Thanks for the guidance re differences in N's - good to know.
Next question is re determination of degrees of freedom, given the "layered" tests I'm going to be doing.
Suppose the regression is ln(cu) on ln(c/L) and I compute it using averaged c for uL (0 < u <= 1), uH (u > 1), and uA (u > 0) for S63, C711, and S63R.
In the first "layer" of significance testing, I test:
1a) uL vs uH for S63 1b) uL vs uA for S63 1c) uH vs UA for S63
2a) uL vs uH for C711 2b) uL vs uA for C711 2c) uH vs UA for C711
3a) uL vs uH for S63R 3b) uL vs uA for S63R 3c) uH vs UA for S63R
In the second "layer", I test:
4a) uL for S63 vs uL for C711 4b) uL for S63 vs uL for S63R 4c) uL for C711 vs UL for S63R
5a) uH for S63 vs uH for C711 5b) uH for S63 vs uH for S63R 5c) uH for C711 vs UH for S63R
6a) uA for S63 vs uA for C711 6b) uA for S63 vs uA for S63R 6c) uA for C711 vs UA for S63R
Should the degrees of freedom for tests (1a-c) thru (3a-c) take into account the fact that tests (4a-c) thru (6a-c) will also be run?
Or should the degrees of freedom for tests (1a-c) thru (3a-c) be determined independently of the fact that tests (4a-c) thru (6a-c) will follow?
In addition to a general (theoretical) answer), it would very much help cut down on the possibility of my misunderstanding your theoretical ansswer if you would also take a moment to specify the degree of freedom that you want me to use for each of the 18 tests (1a- c) thru (3a-c) and (4a-c) thru (6a-c).
Thanks very much - this is another question that occurred to me and I knew I didn't know the answer to it.