Motivation
Analytical data are never precise points but probability distributions represented by mean and standard deviation. Plotting data with their analytical uncertainties as error bars often results in noisy and difficult-to-read plots. However, omitting this information creates a wrong impression of the data.
Suggested function
As a compromise, a convex hull geom is suggested that includes the analytical uncertainty, i.e. rather than using the mean, it uses the mean + the analytical precision as the reference to where to draw the line. This would create any outline around the points, reducing noise while providing a more realistic visualisation of the data.
Potential existing approaches
Buffer functions (e.g. in geospatial packages) might provide a good starting point for such a geom. Functions for multi-variate statistics including analytical uncertainties might exist and could provide some inspiration.
Additional considerations
A first approximation with multi-dimensional analytical uncertainty(e.g., biplots, 3D plots) could be an ellipsoid with the respective axes as the maximum values. In a 2D space, this would result in an ellipsis. However, the true statistical distribution is more complex and it should be checked first if the respective models can be implemented.
Motivation
Analytical data are never precise points but probability distributions represented by mean and standard deviation. Plotting data with their analytical uncertainties as error bars often results in noisy and difficult-to-read plots. However, omitting this information creates a wrong impression of the data.
Suggested function
As a compromise, a convex hull geom is suggested that includes the analytical uncertainty, i.e. rather than using the mean, it uses the mean + the analytical precision as the reference to where to draw the line. This would create any outline around the points, reducing noise while providing a more realistic visualisation of the data.
Potential existing approaches
Buffer functions (e.g. in geospatial packages) might provide a good starting point for such a geom. Functions for multi-variate statistics including analytical uncertainties might exist and could provide some inspiration.
Additional considerations
A first approximation with multi-dimensional analytical uncertainty(e.g., biplots, 3D plots) could be an ellipsoid with the respective axes as the maximum values. In a 2D space, this would result in an ellipsis. However, the true statistical distribution is more complex and it should be checked first if the respective models can be implemented.