Reverse multilevel Model
Locked
 
|
My friend is asking about a model appropriate for his data. He wants to test a hypothesis that Teacher Emotional Intelligence(TEI) is a predictor of Teaching Performance(TP), both are latent variables. TEI is rated by 100 teachers, while TP is rated by 30 students within each teacher. The analysis is at the teacher level. I think its a bad idea if we will use the aggregates of the student ratings on TP and test the model TEI-> TP. By doing this way, I think we throw away the variability of TP within each teacher before we do the analysis. Can we solicit your expert advise on how to go about the model? Eins |
Re: Reverse multilevel Model
|
|
You are correct that the aggregate of TP would throw away its variability and lose potentially interesting information. The Level of the analysis doesn’t make
this a reverse MLM. It’s still an MLM problem, but the outcome of interest is about teachers. In fact, if I understand you correctly, it’s still at the student level. The Student level outcome is their perception of the teaching performance. You will basically
need to setup a stacked data set with TEI repeated for each student record. If I understand correctly, you have 3000 records (100 teachers by 30 students each). The 3000 records will each need to contain a variable that reflects the teacher they have, something
like a class indicator. Then you will want to have the TEI variable, and this variable will be a fixed value for each class, and thus repeated within each of the 30 students per class. Then you would have your DV, TP for each student record. If there isn’t a lot of within teacher variability (necessitating the MLM), I would consider approaching this via a structural equation model instead, which
I believe will help you glean more valuable information about the data. This would allow you to assess the latent structures themselves as well. Matthew J Poes Research Data Specialist Center for Prevention Research and Development University of Illinois 510 Devonshire Dr. Champaign, IL 61820 Phone: 217-265-4576 email:
[hidden email] From: SPSSX(r) Discussion [mailto:[hidden email]]
On Behalf Of E. Bernardo My friend is asking about a model appropriate for his data. He wants to test a hypothesis that Teacher Emotional Intelligence(TEI) is a predictor of Teaching Performance(TP), both are latent variables. TEI is rated by 100 teachers, while TP is rated by 30 students within
each teacher. The analysis is at the teacher level. I think its a bad idea if we will use the aggregates of the student ratings on TP and test the model TEI-> TP. By doing this way, I think we throw away the variability of TP within each teacher before we
do the analysis. Can we solicit your expert advise on how to go about the model? Eins |
|
|
Yes, you understand correctly.
Yes, you are right that we have a total of 3000 records if the setup of the data is similar to what you have described. It means that the 30 students within each teacher will have the same values on the TEI (teacher emotional intelligence). Is the independence assumption on the 3000 records not a problem? Eins From: "Poes, Matthew Joseph" <[hidden email]> To: [hidden email] Sent: Friday, November 30, 2012 6:28 AM Subject: Re: Reverse multilevel Model You are correct that the aggregate of TP would throw away its variability and lose potentially interesting information. The Level of the analysis doesn’t make
this a reverse MLM. It’s still an MLM problem, but the outcome of interest is about teachers. In fact, if I understand you correctly, it’s still at the student level. The Student level outcome is their perception of the teaching performance. You will basically
need to setup a stacked data set with TEI repeated for each student record. If I understand correctly, you have 3000 records (100 teachers by 30 students each). The 3000 records will each need to contain a variable that reflects the teacher they have, something
like a class indicator. Then you will want to have the TEI variable, and this variable will be a fixed value for each class, and thus repeated within each of the 30 students per class. Then you would have your DV, TP for each student record.
If there isn’t a lot of within teacher variability (necessitating the MLM), I would consider approaching this via a structural equation model instead, which
I believe will help you glean more valuable information about the data. This would allow you to assess the latent structures themselves as well.
Matthew J Poes
Research Data Specialist
Center for Prevention Research and Development
University of Illinois
510 Devonshire Dr.
Champaign, IL 61820
Phone: 217-265-4576
email:
[hidden email]
From: SPSSX(r) Discussion [mailto:[hidden email]]
On Behalf Of E. Bernardo
Sent: Thursday, November 29, 2012 11:26 PM To: [hidden email] Subject: Reverse multilevel Model My friend is asking about a model appropriate for his data.
He wants to test a hypothesis that Teacher Emotional Intelligence(TEI) is a predictor of Teaching Performance(TP), both are latent variables. TEI is rated by 100 teachers, while TP is rated by 30 students within
each teacher. The analysis is at the teacher level. I think its a bad idea if we will use the aggregates of the student ratings on TP and test the model TEI-> TP. By doing this way, I think we throw away the variability of TP within each teacher before we
do the analysis. Can we solicit your expert advise on how to go about the model?
