The purpose of this study is to suggest a research-based conceptual accuracy model of best practices for appraising and improving virtual chat reference service. The model draws on a statistical analysis of 16 independent variables and their effect on the dependent variable of an accurate answer based on two-and-a-half years of archived data found within 333 randomly selected academic reference chat transcripts using Pearson correlations and ANOVA tests for variance.

While the theoretical methodology is grounded in Krippendorff's (2004) content analysis, this project also builds on the RUSA's behavioral guidelines as well as studies and methods found in Richardson's (2002a) annotated synthesis of 87 reference transaction research studies. A content analysis and literature review approach resulted in the following 16 qualitative and quantitative independent variables: librarian's initial contact time, total time of transaction, longest time gap between messages by librarian, number of URLs co-browsed with the patron, key strokes by librarian, key strokes by patron, key strokes by both, the question's difficulty, response to a patron's "are you there" statements, librarian's friendliness, lack of jargon, use of open-ended questions, use of closed-ended questions, objectivity, use of a follow-up question by librarian, and the type of question.

Inferential statistical results indicate 30 incidences of significance between nine of the 16 independent variables and chat reference answer accuracy, helping build a research-based conceptual model of online chat reference. The development of a conceptual model of best practices to improve chat reference service can benefit a number of professionals including chat reference librarians, chat reference supervisors, library educators, library studies scientists, library trustees, chat software vendors and most importantly, the library patron. It can also have implications toward other forms of the reference transaction including phone, email and traditional face-to-face.

This study contributes a conceptual model founded on original research-based findings lacking in a mostly anecdotal professional literature on library reference research. In addition, innovations supporting the findings include an original transaction assessment instrument, a student-designed keystroke/time counter for chat transcripts, both qualitative and quantitative coding techniques, and the use of professional librarians as analysts as well as student assistants as coders.