I am planning a multiple regression model on a large customer satisfaction
database and am challenged by a fairly classic (I believe) problem: some
respondents weren't asked on different independent variables/predictors due
to questionnaire design, for example consider a dataset with 1000
observations:
Y- measure of overall satisfaction, totally observed.
and three predictors X1,X2,X3 all asking on a specific area of the client
experience. all three independents are 1-10 likert scale variables
(1-disageree strongly to 10 strongly agree).
The problem is that a large number of respondents weren't exposed to all
three area and thus can answer 'not relevant'.
I am interested in the case where Y is continuous (1-100) but also Binary
(0/1).
Any recommendations (adding dummy codes?, pairwise deletion?), references
and help on interpretation will be greatly appreciated.
|