Good hypotheses make good
methods.
Hypothesis must be testable.  (Exploratory/pilot
analyses are OK as long as you acknowledge
limitations.)
Think in terms of regression model:
Y=1X1+ 2X2+ 3X3+E
Think about directionality:
Y when X
Y when X