Don’t have a formal source, but I can give you a quick rundown of the advice my group ends up giving to every student we work with:
Label the dang axes.
Make the axis labels bigger.
Make histogram lines thicker; make dots larger.
If the dots are very dense, don’t use dots, use a color scale.
For the sake of the absent gods, don’t make your colour scale brown-yellow-lightgray-black-darkbrown-darkgray-darkyellow, as one often-used plotting package did by default. (It was an inheritance from the early nineties, and honestly it was still weird.) Make it something that humans naturally read as a scale, eg blue to red by way of violet, dark green to light green, or blue to red by way of the rainbow.
On a white background, do not use yellow or bright green unless the individual dots or areas are large. Lines, generally speaking, are not large.
Put a legend in one corner, explaining what the line styles mean.
If you’re using (eg) triangles for one data type and circles for another, make the points bigger. Yes, it likely looks perfectly clear on your screen, to your young eyes, at a distance of a foot. You will eventually present it on a crappy twenty-year-old projector to men of sixty and seventy sitting at the back of a large auditorium. EMBIGGEN THE DANG POINTS. Also, use colours to further clarify the difference, unless colour is indicating a different dimension of information.
Make bin sizes a round number − 1, 2, or 5 - in a unit of interest.
If plotting numbers of something, indicate the bin size by labeling the y axis (for example) “Events / 2 MeV”.
As a general rule, make both a linear and a semilog plot. You can skip the linear if there are no features of interest at high densities, and the semilog if there are no features of interest at low densities.
Well, the old default does make local variations more visible, especially for the colour-blind. OTOH I agree that telling at a glance which of two widely separated spots on the graph has a higher value is all but outright impossible with it.
When the seventy-year-old at the back of the large auditorium with the cheap, ancient projector can read them. Alternatively, when your boss stops complaining. Lines are too thick if they overlap; dots are too big when you can’t easily tell the difference between high and medium density. (And if this happens at the default dot size, switch to a colour scale.)
If you’re doing PowerPoint or similar presentation tools, you want your axis labels to be the same size as your bullet-point text. One trick I sometimes use is to whiteout the axis labels in the image file of my plot, and put them back in using the same text tool that’s creating my bullets.
Don’t have a formal source, but I can give you a quick rundown of the advice my group ends up giving to every student we work with:
Label the dang axes.
Make the axis labels bigger.
Make histogram lines thicker; make dots larger.
If the dots are very dense, don’t use dots, use a color scale.
For the sake of the absent gods, don’t make your colour scale brown-yellow-lightgray-black-darkbrown-darkgray-darkyellow, as one often-used plotting package did by default. (It was an inheritance from the early nineties, and honestly it was still weird.) Make it something that humans naturally read as a scale, eg blue to red by way of violet, dark green to light green, or blue to red by way of the rainbow.
On a white background, do not use yellow or bright green unless the individual dots or areas are large. Lines, generally speaking, are not large.
Put a legend in one corner, explaining what the line styles mean.
If you’re using (eg) triangles for one data type and circles for another, make the points bigger. Yes, it likely looks perfectly clear on your screen, to your young eyes, at a distance of a foot. You will eventually present it on a crappy twenty-year-old projector to men of sixty and seventy sitting at the back of a large auditorium. EMBIGGEN THE DANG POINTS. Also, use colours to further clarify the difference, unless colour is indicating a different dimension of information.
Make bin sizes a round number − 1, 2, or 5 - in a unit of interest.
If plotting numbers of something, indicate the bin size by labeling the y axis (for example) “Events / 2 MeV”.
As a general rule, make both a linear and a semilog plot. You can skip the linear if there are no features of interest at high densities, and the semilog if there are no features of interest at low densities.
Here’s a few reason not to do that. (Not to mention the possibility of colour-blind viewers.)
Thanks for the link. I recommend reading it to anyone who’s interested in how data gets (mis)represented.
Well, you have to admit it’s still a big improvement over the old ROOT default. :)
Well, the old default does make local variations more visible, especially for the colour-blind. OTOH I agree that telling at a glance which of two widely separated spots on the graph has a higher value is all but outright impossible with it.
How do I know that they are big enough?
When the seventy-year-old at the back of the large auditorium with the cheap, ancient projector can read them. Alternatively, when your boss stops complaining. Lines are too thick if they overlap; dots are too big when you can’t easily tell the difference between high and medium density. (And if this happens at the default dot size, switch to a colour scale.)
If you’re doing PowerPoint or similar presentation tools, you want your axis labels to be the same size as your bullet-point text. One trick I sometimes use is to whiteout the axis labels in the image file of my plot, and put them back in using the same text tool that’s creating my bullets.
How many of those suggestions could be replaced by “use ggplot2”?
Within our group, none, because then we’d have to learn R. For ChristianKI, quite possibly all of them.