The data I am working with was collected from 300 persons, each reporting an average of 12 contacts (3 to 49), for each they specified the number of times they contact them face-to-face, by phone, by email and by SMS (300 * 12 * 4). I'm not sure how to model both the interactions (error correlations) between the modes of contact, and account for the panel effect of having multiple observations for each of respondents (as in multilevel or error-component models). Any suggestions? I tried to model it as a twolevel analysis where the between variables measure on the cluster/panel level and the within on the individual level. Is it possible just to include a model where I regress the depended variables (modes) on them selfs?