• Posted by Konstantin 26.10.2012

    It is annoying how popular it is to ignore the Y-axis limits on bar charts nowadays. Unfortunately, this is also the default mode for most plotting packages, so no one wants to do anything about it. But something must be done.

    Barplots

    Posted by Konstantin @ 4:42 pm

    Tags: , ,

  • 3 Comments

    1. Rudolf on 08.11.2012 at 22:47 (Reply)

      I would add 5th graphic with units and put “Most correct” next to it. The whiskers are for showing minimum/maximum?

      1. Konstantin on 08.11.2012 at 23:24 (Reply)

        By units you mean the y-axis label? True, those are nice to have, but to my mind are not crucial. They are usually clear from the context and do not contribute much to the main aspect of the boxplot, namely that of “comparison”.

        My main point here is that barplots are abused so often nowadays, that when I see two bars on a plot, one twice as high as the other, I must consciously stop myself from believing my eyes, because I know the y-axis is almost certainly flawed and one value is really not twice as large as the other.

        The whiskers denote the standard error of the estimate (which is nearly always appropriate), and those are actually the things which let you decide, whether one bar being visually higher than the other is worth any attention at all.

        In this particular example the title “Average quality” hints that the values are an estimate obtained by averaging. Such estimates are not precise and have an associated standard error.

    2. Rudolf on 10.11.2012 at 00:46 (Reply)

      Great tip, thanks! I’m wondering how many wrong decisions are being made based upon flawed plots in presentation slides…

    Leave a comment

    Please note: Comment moderation is enabled and may delay your comment. There is no need to resubmit your comment.