Eins
|
|
|
Eins, Your model appears to be one in which a level -2 IV (TEI) is used to predict a level-1 DV (student perceptions of TP). The level-2 variable would have the same value for all cases that are nested within the same level-2 unit (teacher). In HLM terms, the level 2 variable TEI is being used to predict the intercept of TP among students in the same classroom. Am I correct in assuming that each student rates TP for only one teacher (as would be the case in self-contained classrooms)? If so, students are truly nested within teachers. Best, Stephen Brand, Ph.D. www.StatisticsDoc.com From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of E. Bernardo Yes, you understand correctly. From: "Poes, Matthew Joseph" <[hidden email]> You are correct that the aggregate of TP would throw away its variability and lose potentially interesting information. The Level of the analysis doesn’t make this a reverse MLM. It’s still an MLM problem, but the outcome of interest is about teachers. In fact, if I understand you correctly, it’s still at the student level. The Student level outcome is their perception of the teaching performance. You will basically need to setup a stacked data set with TEI repeated for each student record. If I understand correctly, you have 3000 records (100 teachers by 30 students each). The 3000 records will each need to contain a variable that reflects the teacher they have, something like a class indicator. Then you will want to have the TEI variable, and this variable will be a fixed value for each class, and thus repeated within each of the 30 students per class. Then you would have your DV, TP for each student record. If there isn’t a lot of within teacher variability (necessitating the MLM), I would consider approaching this via a structural equation model instead, which I believe will help you glean more valuable information about the data. This would allow you to assess the latent structures themselves as well. Matthew J Poes Research Data Specialist Center for Prevention Research and Development University of Illinois 510 Devonshire Dr. Champaign, IL 61820 Phone: 217-265-4576 email: [hidden email] From: SPSSX(r) Discussion [[hidden email]] On Behalf Of E. Bernardo My friend is asking about a model appropriate for his data. He wants to test a hypothesis that Teacher Emotional Intelligence(TEI) is a predictor of Teaching Performance(TP), both are latent variables. TEI is rated by 100 teachers, while TP is rated by 30 students within each teacher. The analysis is at the teacher level. I think its a bad idea if we will use the aggregates of the student ratings on TP and test the model TEI-> TP. By doing this way, I think we throw away the variability of TP within each teacher before we do the analysis. Can we solicit your expert advise on how to go about the model? Eins |
|
|
|
|
|
Eins, You do not have a Level 1 IV (unless you also have a student-level covariate). You have a level 2 IV, which is TEI. TEI is treated as a level 2 variable according to the way that you set up the MIXED syntax. Values of the level 2 IV will appear in the record for each student, but the way this information is handled is driven by the syntax. Best Regards, Stephen Brand www.StatisticsDoc.com From: E. Bernardo [mailto:[hidden email]]
|
|
|
In reply to this post by Poes, Matthew Joseph
There’s another question that, in principle, could be investigated in this dataset, which is that classroom score variance is related to TEI. I’m not sure how to do that (and I’m curious to see how) but I’d guess that several people on the list would know whether and how to do this with spss. Gene Maguin From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Poes, Matthew Joseph You are correct that the aggregate of TP would throw away its variability and lose potentially interesting information. The Level of the analysis doesn’t make this a reverse MLM. It’s still an MLM problem, but the outcome of interest is about teachers. In fact, if I understand you correctly, it’s still at the student level. The Student level outcome is their perception of the teaching performance. You will basically need to setup a stacked data set with TEI repeated for each student record. If I understand correctly, you have 3000 records (100 teachers by 30 students each). The 3000 records will each need to contain a variable that reflects the teacher they have, something like a class indicator. Then you will want to have the TEI variable, and this variable will be a fixed value for each class, and thus repeated within each of the 30 students per class. Then you would have your DV, TP for each student record. If there isn’t a lot of within teacher variability (necessitating the MLM), I would consider approaching this via a structural equation model instead, which I believe will help you glean more valuable information about the data. This would allow you to assess the latent structures themselves as well. Matthew J Poes Research Data Specialist Center for Prevention Research and Development University of Illinois 510 Devonshire Dr. Champaign, IL 61820 Phone: 217-265-4576 email: [hidden email] From: SPSSX(r) Discussion [[hidden email]] On Behalf Of E. Bernardo My friend is asking about a model appropriate for his data. He wants to test a hypothesis that Teacher Emotional Intelligence(TEI) is a predictor of Teaching Performance(TP), both are latent variables. TEI is rated by 100 teachers, while TP is rated by 30 students within each teacher. The analysis is at the teacher level. I think its a bad idea if we will use the aggregates of the student ratings on TP and test the model TEI-> TP. By doing this way, I think we throw away the variability of TP within each teacher before we do the analysis. Can we solicit your expert advise on how to go about the model? Eins |
| Free forum by Nabble | Edit this page |