I am doing a simple study that involved taking a measure at time point 1 and time point 2 (12 weeks later).  While the sample was a class, not all members were present at both time points, so I have 20 date points at time 1 and 21 date points at time 2.  The measure has a score, and I am taking the means and doing a simple t-test to determine if the intervention caused any increase in the measure at time point 2.  

Questions:  Do I need to throw out outliers if they are more than 2 standard deviations higher than the mean? 

When I do the t-test, do I need to look at one- or two-tailed distributions?  My hypothesis is that the intervention will increase the mean at time 2, so I think I should consider a one-tailed distribution.

Lastly, I am assuming that I have to do a paired t-test since it is a repeated measures design.