Figure 2.

An Average and SD chart of the dataset, with Tukey’s Fences superimposed; as well as control limits based on<a onClick="popup('http://www.biomedcentral.com/1472-6947/12/86/mathml/M37','MathML',630,470);return false;" target="_blank" href="http://www.biomedcentral.com/1472-6947/12/86/mathml/M37">View MathML</a>and<a onClick="popup('http://www.biomedcentral.com/1472-6947/12/86/mathml/M38','MathML',630,470);return false;" target="_blank" href="http://www.biomedcentral.com/1472-6947/12/86/mathml/M38">View MathML</a>. This highlights the problem of using <a onClick="popup('http://www.biomedcentral.com/1472-6947/12/86/mathml/M39','MathML',630,470);return false;" target="_blank" href="http://www.biomedcentral.com/1472-6947/12/86/mathml/M39">View MathML</a> for datasets with no inherent order. Here the data are placed in “original” order. Limits falling outside of the range 0%-100% are not plotted. This shows the consequences of the issue highlighted in Figure 1, and suggests that one should instead fall back on classical outlier analyses in these instances.

Poots and Woodcock BMC Medical Informatics and Decision Making 2012 12:86   doi:10.1186/1472-6947-12-86
Download authors' original image