Anonymous posted on Tuesday, June 10, 2003 - 12:36 pm
I would like to run a path analysis of a model in which the final dependent variable is nomimal in nature; however, the dv has four categories and it is not ordinal in nature. Would this be possible to do with MPlus?
I am running a path model (mediating) with one ultimate dichotomous dependent variable (drug use/no use), a mediating variable (intentions to use) using a likert scale with six categories that has a ceiling effect, and several exogenous variables measured on a likert scale. I am treating the skewed mediating dependent variable as censored (I attended the MPlus seminar in Baltimore last week and realized I needed to do this). Can you direct me to an article or reading that tells me how the censored variable is being treated and why? Peggy Tonkin
bmuthen posted on Friday, March 25, 2005 - 3:51 pm
When you use ML, the censored variable can be considered as arising from an underlying continuous latent response variable that is normal given the exogeneous variables. When censored from above, all people above the censoring point are at the censoring point with a probability derived from this normal distribution. You have a linear regression of the latent response variable on the covariates. To read about how this translates to the observed censored variable, you may want to look at my 1989 (?) factor analysis article in BJMSP and/or in G.S. Maddala's econometrics text from 1983, "Limited-dependent and qual vbles in econometrics". Perhaps Dan Nagin's articles has some pedagogical writings on this as well.
Anonymous posted on Thursday, June 09, 2005 - 1:27 pm
In MPLUS, for a nominal dependent variable u1 (NOMINAL=u1) with 4 categories (u1#1, u1#2, u1#3, u1#4), is the default comparison baseline the highest category (i.e., u1#4; MODEL u1#1 u1#2 u1#3 ON x1 x2)? Can I change the comparison baseline to another category (e.g., u1#2; MODEL u1#1 u1#3 u1#4 ON x1 x2)?
bmuthen posted on Thursday, June 09, 2005 - 6:41 pm
Your 2 questions have the answers Yes and No. Regarding question 2, it would be good to print out all comparisons, and we should add that. But you can always give rescore the variable.