

Clustering with Dummy Variables 

Message/Author 


I have data for participants from 12 different countries. As this number is too small for TYPE=COMPLEX or TYPE=TWOLEVEL analyses, I have been trying to account for the clustering using dummy variables. ANALYSIS: TYPE = GENERAL; MODEL: y ON x1 x2 x3 x1byx2 x1byx3 x2byx3 d1d11; However, when I add the 11 dummy variables to my model, as above, I get the following warnings/errors: WARNING: THE SAMPLE COVARIANCE OF THE INDEPENDENT VARIABLES IS SINGULAR. PROBLEM INVOLVING VARIABLE D11. THE MODEL ESTIMATION TERMINATED NORMALLY THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES COULD NOT BE COMPUTED. THE MODEL MAY NOT BE IDENTIFIED. CHECK YOUR MODEL. PROBLEM INVOLVING THE FOLLOWING PARAMETER: Parameter 18, Y ON D11 Thanks in advance. 


Here are a couple of things you can do to try to figure out why you have this problem in the data. 1. Think about why you might have an x variable that is a linear combination of other x variables. 2. Delete the d1d11 vbles to see if you get the problem in your regression run. If it doesn't happen, add 1 dummy at a time to see when the problem happens. 3. do a type=basic run and inspect the correlation matrix. 


That is very helpful, thank you. Thinking more about your point #1, it makes sense that any clusterlevel predictors will be collinear with the dummy variables. In this example, x3 was a countrylevel variable and this was what was causing problems. I'm afraid the answer is no, but is there any way (without COMPLEX or TWOLEVEL) to include clusterlevel predictors when accounting for clustering with dummy variables? 


You can do that without complex or twolevel. 

anonymous posted on Friday, November 08, 2019  12:24 am



In the MLM model, I am interested contextual effect of the dummy variable at the class level. At the individual, the predictor variable is this dummy variable without centering (no grandmean). At the classroom level, the predictor variable is proportion of this dummy variable in classrooms (i.e., cluster_mean). Is the classroomlevel effect of this cluster_mean dummy variable contextual effect? 

Back to top 

