For any response that could not be categorized with the initial coding scheme, a new code was generated. All responses were coded independently by the lead author, and a subsample (20%) was also coded by a second researcher. Inter-rater agreement was satisfactory (Cohen's Kappa¼0.79).