6
$\begingroup$

enter image description here

For a graph like this what would be a smart way to go about making the error bars distinguishable without using colour coding?

$\endgroup$
2
  • 1
    $\begingroup$ As an amateur I'd suggest splitting each graph into three graphs, each with the same Y minimum/maximum, and putting them side by side for easy comparison. (CAPTCHAs don't work, so I can't post an answer) $\endgroup$
    – l0b0
    Feb 3, 2014 at 8:27
  • 3
    $\begingroup$ Making the bar ends much narrower and offsetting the values at each x just a tad (say by 0.1 on the x-scale as a first try) $\endgroup$
    – Glen_b
    Feb 3, 2014 at 8:32

1 Answer 1

7
$\begingroup$

You've received some good comments so far. Offsetting as Glen_b mentions is what is used in clustered bar charts, and in the grammer of graphics is referred to as dodging. Below is an example of similar data (using SPSS). Also note that I took away the handle bars at the ends, they are typically distracting and wasted ink even when they don't cause overplotting (see similar advice from Andrew Gelman & Thom Baguley).

enter image description here

Your data has better groupings between the type and ordinate pairs, but you can see here a problem in my example. Our vision fails us and it is hard to associate the type with the correct location on the ordinate. Some software allows you to manipulate the degree of grouping better than SPSS does (particularly the ggplot2 R package), but this will generally be a problem with the clustered error bar charts when the groups are disparate. (With this data I would use color for the ordinate value if I wanted to use this chart - given your example though I'm intentionally giving further examples in black-white).

For this reason, I agree l0b0's comment is a good one as well, and you can typically make a more informative plot by using small multiples and constructing the panels to highlight the most informative comparisons you want to make.

enter image description here

Here you can make comparisons (listed in order of the easiest types of comparisons to make).

  1. Within a panel
  2. Across panel columns
  3. Down panel rows

You have enough flexibility to rearrange components of the panels to display the comparisons you are interested in (and if your ordinate values are actually other categories you can use them as panels as well and use one of the other categories as the ordinate value).

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.