Three level model with complex data PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
Message/Author
 Christos Giannoulis posted on Thursday, May 29, 2014 - 10:29 am
Dear Mplus experts:

I have a typical case:
900 students nested in 83 classes(teachers). Teachers were also nested in 33 schools.

I have 25 school level variables: 15 categorical and 10 continuous;
16 teacher level variables: 16 categorical and 3 continuous;and
4 student level variables all categorical

and
(a) one dependent continuous variable: student achievement OR
(b) four dependent continuous variables (i.e. four components of student's achievement)

The sample size was derived using a stratified random procedure. I have the weights by region of school and demographic weights by race and gender of students.

My questions are:

* Do I need to dummy code categorical variables that have more than two categories or is this automatically done in Mplus?
* How can I select predictors across different levels using Mplus? if not, what is the best working method from your experience?
* How can I perform post-stratification weights to my results so they are not biased using Mplus?
* Could you share with me any similar examples available (i.e Mplus syntax) with complex data or examples that are closer to this kind of analyses?
* Could you share with me and the forum, additional documentation (papers, reports, video) that could help me elaborate a little bit on your answers?

Thank you VERY much
 Linda K. Muthen posted on Friday, May 30, 2014 - 11:36 am
You need to dummy code using the DEFINE command.

Check the Raudenbush and Bryk book or ask on MultilevelNet.

Mplus does not do post-stratification.

Example 9.20 through 9.23 show three-level analysis. See also the course video and handout from Utrecht on the website.

See Joop Hox's multilevel book and Raudenbush and Bryk.
 Christos Giannoulis posted on Friday, May 30, 2014 - 1:57 pm
Hi Dr. Muthen,

Thank you VERY VERY much for your answer. I deeply appreciate your meticulous effort to answer every question.

Please allow me one last a silly question for this thread. When you have data derived from a stratified sampling procedure and you want to analyze it in two or three levels how can someone account of the type of sampling regardless of the levels? In other words, what is the alternative of the post-stratification if Mplus does not do it. Does it offer another alternative?

Again, please forgive me for my silly question.
 Linda K. Muthen posted on Friday, May 30, 2014 - 4:14 pm
Typically poststratification information is included in the sampling weights so post-stratification is not needed.
 Christos Giannoulis posted on Friday, May 30, 2014 - 5:25 pm
Thank you! I have the sampling weights from Survey Specialists about the weights of the geographical region by the region and the demographic weights by race and gender
These weights are different from my levels: school, classrooms and students.
How can I embed these weights in my analysis???

This is what I am struggling to figure out how to do it using MPLUS...and I was erroneously thinking that this post-stratification explains it. Thank you for clarifying that.

Hopefully the examples you suggested (9.20 and 9.23) might be of help.

By the way I was looking for the Mplus video and handout in Ultrecht. I found only the courses from John Hopkins University. Are these courses the one you really meant?

Thank you again very very very much.
 Linda K. Muthen posted on Saturday, May 31, 2014 - 10:10 am
Use the WEIGHT option to specify the weight.

The clustering of the data is specified using the CLUSTER option and TYPE=TWOLEVEL or THREELEVEL. See examples in Chapter 9 of the user's guide.

It is the second to last listing under Recent Presentations in the middle of the home page.
 Christos Giannoulis posted on Saturday, May 31, 2014 - 12:19 pm
Thank you VERY much!
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: