Arnold, Julie "Evaluating the Quality of a Chat Service"
portal: Libraries and the Academy - Volume 5, Number 2, April 2005, pp. 177-193
Interesting article with some results that oppose earlier studies. The authors examined 351 virtual reference transactions over an 8-month period and found that correct answers occurred nearly 92% of the time, which is a much higher percentage than previous studies. One of the suggested reasons is that many of the earlier studies used "secret shopper" planted questions while this study looked at real patron questions.
They also found that different patron types (students, faculty, etc) definately ask different types of questions, and they suggest this has significant implications when marketing the service to each of these groups.
Lots of good data here and and interesting read.