Slope as a predictor? PreviousNext
Mplus Discussion > Structural Equation Modeling >
Message/Author
 Jiangang Xia posted on Thursday, December 01, 2016 - 9:31 pm
I have been trying to model a cross-level relationship's effect on level-1 or level-2 outcomes. For example, the relationship is between school principal's influence on decision-making and teacher's influence on decision-making. Here the relationship could be estimated through a 2-level model. My question is, how to if I want to model the relationship's effect on teacher job satisfaction? The relationship is already a slope. I am considering a random slope model that the relationship varies by a higher level such as school district. Could we use this random slope to predict other outcomes such as teacher job satisfaction? If so, how? Should I save the random slopes as a level-3 variable and then run a new regression, or I could directly model this slope's effect? Would appreciate for any thoughts.

Jiangang
 Bengt O. Muthen posted on Friday, December 02, 2016 - 5:32 pm
If you have 3-level modeling a random slope defined on level 2 can be used to predict outcomes on level 3.
 Jiangang Xia posted on Friday, December 02, 2016 - 8:54 pm
If I have a 3-level modeling, could I use this random slope to predict a level-1 or level-2 outcomes?
 Bengt O. Muthen posted on Saturday, December 03, 2016 - 11:41 am
The random slope defined on level 2 varies across the units of level 3. This means that this random slope can predict level 3 variables, for instance the level-3 part of the variation in variables measured on level 1 or level 2. That's how multilevel modeling works.
 Jiangang Xia posted on Saturday, December 03, 2016 - 4:04 pm
I just want to confirm whether a random slope defined on level 2 varies across the units of level 3 could predict level-1 or level-2 variables. You know the focused outcomes are usually measured at lower levels.
 Bengt O. Muthen posted on Saturday, December 03, 2016 - 4:17 pm
My answer stands.
 Kirill Fayn posted on Tuesday, February 20, 2018 - 4:56 am
Dear Bengt,

I am trying to use a slope as a predictor in a three-level model. My model is set up as follows:

USEVARIABLES ARE happy TIME swl level2 level3;
CLUSTER = level2 level3;
WITHIN = TIME;
BETWEEN =(level2) swl ;
MISSING ARE all (-9999);
ANALYSIS: TYPE = THREELEVEL RANDOM;
MODEL: %WITHIN%
s1 | happy ON TIME;
%BETWEEN level2%
s1 ON swl;
%BETWEEN level3%
OUTPUT: TECH1 TECH8;

I get the following error:

THE MODEL ESTIMATION DID NOT TERMINATE NORMALLY DUE TO AN ILL-CONDITIONED
FISHER INFORMATION MATRIX....

THE MODEL ESTIMATION DID NOT TERMINATE NORMALLY DUE TO A NON-POSITIVE
DEFINITE FISHER INFORMATION MATRIX. THIS MAY BE DUE TO THE STARTING VALUES
BUT MAY ALSO BE AN INDICATION OF MODEL NONIDENTIFICATION. THE CONDITION
NUMBER IS 0.650D-16.

THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES COULD NOT BE
COMPUTED.
PROBLEM INVOLVING THE FOLLOWING PARAMETER:
Parameter 10, %BETWEEN LEVEL2%: HAPPY2


I have tried to play around with some random starting values, but the error persists.

Any advice would be greatly appreciated.

Thanks for your time.
 Bengt O. Muthen posted on Tuesday, February 20, 2018 - 2:39 pm
We need to see your full output - send to Support along with your license number.
 Sam McQuillin posted on Friday, June 28, 2019 - 1:14 pm
I have a repeated measures model that includes age of participate (in years), a continuous "risk" variable measured at the corresponding age, and then a level 2 binary outcome (if the participant has a disease or not at final age measure).

I want to predict the level 2 binary outcome (disease or not) from the slope and intercept of a level 1 regression: risk on age.

My first pass at this model is:
...
BETWEEN = disease;
WITHIN = age;
CLUSTER = id;
CATEGORICAL = disease;
...
ANALYSIS: TYPE = TWOLEVEL RANDOM;
ESTIMATOR = BAYES;

MODEL:
%WITHIN%
s | risk on age;
%BETWEEN%
disease on risk s;


Does this model seem reasonable?

If so, am I right to interpret the "Means" output under Between Level for s as the average slope of risk on age (i.e. the average relative linear change in risk per year)? And, the disease on s as the change in probit index (z-score) of disease per unit change in the slope from the within model? Thank you for your help.
 Bengt O. Muthen posted on Friday, June 28, 2019 - 3:27 pm
Q1: Yes.

Q2: Yes.

Q3: Not exactly because the underlying continuous latent response variable for the disease variable does not have mean zero and variance 1. Both the mean and variance are functions of risk and s means, variances, and covariance. The residual variance is 1.
 Sam McQuillin posted on Friday, June 28, 2019 - 4:02 pm
Thank you for the clarification.
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: