Missing data for a paired t-test

#1
Hi everyone. My question is how to handle missing data from a paired T-Test.

I am looking at pre-post survey data for a college course. The survey question are looking at students attitudes on multiple subjects, from the beginning of the course to the end of the course. I used a likert scale and course instructors were hoping that the attitudes towards the topics improved.

In the pre-test I have about 598 responses, in the post-test I have about 363 responses. Some of the students who did the pre-survey dropped the course, and some of the students who did the post survey added the course later in the quarter. My pre-post sample with completed responses for both pre-post are about 250 responses. When I do paired T-Test it seems that attitudes towards the topics improved. I am worried about non-response bias.

After conducting expectation maximization I have complete data for 700 pre-post responses. When I do the paired T-Test it appears most of the attitudes towards the topics became negative. I am worried that this method skewed the data with too many missing responses.

What would be your recommended course of action?
 

Karabiner

TS Contributor
#2
First you could decide whether you want to perform a completer analysis, or an intention-to-treat analysis, or both.

The inclusion of the posttetst-only subjects is a bit unusual, though. What do you expect to gain by this?

With kind regards

Karabiner