Save residuals PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
 Katrin Lintorf posted on Saturday, June 27, 2009 - 5:35 am
I am working on a multilevel analysis and I am interested in the group-specific betas. So I had the idea to save the group-level residuals. As far as I read in some above posted threads this was not possible in earlier versions of Mplus. Is this still true of Mplus 5.2 (5.21)?
 Linda K. Muthen posted on Saturday, June 27, 2009 - 8:06 am
Between-level factors scores can be estimated and saved.
 Katrin Lintorf posted on Saturday, June 27, 2009 - 9:44 am
Thank you for your reply! I had already tried the FSCORES-option but had difficulties in understanding what the dat-file shows. Now I see what you mean.
 regina hackbarth posted on Wednesday, March 09, 2011 - 2:55 pm

i am running twolevel-regression-models w/ manifest variables using the complex option. i've been wondering whether it is possible to check for model assumptions (i. e. normality of residuals) while or after running the analyses? if so, could you please let me know how this can be done? is there a way to save level 1- and level 2-residuals?

thank you very much for your help!
 Bengt O. Muthen posted on Wednesday, March 09, 2011 - 6:05 pm
Type=Complex uses the MLR estimator which is already robust to deviations from normality.

There is no option for saving or checking normality of residuals.
 regina hackbarth posted on Thursday, March 10, 2011 - 2:41 am
thank you very much for your prompt reply. that helps!
 regina hackbarth posted on Wednesday, March 30, 2011 - 3:33 am
dear professors,

i realize my questions are kind of naive and i apologize - i'm only getting started working with mplus.

since my last post, i've come to realize that it's actually the "twolevel"-option i'm using in my analyses.

am i using the mlr estimator all the same and how about the fiml algorithm? i was told it was working automatically if i wrote nothing but "type = twolevel" in my analysis-command. is that true? the reason i'm asking is that i get the following warning:

Data set contains cases with missing on x-variables.
These cases were not included in the analysis.
Number of cases with missing on x-variables: 20

I get a similar warning concerning the dependent variable.

is fiml replacing the missing values still or are they in fact excluded?

again - sorry to bother you if this seems all too stupid to ask.

Thank you very much for your time!
 Linda K. Muthen posted on Wednesday, March 30, 2011 - 8:08 am
Observations with missing values on the observed exogenous variables in the model are excluded because the model is estimated conditioned on these variables. Missing data theory applies only to observed endogenous variables. If you want the observations with missing on the observed exogenous variables to be included, mention their variances in the MODEL command. They will then be treated as endogenous variables and distributional assumptions will be made about them. Observations will missing on all dependent variables are also excluded because they contribute no information.
 regina hackbarth posted on Thursday, March 31, 2011 - 5:30 am
i'm sorry, but i still don't seem to understand. it's a multilevel regression model with manifest variables and i'm not specifying anything but "type = twolevel" in the model command.
in this case - am i using mlr (because it's the default) and is fiml operating, too although i get the warning menttioned above, that variables with missings on x-variables are excluded from the analysis? i mean are they estimated by fiml despite the warning or are they really not included?

thank you!!
 Linda K. Muthen posted on Thursday, March 31, 2011 - 5:57 am
They are not included as stated in the message. FIML is used for estimating the model. Means, variances, and covariances of observed exogenous variables are not parameters in the model.
 regina hackbarth posted on Saturday, April 02, 2011 - 1:48 pm
thank you very much for your answer! but why is it that i don't find fiml in the "twolevel"-section of the "estimator"-list in the user's guide (p. 483)?

i still don't seem to get that straight: with type = twolevel and nothing else changed - am i using mlr as stated in the user's guide or am i using fiml? or are they in fact the same? where can i find a reference to better understand what fiml is doing? how can i best describe to a reader how my models are being estimated?
 Linda K. Muthen posted on Saturday, April 02, 2011 - 4:47 pm
The current Version 6 user's guide has the information on page 532. The estimators ML, MLR, and MLF are full-infomration maximum likelihood estimators (FIML). See the Little and Rubin book on the reference list in the user's guide for a description of how Mplus handles missing data estimation.
 Jak posted on Tuesday, May 24, 2011 - 5:49 am
Dear prof. Muthen,

Does FIML (with MLR estimation for example), mean that the parameters are estimated by optimizing the likelihood of the raw data? Or the likelihood of observed covariances and means?

Thanks in advance!
 Linda K. Muthen posted on Tuesday, May 24, 2011 - 4:39 pm
Raw data.
 Jak posted on Wednesday, May 25, 2011 - 2:54 am
Thank you very much for your quick reply.

Is this the case for both continuous and categorical variables?

I am wondering because computational time is so much shorter for continuous data.
 Linda K. Muthen posted on Wednesday, May 25, 2011 - 5:35 am
Yes. With categorical outcomes and maximum likelihood estimation, some models require numerical integration. This is computationally very demanding. There is a brief description of this in the user's guide.
 Josh Klugman posted on Tuesday, April 22, 2014 - 7:49 pm
Do I understand correctly that MPlus version 7 does not allow analysts to save the level 2 residuals?

This is especially an issue for multilevel logistic regression. Researchers argue that average marginal effects is the best way to convey effects in a logistic regression. My understanding is that in a multilevel context to calculate AMEs the level 2-specific intercept needs to be included to calculate each case's predicted probability.
 Linda K. Muthen posted on Wednesday, April 23, 2014 - 10:44 am
You can save the random intercept and slope using the FSCORES option and compute residuals from them.
 Josh Klugman posted on Wednesday, April 23, 2014 - 11:55 am
Professor Muthen, thank you for your response. I kept on seeing your references to FSCORES but I assumed they would not apply to a model with no latent variables. If I was wrong that is good news!
 Linda K. Muthen posted on Wednesday, April 23, 2014 - 4:15 pm
Random intercepts and slopes are latent variables.
Back to top
Add Your Message Here
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Options: Enable HTML code in message
Automatically activate URLs in message