# Mplus multilevel model with variables of different length

Let's say I have 4 variables on the within person level - xa, ya, xb, yb. xa and ya have each 100 trials, xb and yb have 200, and there are 150 subjects. I want to build the following multilevel model in Mplus:

CLUSTER = subject;
WITHIN = ya, yb;

ANALYSIS:
TYPE = TWOLEVEL RANDOM;

MODEL:
%WITHIN%
sa | ya ON xa;
sb | yb ON xb;

%BETWEEN%
G BY sa sb;


So basically, I extract the random slopes from the within person level regression a and b and use them to model another factor G on the between person level. The issue is that Mplus calculates the whole model only on 100 rows where all 4 variables have non-missing values, although regressions ya on xa, and yb on xb are done independently of each other. I hoped that Mplus will calculate the a regression on 100 trials and extract 150 random slopes for all people, and that it will calculate the b regression on 100 trials, and again extract 150 random slopes for all people. Then, on the between person level we would be able to model the latent variable G with two variables that each have 150 values....

However, Mplus always calculates the random effects only on 100 trials. How can I correct this?

This is the warning I get

*** 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:  100
1 WARNING(S) FOUND IN THE INPUT INSTRUCTIONS

SUMMARY OF ANALYSIS

Number of groups                                                 1
Number of observations                                        100


## 1 Answer

I'm not sure I understand your data and problem. So your intention was that mplus would use the cases with no missing values but give results about cases with variables missing the x? In a regular regression the estimation will only include the cases with no missing values in the x variable. The model is estimated conditioned on it. Could you mention the variance of x variables in the Model? (xa; xb;) That can bring cases otherwise exclude, i.e., allowing the exogenous variables to covary and so forth would extend the amount of cases covered by the estimator. But if you have a complex pattern of missing cases and you want, say, your yb variable to help you fill in all the xa missing values, perhaps multiple imputation might help. It might take very long to run the analysis though.

• If the question is not clear you probably shouldn't attempt an answer. Comments can be used for clarification. – Michael Chernick Jul 21 '18 at 3:54