I have five point likert scale responses ((1-most imp, 5-least imp) for 14 items that I want convert them into 5 factors with confirmatory factor analysis. I already have designed which item should go into what factor, so I can't use exploratory FA. Once I get the factor scores, I will use them in a logistic regression. I have the problem of skewed responses as well, with more people responding for certain facilities to be more important in their neighborhood. My question is, shall I go for CFA, or shall I just calculate the mean value for the item responses within each amenity (e.g. Public transportation-bus, metro) and use these mean values in the logistic regression? Also for CFA, Can I use SPSS adding only the items that will be in that factor by forcing it to extract only one factor? If so, which method should I use-PC, ULS, GLS, ML, PAF? Thanks, aruna
If you are sure which variables go into which factors, then, given that the variables are ordinal and skewed, I suggest not doing CFA or any kind of FA. First, methods for FA with ordinal data are at least somewhat controversial. Any FA will rely on multiple choices, any of which can be objected to and all of which should be justified.
You mention taking the mean. This may be good. However, you also mention that the data are skewed. Depending on the nature of the skew, both in each variable and across variables in a set, it may be better to take the median or perhaps the trimmed mean (a nice compromise between median and mean).
This is much simpler than CFA, and, unless some variables are much more important than others, will give quite similar results.