Hi All,
I wonder if you can help me, I'm very new to multivariate statistics and really confused.
I have a large dataset (n=344) of responses to a questionnaire containing measures of various aspects of social life. Most measures are Likert response scales (different response scales) but some are continuous scales (e.g. hours spent doing different activities).
I am planning on doing an exploratory factor analysis on my data but prior to that I have some problems with missing data. Missing data accounts for less than 5% for all variables but I want to impute the missing values rather than delete cases with missing data.
I find the whole PASW MVA quite confusing but it looks as though my data are not missing completely at random as Little's MCAR test is significant (.003). So, I think this means I use EM to impute missing values.
I have really tried to find information on these issues but feel more confused than when I started.
My main questions are
a) am I supposed to round up the imputed values for my ordinal (Likert) variables?
b) I have some other variables in my data set which won't go into the factor analysis, but I aim to use these variables in a later analysis (for example, a measure of self-esteem- later I will analyse whether score on self-esteem measure seems to be related to factor scores from EFA analysis).
Am I supposed to enter these variables also into the MVA and EM imputation? I've done it both ways and the imputed values are slightly different.
If these questions are too hard to answer then please direct me to any papers/websites that may help me to understand these issues better.
Or any general advice at all much appreciated!
Thank you.