It has been drummed into me since junior school that graphs are useless without error bars. I can't see anything in the docs about how to generate these. Can it really be true that a whole data analysis platform has been built without this feature? Related to this, error bars are there to give an indication of the tightness of the data and whether one can really claim a difference between two aggregate values. Alternative approaches would include box plots, or violin plots. These don't feature either as far as I can see. Have I missed something?
Leave a thumbs-up reaction to this GitHub issue to get this feature request noticed:
+1 for box plots. They can be (ab)used to create visualizations difficult to model with other methods, i.e. bar charts. I have a climate chart that plots average high and low temperatures that looks great in CanvasJS but can’t be represented in Metabase (as of v57 anyway).
As far as the lack of error bars goes, Metabase was built primarily as a business intelligence tool, which uses descriptive statistics and not inferential statistics (for the most part). You don’t need to sample the populations of your monthly revenue and display error bars to indicate the uncertainty; you have all the data. That changes when making predictions, of course.
Metabase also doesn’t print very well, so you aren’t using it to generate charts to put in your research paper; Excel, LaTeX, etc. are better suited for that job. That might change if Metabase’s Documents feature takes off (and starts displacing Tableau as a narrative tool).