I have a hypothetical set of diet composition, showing how much of each fruit does a particular group of people eat something like:

table percentages

  • How can I visualize in one page, all the compositions for groups A to G and more importantly, their change over the past year?
  • I know having more than 4 colors in design is a no-no, but what can I do here?

What is more important for you - between group comparison or the intra-group composition? For the former, a parallel coordinates plot seems to be a natural choice: http://charliepark.org/slopegraphs/ For the latter, a time series of percent stacked charts might look fine - you do not have to use 7 colors, just alternate them to emphasize the pattern.

  • 1
    $\begingroup$ Slope graphs! Ok that is a great tool! Worth looking at this...much thanks!! The design challenge for me now is that the audience wants to do both between and intra group comparison....Maybe I will do two pages? Or is there another option? $\endgroup$
    – Michael
    Jul 22 '11 at 2:53

To me the slope graph looks really messy and I think I'd have trouble looking at it, especially across eight time series.

I am not an expert in graph design, so this may also be a no-go, but have you considered four colors with three types of plot type?

Though, I think there is an even better approach. I know you say that 4 colors is a no-go, I'm about to ignore that. It is probably canonically true... but you are describing fruits. These have canonical colors as well as shapes associated with them. If you use those colors and shapes I think it would be hard to go wrong.

Using colors alone there might be some confusion, e.g. green apple vs honeydew, red apple vs watermelon, etc. But using colors poses an additional problem, color blind individuals. You can test for the extent to which this would be a problem by creating an image of your different colors and looking here: http://www.vischeck.com/vischeck/. Protanopia and Deuteranopia, forms of red-green color blindness, are by far the most common (occurring almost always in males). Even so, color blindness is a misnomer and if you select your pallet carefully the differences in shades may be sufficient.

In conjunction with a color approach, you want to use fruit shaped points. These are unlikely to be a default in a plotting program and you may have to spend some time in photoshop to make it look right. Even if you can't take that time, differing geometric plot points AND color should make things reasonable.

Moreover, you could use the approach I suggest with a slope graph.

As a side note, if you have a technically/mathematically astute audience, a Y-axis in log odds might mean more to them that percentages alone.

  • $\begingroup$ Still kind of messy, I suppose it is kind of inevitable, afterall there are that many fruits to compare...The amount of variables makes for quite a messy slope graph, if not one that extends for pages...Aside from slope graphs, are there any other alternatives? Using more than one chart to tell the message is alright... $\endgroup$
    – Michael
    Jul 24 '11 at 11:16
  • $\begingroup$ The problem, as the answer your comment on the accepted answer indicates, is that you want to be able to do between and intragroup comparisons. Between group (over the entire time series) is always going to be messy. Intragroup comparisons are easy, a different slide for each fruit. Between group comparisons across a time series are difficult (if the time series actually matters). If the time series didn't matter, you could do a separate bar graph for each time point. $\endgroup$ Jul 24 '11 at 14:09
  • $\begingroup$ I don't know your limitations, but if you aren't bound to paper this would be an excellent place to use an interactive graph. E.g. follow my design suggestion but then have check marks to include/exclude various fruits from the visualization; and/or, allow a mouse-over on a given fruit to bold it's line and dim the others. $\endgroup$ Jul 24 '11 at 14:10
  • $\begingroup$ ... or - if it doesn't drive you crazy - you could use my suggestion or a slope graph AND include sparklines (a link from Alex's answer). Sparklines would give you the intragroup comparison you are seeking where as some variant of a regular line chart / slope graph would give you the between groups comparison. $\endgroup$ Jul 24 '11 at 14:13

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.