Comparing responses from web and paper-based collection modes in a choice modelling experiment