Using confidence intervals around individual means to assess statistical significance between two means

In summarizing individual means by treatment in tables and figures it is recommended that 84% confidence intervals be produced. Doing this will add an extra dimension to interpretation - allowing an assessment of statistical significance at the 5% level. With 84% confidence intervals, in terms of a range of plausible values for the population mean, interpretation would be different compared to the standard 95%. However, 84% confidence intervals do still also describe a plausible range for the means. In a context with plots by time a multiplicity issue may be raised which will need to be accounted for. However, such graphs are often produced only for exploratory purposes and so any assessment of statistical significance may be made in this context. Relaxing the confidence intervals around individual means is something that has been discussed for sometime now [1-3] and in a context with figures such as Figure 1 does add to the value of diagrammatic representation of studies [4]. When quoting the difference between two means 95% should still be used. Copyright © 2004 John Wiley & Sons, Ltd.