Researchers should make data freely accessible.

3 Editorial If you have ever had the pleasure of leafing through old journals you would appreciate that the journals of yesteryear were quite different. A typical article from a scientific journal of the early 20th Century was much longer than those of today. Many articles in old journals contained lots of tables but few graphs. (The graph is a remarkably recent innovation.) And, surprisingly to the modern reader, it was not uncommon for researchers to publish a complete record of all of the data from a particular experiment. That is, researchers often reported data from each subject. These days journal articles rarely report data from individual subjects. Instead, it is more usual to see summary statistics reported. For example, researchers may report the mean and standard deviation of a distribution. Descriptive statistics such as means and standard deviations, when used appropriately, provide a concise summary which substitutes for a tedious enunciation of each datum. The convention of reporting summary statistics, rather than 'raw' data, is a pragmatic one. Modern readers are faced with unmanageably large amounts of research data so they prefer to read concise research reports. Moreover, contemporary clinical studies are very large. Obviously it would be impossible to provide, in hard copy, data for each of the 38 050 participants in the observational study of low back pain reported by Smith and colleagues in this journal (Smith et al 2006). The sensible shortcut is to report only summaries of data. There are, however, reasons why some readers might want access to raw data. Access to raw data makes it possible to: 1. Scrutinise data. By inspecting raw data readers can ascertain how complete the data set is and identify anomalies in the data such as outliers. This provides an indication of data quality that may not be apparent in summary statistics. 2. Re-analyse data. Some published statistical analyses (perhaps most often the simplest analyses) are performed incorrectly. Even when the analysis is conducted correctly, it may be suboptimal. When raw data are available it is possible to check the accuracy of an analysis or to subject the data to better analyses. 3. Incorporate data in meta-analyses. Ideally most quantitative research data would eventually be incorporated in a meta-analysis. But meta-analysis is often thwarted by incomplete reporting of data. This problem could be circumvented if meta-analysts routinely had access to raw data. Access to raw data also opens …