model is not
well balanced, which applies to most of them. And
none are weighted.

Keep in mind that these contrasts are simply sums
of betas.

A. One often wants a GLT to be balanced so that
main effects are not shown. By that I mean a GLT
of "A B -C" might show a big effect, even if A, B
and C are all basically the same.

It is not balanced with a sum of zero. Consider
the GLT of "A A -A" similarly. It would show
the exact same result as A, not as a zero result
contrast between A and itself.


B. A smaller point is that since these contrasts
are sums, adding them up is magnifying the betas.
While that does not affect their statistics, it
might prove a little confusing when discussing
the magnitudes of the contrasts.


So getting back to latex-inc for example, let me
suppose that you really want to compare the average
inc beta against the average ex beta (rather than
the sum of 6 inc betas against only 3 ex betas).
Then the GLT might read:

'SYM: -0.1667*inc1 -0.1667*inc3 ... 0.333*ex9 ...'


As it stands, summing 6 against 3, the 'inc' betas
might "win out", giving a result that looks much
like the 'inc' result to begin with.

At least the first ex-inc_full contrast has a sum
of 0 for the weights (6-6). So while it is inflated,
it seems quite reasonable.


I would only expect 2 differences between the first
ex-inc and the second ex-inc_full. One is because
of the scaling by 6 in the second case. The other
is from a degree of freedom difference. The second
model gives up more degrees of freedom to the
regressors.

- rick