"To test the average difference between 1S-1C and R1S-R1C at uL, let Y = the sum of the values in the uL:1S-1C column minus the sum of the values in the uL:R1S-R1C column, and use the expanded point 4 to get var_Y and df_Y. Then do the same thing for uH: let Z = the sum of the values in the uH:1S-1C column minus the sum of the values in the uH:R1S-R1C column, and use the expanded point 4 to get var_Z and df_Z. Then there are three t-tests: Y/sqrt[var_Y], Z/sqrt[var_Z], and (Y-Z)/sqrt[var_Y + var_Z], where the df for the last test is obtained using the expanded point 4."
I executed this protocol for the case of 1:R1 and got:
for Y: p ~ .95 (at uL) for Z: p ~ .33 (at uH) for(Y-Z): p ~ .23 (across uL,uH)
Don't know whether you'll think these are strong enough to keep going.
What I'm hoping you'll say is that the protocol indicates that the S-C effect is demonstrably non-existent at uL, at least possibly present at uH, and at least possibly different between uL and uH.
But of course, it's your call.
While waiting for your evaluation, I am going to execute the same protocol for 2:R2 and 3:R3, just to get a sense of how 1:R1 differs from 2:R2 and 3:R3, if at all.
Also, if you still think there's reason for me to keep going and execute the protocol on the regression coefficients, should I use the SE's or var's of the coefficients? In the original CHTT's on 12 rows, we used the SE's of the coefficients, but shouldn't I be using the var's themselves in this particular protocol?
Thanks as always for the time you spend considering these matters. I'd rather have a (0.95,0.33,0.23) that you think may be worth something than a baloney (1,0.01,0.04), any day.