Message/Author 


Hello, I ran an LGM with a ZIP smoking variable (number of cigarettes smoked in the past 30 days). The outcome gave me the following feedback, WARNING: COUNT VARIABLE HAS LARGE VALUES. IT MAY BE MORE APPROPRIATE TO TREAT SUCH VARIABLES AS CONTINUOUS. Would it be more appropriate in such case to log transform the variable instead of using ZIP? 


You may want to do a log transformation. You could also consider a twopart model. See Example 6.16 in the user's guide. 


Hi Linda, I did use the twopart already. One problem though, I am doing a mediation analysis but this analysis strategy won't do the model indirect because it employs integration. Is there another way to assess the indirect effect using a twopart model? 


You can define indirect effects using MODEL CONSTRAINT. This will give you a standard error for each indirect effect you define. 


Thank you. 

Back to top 