Hi there,
I had a question about 3dANOVA2 and how it calculates means.
We used 3dANOVA2 to perform an ANOVA on 12 subjects, using sub-briks of interest of the coefficient files generated via 3dDecovolve for each subject as inputs to the ANOVA.
Here's the ANOVA command we used:
3dANOVA2 -type 3 -alevels 4 -blevels 12 \
-dset 1 1 'gp_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[6]' \
-dset 2 1 'gp_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[7]' \
-dset 3 1 'gp_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[8]' \
-dset 4 1 'gp_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[9]' \
\
-dset 1 2 'sk_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[6]' \
-dset 2 2 'sk_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[7]' \
-dset 3 2 'sk_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[8]' \
-dset 4 2 'sk_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[9]' \
\
-dset 1 3 'wt_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[6]' \
-dset 2 3 'wt_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[7]' \
-dset 3 3 'wt_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[8]' \
-dset 4 3 'wt_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[9]' \
\
-dset 1 4 'cg_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[6]' \
-dset 2 4 'cg_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[7]' \
-dset 3 4 'cg_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[8]' \
-dset 4 4 'cg_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[9]' \
\
-dset 1 5 'ja_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[6]' \
-dset 2 5 'ja_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[7]' \
-dset 3 5 'ja_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[8]' \
-dset 4 5 'ja_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[9]' \
\
-dset 1 6 'rd_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[6]' \
-dset 2 6 'rd_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[7]' \
-dset 3 6 'rd_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[8]' \
-dset 4 6 'rd_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[9]' \
\
-dset 1 7 'jd_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[6]' \
-dset 2 7 'jd_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[7]' \
-dset 3 7 'jd_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[8]' \
-dset 4 7 'jd_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[9]' \
\
-dset 1 8 'gm_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[6]' \
-dset 2 8 'gm_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[7]' \
-dset 3 8 'gm_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[8]' \
-dset 4 8 'gm_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[9]' \
\
-dset 1 9 'hg_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[6]' \
-dset 2 9 'hg_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[7]' \
-dset 3 9 'hg_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[8]' \
-dset 4 9 'hg_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[9]' \
\
-dset 1 10 'rh_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[6]' \
-dset 2 10 'rh_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[7]' \
-dset 3 10 'rh_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[8]' \
-dset 4 10 'rh_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[9]' \
\
-dset 1 11 'sl_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[6]' \
-dset 2 11 'sl_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[7]' \
-dset 3 11 'sl_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[8]' \
-dset 4 11 'sl_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[9]' \
\
-dset 1 12 'nr_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[6]' \
-dset 2 12 'nr_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[7]' \
-dset 3 12 'nr_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[8]' \
-dset 4 12 'nr_coefSubMemSTIMONLYxREWARDRegscaled_RMTB6+tlrc[9]' \
\
-amean 1 stimHRc12 \
-amean 2 stimLRc12 \
-amean 3 stimHF \
-amean 4 stimLF \
-adiff 1 2 stimhi-loRc12 \
-adiff 3 4 stimhi-loF \
-acontr 1 1 1 1 All \
-acontr 1 0 1 0 StimHigh \
-acontr 0 1 0 1 StimLow \
-acontr 1 1 0 0 StimRem \
-acontr 0 0 1 1 StimFor \
-acontr 1 -1 0 0 StimRemHvL \
-acontr 0 0 1 -1 StimForHvL \
-acontr 1 0 -1 0 StimHighRvF \
-acontr 0 1 0 -1 StimLowRvF \
-acontr 1 -1 1 -1 StimHvL \
-acontr 1 1 -1 -1 StimRvF \
-fa TaskEffect \
-bucket 12-subjANOVA_SubMemSTIM_6B
Our problem is this: Using the "-amean" flag, we get mean values for the conditions in question (each of the levels) which, in theory, should be exactly the same (or very, very close) as a mean coefficient value generated by taking each subject's coefficient file and performing a mean outside of the ANOVA.
We actually wanted to test only four independent points in an ANOVA, and thus went into each coefficient file for each subject, obtained a value for each point (all using 3dROIstats) and performed a simple mean in Excel for the values we got for each subject. These means varied by a small yet substantial amount from the value we got for the means from the output BRIK of our 3dANOVA2 command. The variation was enough to not be due to rounding errors, etc. (I think) but not enough for us to question the values we got from either methodology. Is there a particular way 3dANOVA2 calculates a mean that we are not aware of?
Thanks!
Arul
