Message/Author |
|
Laura posted on Thursday, August 09, 2012 - 1:39 pm
|
|
|
I would like to know why the number of free parameters in Mplus change when I run a regression in one of two ways: 1) I can specify a model such as Y ON X1 X2; 2) Alternatively I can do this (I may pursue this option if I have missing data): Y ON X1 X2; Y; X1; X2; The number of parameters estimated differs depending on the option I choose (Mplus output for option 1 will say 4 free parameters, but for option 2 will say 9). However, aren't these two models identical? Is the model specified in option 1 actually more parsimonious and computationally effective? |
|
|
No. A regression model is estimated conditioned on the observed exogenous variables. In 1, the means, variances, and covariance of x1 and x2 are not model parameters. In 2, they are and distributional assumptions are made about x1 and x2. |
|
Joan W. posted on Monday, September 10, 2012 - 10:34 am
|
|
|
Dear Drs. Muthen, I have a dependent variable that is estimated probability, so it ranges from 0 to 1. In SAS, I can simply use genmod procedure and specify a logit link. How I may handle this in Mplus? Thanks. Joan |
|
|
You can add to your input define: y=log(y/(1-y)); |
|
|
...actually ... if you are trying to replicate something like example 29.1 http://www.math.wpi.edu/saspdf/stat/chap29.pdf you really have to enter the data in binary format in Mplus rather than "estimated probability" so observation A .1 1 10 in Mplus should be coded as A 0 A 0 A 0 A 0 A 0 A 0 A 0 A 0 A 0 A 0 A 1 and ff course replace A by a binary dummy |
|
Back to top |