
Message/Author 


Hello, I am working with data from 4 different sites in a merged dataset. I am trying to follow the multiple cohort example in 6.18 using the site id as a grouping variable. However, I continue to get the following error: "One or more variables in the data set have no nonmissing values." The number of measurements varies by site, so I am guessing the problem has to do with some sites having data for a specific time point, while other locations do not have any data for that timepoint. I am unsure of how I should handle the missing data. Any suggestions would be appreciated. 


I suggest you check your conjecture: ``some sites having data for a specific time point, while other locations do not have any data for that timepoint. " You can try to limp together time points to avoid this problem. 


Thank you for your prompt reply. I am a new MPlus user and I appreciate your assistance. Similar to the example 6.18, each site has been given a site number (1,2,3,4). Due to variation in age at assessment and time between assessments, scores were grouped by child age into threemonth windows in each data set. The first age band extends beyond 3 mos. to reflect 3748 mos., the remaining age bands are 4951, 5254, 5557, etc. to 8587 mos. To approximate the multiple group model, my model statement is: Model: i s q htoes1@0 htoes2@1 htoes3@2 htoes4@3... htoes14@13; This model applies to site 1. Site 1 has scores in all age bands. I set up similar model statements for each additional site, but only included the age bands for which they had scores. For example, the model statement for site 3, which does not have any scores in the first age band, looks like this: Model Site3: i s q htoes2@1 htoes3@2...htoes14@13; This set up generates the error: "One or more variables in the data set have no nonmissing values." I am not sure what I am doing wrong. 


Sounds like your different groups have different number of observed variables. If so, that requires special handling. 


Thank you. Are you able to recommend any articles, reference material, or sections of the MPlus user guide that discuss this speacial situation? 


You can try an alternative approach where you don't divide the sample into groups (cohorts). Then all but Site 1 will have missing data for some variables  this is ok. Sometimes this approach has difficulties due to so much missing data but it is worth trying for you. So just arrange the data so that all records have the number of observed variables the same as for site 1  inserting missing data where needed. 


I will try that approach. Thank you! 

Back to top 

