Confusing one this, even our senior biostats guy didn't know the answer.
First, I run a mixed-model ANOVA with a between groups level (2 levels) and a within groups level (3 levels called 'Time'). In the first model, 'Time' is significant under 'Tests of Within-Subjects Effects' and the corresponding post-hoc comparisons are all very significant (i.e. Time 3>Time 2 p<.001, Time 2>Time 1 p<.001, Time 3>Time 1 p<.001).
So far, so good.
Then I add IQ as a covariate.
Suddenly, time is no longer significant BUT the post-hoc tests are still very significant.
What could be the cause of this?!?!?!?