2
$\begingroup$

I have multi-dimensional data in the following form:

|          |              1              |              2              |              3              |
|----------|:-------:|:-------:|:-------:|:-------:|:-------:|:-------:|:-------:|:-------:|:-------:|
|          | Method1 | Method2 | Method3 | Method1 | Method2 | Method3 | Method1 | Method2 | Method3 |
| Dataset1 |    10   |    8    |    7    |    15   |    20   |    18   |    20   |    30   |    28   |
| Dataset2 |    15   |    10   |    9    |    17   |    19   |    17   |    22   |    30   |    29   |
| Dataset3 |    13   |    10   |    8    |    15   |    16   |    15   |    21   |    32   |    29   |

Is there a way to intuitively visualize this data for comparing the three methods for each dataset, for the different values of 1,2,3? What is the name of the proposed chart (so than I can look it up in R or Excel or Matlab documentation)?

I am not sure what this kind of data is called, so pardon my possibly wrong tags.

$\endgroup$
  • 2
    $\begingroup$ I don't think this question should be considered off-topic. The OP is asking for how to visualize data like this / what the name of the appropriate figure would be. That is clearly w/i our purview. $\endgroup$ – gung May 29 '15 at 15:08
3
$\begingroup$

You seem to have a response variable with three categorical controls. Whether the controls are ordered or their labels are arbitrary is not clear. Similarly, whether joining some points (but not others) by lines can be justified is not clear.

Something like a (Cleveland) dot chart would be a starting point. There are several small trade-offs in shuffling the order of the breakdown and between mixing vertical and/or horizontal order.

Similarly, with different symbolism added points could be placed on the same line, thus reducing the number of dimensions.

enter image description here

$\endgroup$
  • $\begingroup$ +1, but you really need a legend to differentiate the outer 1,2,3 from the nested 1,2,3 from the top 1,2,3. $\endgroup$ – gung May 29 '15 at 21:55
  • $\begingroup$ @gung Thanks and quite correct. The data seem too anonymous and perhaps too unreal to be intrinsically interesting, but one should show willing. $\endgroup$ – Nick Cox May 29 '15 at 22:30
  • $\begingroup$ +1 I really like cleveland's dot plot for this problem, one of the most useful yet underused charts. Would you consider using lollipop chart as alternative to dot plots since lollipop charts does not have "junk" lines after the dots ? Thanks $\endgroup$ – forecaster May 29 '15 at 22:51
  • $\begingroup$ If I understand "lollipop chart" correctly, it's certainly an alternative in my view. I think of the lines as just guide lines. It's the function of the dots to convey magnitude. $\endgroup$ – Nick Cox May 29 '15 at 22:57
1
$\begingroup$

You can use x, y, color, facet and line for this:

enter image description here

or you can use pointsize for values:

enter image description here

You can use different shapes if you cannot use color (some publications need black-white images):

enter image description here

Heatmap can also be used but does not look best:

enter image description here

Note that the values are random and not same as your data set.

$\endgroup$
  • $\begingroup$ +1. I thought of something similar. I will wait for a few hours to get some additional feedback from the community before accepting your answer. $\endgroup$ – Alexandros May 29 '15 at 17:31

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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