Message/Author |
|
Joe Swenson posted on Friday, February 29, 2008 - 11:38 am
|
|
|
Hello, I am a new M-Plus user and new to hierarchical modeling in general. I am trying to run a two-level multilevel logistic model with complex survey data. The levels of interest to me are interviewers (random effect) and respondents within interviewers (with fixed respondent covariates: x1-x5). The dependent variable is a 0/1 respondent outcome. I want to adjust my results for the complex design, so I believe a TYPE = COMPLEX TWOLEVEL is appropriate. (Please correct me if I am wrong!) I am having some trouble with the coding. Particularly, I don't think I am defining the 'within' and 'between' variables correctly and was hoping someone could help me on this point. Here is what I have so far: Title: Interviewer effects Data: File is c:\data.dat; Variable: Names are userid iwerid stratum secu pweight y x1-x5; cluster is secu; stratification is stratum; weight is pweight; categorical = y; Missing = all(-1234); within=x1-x5; Analysis: type=complex twolevel; Model: %within% y on x1-x5 %between% iwerid; Output: sampstat; |
|
|
It sounds like you want to take into account the nonindependence of observations due to respondents being nested within interviewers. You can do that by using TYPE=COMPLEX or TYPE=TWOLEVEL. The difference between these two approaches in discussed in the introduction to Chapter 9. |
|
Joe Swenson posted on Friday, February 29, 2008 - 3:46 pm
|
|
|
No, I want to take into account the nonindependence of observations due to interviewers and respondents being nested within psu's, and account for unequal weighting and stratification. In addition I want to estimate the two-level (respondents nested within interviewers) random effects model and estimate the interviewer variance component. According to the User's Guide, I think I need to use both COMPLEX and TWOLEVEL in the same analysis statement. Is this correct? Can M-Plus handle this type of analysis? |
|
|
If you also have nesting within psu's, you can use TYPE= COMPLEX TWOLEVEL and CLUSTER = psu interviewer. This is discussed under the CLUSTER option in the user's guide. You can also have a stratification variable in this situation. |
|
Joe Swenson posted on Saturday, March 01, 2008 - 6:05 am
|
|
|
I am really interested in estimating the between-interviewer variance component (adjusted for the complex design), which I want to use to calculate the intra-interviewer correlation. When I insert the interviewer id variable in the Cluster statement the output does not give interviewer variance component. Could there be something wrong with my model statement, specifically the 'within' and 'between' coding below? Title: Interviewer effects Data: File is c:\data.dat; Variable: Names are userid iwerid stratum secu pweight y x1-x5; cluster are secu iwerid; stratification is stratum; weight is pweight; categorical = y; Missing = all(-1234); within=x1-x5; Analysis: type=complex twolevel; Model: %within% y on x1-x5 Output: sampstat; |
|
|
Any variable measured on the individual level that is placed on the WITHIN list will be modeled only in the within part of the model. Please read about the BETWEEN and WITHIN options in the user's guide for more information about this. |
|
|
I have survey data of 1,000 individuals nested within 63 workgroups and workgroups nested within 5 organizations. I am interested in modeling 4 workgroup measures on individual attitudes but also want to account for the clustering for organizations. I have been using type=twolevel complex, using twolevel to model workgroup effects and complex to account for organizational clustering: 1. Is this an appropriate method being that there are 5 organizations? (even though there are 63 level 2 units) 2. Is this approach similar to using dummy variables as controls in a two-level model? I am hesitant to add 4 more dummy variables in level 2 since I only have 63 level 2 units. Thanks. |
|
|
The number of units required is the same for COMPLEX and TWOLEVEL. I would use dummy variables in your case. |
|
Back to top |