Regression Models

Regression models are among the simplest things that can be done with Mplus. You will perhaps use other software for this purpose; but Mplus still may be commended for some features, e.g. the ease with which you can estimate regression models with more than one dependent variable.

A regression model is specified very simply:

MODEL:
        depvar(s) ON indvar(s);

Do not forget that all independent variables are supposed to be continuous and therefore any declaration of an independent variable as a type other than continous will result in an error message. But of course you may use dummy independent variables; just don't tell Mplus.

Models with nominal dependent variables

In the case of dependent variables that are (declared as) nominal (i.e. unordered categorical), a (binary or multinomial) logit model is estimated. There is nothing special in these models, but one may wish to know how to estimate a null model (for instance, to obtain the log likelihood for this model). Here's how.

Let's assume that there is one dependent variable, aptly called depvar, with three categories. The null model now is estimated as follows:

MODEL:
        [depvar#1] (p1); [depvar#2] (p2);

[depvar#1] (p1); will result in estimation of the first constant of the null model; and so on. Of course the number of categories mentioned is the number of categories of the dependent variable minus one. Don't forget the semicolon that separates the different categories.

© W. Ludwig-Mayerhofer, Mplus Guide | Last update: 10 Apr 2010