An inherent assumption of the coding process was that respondents only mentioned web site features that they personally found significant. Consequently, features not mentioned by a respondent, even those dealing with features that were objectively present on the site, were coded as a zero.Thus, the interpretation of these zero values was either that the binary factor was not observed on the site, or that the tri-coded variable was neutral. Potential coding bias was addressed by having the data coded by three independent experts (the two authors and a graduate student). Initial coding of the surveys resulted in an 83 percent inter-coder agreement. Coding inconsistencies were subsequently reconciled prior to analysis.