Erik Thoonen posted on Thursday, February 21, 2008 - 2:17 am
To answer the question whether differences in perceptions (teachers vs. school leader) of school leaders' leadership influence organizational variables, I would like to perform a multilevel path analysis (intraclass correlation between schools <.25)
The next input I used to perform the first analysis:
VARIABLE: NAMES ARE school A_popul wel07m par07m iv07m perlv perlic perlis; BETWEEN = A_popul; CLUSTER = school; MISSING ARE ALL (-1); ANALYSIS: TYPE = TWOLEVEL; MODEL: %WITHIN% par07m wel07m iv07m ON perlv; (etc.) %BETWEEN% par07m wel07m iv07m ON perlv; (etc.) perlv perlic perlis ON A_popul; However, Mplus could not run this input and gave me the next error messages:
*** ERROR in Model command Variable is a y-variable on the BETWEEN level but is an x-variable on the WITHIN level: PERLV (same for other per-variables)
"A_popul" means the number of students and is equal for all teachers of the same school, and therefore automatically a between-variable.
Mu question is: how can I run this path analyses in which the variables: perlv, perlic, perlis can be the X-variabel on the within-level, but are an intermediate variable on the between-level?
Or is this not possible at all?
Thanks a lot!
Boliang Guo posted on Thursday, February 21, 2008 - 2:46 am
you pls read following paper for reference. Krull, J. L. & MacKinnon, D. P. (1999). Multilevel Mediation Modeling in Group-Based Intervention Studies. Evaluation Review, 23, 418-444. Krull, J. L. & MacKinnon, D. P. (2001). Multilevel modeling of individual and group level mediated effects. Multivariate Behavioral Research, 36, 249-277. MacKinnon, D. P. (2007). Introduction to Statistical Mediation Analysis. Mahwah, NJ: Erlbaum. multilevel path analysis is easy to set up for 2-2-1, 1-1-1 and 2-1-1 model, not available foe 1-2-1 model, which the depend variable(M) is level 2 variable but the predicator is level 1 variable, there is a latest paper in PM2007 show the multilevel equation for level 2 dependent variable predicated by level 1 variable, pls have a look for reference.you can find it in Mplus web.
Erik Thoonen posted on Thursday, February 21, 2008 - 7:18 am
With regard to the mentioned model, my model estimates that A_popul (level-2) has an influence on perlv, perlic, perlis (level-1) which influence organizational variables (level-1). So I consider only A_popul as a between-variable, because there is no variation within schools regarding the number of students. So, isn't it that I have a 2-1-1 model or do I misunderstand something?
Thanks Dr Muthen, I will try SEMNET. I have another question please. When I run my above mentioned 2-level moderated mediated model, I notice that at the within-level only the outcome variables are correlated. I have control variables and 3 predictors at the within-level, which are not correlated. Likewise, at the between-level only 3 outcome variables are correlated, and 3 between-level covariates and 4 predictors are not correlated. Using with statements, I entered the correlation at the appropriate levels. The model runs fine with adding within-level correlations, but gives the following error when I correlate between-level predictors: THE MODEL ESTIMATION DID NOT TERMINATE NORMALLY DUE TO AN ILL-CONDITIONED FISHER INFORMATION MATRIX. CHANGE YOUR MODEL AND/OR STARTING VALUES…… THIS MAY BE DUE TO THE STARTING VALUES BUT MAY ALSO BE AN INDICATION OF MODEL NONIDENTIFICATION. THE CONDITION NUMBER IS -0.104D-12. THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES COULD NOT BE COMPUTED. PROBLEM INVOLVING THE FOLLOWING PARAMETER: Parameter 158, %BETWEEN%: TURECG WITH AGGJD THE MODEL IS NOT IDENTIFIED. THE DEGREES OF FREEDOM FOR THIS MODEL ARE NEGATIVE. CHI-SQUARE TEST AND SAMPLE STATISTICS ARE NOT AVAILABLE.
Does Mplus correlate predictors by default and which would be ideal starting values?
We need to see the full output to comment on this - send to Support along with your license number.
SY Khan posted on Thursday, May 31, 2018 - 2:51 am
Hi Dr Muthen,
I would be very grateful for your input and advice on my sent outputs which give the error message mentioned in the post above.
Additionally, I get the following error when I run the same model with just the within-level correlations:
THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES MAY NOT BE TRUSTWORTHY FOR SOME PARAMETERS DUE TO A NON-POSITIVE DEFINITE FIRST-ORDER DERIVATIVE PRODUCT MATRIX. THIS MAY BE DUE TO THE STARTING VALUES BUT MAY ALSO BE AN INDICATION OF MODEL NONIDENTIFICATION. THE CONDITION NUMBER IS 0.153D-10. PROBLEM INVOLVING THE FOLLOWING PARAMETER: Parameter 11, %WITHIN%: [ JDXMS ]
Can you please advise how to fix these issues? Thanks
I am running a multilevel path analysis and am trying to confirm that the syntax I'm using is correctly specified. An abbreviated version of the syntax is below:
Usevariables are PCLASSID zlang1 zlang2 zlang3 zpmlang1 zpmlang2 zecers1 zecers2 zcmplinc ; between = zecers1 zecers2 zcmplinc ; cluster = PCLASSID ; Analysis: type = twolevel ; Model: %within%
zlang1 on zcage1 ' zlang2 on zpmlang1 zlang1 ; zpmlang2 on zpmlang1 ; zlang3 on zlang2 zpmlang2;
zpmlang1 on zcmplinc ; zecers1 on zcmplinc ; zpmlang2 on zecers1 ; zecers2 on zecers1 zpmlang1 ; zlang2 on zecers1 ; zlang3 on zcmplinc zecers2 ;
In talking to a colleague we discovered that we have been doing different things to accommodate l1 variables that are also being used at l2. I simply did not specify these l1 variables (e.g., zpmlang1, zlang3) as "within level". But my colleague has been creating latent variables of the l1 variables at l2 using the "by" command. She also has not been specifying the originals as either "within" or "between" variables.
Are both of these approaches appropriate? Or is one incorrect?
See the V8 UG ex 9.1 part 2 on page 274. A latent variable decomposition takes place in your model specification. It is typically not necessary to create a latent variable as your colleague did but the results should be the same.
Attached is the syntax from my colleague. The creation of the latent var is a bit different than that described in UG. Does the same conclusion apply? i.e., the "by" command isn't needed in model specified in original post?