johnzeringue
commented
10 years ago Thanks for the info! I had no clue whether to call it error or uncertainty or confidence or whatever. We'll make this happen by Friday's release.
Thoughts on the chart names/semantics are much appreciated. Especially looking forward to uncertainty charts without uncertainty correlations, it's tricky to decide if this belongs in the same chart as the ellipses or in a chart of its own.
By "two-variable plot (x-y)" you're really only specifying some sort of scatter plot, correct? That seems awfully vague to me. Or does x-y mean that the two variables are correlated?
Topsoil at present erroneously uses of the word 'error'. In metrology, or the science of measurements, 'error' and 'uncertainty' have two different meanings. I and others find this distinction to be useful when talking about geochronological data (see http://elements.geoscienceworld.org/content/9/1/19.abstract ).
To metrologists, the error is the difference between the measured value and the true value--if you measured 10 cm and the true length was 9.5, then your error was 0.5 cm. If you don't know the true value (the general case in science--we make measurements to estimate the true value) then you don't know the error.
The uncertainty, on the other hand, describes the distribution of possible values of the thing you measured. It could for instance be stated as a range, or as a portion of a probability distribution (e.g. +/- 2sigma), but not as a single number.
That's all a long-winded way of saying that wherever Topsoil uses the word 'error' right now, it needs to use the word 'uncertainty' : two-sigma uncertainty, uncertainty ellipse, etc. For more information or for trouble with insomnia, see
Section 3 of http://www.bipm.org/utils/common/documents/jcgm/JCGM_100_2008_E.pdf or D.1.1.4 of http://physics.nist.gov/Pubs/guidelines/TN1297/tn1297s.pdf
As an extra note: we should change the name of the 'error ellipse' chart anyway. Uncertainty ellipses are a nice way to visualize data with bivariate Gaussian uncertainties, but all of our data won't have bivariate Gaussian uncertainties, and sometimes ones that do will have uncertainties smaller than the ellipse. A more general description would be a "two-variable plot (x-y)", unless anyone else out there can come up with a better name. More in an upcoming issue about how we might organize the multiple possible plots for a given set of input data.