Method comparisons with quantitative data III: difference plots

In the previous article in this series¹ we discussed the misuse of the correlation coefficient to describe agreement between two methods. For a long time (and some may argue still) it was the first and last port of call for comparing methods. It was not until 1983 that Bland and Altman provided an alternative to Pearson’s correlation coefficient and introduced difference analysis to the field.² It is interesting to note the article was published in a statistical, rather than biomedical journal, followed by an updated version in The Lancet three years later.³

            In this article we will discuss the basics of difference analysis. As the years have passed, the method has developed significantly. It is an extensive topic, covered very deeply in published literature. Here, we will limit our discussion to the most basic of applications of the method.

            Bland and Altman’s Limits of Agreement method introduces a new concept to our statistical repertoire. They recognised that a method comparability should consider two aspects of the comparison – agreement at the experimental (overall) and individual (paired sample) level.