Well you're sample is very small. In my opinion you have to exclude the missing values cases listwise. Yes the decrease in sample will matter. You lose some power to reject Ho as n decreases. You can give a percentage score or an averaged item score or a total item score. This does not matter. The results will be the same. I can't really answer your other questions. I'd use a matched ttest or a random block (repeated measures) anova to test Ho depending on # of groups you have.