Missing data imputation in 3-step? PreviousNext
Mplus Discussion > Latent Variable Mixture Modeling >
 James Lemoine posted on Wednesday, July 31, 2013 - 11:46 am
Thanks in advance for all the help this forum provides. I'm running a latent class analysis with a mixture analysis and using the 3-step procedure to test the latent classes' effects on an outcome variable (DU3STEP). There is some missing data in that outcome variable, so I was wondering if it would be possible to use imputation. I've got the analysis without missing data imputation working fine, but when I add:

IMPUTE = variablename;

to the syntax, I get an error message that there's an "Unknown variable(s) in the IMPUTE option: variablename". The variable name is spelled correctly and introduced earlier in the USEVARIABLES and AUXILIARY lines. This problem only seems to arise for variables I try to impute for which are also in AUXILIARY.

So my question is, is it possible to impute missing values for variables used in the AUXILIARY command? Or is this just a bad idea altogether since this variable is essentially the Y in my formula? What am I doing wrong? Please be gentle, I'm a neophyte!
 Linda K. Muthen posted on Wednesday, July 31, 2013 - 12:03 pm
Varibles on the AUXILIARY list are not used in the data imputation. They are saved with the imputed data. Example 11.5 goes over all of the related options. You might find that helpful.
 James Lemoine posted on Friday, August 02, 2013 - 7:11 am
Thank you for your reply. I have another question within the same general topic. I successfully ran an LCA and 3-step mixture analysis with a distal outcome. I then noticed some data on the outcome was marked as missing that I actually had, so I went back into the data file and added that data back in to the existing cases.

Now when I run the LCA/3-step with this single outcome, I get all 9999's for the results estimates/means/p-values of the distal outcome section. There are no new error messages other than the note that there were "problems with the distal outcome." The only difference between before (when it worked) and now (when it doesn't) is that there are fewer missing cases in the (binary) distal outcome. Do you have any information you could share on what might be causing this, and how I could run the analysis?
 Linda K. Muthen posted on Friday, August 02, 2013 - 8:33 am
Please send the two inputs, data sets, outputs, and your license number to support@statmodel.com.
 Ray Sin posted on Tuesday, June 09, 2015 - 6:48 pm
Dear Drs. Muthen,
I'm running a LCA together with several covariates. Some of the covariates were initially unordered categorical variables which I turn into dummies (and omitting the ref category) in order to run the LCA, using R3step.
I understand that missing data on LCA is handled by FIML. However, listwise deletion is applied to the covariates. The output says that data imputation is a way that can be used for observations that are missing with respect to the covariates.
How do I do that? Could you please help?
This is my code.
Names are
year id age female income family politics child presch pre_bb bb genx
mil white black other lths hs jc college married nevmarr wds sp_fulltime
sp_parttime sp_notworking marrWchld marrNchld singWchld singNchld
cohabWchld cohabNchld hse_others;
Missing are all (-9999) ;
usevariables are family politics child presch ;
categorical are family politics child presch;
classes = c(4);
useobservations = year == 1977 ;
auxiliary=(r3step) female black other
lths hs jc
wds married
sp_fulltime sp_parttime
marrNchld singWchld singNchld cohabWchld cohabNchld hse_others;

Type = mixture;
starts = 1000 250;
stiterations = 20;
lrtbootstrap = 200;
lrtstarts = 20 5 100 25;
 Linda K. Muthen posted on Wednesday, June 10, 2015 - 1:32 pm
See Example 11.5 in the user's guide. See also DATA IMPUTATION.
 Ray Sin posted on Wednesday, June 10, 2015 - 1:46 pm
Sorry but this is me again. Do I impute before running the LCA? Or is the imputation set embedded within the LCA?
 Linda K. Muthen posted on Wednesday, June 10, 2015 - 3:50 pm
You should do this before the LCA.
 R Aben posted on Friday, August 21, 2015 - 5:55 pm
Hello, I am running LCA with continuous indicators and using the Lanza et al. (2013) distal outcome method to examine class differences in continuous auxiliary variables (DCON). When I run this model on a raw dataset, the distal outcome output includes means of the auxiliary variables, approximate standard errors, and significance tests for all class comparisons. When I run this model with 10 imputed datasets (to deal with missingness on the auxiliary variables), the output includes means and approximate standard errors but does not include significance tests for class comparisons. Is it possible to obtain these significance tests in Mplus, or are these not available when imputed datasets are used? Thanks in advance for any help you might be able to give.

My code is as follows (for the imputed datasets):

TITLE: LPA imputed datasets
DATA: FILE IS imputelist.dat;
VARIABLE: NAMES ARE y1-y10 o1-o11;
CLASSES = c (4);
 Tihomir Asparouhov posted on Monday, August 24, 2015 - 10:07 am
These are not available in Mplus yet but you can in principle obtain these comparisons manually by running the imputed data sets separately and using the multiple imputation formula
 Ali posted on Monday, January 18, 2016 - 6:09 am
I am running LCA.There are four nominal outcomes, and each outcome has 35% missing data due to the complex survey design. most missing are because the items are not administered. so, missing are MAR, and I tried to do imputation, but it didn't work.It showed me the message "The DATA IMPUTATION command is not available for analysis with count, continuous-time
survival, censored or nominal variables."

My codes are:
CLASSES = c (3);
NOMINAL = u1-u4;
auxiliary=CNT Student_ID_2 ST04Q01;
IMPUTE = u1-u4(c);

It is possible to impute the data for the nominal variables?
 Bengt O. Muthen posted on Monday, January 18, 2016 - 2:52 pm
No. You can use ML under MAR.
 Daniel Lee posted on Thursday, June 27, 2019 - 1:44 pm
I am running an LPA and I included several covariates using the R3STEP command (Auxiliary = (R3STEPP) X1 X2 X3 X4; )

I am missing on several of the variables (half the sample were dropped in the analysis) and the error message referred me to Type=Imputation. However, when I read user guide 11.5, I noticed that you cannot impute Auxiliary variables.

I was wondering if you can direct me to a resource where I can use Type=Imputation in my scenario. Thank you!
 Bengt O. Muthen posted on Thursday, June 27, 2019 - 5:06 pm
You can do the multiple imputation and then a run reading Type=Imputation data where you use the manual R3step approach described in Web Note 15.
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