We ran a 3-way ANOVA getting a non-significant result, but in doing so we saw that one condition looked like it had a strong effect. We tried a Bonferroni-corrected t-test of just that group, and it turned out significant. Is it valid to consider this? Is there some other correction needed to deal with the fact that we tested that group only after we saw that it "looked" like an effect?
(And does the answer to this sort of question depend on between vs. within factors? We have a mix. I ask this because SPSS doesn't seem to give the option for post hoc tests under the GLM procedure except for between-factor effects. Not the best way to learn, but it's a clue, nonetheless.)