It’s used because most people who make them are poor visual communicators.
1 Like
A contra-hypothesis is that physical tells are more readily communicated via eyes/head/neck then they are via lower extremities.
Any research on that?
(because everything we know about network speed of the nervous system shows the pings to head/neck beat the lower extremities).
This is actually a big issue in the data visualization community. The default colormap in pretty much every set of data analysis tools completely sucks:
http://eagereyes.org/basics/rainbow-color-map
I personally am partial to a heatmap corresponding to the colors of the blackbody spectrum at various temperatures for univariate data, but I’ve discovered that I’m in the distinct minority in that regard, and the point of the visualization is to convey the patterns you’re seeing to other people, so I generally use one color with varying density and saturation for univariate data, and two such colors, fading to white at zero, for bivariate data.
2 Likes
You might want to check the links I posted above, and let us know what you think of the points they deal with. It’s a little bit to long to summarize effectively in a comment here.
Hi SpaceMonkey - I’m not trying to say that rainbow color maps are the “bestest thing ever”. However, they serve a specific purpose. I really have no problem with the color qualitative scheme, but it’s designed for use on maps where clear boundaries are visible.
DigitalArtform wanted the hot-hot and the cold-cold with a blended standard gradient (because to him yellow is warmer than red), and that typically doesn’t work when trying to provide translatable data on a graphic with no internal borders. Take a look at the four skull images in your link. In the default colormap you can easily identify three areas rated differently that all the others. On the isomorphic color map, you can’t see the same information - it’s lost in a white field.
DigitalArtform also didn’t know/care how other people might use colormaps. Some people do need to know where the cold spots are - they may just as important as a hot spot - representing a weakness or a void. That’s something that can’t just get lost in a deep purple blur. To him, me pointing that out was pointless.
I always agree that the best type of mapping (or any design) should be used for a project, so I wasn’t saying “only ever use this one thing”. What I was saying was that one person made a suggestion that had a lot of flaws for specific reasons, and that they probably should examine use of the item before trying to redesign it.
Now about your spectrum - I can understand why some people might have a problem - it puts “warm” colors at low end values and “cool” colors at high end values. Even though it’s realistic, I doubt it would satisfy DigitalArtform who wants to go there, but in a very simplified manner. Many people don’t intuitively realize that blue and white are hotter than red and yellow, so I can understand other people’s resistance to it. I think if you used it in the right applications (for the right audiences) it wouldn’t be a problem. Also, I’ve already explained the problem with using a single color for mapping when trying to find information at certain points, the standard gradient hides it. So, I feel you’d need to include a few steps or all you’d have is a gradient image.
I hope that answers well, and that you’re cool with it.
1 Like
That does answer well. In cases where cold is as important as hot, you’re often dealing with bivariate data, which I like a two-color scheme for, but I can see how rainbow might be useful for situations like that that are inherently univariate. I have a particular animus against the rainbow map though, because the way I found those blog posts was not through anyone pointing them out to me, but because i was plotting data with the default rainbow map, and I kept seeing what I thought were important patterns in the data, which, when I actually ran the stats to test for them, turned out not to be there at all. After this happened a few times, I realized that these were illusions created by the rainbow colormap, and started googling to find better ways of representing the data, and ran across hose two blog posts that had detailed explanations of why I was having that problem.
In short, while I’m willing to concede that there may be some cases where the rainbow map is useful, I think that its unfortunate status as the default results in it being a bad choice in the huge majority of cases in which it is used. For univariate data, something more like this would be a much better default:
http://stanford.edu/~mwaskom/software/seaborn-dev/examples/hexbin_marginals.html
2 Likes
Ah, yeah, that’s my point exactly.
If you were running into problems while using the rainbow mapping scheme, then definitely it wasn’t the right choice for you, and you needed to look for other options. That’s only reasonable. It isn’t always the best choice. It is a widely useful option, and has some benefits built in (like the ability to be read by those who are color blind and application to an image without adding false borders) which make it useful as a standard, but that doesn’t mean it’s universally applicable. You also can’t always use the same design of chart to represent data equally well.
I like your hex example here, but I think your eyes work as well as mine do for reading color, and that’s a hazard you might run into when developing for other viewers. You need to make sure steps between shades are measurable. When not changing hue, a lot of people can’t read minor changes between shades. You may have taken the Farnswell 100 Hue Test when I posted it to another thread (Vermeer). I typically score between 0 and 4 (if rushing), so I have excellent color vision. Most people don’t see color that well, so I always make sure that I’m adjusting for my audience when I design.
