alternatives to pie graphs for proportional data

[#1]

This is less of a JMP question and maybe and more of an overall "best data practices" issue, but I thought I'd throw it out there since it has implications for showcasing Graphbuilder's many perks: I routinely have groupings of proteins that have different functions, and I want to show how their distribution differs across different members of the coral "holobiont:" the coral host and its symbionts. Normally, I just color code the functional categories of the proteins, make two pie graphs (one for the coral and one for the symbiont), then plot them next to each other (see attached image).....but everyone seems to hate pie graphs since they are apparently too hard to interpret! To me, one who stares at these data constantly, the pie graph seems fine, but maybe I am living in the plotting dark ages and there are other, better ways to present proportional/assemblage data.

Some alternatives are: mosaic plots, treemaps, and bar graphs. I basically just want to show that the breakdown of the proteins differ across the two organisms. I've attached the JMP table with a few variations of the same data as scripts. Are the pie graphs really that much worse than any of these alternatives? I guess one could argue that the mosaic plot most directly shows differences. Also, if anyone has any GraphBuilder tricks as to how to best compare assemblage data (since there are surely plenty of ways I haven't thought of), feel free to chime in. I would be happy to receive any input. 

[#2]

Hi @abmayfield ,

You could use something simple like stacked bar charts.  Using Tabulate to get the values as % of Total and then graph build with labels. I tried directly within Graph Builder but is seems that % of total does not show up as a label but value does in 15.2.1. That is why I used tabulate for the compartment>protein with % total and then again for just protein.  I made tables of the two tabulates then made a stacked bar chart of each.

I like the Mosaic better since it is proportional to each other and shows the % of total.

Best,

[#3]

Thanks for the input. Hmmm, having now queried quite a few others, it really does seem like the humble pie chart may go the way of the dodo. I talked to some college students taking their first stats classes, and they said there is a huge body of literature on how poorly they (pie graphs, not the students) convey information! Thankfully, there appears to be a plethora of alternatives in Graphbuilder. I think stacked bars is a good idea. I actually am pretty sure I can get it to do just what you said, but I'd need to create at least one (continuous) "counts" or "frequency" column in all likelihood (the table I posted was 100% categorical). easy. Restructuring to where there are counts for each process is easy, too. 

[#4]

+100 to abandoning the pie charts.  Virtually any other display is better.  However, stacked bar charts are not far behind pie charts in their difficulty in reading.  I'd suggest using side by side bar charts.  While they are unexciting, they directly show the differences between host and sym and there is little ambiguity of interpretation.  I've attached your data with my script added.

[#5]

Dale, thanks for the feedback and the script. It's funny: I PERSONALLY much prefer pie graphs to the side-by-side bar graph (i.e., your script), but since I ultimately want to appease reviewers (and, more importantly, make easy-to-interpret figures), I will set aside my own aesthetic preferences, admit graphing defeat, and try to move away from pie graphs!

The same people who recommended such changes (peer reviewers) also said to turn my bar graphs into a table....I might try to fight that one since I kind of feel like there is the unspoken rule that, if it can be plotted, that is preferable to putting the data in a table if at all possible. Who wants to read a table full of raw data?

[#6]

Another possible visualization would be the parallel element:

Graph Builder(
	Size( 495, 450 ),
	Show Control Panel( 0 ),
	Variables(
		X( :compartment ),
		X( :protein function, Position( 1 ) ),
		Color( :compartment )
	),
	Elements( Parallel( X( 1 ), X( 2 ), Legend( 22 ) ) )
);

[#7]

Wow! Now this is the reason I post these sorts of (seemingly mundane and trivial) questions: that's a type of plot I didn't even know you could make in GraphBuilder! It may not be the "winner" for this particular dataset because I think it's not the simplest (though arguably the prettiest), but I regularly have a need for this sort of graphic display with other datasets, so this is great. I literally never clicked on the parallel plot icon in 10 years of using JMP until five minutes ago (i.e., after reading your suggestion).