|
If you think you know that the difference between your control and
experimental conditions will be in a particular direction, you can
use a "one-tailed" design. If you just hope there will be *some*
difference, but don't want to commit to what direction the difference
will be in, you must use a "two-tailed" design.
One-tailed tests make it easier to reject the null hypothesis, as long as you are correct about which direction the difference lies in. It is intellectually dishonest to do a one-tailed test if you didn't a priori expect the difference to be in the direction that was then observed. |