I would never disagree with someone making the best choice for graphic use when they really understand the problem and all the variables they might encounter. That was never my intent.
we could probably do a clickmap or mousemap pretty easil
1 Like
I’m fine if you call them something other than ‘heatmaps.’ The recent introduction of the term ‘colormap’ into the conversation is much more general and doesn’t imply positions on a scale. You could colormap regions based on categories that exist in no particular order.
That’s pretty much what I did. I started in the middle of the black body spectrum where it’s ‘hot,’ and then I worked my way cooler and cool in Kelvin temperature through yellow, orange and red. Then as a concession to psychology I went into blues, more like the way a setting sun gives way to a dark blue sky. But I’m not in this field, so I haven’t tried to come up with any ideal scale. I just threw that example together for this thread. It just seems obvious to me from what I know about painting and photography, and while it may be confirmation bias wrapped up in an appeal to authority, I can’t help but notice how many apparently notable people from this field say the same thing I do.
I do remember seeing a similar article a while back about men showing statistically significant amounts of ‘crotch stare’ when shown any picture of a person. And even of dogs, if I recall correctly.
Yeah science!
I will respond to you one time because I mentioned you directly in my comments to Space Monkey. In response to your first comment:
As I already explained - if they’re “heat maps” the colors used are purely used for visibility’s sake. They aren’t related to “real heat”. You can’t run colors the way you want to on the type of map you want to do it on - maps in general aren’t “true to color”. Here’s just one of millions of of examples of a “map”. (You could also just visit Google maps.)
One design question when building a world map is, “What is the minimum number of colors that can be used to fill the map will no matching colors touching at any border?” That the type of problem people run into when designing maps. No lie.
In response to the second comment:
No, they aren’t the same thing. Because you aren’t offering any color separation on your mapping, and you’re trying to use that gradient in an inappropriate fashion. Space Monkey said that in those cases, he would be using his bivariate coding - not a single gradient form that would actually hide information. He also admitted that there were cases where the rainbow map was more applicable - heat maps happen to be one of those. (He was working on something else.)
Design isn’t necessarily about how pretty the thing is, but how well it functions, and you seem to completely not understand - because you clearly aren’t a designer of any kind - that design must follow function or it’s bad design! You don’t just pick colors because they’re pretty or realistic. It’s a lot harder than that. You want your design to be beautiful, but first it has to do the job it was designed to do.
It’s not about beauty. It’s about the human visual system and how perception works. These things can be investigated and studied. Nobody has to invent a scheme in advance like ROYGBIV and decide it should work because it should work. All I can say is that I find the rainbow scheme to be horrible, and I can say why I think so.
And notice that’s exactly what I did do.
Actually merely using white in the hot islands of interest and diminishing shades of gray down to black in the areas I don’t give a crap about is probably sufficient. Coloration of any kind is pretty much decorative. That way I can do Boolean operations on two heat maps and the result still makes sense. (And I can colorize that result with a gradient color map of some kind). So now I can take the heat map of cancer hot spots and “and” it (multiply it in 0…1 space) with a heat map of chemical plant hot spots and see by the surviving gray areas in common what the intersection of the two is. I can use Photoshop to do Boolean operations on the maps.
You keep mentioning that there are other kinds of maps in which every point is equally important. I know that. I’m not talking about those kinds of maps.
Try reading. Seriously, try reading. You’re basically going backwards in developmental history.
In other words, you’ll need to use bivariate coding to see the information.
Which is what I’ve been saying all along.
I will not speak to you further.
The paper tells us that accountants have had a long history of playing amateur designer. The trick is to do better than these old, cold, cerebral systems that allow you to puzzle out an answer after a period of thought. I don’t want to look at my digital watch and puzzle out the numbers. I just want to look at my analog watch hands, because like little pie charts, their sweep visualizes data so much better and faster. This map is a completely different kind of map than what we’ve been talking about, btw. The space being mapped is an abstract one, and I gain no rapid information about anything. I have to find the edge value. Then I have to find the other edge value. Then I have to trace my way along until I find an intersection. Then, if I’ve lived that long, I have to ascertain whether or not it’s one of those two identical looking shades of green…
Value trumps hue every time.
This topic was automatically closed after 5 days. New replies are no longer allowed.
