Missing Data

FORUM QUESTION: Julia St George - 22 May 2009 - 11:44

I have done a survey of interns in which each person was asked 12 questions, before and after a course. Some of the questions have missing responses. How should I handle the missing data, if I just want to see if the overall response was more positive after than before the course?

FORUM ANSWER: -- Charity Moore, PhD, MSPH Julia St George - 27 May 2009 - 10:57

First investigate the patterns of missing data so you can describe who is missing what information. You should also see what personal characteristics are related to the missingness. You can do a complete case analysis if there are very few observations with missing data relative to the overall sample (say <10%).

If this is part of a pilot study, then it's informative to know which questions were missing responses and if it is related to how the question was worded and the potential responses.

If you have a high degree of missingness, I would suggest you consult with one of the biostatisticians with your local CTSA.

Also, it would be very important to look at the patterns of missingness across the 12 questions. If some people are missing multiple questions, then further investigation would be needed. There are many things that need to be done to investigate the missing data, the different types/patterns, and assumptions about them.