7.6.2  Who should extract data?

It is strongly recommended that more than one person extract data from every report to minimize errors and reduce potential biases being introduced by review authors. As a minimum, information that involves subjective interpretation and information that is critical to the interpretation of results (e.g. outcome data) should be extracted independently by at least two people. In common with implementation of the selection process (Section 7.2.4), it is preferable that data extractors are from complementary disciplines, for example a methodologist and a topic area specialist. It is important that everyone involved in data extraction has practice using the form and, if the form was designed by someone else, receives appropriate training.

 

Evidence in support of duplicate data extraction comes from several indirect sources. One study observed that independent data extraction by two authors resulted in fewer errors than a data extraction by a single author followed by verification by a second (Buscemi 2006). A high prevalence of data extraction errors (errors in 20 out of 34 reviews) has been observed (Jones 2005). A further study of data extraction to compute standardized mean differences found that a minimum of seven out of 27 reviews had substantial errors (G√łtzsche 2007).