About the Bland&Altman plot tool
The webtool can generate several visualizations of paired data to enable the quantitative comparison of their differences, known as the Bland-Altman plot. The specifics of Bland-Altman plots for comparing methods have been detailed in a paper by Bland and Altman (1986). More advanced uses of these plots have been detailed in a follow-up paper by Bland and Altman (1999). Users are advised to familiarize themselves with the methods prior to generating plots.
- Upload or copy/paste your data so that each row represents a unique individual and each column represents a method. Example data in this format is provided in the webtool. The software will plot the difference on the Y-axis against the average on the X-axis.
The software will examine the residuals of the differences and perform a Shapiro-Wilk test on the differences to ensure these are normally distributed (this is an assumption of the method). The software will also examine the data to determine if fixed and/or proportional bias exists. An alert will be shown if either of these situations exist.
- If the differences show a normal distribution and no proportional bias exists, the software will plot the 95% limits of agreement and the 95% confidence intervals around these limits (and the fixed bias). If the differences are NOT normally distributed, choose an option that normalizes the differences (either the differences divided by the average, or the difference of the natural logs of the values). These should normalize the differences. Alternatively, choose the “Regression” method or "Non parametric" under the “Statistics” section to analyze the data.
- If a proportional bias is detected, choose “regression” method to plot the limits of agreement. With this approach, the confidence intervals around the limits of agreement are not calculated.
- Check “Show 95% confidence intervals” to add confidence intervals to the limits and trend line.
- Select “Label limts on secondary axis” to display the values of the bias and the limits of agreement (not available for the "regression" method).
Data Summary Tab
This provides a table of the relevant values and can be downloaded or copied.
Statistics of the distribution
The Shapiro Wilk Test tests for the normality of the differences. A P<0.05 indicates that the differences are NOT normally distributed and an alternative option for comparing methods (e.g. a normalization or non-parametric method) should be considered.
Coefficient of reproducibility
This value should only be reported if the user is comparing pairs of measurements by the same method (to examine within-method variability or agreement). The coefficient of reproducibility is not applicable if two methods are being compared.
The plot can be saved as a PNG file or a PDF file. The PDF can be opened and edited with Adobe Illustrator to allow for fine adjustments of the lay-out.The data from the summary table can be copied or downloaded as a file.
The webtool is created and maintained by Joachim Goedhart in collaboration with Mark Rishniw
Bug reports and feature requests can be communicated in several ways:
Source code is available at github/JoachimGoedhart
Find our other dataViz apps at https://huygens.science.uva.nl